EFFECTS OF HABITAT FRAGMENTATION ON BIRDS IN WESTERN LANDSCAPES: CONTRASTS WITH PARADIGMS FROM THE EASTERN UNITED STATES T. Luke George and David S. Dobkin, editors Studies in Avian Biology No. 25 A PUBLICATION OF THE COOPER ORNITHOLOGICAL SOCIETY Cover watercolor painting of a Varied Thrush (lxoreus naevius) in a naturally fragmented western landscape and a Kentucky Warbler (Oporornis formosus) in an anthropogenically fragmented eastern landscape, by Wendell Minor STUDIES IN AVIAN BIOLOGY Edited by John T Rotenberry Department of Biology University of California Riverside, CA 92521 Artwork by Wendell Minor Wendell Minor Designs 15 Old North Road Washington, CT 06793 Studies in Avian Biology is a series of works too long for The Condor, published at irregular intervals by the Cooper Ornithological Society. Manu- scripts for consideration should be submitted to the editor. Style and format should follow those of previous issues. Price $22.00 for softcover and $35.00 for hardcover, including postage and handling. All orders cash in advance; make checks payable to Cooper Orni- thological Society. Send orders to Cooper Ornithological Society, % Western Foundation of Vertebrate Zoology, 439 Calle San Pablo, Camarillo, CA 93010. ISBN: 1-891276-34-4 Library of Congress Control Number: 2002114416 Printed at Allen Press, Inc., Lawrence, Kansas 66044 Issued: December 18, 2002 Copyright ¸ by the Cooper Ornithological Society 2002 CONTENTS LIST OF AUTHORS ............................................... 1 PREFACE ......................................................... 3 INTRODUCTION: Habitat fragmentation and western birds ............. .............................. T. Luke George and David S. Dobkin 4 THEORY AND CONTINENTAL COMPARISONS A multi-scale perspective of the effects of forest fragmentation on birds in eastern forests ........ Frank R. Thompson, III, Therese M. Donovan, Richard M. DeGraaf, John Faaborg, and Scott K. Robinson 8 What is habitat fragmentation? ....................................... ............... Alan B. Franklin, Barry R. Noon, and T Luke George 20 Habitat edges and avian ecology: geographic patterns and insights for west- ern landscapes .................... Thomas D. Sisk and James Battin 30 Effects of fire and post-fire salvage logging on avian communities in conifer- dominated forests of the western United States ..................... Natasha B. Kotliar, Sallie J. Hejl, Richard L. Hutto, Victoria A. Saab, Cynthia P. Melcher, and Mary E. McFadzen 49 Geographic variation in cowbird distribution, abundance, and parasitism .. ......................... Michael L. Morrison and D. Caldwell Hahn 65 Effects of forest fragmentation on brood parasitism and nest predation in eastern and western landscapes .................................... ............................... John E Cavitt and Thomas E. Martin 73 Effects of forest fragmentation on tanager and thrush species in eastern and western North America ...... Ralph S. Hames, Kenneth V. Rosenberg, James D. Lowe, Sara E. Barker, and Andr6 A. Dhondt 81 EFFECTS OF FRAGMENTATION ON WESTERN ECOSYSTEMS The effects of habitat fragmentation on birds in coast redwood forests .... ................................. T Luke George and Arriana Brand 92 Effects of habitat fragmentation on birds in the coastal coniferous forests of the Pacific Northwest ..... David A. Manuwal and Naomi J. Manuwal 103 Birds and changing landscape patterns in conifer forests of the north-central Rocky Mountains ... Sallie J. Hejl, Diane Evans Mack, Jock S. Young, James C. Bednarz, and Richard L. Hutto 113 Effects of habitat fragmentation on passerine birds breeding in intermountain shrubsteppe ................. Steven T Knick and John T. Rotenberry 130 Habitat fragmentation effects on birds in southern California: contrast to the "top-down" paradigm ........................... Douglas T. Bolger 141 Effects of anthropogenic fragmentation and livestock grazing on western riparian bird communities ...... Joshua J. Tewksbury, Anne E. Black, Nadav Nur, Victoria A. Saab, Brian D. Logan, and David S. Dobkin 158 STUDIES ON FOCAL SPECIES Spotted Owls, forest fragmentation, and forest heterogeneity ............. ............................... Alan B. Franklin and R. J. Gutierrez 203 Effects of forest fragmentation on populations of the Marbled Murrelet . .. ........... Martin G. Raphael, Diane Evans Mack, John M. Marzluff, and John M. Luginbuhl 221 LITERATURE CITED .............................................. 236 LIST OF AUTHORS SARA E. BARKER Laboratory of Ornithology Cornell University Ithaca, NY 14850 JAMES BATTIN Department of Biological Sciences Northern Arizona University Flagstaff, AZ 86011-5694 JAMES C. BEDNARZ Department of Biological Sciences Arkansas State University State University, AR 72467 ANNE E. BLACK Colorado National Heritage Program Fort Collins, CO, and Point Reyes Bird Observatory 4990 Shoreline Highway Stinson Beach, CA 94970 DOUGLAS T. BOLGER Environmental Studies Program HB6182 Dartmouth College Hanover, NH 03755 L. ARRIANA BRAND Department of Fishery and Wildlife Biology Colorado State University Fort Collins, CO 80523 JOHN E CAVITI' U.S. Geological Survey Montana Cooperative Wildlife Research Unit University of Montana Missoula, MT 59812 (Present address: Department of Zoology Weber State University 2505 University Circle Ogden, UT 84408-2505) RICHARD M. DEGRAAF USDA Forest Service Northeastern Research Station Holdsworth Hall University of Massachusetts Amherst, MA 01003 ANDRI A. DHONDT Laboratory of Ornithology Cornell University Ithaca, NY 14850 DAVID S. DOBKIN High Desert Ecological Research Institute 15 SW Colorado Avenue, Ste. 300 Bend, OR 97702 THERESE M. DONOVAN SUNY College of Environmental Science and Forestry 1 Forestry Drive Syracuse, NY 13210 (Present address: Vermont Cooperative Fish and Wildlife Research Unit 311 Aiken Center University of Vermont Burlington, VT 05405) JOHN FAABORG Division of Biological Sciences 110 Tucker Hall University of Missouri Columbia, MO 65211 ALAN B. FRANKLIN Colorado Cooperative Fish and Wildlife Research Unit Department of Fishery and Wildlife Biology Colorado State University Fort Collins, CO 80523 T. LUKE GEORGE Department of Wildlife Humboldt State University Arcata, CA 95521 R. J. GUTIIRREZ Department of Wildlife Humboldt State University Arcata, CA 95521 (Present address: Department of Fisheries and Wildlife University of Minnesota St. Paul, MN 55108) D. CALDWELL HAHN U.S. Geological Survey Patuxent Wildlife Research Center 11410 American Holly Drive Laurel, MD 20708-4015 RALPH S. HAMES Laboratory of Ornithology Cornell University Ithaca, NY 14850 SALLIE J. HEJL USDA Forest Service Rocky Mountain Research Station P.O. Box 8089 Missoula, MT 59807, and Sierra Nevada Framework Project 801 I St., Rm. 419 Sacramento, CA 95814 (Present address: Department of Wildlife and Fisheries Sciences 2258 TAMU Texas A&M University College Station, TX 77843-2258) RICHARD L. HUTTO Division of Biological Sciences University of Montana Missoula, MT 59812 STEVEN T. KNICK U.S. Geological Survey Forest and Rangeland Ecosystem Science Center Snake River Field Station 970 Lusk Street Boise, ID 83706 NATASHA B. NOTLIAR U.S. Geological Survey Fort Collins Science Center 2150 Centre Avenue, Bldg C Fort Collins CO 80526-8818 2 STUDIES IN AVIAN BIOLOGY NO. 25 BRIAN D. LOGAN U.S. Geological Survey Montana Cooperative Wildlife Research Unit University of Montana Missoula, MT 59812 JAMES D. LOWE Laboratory of Ornithology Cornell University Ithaca, NY 14850 JOHN g. LUGINBUHL College of Forest Resources University of Washington Seattle, WA 98195-2100 DIANE EVANS MACK USDA Forest Service Pacific Northwest Research Station 3625 93rd Ave SW Olympia, WA 98512-9193 DAVID A. MANUWAL College of Forest Resources Box 352100 University of Washington Seattle, WA 98195 NAOMI J, MANUWAL 19420 194th Ave NE Woodinville, WA 98072 THOMAS E. MARTIN U.S. Geological Survey Montana Cooperative Wildlife Research Unit University of Montana Missoula, MT 59812 JOHN M. MARZLUFF College of Forest Resources University of Washington Seattle, WA 98195-2100 MARY E. MCFAZEN USDA Forest Service Rocky Mountain Research Station EO. Box 8089 Missoula, MT 59807 CYNTHIA P, MELCHER U.S. Geological Survey Fort Collins Science Center 2150 Centre Avenue, Bldg C Fort Collins CO 80526-8818 MICHAEL L. MORRISON University of California White Mountain Research Station 3000 East Line Street Bishop, CA 93514 BARRY R. NOON Department of Fishery and Wildlife Biology Colorado State University Fort Collins, CO 80523 NADAV NUR Point Reyes Bird Observatory 4990 Shoreline Highway Stinson Beach, CA 94970 MARTIN G. RAPHAEL USDA Forest Service Pacific Northwest Research Station Olympia, WA 98512-9193 SCOTF K. ROBINSON Department of Animal Biology 172 Natural Resource University of Illinois Champaign, IL 61820 KENNETH V. ROSENBERG Laboratory of Ornithology Cornell University Ithaca, NY 14850 JOHN T. ROTENBERRY Center for Conservation Biology and Department of Biology University of California Riverside, CA 92521 VICTORIA A. SAAB USDA Forest Service Rocky Mountain Research Station 316 E. Myrtle St. Boise, ID 83702 THOMAS D. SISK Center for Environmental Sciences and Education Northern Arizona University Flagstaff, AZ 86011-5694 JOSHUA J. TEWKSBURY Biological Sciences University of Montana Missoula, MT 59812 (Present address: Department of Zoology Box 118525 University of Florida Gainesville, FL 32611) FRANC R. THOMPSON, III USDA Forest Service North Central Research Station 202 Natural Resources Bldg. University of Missouri Columbia, MO 65211 JOCK S. YOUNG Division of Biological Sciences University of Montana Missoula, MT 59812 Studies in Avian Biology No. 25:3, 2002. PREFACE This volume grew from recognition of the need for a forum to address explicitly the con- trasts and similarities of fragmentation processes and fragmentation effects in eastern and western landscapes. That recognition arose over the course of several years in informal discussions between the editors, which crystallized at the second North American Ornithological Confer- ence in 1998 in St. Louis, where we conceived of a symposium and outlined the areas that should be covered. A one-day symposium organized by the edi- tors was held the following year in Portland, Oregon, at the annual meeting of the Cooper Or- nithological Society. The central focus of the symposium was to contrast patterns in the west- ern versus eastern United States, and to differ- entiate and contrast natural versus human- caused fragmentation patterns and associated ef- fects. From the outset, the symposium was in- tended to serve as the basis for a monograph in the STUDIES IN AVI^N BIOLOGY series. Nearly all of the 16 chapters contained in this volume are based on symposium presentations, although not all topics covered in the symposium are repre- sented here. Each chapter has been peer-re- viewed and reviewed by the editors, as well. We are grateful to the Cooper Ornithological Society for providing logistic support and an ex- cellent venue for the symposium, and to our col- leagues who graciously agreed to serve as peer- reviewers for the chapters in this volume. We thank the United States Environmental Protec- tion Agency's Ecosystem Science Branch for generously providing funds to support publica- tion of this volume through Assistance Agree- ment No. 82772001 to the High Desert Ecolog- ical Research Institute. The research contained herein has not been subjected to Agency review, and therefore does not necessarily reflect the views of the Environmental Protection Agency. Additional funds in support of the symposium were provided by the Oregon/Washington office of the United States Bureau of Land Manage- ment and the Cooper Ornithological Society. The editors thank Wendell Minor for providing the artwork that graces the cover. David S. Dobkin T Luke George Studies in Avian Biology No. 25:4-7, 2002. INTRODUCTION: HABITAT FRAGMENTATION AND WESTERN BIRDS T. LUKE GEORGE AND DAVID S. DOBKIN Habitat fragmentation and loss due to human activities has been identified as the most impor- tant factor contributing to the decline and loss of species worldwide (Noss and Cooperrider 1994). Although the response of species to hab- itat loss generally is clear, the effects of habitat fragmentation are much more complex (Fahrig 1997, Bunnell 1999). Over the last two decades, our understanding of the effects of habitat frag- mentation on bird populations has increased tre- mendously. Early studies viewed habitat frag- ments as islands and interpreted patterns of spe- cies richness in the context of island biogeog- raphy theory (Forman et al. 1976, Galli et al. 1976). It soon became apparent, however, that in contrast to oceanic islands, the habitat or ma- tfix surrounding fragments profoundly influ- enced the ecological conditions within those fragments. In particular, rates of nest predation and cowbird parasitism of ground-nesting and cup-nesting birds were found to be extremely high close to forest edges (Ambuel and Temple 1983) and in small forest fragments (Wilcove 1985, Robinson 1992). Further study revealed that patterns of nest predation, and especially nest parasitism, were influenced by forest cover in the surrounding landscape (Andr6n and An- gelstam 1988; Andr6n 1992, 1994, 1995; Rob- inson et al. 1995, Donovan et al. 1997). Taken together, these results suggested that declines and losses of birds from small forest fragments were related to elevated rates of nest predation and parasitism. These observations led to the de- velopment of a top-down hierarchical model that included regional, landscape-level, and local ef- fects to explain variation in nesting success across the landscape and subsequent changes in abundance and distribution of the affected spe- cies (Thompson et al. this volume). Because much of the empirical support for this model derives from studies conducted in the eastern United States (i.e., east of the Rocky Moun- tains), this model embodies what can be viewed as the "eastern paradigm." As better understanding of the human-im- posed dynamics and the natural ecological pro- cesses that govern western landscapes has ac- crued in recent years, applicability of the eastern paradigm to landscapes of the western United States has become more tenuous. First, the na- ture of the matrix in most western ecosystems differs dramatically from the East. Habitat frag- ments studied in the eastern United States fre- quently are embedded in agricultural or urban landscapes, but most studies of habitat fragmen- tation in the West have focused on forest frag- ments created by timber harvest. Logging op- erations result in fragments of mature or old- growth forest that are embedded in a matrix of young, regenerating forest. Landscapes com- posed of young forest, in contrast to agricultural and exurban landscapes, may not harbor high densities of predators and brood parasites, and consequently birds inhabiting fragments may not suffer the high rates of nest predation and par- asitism observed in the East. While the extent of urban and agricultural development is in- creasing in the West, it is substantially less than in the East (Fig. 1). As a result, fragments of natural vegetation generally are embedded in a matrix of agricultural and urban land in the East, but urban and agricultural lands generally are isolated in a matrix of unconverted habitat in the West (Fig. 2). Clearly there are some regions in the western United States that exhibit patterns similar to the East. For instance, 71% of Cali- fomia's Central Valley and 63% of Oregon's Willamette Valley have been converted to agri- cultural or urban uses, which is similar to the high levels of conversion in many eastern and Midwestern regions (T. L. George, unpubl. data). A second suite of fundamental differences be- tween eastern and western landscapes results in a higher degree of natural heterogeneity in the West. Greater aridity, the greater spatial extent and temporal frequency of fires, and greater to- pographic diversity made western landscapes in- herently more patchy than eastern landscapes long before European settlement (Hejl et al. this volume, Kotliar et al. this volume). Having con- tended with the natural heterogeneity of western landscapes for thousands of generations, avian populations inhabiting this region may be less affected by fragmentation processes and conse- quences than avian populations of the relatively more homogeneous landscapes of the pre-Eu- ropean-settlement eastern United States. If noth- ing else, these differing selective milieus make it difficult to predict the responses to disturbance (whether natural or anthropogenic) by species inhabiting western landscapes. The primary objective of this volume was to INTRODUCTIONsGeorge and Dobkin 5 Percent Converted Land by Ecoregion P½½nt Converted Lnd /0-10 10-20 20 -30 90 - 100 20t 2t- 400 Iv'files FIGURE I. Proportion of land converted to agriculture or man-made structures in the conterminous United States in 66 physiographic regions. Proportions were calculated l¾om the U.S. Geological Survey Land Use and Land Cover (LULC) database compiled between 1975-1985 (Mitchell et al. 1977). The LULC database included 45 categories (Anderson et al. 1975); we combined all agricultural and developed land into an "altered" category (see Appendix) and calculated the proportion of altered and unaltered land within each region. The physiographic regions are those used by Robbins et al. (1986) for analyses of the Breeding Bird Survey data. examine the effects of habitat fragmentation on western bird populations, particularly in the con- text of predictions derived from eastern para- digms. We defined the western United States as the area from the Rocky Mountains west to the Pacific Coast in the conterminous United States. The lUllowing chapters are grouped into three sections covering theory and continental-scale comparisons, effects of fragmentation in specific western ecosystems, and studies of focal species. Thompson et al. begin by describing and sum- marizing evidence for the eastern paradigms and provide a multi-scale working hypothesis for the effects of habitat fragmentation on birds. Frank- lin et al. provide a definition of habitat fragmen- tation. paying particular attention to the distinc- tion between habitat fragmentation and habitat heterogeneity, and Sisk and Battin review the concept of habitat edge as it applies to western landscapes. The ubiquitous role of fire in shap- ing western landscapes and their associated avi- faunas is addressed by Kotliar et al. Studies that span the continent offer a unique opportunity to compare the response of birds and their nest predators and parasites to frag- mentation in the East and the West. Morrison and Hahn summarize studies of the response of Brown-headed Cowbirds (Molothrus ater) to fragmentation in the East and the West. Cavitt and Martin examine differences in rates of nest predation and parasitism between fragmented and unfragmented areas in the East and the West using data on the outcome of tens of thousands of nests in the BBIRD database (Martin et al. 1997). Employing data from the Cornell Labo- ratory of Ornithology's "Birds in Forested Landscapes" project, Hames et al. compare the responses of tanagers, thrashes, and Brown- headed Cowbirds to forest fragmentation across the United States. Six chapters focus on individual western eco- systems selected to reflect both the relative im- portance of specific vegetation communities and the constraint of where fragmentation-related re- 6 STUDIES IN AVIAN BIOLOGY NO. 25 I NonConverted ,'x, U.S. State Boundaries Contrasting Landscapes: West rs. Midwest ;' ,. , '.. ,- r "'ø' l, ," I ß , , -.. , / ',,, ;I . '..  ' !l ,, ' FIGURE 2. Examples of the distribution of altered and unaltered habitat in the midwestern and the western United States. Land cover data were obtained from U.S. Geological Survey Land Use and Land Cover (LULC) database compiled between 1975-1985 (Mitchell et al. 1977). search has been conducted in the West. Three chapters focus on coniferous forests. George and Brand summarize studies in redwood (Sequoia sempervirens) forests, Manuwal and Manuwal summarize research in the wet coniferous forests of the Pacific Northwest, and Hejl et al. examine forests of the northern Rocky Mountains. Knick and Rotenberry describe avian responses to frag- mentation in the Intermo,untain shrubsteppe, Bolger summarizes a wealth of studies that have been conducted in the highly urbanized coastal sage scrub and chaparral regions of southern California, and Tewksbury ½t al. analyze riparian bird communities across seven riparian systems in five western states. Notably lacking are sum- maries of the effects of fragmentation on birds in the southern Rocky Mountains and the desert Southwest. There were too few studies on the effects of habitat fragmentation on birds in these regions to warrant reviews. A recent publication by Knight (2000) provides an overview of the effects of habitat fragmentation in the southern Rocky Mountains. Finally, as a reflection of the relatively great attention paid to loss and fragmentation of old- growth forests in the western United States, two chapters are devoted to multi-scale assessments of focal species in the context of loss and IYag- mentation of their old-growth forest habitats. Franklin and Guti6rrez synthesize information across subspecies of Spotted Owls (Strix occi- dentalis), and Raphael et al. examine Marbled Murrelets (Brachyramphus marmoratus). Both of these species have had a significant impact on management of western forests. Although the picture is far from complete, the contents of this monograph illustrate the state of our knowledge regarding fragmentation effects on western bird populations at the beginning of the 21st century. We hope this volume will serve as a landmark contribution to the ecological and conservation literature by presenting a solid syn- thesis and foundation to buttress future research, and by conveying important policy implications for public land management in the western Unit- ed States. INTRODUCTION--George and Dobkin 7 APPENDIX. LAND USE CATEGORIES IN USGS DATABASE DESIGNATED AS ALTERED (1) OR UNALTERED (0) FOR FIGURES 1 AND 2 Anderson a land use category Altered Urban or built-up land 1 Residential 1 Commercial and services 1 Industrial 1 Transportation, communication, utilities 1 Industrial and commercial complexes 1 Mixed urban or built-up land 1 Other urban or built-up land 1 Agricultural land 1 Cropland and pasture 1 Orchards, groves, vineyards, nurseries, and ornamental horticultural 1 Confined feeding operations 1 Other agricultural land 1 Rangeland 0 Herbaceous rangeland 0 Shrub and brush rangeland 0 Mixed rangeland 0 Forest land 0 Deciduous forest land 0 Evergreen forest land 0 Mixed forest land 0 Water 0 Streams and canals 0 Lakes 0 Reservoirs 0 Bays and estuaries 0 Wetland 0 Forested wetland 0 Nonforested wetland 0 Barren land 0 Dry salt flats 0 Beaches 0 Sandy areas not beaches 0 Bare exposed rock 0 Strip mines, quarries, gravel pits 0 Transitional areas 0 Tundra 0 Shrub and brush tundra 0 Herbaceous tundra 0 Bare ground 0 Wet tundra 0 Mixed tundra 0 Perennial snow or ice 0 Perennial snowfields 0 Glaciers 0 a From Anderson et al. (1922). Studies in Avian Biology No. 25:8-19, 2002. A MULTI-SCALE PERSPECTIVE OF THE EFFECTS OF FOREST FRAGMENTATION ON BIRDS IN EASTERN FORESTS FRANK R. THOMPSON, III, THERESE M. DONOVAN, RICHARD M. DEGRAAF, JOHN FAABORG, AND SCOTT K. ROBINSON Abstract. We propose a model that considers forest fragmentation within a spatial hierarchy that includes regional or biogeographic effects, landscape-level fragmentation effects, and local habitat effects. We hypothesize that effects operate "top down" in that larger scale effects provide constraints or context for smaller scale effects. Bird species' abundance and productivity vary at a biogeographic scale, as do the abundances of predators, Brown-headed Cowbirds (Molothrus ater), and land-use patterns. At the landscape scale the level of forest fragmentation affects avian productivity through its effect on predator and cowbird numbers. At a local scale, patch size, amount of edge, and the effects of forest management on vegetation structure affect the abundance of breeding birds as well as the distribution of predators and Brown-headed Cowbirds in the landscape. These local factors, along with nest-site characteristics, may affect nest success and be important factors when unconstrained by processes at larger spatial scales. Landscape and regional source-sink models offer a way to test various effects at multiple scales on population trends. Our model is largely a hypothesis based on retroduction from existing studies; nevertheless, we believe it has important conservation and research implications. Key Words: Brown-headed Cowbirds; eastern forests; edge-effects; fragmentation; landscape; Mol- othrus ater; multi-scale; nest predation; predators; songbirds. Much recent research has focused on the effects of forest fragmentation on breeding neotropical migrant birds and recent reviews have concluded that forest fragmentation generally results in in- creased nest predation and brood parasitism (Robinson and Wilcove 1994, Faaborg et al. 1995, Walters 1998). For example, numbers of Brown-headed Cowbirds (Molothrus ater), brood parasitism, and nest predation are nega- tively correlated with the amount of forest cover in landscapes in the midwestern U.S. (Donovan et al. 1995b, Robinson et al. 1995a, Thompson et al. 2000). Enough variation or inconsistency exists among studies, however, that it is difficult to develop a general model of the effects of for- est fragmentation on songbirds that addresses spatial scale, accounts for local and regional var- iation in observed effects, and describes mech- anisms for observed effects. Most research has been conducted in eastern forests. Differences in ecological patterns and land use between eastern and western North America, however, has led to speculation that the effects of fragmentation on birds may differ among these regions (George and Dobkin this volume). We have been developing a conceptual model that places the effects of landscape-level forest fragmentation within a spatial hierarchy that ranges from biogeographic or regional effects to local effects (Freemark et al. 1995, Donovan et al. 1997, Robinson et al. 1999, Thompson et al. 2000). Our purpose in developing this model is to provide a synthesis of the current understand- ing of forest fragmentation effects in eastern landscapes, and to stimulate research that will enhance that understanding in both eastern and western North America. Our model is a simple framework within which factors affecting spe- cies viability can be examined. We present the model as a series of hypotheses organized by this framework, and then review key studies that we used to formulate these hypotheses. We pre- sent the model as series of hypotheses because it is formed largely by retroduction. Retroduc- tion is the construction of a hypothesis about a process that provides an explanation for ob- served patterns or facts (Romesburg 1981). Models of this type are often most useful as hy- potheses for hypothetico-deductive research (Romesburg 1981), and we review a few studies of this type that test our hypotheses. We do not provide an exhaustive literature review because recent reviews exist (e.g., Robinson and Wilcove 1994, Faaborg et al. 1995, Walters 1998, Heske et al. 2001). We primarily review fragmentation effects at a landscape scale and edge effects at a habitat scale. However, we also discuss effects at larger and smaller scales because of important interactions with edge and landscape effects. For brevity and because of the focus of this volume we focus on biogeographic, landscape, and hab- itat effects on songbird reproductive success. The context for our review is the eastern decid- uous forest, although where possible we make comparisons to western landscapes. THE MODEL From a breeding ground perspective, habitat characteristics associated with reproductive suc- cess of forest passefines can be evaluated at sev- eral spatial scales: (1) the nest-site scale--the FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 9 micro-habitat characteristics directly around the nest or the immediate vicinity of the nest; (2) the habitat scale--the features of the habitat patch in which the nest is located; (3) the land- scape scale--the collection of different habitat patches and the position of a particular habitat within a landscape, the matrix within which the habitat is embedded, and the juxtaposition and proximity of other habitats in the landscape (Freemark et al. 1993); and (4) biogeographic scales. For example, vegetation structure at a habitat scale, or location within a landscape, may be more important than nest site characteristics such as concealment in reducing nest depreda- tion (Bowman and Harris 1980, Leimgruber et al. 1994, Donovan et al. 1997, Burhans and Thompson 1999) or parasitism (Best 1978, Johnson and Temple 1990, Burhans 1997, Morse and Robinson 1999). Furthermore, nest preda- tion or brood parasitism may be related to land- scape composition and structure (Robinson et al. 1995a, Donovan et al. 2000, Thompson et al. 2000). Finally, geographic location and abiotic and biotic characteristics at multiple scales can directly impact a population's growth (Hoover and Brittingham 1993, Leimgruber et al. 1994, Thompson 1994, Coker and Capen 1995, Thompson et al. 2000). The essence of our mod- el is that all spatial scales may contribute to the ability of a local subpopulation to replace itself (Sherry and Holmes 1992), but the importance of each may depend on habitat features at other scales or the geographic location within the breeding or non-breeding range. These effects can be arranged in a hierarchy in which larger scale effects provide constraints or context for smaller scale effects (Fig. 1). What types of evidence directly support this model? Evidence of top-down constraints comes from observational, experimental, and meta- analysis studies across eastern North America. Although we provide several examples of cor- relative evidence for such constraints, we em- phasize that experimental and meta-analysis ap- proaches that directly test the top-down con- straint hypothesis have been very instructive be- cause they attempt to control for factors operating at other spatial scales. For example, we tested the hypothesis that landscape effects are more significant than local edge effects, and that edge effects are dependent on landscape context, in a rigorously-designed, large-scale, randomized field experiment. We found strong evidence that edge effects in nest predation are dependent on landscape context, and that land- scape context is a better predictor of cowbird abundance than any other local-scale affect mea- sured (Fig. 2; Donovan et al. 1997). In land- Large Scale, Biogeographic Effects Abundance and demographics of songbirds, cowbirds, and predators vary at a geographic scale. Landscape-Level Effects Land cover and use affect the abundance of breeding birds, predators and nest predation, and cowbirds and brood arasitism. Habitat and Local Effects Habitat type, patch size, proximity to edge, and forest management affect predator and cowbird activity, nest )redation, and brood )arasitism Nest-Site Effects Characteristics such as nest type, height, and concealment affect the probability of predation and parasitism FIGURE 1. Conceptual model of factors at multiple spatial scales affecting reproductive success of song- birds. Larger scale factors are hypothesized to be more important determinants of species viability because they provide context or constraints for smaller scale effects. scapes with < 15% forest, predation was high in forest edge and interior; at 45-55% forest cover, predation was high in forest edge and low in forest interior; and at >90% forest cover, pre- dation was low in forest edge and interion Cow- bird abundance was much greater in landscapes with high levels of forest fragmentation than those with low levels of fragmentation (Fig. 2). While we could not randomly assign landscape treatments in this study (because the landscape patterns already existed), study sites were ran- domly selected from a three-state area. As a re- sult, we believe these results allow strong infer- ences for at least Missouri, Illinois, and Indiana. The results of this research were also confirmed by a meta-analysis of nest depredation studies in which researchers compared the landscape con- text for studies that documented edge effects on predation patterns with those that failed to find edge effects (Bayne and Hobson 1997, Hartley and Hunter 1998). We believe that these large-scale analyses are 10 STUDIES IN AVIAN BIOLOGY NO. 25 60 so 40 30 20 lO 1.0 0.8 0.6 0.4 0.2 o.o A AB B / A AB B High Medium Low Level of fragmentation and edge (E) or interior (I) FIGURE 2. Effects of landscape level of fragmen- tation and local edge effects on nest predation and cowbird abundance in the midwestern United States. Fragmentation levels were measured as the amount of forest cover and were: high, < 15% forest; medium, 45-55% forest; and low, > 90% forest. Edge (E) and interior (I) treatments were 50 rn and > 250 m from forest edge, respectively. Levels of forest cover with different letters, and edge and interior treatments with an asterisk are significantly different (ANOVA, P < 0.05). Data and figures adapted from Donovan et al. (1997). critical for understanding how forest fragmen- tation impacts songbird populations. Although artificial nest experiments at large spatial scales may provide some insights, our hypothesis that larger scale effects provide constraints or con- text for smaller scale effects depends on obser- vations of nesting success at numerous locations across a species' range. Obviously, collection of these data is not an easy task, and significant advances will likely be made through large-scale collaborations (e.g., Robinson et al. 1995a), large-scale research programs with standardized methodology (e.g., BBIRD; Martin et al. 1997), or through meta-analyses (e.g., Hartley and Hunter 1998, Chalfoun et al. 2002). We have focused on direct measures of nesting success, nest predation, and predator abundance; how- ever, we recognize that indirect measures will be necessary and provide insight at large spatial scales (e.g., Project Tanager; Rosenberg et al. 1999). LARGE-SCALE, BIOGEOGRAPHIC EFFECTS Hypothesis: Breeding birds exhibit geograph- ic patterns in their demographics. These are in part the result of geographic patterns in the dis- tribution of predators and cowbirds, and pro- vide the context for smaller scale effects and can affect local reproductive success. PREDATOR DISTRIBUTION Predator abundance and species richness vary across North America. Levels of nest predation could be higher where the total abundance and diversity of predators is higher. For example, Rosenberg et al. (1999) documented biogeo- graphic patterns in predator communities as part of Project Tanager. Tanagers (Piranga spp.) were exposed to different combinations of pred- ators across their range, and predators responded differently to forest fragmentation. The highest incidence of the predators they surveyed oc- curred in the Midwest. General patterns in the distribution of avian predators can be generated from Breeding Bird Survey (BBS) data (Sauer et al. 1997). Detecting biogeographic patterns in nest predation related to predator abundance or diversity will be difficult because of the large number of potential nest predators and variation in their distributions across North America. Fur- ther complicating these patterns is the interac- tion between diversity and abundance; even in areas of low predator diversity a single predator may be very abundant. BROWN-HEADED COWBIRD DISTRIBUTION Cowbirds demonstrate strong geographic pat- terns in abundance; therefore, the potential ef- fects of fragmentation or habitat effects are con- strained by this larger-scale effect. More simply put, in regions of the country where cowbirds are rare it is unlikely that fragmentation or local factors will have a strong effect on parasitism levels. The strongest evidence of this geographic ef- fect comes from BBS data. A distribution map generated from BBS data shows a general pat- tern of high abundance of cowbirds in the Great Plains and decreasing abundance with distance from the Great Plains (Sauer et al. 1997). Thompson et al. (2000) examined patterns from the BBS data by regressing mean statewide cow- bird abundance on distance from the center of their range in the Great Plains and the percent of forest cover in that state. Mean statewide cowbird abundance was negatively related to forest cover in a state and a state's distance from FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 11 the center of the cowbird's breeding range (R 2 = 0.67). Regression coefficients for distance to center of range and forest cover were both sig- nificant. However, the partial correlation of dis- tance to center of range with cowbird abundance was greater than that for forest cover and cow- bird abundance. While both partial correlations were significant, the effect of distance to the center of the range was stronger and provides some indication of the importance of biogeo- graphic constraints. Additional evidence of this effect is seen in parasitism levels. Wood Thrush (Hylocichla mustelina) parasitism levels de- crease from Midwest to Mid-Atlantic to New England (Hoover and Brittingham 1993; see also Smith and Myers-Smith 1998). LANDSCAPE-LEVEL EFFECTS Hypothesis: Nest predation and cowbird par- asitism increase with forest fragmentation at the landscape scale. Predation and parasitism is greater in fragmented landscapes because of a positive, numerical response by predators and cowbirds that is the result of increase in the availability and interspersion of food, hosts, or other resources. A landscape is a heterogeneous mosaic of habitat patches in which individuals live and dis- perse (Dunning et al. 1992), usually ranging in size from a few to hundreds of square kilome- ters. Most research on landscape-level effects and fragmentation has occurred in the last de- cade; understanding the logical importance of these factors required a major shift in our con- cepts of habitat relationships. Biologists, how- ever, have been documenting the distribution of forest passerines in relation to habitat and hab- itat-patch characteristics for literally decades (e.g., Robbins et al. 1989b; reviewed by Free- mark et al. 1995), often using the MacArthur and Wilson (1967) model of island biogeogra- phy as a guiding framework (reviewed in Faa- borg et al. 1995). Patch size, patch shape, and interpatch distances, as well as forest type, have important effects on bird community composi- tion. However, there is ample evidence to sug- gest that these local patterns are driven in part by habitat characteristics at the landscape scale, and also vary regionally. Most investigators of fragmentation effects recognized that habitat fragments differed from true islands because the matrix between the fragments was not ocean, but was a different habitat that supported its own set of species. The inclusion of "edge" species in counts on fragments was certainly one form of recognition that effects from the surroundings of the study site could be important. However, to truly understand all the effects of landscape-lev- el processes upon forest birds we needed to study a variety of landscapes, as opposed to a variety of patches. PATTERNS OF LAND COVER AND THEIR EFFECTS ON THE ABUNDANCE OF PREDATORS AND NEST PREDATION Land cover can significantly influence the number and diversity of predators, as well as constrain the importance of more local-scale habitat factors such as patch size, vegetation structure, or distance to edge effects on nest pre- dation. We begin by reviewing the main effects of landscape pattern, and then discuss how land- scape factors potentially constrain more local- scale effects on nest predation. Detection of this constraint, however, may be difficult because predators throughout North America vary great- ly in habitat use, foraging behavior, and how they collectively contribute to observed nest pre- dation patterns in forest passerines (e.g., Gates and Gysel 1978, Andrdn and Angelstam 1988, Yosef 1994, Tewksbury et al. 1998, Marzluff and Restani 1999, Dijak and Thompson 2000). Robinson et al. (1995a) and Donovan et al. (1995b) were the first to use empirical data from real nests to relate nest predation to forest frag- mentation at a landscape scale. They measured many landscape variables but used the percent of forest cover within a 10-km radius as a simple measure of forest fragmentation and examined its correlation with daily nest predation. Corre- lations for all nine species were in the predicted direction, three correlations were significant (P < 0.05), and two additional species had P-values between 0.05 and 0.20. A combined probabili- ties test on all nine species indicated the overall effect of percent forest cover was significant (P < 0.02). Here we present data points and re- gression lines for two of the species with sig- nificant effects, and two with marginally signif- icant effects (Fig. 3). For all these species the highest nest predation rates occurred in land- scapes with less than 40% forest cover. Given the high variability in nest predation rates over both time and space, we believe these results are indicative of an important relationship even though some of the correlations were not statis- tically significant by the conventional criterion. Two studies have since corroborated the hy- pothesis that nest predation increases with forest fragmentation in eastern forests. In a rigorously designed observational study, Donovan et al. (1997) tested hypotheses concerning edge and landscape effects on nest predation and parasit- ism. They randomly selected 18 landscapes from three states with high, moderate, or low levels of fragmentation and determined predation rates of artificial nests in interior and edge habitat. 12 STUDIES IN AVIAN BIOLOGY NO. 25 0.12' 0.10. 0.08- 0.06- 0.04- 0.02' o 0.12 0.0 0.08 0.06 0.04 0.02 Wood Thrush R 2 = 0.54, P=0.02 Indigo Bunting Ovenbird R 2 = 0.24, e=0.21 20 40 60 80 100 Kentucky Warbler ß R 2 = 0.55, P=0.09 20 40 60 80 100 Percent forest cover FIGURE 3. Relationship of daily nest predation to the amount of forest cover in landscapes defined by a 10-km radius in the Midwestern United States. Data are from Robinson et al. (1995a). Predation rates increased with forest fragmen- tation, and fragmentation (landscape) effects overwhelmed local edge effects (Fig. 2). Hartley and Hunter (1998) conducted a meta-analysis of a set of artificial nest experiments and showed that predation rates increased as forest cover de- creased at 5-, 10-, and 25-km scales of forest cover. Both Donovan et al. (1997) and Hartley and Hunter (1998) addressed factors at multiple scales by investigating the interaction between local edge effects and landscape fragmentation effects, and we discuss this later under edge ef- fects. Many of the previous studies used percent forest cover in a defined landscape as the inde- pendent variable. Most, however, used this mea- sure because it was a convenient index of frag- mentation, and hypothesized predation and par- asitism were high in fragmented landscapes as a result of increases in the abundance of generalist predators and cowbirds (Donovan et al. 1995b, Robinson et al. 1995a, Thompson et al. 2000). Tewksbury et al. (1998) reported levels of predation at real nests increased with higher landscape-levels of forest cover. While their re- sults are contrary to our hypothesis and findings for eastern forests, nevertheless they found a landscape effect on nest predation. They be- lieved the primary predator in their landscape was the red squirrel (Tamiasciurus hudsonicus), and red squirrels were more abundant in heavily forested landscapes. We believe this difference can be explained by our overall model as a dif- ference in predator communities resulting from biogeographic and habitat differences in preda- tor communities. Another study (Friesen et al. 1999) found relatively high nesting success in a highly fragmented landscape in Ontario, but it is not possible to conclude if this difference was due to annual variation, biogeographic context, or a lack of generality of the fragmentation ef- fect. The effects of landscape composition on pred- ator abundance and distribution have received much less attention than patterns in nest success (Chalfoun et al. 2002). Raccoons (Procyon lo- tor) and opossums (Didelphis virginiana) reach their highest densities in highly fragmented landscapes (Andrn 1992, Dijak and Thompson 2000), potentially because their distributions are associated with developed and agricultural hab- itats that are interspersed with forest habitat. In eastern North America Blue Jays (Cyanocitta cristata) are significantly more abundant in highly fragmented landscapes with < 15% forest cover than in landscapes with moderate or high forest cover (T M. Donovan, unpubl. data). Ro- senberg et al. (1999) surveyed occurrence of FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 13 some potential nest predators along with tanager species; they generally found positive relation- ships between predators and fragmentation, but responses were often region or species specific. Abundance of some other predator species, how- ever, may not be affected by forest patterns at a landscape scale, but by more local habitat effects such as edge. PATTERNS OF L^ND COVER ^ND THEIR EFFECT ON THE ABUNDANCE OF COWBIRDS AND BROOD PARASITISM Landscape considerations seem logical for cowbirds because cowbirds utilize different hab- itats for feeding and breeding activities in the midwestern U.S. (Thompson 1994). Cowbirds generally feed in open grassy or agricultural ar- eas, whereas breeding resources (hosts) are often distributed in forested areas (Rothstein et al. 1984, Thompson 1994, Thompson and Dijak 2000). Telemetry studies in Missouri and New York show that although feeding and breeding resources can overlap spatially, cowbirds move between them to optimize the use of each re- source (Thompson 1994, Hahn and Hatfield 1995). In Missouri, female cowbirds tend to par- asitize nests in host-rich forests in the early morning and move to open grassy or agricultural areas to feed as the day progresses (Thompson 1994, Morris and Thompson 1998, Thompson and Dijak 2000). Also, cowbirds are common in hayfields and mowed roadsides in the White Mountains of New Hampshire, but do not occur in adjacent forest even though permanent open- ings and clearcuts exist in the forest (Yamasaki et al. 2000). Cowbirds are also more abundant along corridors such as roads that include mowed grass, than in forest interior in New Jer- sey (Rich et al. 1994). While the specific habi- tats used differ, the same landscape relationships between feeding and breeding habitat exist in western landscapes (Rothstein et al. 1984). The probability that a cowbird occurs in a forest, therefore, depends at least partly upon the prob- ability that a feeding area is nearby. As areas become more forested, cowbird breeding oppor- tunities may increase but feeding opportunities may decline. Hence, in heavily forested environ- ments such as the Missouri Ozarks, cowbird densities are low and parasitism rates of forest birds have been recorded in the 2-4% range (Clawson et al. 1997). In contrast, fragmented agricultural regions can support massive cow- bird populations that attack the limited number of forest breeding birds, resulting in parasitism rates approaching 100%, with high rates of mul- tiple-parasitism in a single nest (Robinson 1992). In this case, cowbirds are probably not ß r = -0.72 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 80 70 * ß 4O 30 20 10 ß 0 ß ' 0 20 40 60 80 100 % forest cover in landscape FIGURE 4. Correlation of the amount of forest cover in a 10-km radius with cowbird relative abundance and level of brood parasitism in the Midwestern United States. Data and figures are adapted from Thompson et al. 2000. food limited but may be constrained by the num- ber of available host nests. Cowbird abundance and levels of parasitism are closely correlated with landscape statistics reflecting the amount of forest fragmentation, the percent of forest cover, and the amount of potential feeding habitat (agricultural land uses) in the landscape. For example the number of cowbirds and level of brood parasitism are both highly negatively correlated with the amount of forest cover in a 10-km radius (Fig. 4). Land- scapes have been defined by 5- to 10-km radii in these studies (Robinson et al. 1995a, Donovan et al. 2000, Thompson et al. 2000), which relates well to the distances most cowbirds commute between breeding and feeding areas (<5 km; Thompson 1994, Thompson and Dijak 2000). Hochachka et al. (1999) combined numerous data sets from across the United States to test the generality of the midwestern pattern at two different spatial scales. They found that increas- ing amounts of forest cover within 10 km of study sites was correlated with reduced parasit- ism rates across the continent. In contrast, when they analyzed the data using forest cover within 50 km of the study site, they found that increas- ing forest cover resulted in slightly increased parasitism rates in sites west of the Great Plains. 14 STUDIES IN AVIAN BIOLOGY NO. 25 Although there are still details that we do not understand, it appears quite clear that there are landscape-level effects on cowbird densities that affect parasitism rates throughout the range of the Brown-headed Cowbird. We have suggested that the importance of landscape composition in limiting cowbird num- bers is constrained by biogeographic location. Is there evidence that landscape composition con- strains the importance of local-scale effects such as host density, nest concealment, or other fac- tors? Several studies suggest that cowbirds se- lect habitats with high host densities (Verner and Ritter 1983, Rothstein et al. 1986, Thompson et al. 2000). However, this relationship may de- pend upon whether landscapes offer both breed- ing and feeding opportunities for cowbirds. In Missouri, cowbirds are more abundant in frag- ments than in contiguous forest with a compar- atively greater abundance of hosts (Donovan et al. 2000). We found evidence that cowbird and host abundances were correlated in fragmented landscapes, but not in contiguous forest land- scapes, suggesting that landscape composition may constrain the influence of local host abun- dance on local cowbird abundance. If food or host resources are scarce at the landscape scale, local habitat characteristics may not explain ei- ther cowbird abundance or parasitism levels. Landscape composition may also constrain the importance of local-scale habitat features such as edge or patch size in determining cow- bird numbers and parasitism levels. For exam- ple, in a heavily forested landscape in Vermont (94% forest cover), cowbird distribution at the patch level was best explained by examining one local-scale habitat characteristic (patch area) and two landscape-scale habitat characteristics (dis- tance to the closest opening and the number of livestock areas [known feeding areas] within 7 km of the patch; Coker and Capen 1995). Sim- ilarly, in Missouri the distribution of cowbirds is not as well correlated with patch level statis- tics such as area or the ratio of perimeter to area, but by landscape-level measures that encompass the known daily movements of cowbirds (Don- ovan et al. 2000). HABITAT-SCALE EFFECTS Hypothesis: Habitat-scale factors affect the probability a nest is depredated or parasitiged because of effects on predator and cowbird abundance and activity patterns or nest detect- ability. The strength of these effects depends on the biogeographic and landscape context. Within a given biogeographic and landscape context, nest predation and brood parasitism should be related to habitat effects. Species de- mographics vary among habitats as a reflection of habitat quality. The question of interest here is whether there are consistent features or pro- cesses at the habitat scale, or interactions with landscape and biogeographic processes that el- evate predation and parasitism. Several possibil- ities of habitat effects are patch size, proximity to edge, forest management, and nest conceal- ment. These effects have been widely studied, yet there are substantial gaps in our knowledge and inability to explain known effects within a conceptual model. Recent reviews (Martin 1993, Paton 1994, Robinson and Wilcove 1994, Faa- borg et al. 1995, Heske et al. 2001) have ad- dressed these topics to various degrees. Here we address edge and forest management effects and how they fit within our general model. EDGE EFFECTS Edge effects are not uniform within or among regions (cf. Bolger this volume). Many studies show no edge effects or only such effects very close (<50 m) to edges (Paton 1994, Hartley and Hunter 1998). Parasitism levels remain high in forest far from edge in some landscapes (Marini et al. 1995, Thompson et al. 2000), and in at least one landscape parasitism in forest declined gradually from 70% to 5% over a gradient of 1500 m from an agricultural edge (Morse and Robinson 1999). At least four hypotheses have been suggested for higher predation rates near edges: (1) pred- ators may be attracted to edges because of abun- dant prey (a functional response; e.g., Gates and Gysel 1978, Ratti and Reese 1988); (2) predator density may be greater near edges than in forest interiors (a numerical response; e.g., Bider 1968, Angelstam 1986, Pedlar et al. 1997); (3) the predator community may be richer near edges (Bider 1968, Temple and Cary 1988, Marini et al. 1995); and (4) predators may forage along travel lanes such as edges (Gates and Gysel 1978, Yahner and Wright 1985, Small and Hunt- er 1988, Marini et al. 1995). Results of edge-effects studies have been in- consistent and comparisons among studies have been confounded by lack of experimental con- trol of landscape or habitat context, differences in predator communities, and methodological bi- ases. Problems associated with artificial nests exist (e.g., nest appearance, lack of parental and nestling activity), but even the types of eggs used in artificial nests may bias results. Large eggs (i.e., quail or chicken) exclude predation by some small predators and predation rates are greater when small eggs are used (Haskell 1995a, DeGraaf and Maier 1996). Lack of a mechanistic approach that addresses hypotheses for why predation should be higher near edges FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 15 has also hampered research. A more mechanistic approach requires studies of predator activities or abundances, not just nest predation patterns. Equally variable are the results of nest place- ment studies (i.e., ground vs. shrub/elevated nests). Major and Kendal (1996) reported higher predation at elevated nests in six studies, higher predation at ground nests in four studies, and equal predation rates in three studies. Ground nests containing Japanese Quail (Coturnix spp.) and plasticine eggs exhibited increased preda- tion along farm edge and interior in Saskatche- wan, but there were no detectable differences in predation rate between ground and shrub nests in logged edge, in logged interior, or in contig- uous forest (Bayne and Hobson 1997). Although two studies in the northeastern U.S. did not de- tect any difference in predation rates between ground and shrub nests (Vander Haegan and DeGraaf 1996, Danielson et al. 1997), DeGraaf et al. (1999) found a strong placement effect (high predation on ground nests) using small eggs, as did Marini et al. (1995). Our perspective on edge effects is from stud- ies in eastern forests that largely investigated predation of forest bird nests by medium sized mammals such as raccoons and opossums, and corvids such as Blue Jays and American Crows (Corvus brachyrhynchos). Based on our studies and others, we offer two predictions that may help account for the variability among previous studies. Edge effects are dependent on landscape and habitat context The importance of landscape context is emerging as perhaps one of the few generalities that can be made concerning edge effects. Our hypothesis is that the occurrence of local edge effects is dependent on landscape composition and pattern because of dependence of predators and cowbirds on landscape-level factors. Some evidence exits to support this hypothesis. Edge effects tend not to exist in mostly forested land- scapes (Heske 1995, Marini et al. 1995, Bayne and Hobson 1997, Hartley and Hunter 1998, DeGraaf et al. 1999, Chalfoun et al. 2002). Some level of forest fragmentation is necessary to support high numbers of generalist predators in eastern forests. At moderate levels of frag- mentation elevated predation rates will be lim- ited to edges because predators depend on ag- ricultural habitats or human settlements. At ex- treme levels of fragmentation all forest habitat is within close proximity to these habitats and predation is high throughout the forest. We be- lieve edge effects are a result of increases in abundance of predators due to landscape effects (fragmentation) and activity patterns of preda- tors in fragmented landscapes (Andrn 1995, Chalfoun et al. 2002). As previously discussed, Donovan et al. (1997) directly tested this hypothesis with a rig- orous field experiment using artificial nests, and found strong support for it. Hartley and Hunter (1998) detected the same effects in a meta-anal- ysis of artificial nest studies. In a different meta- analysis Chalfoun et al. (2002) determined that predator responses to edges, patch size, or frag- mentation were not independent of landscape context. Predator abundance or activity was re- lated to edge, patch area, or fragmentation in 66.7% of tests when adjacent land use was ag- ricultural, 5.6% when forest, 16.7% when grass- land, 5.6% when clearcut forest. In addition to the effect of landscape context on predator abundance, landscape and habitat contexts also affect the species of predators pre- sent. The variability in results among studies of egg predation may reflect diflrences in nest predator communities or the abundance of par- ticular species in study areas (e.g., Picman 1988). For example, in New England Blue Jays and raccoons were predominant predators of ar- tificial nests in suburban forests, whereas fishers (Mattes pennanti) and black bears (Ursus amer- icanus) were important in extensive forest (DeGraaf 1995, Danielson et al. 1997), and no avian nest predators were detected in the inte- riors of extensive forest (DeGraaf 1995). Attempts to identify egg predators include characterizations of predation remains of real eggs (Gottfried and Thompson 1978; but see Marini and Melo 1998), impressions in plasti- cine (Bayne et al. 1997) and clay eggs (Donovan et al. 1997), hair catchers (Baker 1980), and re- motely triggered cameras (DeGraaf 1995). The most promising technique, however, may be the use of subminiature video cameras with infrared illumination at real nests (Thompson et al. 1999, Bolger this volume). For example, F. Thompson and D. Burhans (pers. comm.) used this tech- nique and determined 85% of nest predation events in old fields were by snakes, whereas 60% of predation events in forests were by rac- coons. Not all edges are the same We suggest that negative edge effects are most likely to occur where land-use patterns or topography concentrate activities of predators, and are therefore a functional response by pred- ators. Edge effects are most likely to occur where forest abuts habitats that provide key re- sources for predators. Agricultural edges gener- ally have stronger edge effects than other types of edge (e.g., regenerating forest, grassland) on nesting success (Hanski et al. 1996, Hawrot and 16 STUDIES IN AVIAN BIOLOGY NO. 25 Neimi 1996, Darveau et al. 1997, Hartley and Hunter 1998, Marzluff and Restani 1999, Morse and Robinson 1999; but see King et al. 1996, Suarez et al. 1996) and on predators (Chalfoun et al. 2002). Differences in results among studies likely are due at least partly to differences in habitat use among predators. In one of the few studies of predator distri- butions relative to edges, Dijak and Thompson (2000) showed that raccoons respond differently to different edge types. Raccoon activity was significantly greater in forest adjacent to agri- cultural fields and riparian areas than in forest adjacent to roads, clearcuts, or forest interior. Studies of raccoon foraging behavior show that the degree of nest cover is much less important than local habitat heterogeneity in preventing depredation (Bowman and Harris 1980). In Illi- nois Blue Jays used edges differently and pre- ferred gradual shrubby edges (J. Brawn, unpubl. data). Avian predators were more abundant in forest-dividing corridors composed of shrub- sapling vegetation than grass in New Jersey (Rich et al. 1994). Heske (1995), however, found no significant difference in predator activity ad- jacent to and >500m from edges. Recent work in New England oak forests showed that six spe- cies of small mammals represented 99% of cap- tures at both forest edge and interior and their abundance and nest predation rates did not differ between edge and interior (DeGraaf et al. 1999). We believe these differences in edge effects are a result of differences in predator species, type of edge, and landscape context. SILvICULTURAL PRACTICES Silvicultural practices such as tree harvest and regeneration of stands (habitat patches) dramat- ically affect habitat scale characteristics. Bird communities can change greatly in response to these practices, and balancing the needs of spe- cies with diverse habitat needs in managed for- ests is a challenge for land managers and plan- ners (see review by Thompson et al. 1995). Here we focus on two aspects of silvicultural practices that are related to concerns for forest fragmen- tation: fragmentation of old forests by young forests, and creation of edges between old and young forests. Fragmentation of mature forest by young forest Fragmentation of mature forest by young for- est created by timber harvest has raised conser- vation concerns because of the loss of mature forest habitat and potential fragmentation ef- fects. Both even-aged forest management and uneven-aged forest management result in chang- es in the bird community (Thompson et al. 1992, Annand and Thompson 1997, Robinson and Robinson 1999). These changes in the bird com- munity can be interpreted as good or bad de- pending on management objectives. Habitat needs of forest breeding birds need to be ad- dressed by identifying conservation objectives and then evaluating the effects of land manage- ment practices on these. Young forests in the East provide habitat for at least some species acknowledged as management priorities (e.g., Kirtland's Warbler [Dendroica kirtlandii], Prai- rie Warbler [Dendroica discolor], Golden- winged Warbler [Vermivora chrysoptera]); therefore the needs of early and late successional species need to be addressed in forest manage- ment plans. We are aware of no evidence in eastern forests that fragmentation of mature forest by young forest creates the type of negative fragmentation effects that fragmentation by agricultural or de- veloped land uses do. We have suggested that cowbirds and generalist predators benefit from interspersion of agricultural and developed land use in forests because they provide rich food sources, but this would not seem to apply to young forests. For example, in extensively for- ested northern New England, predation rates on artificial ground and shrub nests were not dif- ferent among timber size-classes (DeGraaf and Angelstam 1993). Likewise, predation rates on artificial ground and shrub nests were similar in managed and reserved large forest blocks (DeGraaf 1995). Edge eJfkcts between mature and young forest Not many studies have directly addressed edge effects in managed eastern forests. The ev- idence for edge effects between mature forest and recently harvested stands is highly variable and suggests results vary locally. In a study of Ovenbird (Seiurus aurocapillus) reproductive success in northern New Hampshire in relation to clearcutting (King et al. 1996), nests, territo- ries, and territorial males obtaining mates were equally distributed in edge (0-200 m) and inte- rior (201-400 m) mature forest. Nest survival was higher in forest interior in year 1, but not in year 2. The proportion of pairs fledging at least one young, fledgling weight, and fledgling wing-chord did not differ between edge and in- terior in either year, nor did the number of young fledged per pair. In another study artificial nests were placed in edge areas (0-5 m from edges) and interior areas (45-50 m from edges) adja- cent to clearcuts and groupcuts. The probability of a nest being depredated was higher in edge than interior, and was independent of nest con- cealment, nest height, or whether adjacent to clearcuts or group-selection cuts (King et al. FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 17 1998). In Illinois forest predation of Kentucky Warbler (Oporornis formosa) nests was not re- lated to clearcut edges (Morse and Robinson 1999). Nest predation, however, was significant- ly higher in clearcuts than adjacent older forests, suggesting differences in vegetation structure were important while edge was not. Edge effects can differ among species nesting in the same habitat patch as well. Woodward et al. (2001) determined that nest success of songbirds nest- ing in regenerating forests and cedar glades var- ied with distance to mature forest edge, but that patterns were different among species and did not generally increase monotonically with dis- tance from edge. Given that edge effects seem to vary locally it is important to remember the top down nature of our model. Landscape level fragmentation of forests by habitats that elevate predator and cowbird numbers is likely a more important de- terminant of nest success at a population level than are local edge effects. While some studies have demonstrated edge effects, no studies have shown a population-level effect on viability. POPULATIONS ARE STRUCTURED AS SOURCES AND SINKS Hypothesis: Top-down spatial constraints lim- it reproductive success in some fragmented landscapes in the Midwest to the point where populations in such landscapes will either de- cline to extinction or will persist as part of a larger, source-sink system The presence of sink populations may or may not be a detriment to the larger population, depending on the amount of sink habitat in the landscape and to what de- gree individuals select sink habitat for breeding. AT A POPULATION SCALE, SINKS EXIST IN HIGHI. Y FRAGMENTED HABITATS Source-sink theory (Pulliam 1988) has be- come a popular framework for describing the population dynamics of organisms that are af- fected by habitat fragmentation. Pulliam (1988) used models based on births, immigration, deaths, and emigration (BIDE models; Cohen 1969, 1971) to describe geographic subpopula- tions that are connected by dispersal. All sub- populations contribute individuals that make up the greater population, or the entire source-sink system. At equilibrium, a subpopulation is a source when B > D and E > I; and is a sink when B < D but E < I. The greater population is at dynamic equilibrium (not changing) when B (all the births) + I (all the immigrants from outside the greater population) - D (all the deaths) - E (all the emigrants that leave the greater population) = 0. If habitat fragmentation subdivides populations into more or less inde- pendent breeding subpopulations, then source- sink structure may be an appropriate demo- graphic model. Is there any evidence that forest passerines exhibit source-sink population structure that is linked to the degree of habitat fragmentation? Several field studies document that reproductive success of neotropical migrant birds varies across a species' range (Probst and Hayes 1987, Robinson et al. 1995a), but few studies examine the interaction of subpopulations from a source- sink viewpoint. One must know the BIDE pa- rameters of each subpopulation to evaluate source-sink dynamics. Measurement of these pa- rameters is extremely field intensive and poten- tially unachievable with current techniques be- cause of the dispersal capabilities of birds. Sur- veys of bird abundance may not be capable of establishing source-sink status (Brawn and Rob- inson 1996). Most empirical studies documenting sink pop- ulations use nesting data and mortality data from the subpopulation, and model population persis- tence over time in the absence of immigration or emigration (Ricklefs 1973, King and Mewaldt 1987, Stacey and Taper 1992, Pulliam and Dan- ielson 1991, Donovan et al. 1995b). Without im- migration, sink populations decline over time and go extinct. With immigration, however, sinks can persist with no detectable declines in numbers over time (Pulliam 1988). What evidence is there, then, that birds are structured as sources and sinks, and that source- sink status is related to level of landscape-scale fragmentation? The evidence is very weak at this time, in part because we do not yet know the geographic scale that encompasses dispersal movements among sources and sinks. However, there is evidence that reproductive success in fragmented landscapes is too low to compensate for adult mortality (e.g., Donovan et al. 1995b, Trine 1998), and that dispersal occurs among habitat patches. For example, Trelease Woods is an isolated woodlot in central Illinois where bird populations have been censused since 1927 (Kendeigh 1982). In most years, several breed- ing pairs of Wood Thrush occurred in the wood- lot, but three extinction events were recorded that were followed by three colonization events, suggesting that the colonists of unknown origin were not produced locally (Brawn and Robinson 1996). Although direct evidence to support source- sink structure is weak, predictions generated from population modeling may offer some sup- porting evidence. Source-sink models suggest that sinks should show relatively higher year to year variation in abundance than source popu- lations (Davis and Howe 1992). As predicted, 18 STUDIES IN AVIAN BIOLOGY NO. 25 recent empirical studies demonstrate that popu- lations in fragmented landscapes have greater annual variation than populations in continuous landscapes, which may also affect turnover rates and local extinction (Boulinier et al. 1998). However, it is still unclear whether such vari- ability is due to local processes (such as vari- ability in source-sink status over time), to source-sink dispersal dynamics, or other causes. THERE IS NO EVIDENCE THAT SINKS OR EDGES FUNCTION AS ECOLOGICAL TRAPS AT A LOCAL SCALE Although reproductive and survival rates are too low to maintain numbers in sinks, these hab- itats may benefit the greater source-sink system by "housing" a large number of individuals at any given time. Additionally, a significant num- ber of young may be produced in low-quality habitats, depending on the number of individuals breeding there (Pulliam 1988, Howe et al. 1991). Is there evidence, however, that maintenmce of sink habitat is a detriment to population per- sistence? Animals often have the opportunity to select among a variety of habitats that vary in quality; preferred habitats are those that are se- lected disproportionately to other available habi- tats (Johnson 1980). If individuals avoid low- quality areas, the presence of low-quality habitats may not negatively influence population persis- tence. However, if individuals select low-quality habitats over available, high-quality habitats for reproduction and survival, then low-quality hab- itats may function as ecological traps, and their presence may lead to population extirpation (Gates and Gysel 1978, Ratti and Reese 1988, Pulliam md Dmfielson 1991). Edges have been suggested to be an ecologi- cal trap because they are potentially food rich and have high abundances and diversity of birds, which in turn potentially attract predators searching for food-rich areas (Gates and Gysel 1978, Ratti and Reese 1988). Woodward et al. (2001) examined the ecological trap hypotheses for several species of shrubland-nesting song- birds, and while nesting success varied with dis- tance to edge, they found no evidence that edges acted as ecological traps. Observations of high densities of Wood Thrushes in fragmented Mid- west landscapes (Donovan et al. 1995b) have led us to speculate that fragments are similarly act- ing as traps. High densities of birds in poor-qual- ity fragmented landscapes and low densities in high-quality contiguous landscapes may be the result of: (1) absence of suitable habitat features such as nest sites in contiguous landscapes; (2) displacement of individuals from high quality contiguous landscapes through interspecific competition; or (3) innate preference for habitat characteristics that more commonly occur in fragmented landscapes, such as edge. Population models suggest that when individ- uals in the population selected high- and low- quality habitats in proportion to habitat avail- ability in the landscape, landscapes could con- tain up to 40% low-quality habitat and still pro- mote population persistence. However, when individuals preferred low-quality habitats over high-quality habitats, populations on landscapes containing > 30% low-quality habitat were ex- tirpated, and the low-quality habitat functioned as an ecological trap (Donovan and Thompson 2001). Clearly, much more work is needed to determine the effect of sink habitats on popula- tion persistence. POPULATIONS STRUCTURED AS SOURCES AND SINKS CAN GROW OR DECLINE Populations structured as sources and sinks can grow or decline depending on the amount of sink habitat, the selection and use of sinks for breeding, and the magnitude of spatial and tem- poral variation in demographic parameters. It is critical that we examine how our observations of reduced fecundity or density in fragmented landscapes may impact population trends of a source-sink system. We believe our observations of correlations between nesting success and for- est cover at the landscape level in the Midwest (e.g., Robinson et al. 1995a) have been uncriti- cally cited as strong evidence that habitat frag- mentation causes bird populations to decline. The negative correlation between fragmentation and nesting success offers support for the hy- pothesis that fragmentation of breeding habitat is causing declines in some songbird population. No one, however, has attempted to evaluate the number of source and sink populations and their effect on a regional population. For example, Ovenbirds in the Midwest U.S. are thought to be impacted by habitat fragmen- tation in several ways: they are area-sensitive (Hayden et al. 1985, Burke and Nol 1998), their pairing success on fragments is often signifi- cantly lower compared with larger, contiguous patches (Gibbs and Faaborg 1990, Villard et al. 1993), and they have higher daily nest-mortality and parasitism levels in fragments compared with larger patches (Donovan et al. 1995b, Rob- inson et al. 1995a). Yet, Breeding Bird Survey data suggest that Ovenbirds are maintaining numbers and even increasing in many areas in the Midwest (Sauer et al, 1997). Overall popu- lation growth (the growth rate of the entire source-sink system on the landscape) may not be impacted by the poor reproductive success of birds in fragments if breeding individuals gen- erally avoid small patches or if the landscape is FRAGMENTATION IN EASTERN FORESTS--Thompson et al. 19 dominated by larger patches that are used for breeding. We have used modeling approaches to test how landscape composition, habitat selection, and nesting success interact to produce popula- tion increases or declines at a regional scale (Donovan and Lamberson 2001). The model combined (1) the frequency distribution of patch sizes in the landscape (e.g., highly fragmented landscapes vs. continuously forested land- scapes), (2) the distribution of individuals across the range of patches in the landscape (e.g., area sensitive vs. area insensitive vs. edge distribu- tion patterns), and (3) the fecundity of individ- uals as a function of patch size in the landscape (e.g., fragmentation effects on fecundity vs. no fragmentation effects on fecundity). We used this model to examine population growth under various landscape, distribution, fecundity, and survival scenarios. Results from the model indicate that the high- ly cited observation that fecundity decreases as patch size decreases does not necessarily cause landscape level population declines in songbirds. When total habitat in the landscape is held con- stant, reduced fecundity associated with patch size could lead to population declines when landscapes are highly fragmented, or when land- scapes are more continuous, but individuals oc- cur in high densities in small patches and low densities in large patches. Thus, when land- scapes offer both large and small patches for breeding (a more contiguous landscape), area- sensitive species can maintain population sizes in spite of decreased fecundity in small patches because birds achieve their highest densities in patches where fecundity is greatest, and high re- production in such source habitats can maintain sinks within the landscape (Donoran and Lam- berson 2001). Two recent large scale analyses of Breeding Bird Survey data have linked popula- tion change to fragmentation. Donoran and Flather (2002) found a significant negative cor- relation between the proportion of a population occupying fragmented habitat and population trend. Boulinier et al. (2001) found that richness of forest area-sensitive species was lower, and year-to-year rates of local extinction higher, on Breeding Bird Survey routes surrounded by landscapes with lower mean forest-patch size. RESEARCH AND CONSERVATION IMPLICATIONS We believe there is adequate corroborative ev- idence for this multi-scale approach to fragmen- tation to use this as a working model for re- search and conservation. We believe one of the most important conclusions from our work in eastern forests is that landscape composition is an important determinant of reproductive suc- cess, even at a local scale. In eastern forests where concerns are focused on the effects of cowbird parasitism and on generalist predators associated with agricultural and other human- dominated land uses, fragmentation of forests and a reduction in the amount of forest in the landscape results in increased levels of predation and parasitism. Future research should directly test our hypotheses of top-down constraints on reproductive success as well as hypothesized mechanisms for effects at each scale. Research should address the larger scale context of studies and potential differences among predators. There is already evidence that landscape level effects of fragmentation differ between the west- ern and eastern United States (Tewksbury et al. 1998), which is further indication of the impor- tance of top-down constraints and a multi-scale approach. This model has important conservation impli- cations as well. The importance of large-scale effects suggests that at high levels of fragmen- tation, conservation efforts should be focused on restoration of the landscape matrix and a reduc- tion in fragmentation. At some level, where the landscape-level effects of fragmentation are no longer critical, local habitat management prac- tices become important. Local management con- siderations could include management practices to provide appropriate habitat types, minimize edge, or manage habitat structure. Finally, while we believe fragmentation is a major conserva- tion issue in eastern forests, we caution that not all fragmentation needs to be mitigated. Frag- mentation of one habitat provides other habitats, and source-sink dynamics suggest that some proportion of a population can reside in sink habitat. A challenge for researchers, land man- agers, and policy-makers is to determine when fragmentation at a regional or population level is severe enough to drive population declines, and to balance competing species conservation objectives and land use. ACKNOWLEDGMENTS We thank the numerous graduate students, techni- cians, colleagues, and supporting agencies who have assisted or supported the work that led to the ideas presented in this paper. Studies in Avian Biology No. 25:20-29, 2002. WHAT IS HABITAT FRAGMENTATION? ALAN B. FRANKLIN, BARRY R. NOON, AND T. LUKE GEORGE Abstract. Habitat fragmentation is an issue of primary concern in conservation biology. However, both the concepts of habitat and fragmentation are ill-defined and often misused. We review the habitat concept and examine differences between habitat fragmentation and habitat heterogeneity, and we suggest that habitat fragmentation is both a state (or outcome) and a process. In addition, we attempt to distinguish between and provide guidelines for situations where habitat loss occurs without frag- mentation, habitat loss occurs with fragmentation, and fragmentation occurs with no habitat loss. We use two definitions for describing habitat fragmentation, a general definition and a situational definition (definitions related to specific studies or situations). Conceptually, we define the state of habitat frag- mentation as the discontinuity, resulting from a given set of mechanisms, in the spatial distribution of resources and conditions present in an area at a given scale that affects occupancy, reproduction, or survival in a particular species. We define the process of habitat fragmentation as the set of mechanisms leading to that state of discontinuity. We identify four requisites that we believe should be described in situational definitions: what is being fragmented, what is the scale of fragmentation, what is the extent and pattern of fragmentation, and what is the mechanism causing fragmentation. Key Words: forest fragmentation; habitat; habitat fragmentation; habitat heterogeneity. Habitat fragmentation is considered a primary issue of concern in conservation biology (Meffe and Carroll 1997). This concern centers around the disruption of once large continuous blocks of habitat into less continuous habitat, primarily by human disturbances such as land clearing and conversion of vegetation from one type to an- other. The classic view of habitat fragmentation is the breaking up of a large intact area of a single vegetation type into smaller intact units (Lord and Norton 1990). Usually, the ecological effects are considered negative (Wiens 1994). In this paper, we propose that this classic view pre- sents an incomplete view of habitat fragmenta- tion and that fragmentation has been used as such a generic concept that its utility in ecology has become questionable (Bunnell 1999a). In attempting to quantify the effects of habitat fragmentation on avian species, there is consid- erable confusion as to what habitat fragmenta- tion is, how it relates to natural and anthropo- genic disturbances, and how it is distinguished from terms such as habitat heterogeneity. Here, we attempt to provide sufficient background to define habitat fragmentation adequately and, as a byproduct, habitat heterogeneity. This paper was not intended as a complete review of the existing literature on habitat fragmentation but merely as a brief overview of concepts that al- lowed us to arrive at working definitions. There are two ways to define habitat frag- mentation. First, there is a conceptual definition that is sufficiently general to include all situa- tions. We feel a conceptual definition is needed for theoretical discussions of habitat fragmenta- tion. Second, there is a situational definition that relates to specific studies or situations. In this paper, we review current definitions and offer a revised conceptual definition of habitat fragmen- tation. In addition, we propose four requisites for building situational definitions of habitat fragmentation: (1) what is being fragmented, (2) what is the scale(s) of fragmentation, (3) what is the extent and pattern of fragmentation, and (4) what is the mechanism(s) causing fragmen- tation. To define habitat fragmentation, it is first necessary to review current understanding of how habitat is defined, and to contrast fragmen- tation and heterogeneity. FRAGMENTATION--THE HABITAT CONCEPT Prior to understanding fragmentation of hab- itat, the term habitat must be properly defined and understood. Habitat has been defined by many authors (Table 1) but has often been con- fused with the term vegetation type (Hall et al. 1997; see Table 1). As Hall et al. (1997) point out, habitat is a term that is widely misused in the published literature. The key features of the definitions of habitat in Table I are that habitat is specific to a particular species, can be more than a single vegetation type or vegetation struc- ture, and is the sum of specific resources needed by a species. Habitat for some species can be a single vegetation type, such as a specific seral stage of forest in a region (e.g., old forest in Fig. 1 a). This might be the case for an interior forest species where old forest interiors provide all the specific resources needed by this species. How- ever, habitat can often be a combination and configuration of different vegetation types (e.g., meadow and old forest in Fig. lb). In the ex- ample shown in Figure lb, a combination of old forest and meadow are needed to provide the specific resources for a species. Old forest may 2O WHAT IS HABITAT FRAGMENTATION?--Franklin et al. 21 ¸ ¸ 22 STUDIES IN AVIAN BIOLOGY NO. 25 POOR Old forest Meadow Non-habitat e,, IG m FIGURE 1. Example of habitat represented as (a) a single vegetation type, (b) a mosaic of different vegetation types, and (c) different mosaics of vegetation types representing different degrees of habitat quality. provide some resources necessary for survival, whereas meadow might provide resources nec- essary for reproduction. In addition to considering habitat versus non- habitat (the intervening matrix), habitat can have a gradient of differing qualities (Van Horne 1983) where habitat quality is defined as the ability of the environment to provide conditions appropriate for individual and population persis- tence (Hall et al. 1997). The idea that habitat can be a specific combination and configuration of vegetation types can be extended further to different combinations and configurations rep- resenting different levels of habitat quality (Fig. lc). Poor habitat quality may result from too much of one vegetation type relative to another. Returning to the example from Figure lb, too much meadow may provide sufficient resources for reproduction, but not enough for survival (Fig. lc). Habitat quality is influenced by the mix and configuration of the two vegetation types (Fig. lc). An important consideration in both defining and understanding habitat fragmentation is that it ultimately applies only to the species level be- cause habitat is defined with reference to a par- ticular species. Habitat is proximately linked to communities and ecosystems only because these levels are composed of species. There is no con- cept of community or ecosystem habitat. For ex- ample, one cannot take a vegetation map and assess habitat fragmentation without reference to a particular species. Therefore, habitat fragmen- tation must be defined at the species level and those levels below (e.g., populations and indi- viduals within species). FRAGMENTATION VERSUS HETEROGENEITY Based on existing definitions (Table 1), frag- mentation can be viewed as both a process (that which causes fragmentation) and an outcome (the state of being fragmented; Wiens 1994). The definitions in Table 1 suggest that fragmen- tation represents a transition from being whole to being broken into two or more distinct pieces. The outcome of fragmentation is binary in the sense that the resulting landscape is assumed to be composed of fragments (e.g., forest) with something else (the non-forest matrix) between the fragments. In contrast, heterogeneity implies a multi-state outcome from some disturbance process. For example, contiguous old-growth forest can be transformed into a mosaic of dif- ferent seral stages by some disturbance such as fire (e.g., Fig. lb). If each seral stage, as viewed by a species, is a distinct habitat, then the result of the disturbance is an increase in habitat het- erogeneity. In addition, if habitat is a combina- WHAT IS HABITAT FRAGMENTATION?--Franklin et al. 23 tion of different vegetation types, then hetero- geneity in vegetation types may influence habitat quality (e.g., Fig. lc), but does not represent fragmentation. Habitat fragmentation is heterogeneity in its simplest form: the mixture of habitat and non- habitat. However, the effects of habitat fragmen- tation is also dependent on the composition of non-habitat. The matrix of non-habitat may have a positive, negative, or neutral effect on adjacent habitat. For example, non-habitat consisting of agricultural fields may have a very different ef- fect than non-habitat consisting of younger for- est. The key point is whether intervening non- habitat affects the continuity of habitat with re- spect to the species. We argue that habitat frag- mentation has not occurred when habitat has been separated by non-habitat but occupancy, re- production or survival of the species has not been affected. Under this argument, key com- ponents in defining habitat fragmentation are scale, the mechanism causing separation of hab- itat from non-habitat (i.e., the degree to which connectivity is affected), and the spatial arrange- ment of habitat and non-habitat. For example, a narrow road dividing a large block of habitat may not affect occupancy, reproduction or sur- vival for a wide-ranging species, such as a rap- ton However, the road may affect a species with a narrower range, such as a salamander. Thus, fragmentation is from the species' viewpoint and not ours. We discuss these points in more detail further on. The analogy of habitat fragmentation as equivalent to the breaking of a plate into many pieces (Forman 1997:408)is of limited utility. First, habitat fragmentation generally occurs through habitat loss; unlike the broken plate, the sum of the fragments is less than the whole. For example, in a uniform landscape composed en- tirely of a single habitat, fragmentation is only possible if accompanied by habitat loss. Thus, fragmentation usually involves both a reduction in area and a breaking into pieces (Bunnell 1999b). Second, the transition from being whole to being in pieces may lead to a change in qual- ity of one or more of the fragments if habitat quality is a function of fragment size. For ex- ample, fragmentation of continuous forest (ac- companied by an inescapable reduction in forest area) may change the quality of the fragments; habitat quality may increase for edge species and decrease for forest interior species (Bender et al. 1998). When the effects of habitat loss and fragmen- tation are addressed independently, habitat loss has been suggested as having the greatest con- sequences to species viability (e.g., McGarigal and McComb 1995, Fahrig 1997). This obser- vation led Fahrig (1999) to suggest the need to distinguish three cases: (1) habitat loss with no fragmentation; (2) fragmentation arising from the combined effects of habitat loss and break- ing into pieces; and (3) fragmentation arising from the breaking apart but with no loss in hab- itat area. These three cases are illustrated in Fig- ure 2. It is possible to illustrate these cases with reference to a common landscape only if the ref- erence landscape is composed of at least one habitat and a surrounding matrix within the bounded landscape (Fig. 2). This occurs because case (3) requires the ability to shift the location of the focal habitat within the landscape bound- aries. If there was no matrix within the land- scape boundaries (e.g., the landscape was com- posed entirely of the single habitat), then only cases (1) and (2) in Fig. 2 would apply. The possibilities illustrated in Fig. 2 are not artificial constructs. Conservation planning usu- ally occurs in a context of habitat mosaics with a diversity of land uses and land ownerships. As such, case 3 is a common result of conservation tradeoffs. For example, wetland mitigation in the U.S. often requires no net loss in wetland area but allows a change in the spatial pattern and location of wetlands. Thus, it is possible to break one large wetland into two or more pieces, mit- igate this loss somewhere else on the landscape by creating additional wetlands, and claim no net loss in area. Fragmentation arising from habitat loss un- avoidably leads to an increase in heterogeneity in habitat quality because the fragments may un- dergo a change in state either directly (through conversion) or indirectly through edge effects (see Bolger this volume, Sisk and Batten this volume). In light of the previous discussion, this possibility suggests that we need another case in addition to those discussed by Fahrig (1999). This case (case 4 in Fig. 2) includes changes in the spatial pattern of a habitat that are, or are not, accompanied by a change in the quality of the habitat. Case (4) would occur as a byproduct of case (2) depending on the habitat require- ments of the species in question. We attempt to capture these differences in outcome in a dichotomous flow diagram (Fig. 3). Following the diagram from top to bottom requires the investigator to answer a series of questions: "Has there been a reduction in area of the focal habitat? .... Has there been a change in spatial continuity of the habitat? .... Has there been a change in quality of the focal habitat?" Answering this progression of questions allows one to discriminate habitat loss from fragmen- tation, and to recognize cases where habitat quality has changed. A final point is that fragmentation of vegeta- 24 STUDIES IN AVIAN BIOLOGY NO. 25 Original habitat boundary /'1 i __ ', ................... Landscape boundary/i ! _ _ _  Original Landscape / [,/ ...... "/ I L __ with Focal Habitat | .a.ja.t/"'x,   1. Habitat loss + no fragmentation 2. Habitat loss + fragmentation 3. No habitat loss + fragmentation 4. Habitat loss + fragmentation + change in habitat quality FIGURE 2. Four cases illustrating the relationship between habitat loss, habitat fragmentation, and change in habitat quality in a bounded landscape. tion type and habitat fragmentation are often considered synonymous (e.g., the definition by Faaborg et al. (1993) in Table 1). However, the extent and effects of fragmentation can be very different when habitat is considered a single vegetation type or a combination of vegetation types (Fig. 4). Starting with the landscape in Figure 4, forest fragmentation would only be I Contiguous Habitat Area Reductin? YES Aea Re] NO I Ch; in Spatial Continuity?) (chane in Spatial Continuity?) YES NO NO ß YES i   Fragmented I Habitat I Fragmented I I Contiguous (in Quality?) (in Ouality?/ inQ YES NO YES % NO YES NO Habitat Loss Habitat Loss Habitat Loss Habitat Loss Fragmentation Fragmented + + + + Habitat Fragmentation Fragmentation Change in Quality Change in Quality + Change in Quality FIGURE 3. Flow diagram to differentiate between landscapes experiencing habitat loss, habitat fragmentation, and changes in habitat quality. WHAT IS HABITAT FRAGMENTATION?--Franklin et al. 25 I Old forest Forest Fragmentation Meadow Habitat Fragmentation i .ance FIGURE 4. Schematic differences in forest fragmentation and habitat fragmentation in a landscape composed of a habitat consisting of two vegetation types (old forest and meadow). considered as habitat fragmentation for a species whose habitat was solely defined as interior old forest (a single vegetation type). However, for the hypothetical example used previously where a species' habitat is composed of two vegetation types (meadow and old forest), habitat fragmen- tation would occur when some disturbance (such as a flood) disrupted the continuity in the con- figuration of these two vegetation types (Fig. 4). Thus, to define habitat fragmentation adequately, habitat must first be defined at a scale relevant to the species being examined. WHAT IS THE SCALE OF FRAGMENTATION. 9 The second requisite for defining habitat frag- mentation is determining the scale at which frag- mentation is occurring. Wiens (1973) and John- son (1980) recognized different scales in under- standing distributional patterns and habitat se- lection, respectively. For example, Johnson (1980) proposed first-order selection at the geo- graphical range of a species, second-order at the home range of individuals or social groups, and third-order at specific sites within individual home ranges. A similar hierarchical scaling can be used in defining and understanding habitat fragmentation. For example, habitat tYagmenta- tion could be considered at a range-wide scale for fragmentation that occurs throughout a spe- cies geographic distribution, a population scale where fragmentation occurs within populations connected by varying degrees by animal move- ment, and a home-range scale for fragmentation that occurs within home ranges of individuals (Fig. 5). While this scaling can be subdivided into finer intermediate levels, the idea remains the same; habitat tYagmentation is scale-depen- dent with different processes predominating at the different scales for a given species. For ex- ample, lYagmentation at the range-wide scale can affect dispersal between populations, frag- mentation at the population scale can alter local population dynamics, and fragmentation at the home range scale can affect individual perfor- mance measures, such as survival and reproduc- tion. Clearly, the different scales are not mutu- ally exclusive, but provide a unifying nested re- lationship that allows for understanding mecha- nisms and processes at different levels (Johnson 1980). Rather than a hierarchical scale. Lord and Norton (1990) proposed a continuous gradient of scale. At one end of the gradient, they defined geographical fragmentation where fragments are large relative to the scale of the physiognom- ically dominant plants (Fig. 6a) and. at the op- posite end, they defined structural fragmentation where tYagments are individual plants or small 26 STUDIES IN AVIAN BIOLOGY NO. 25 Range-wide Scale Population Scale FIGURE 5. Home Ranme Scale Example of three different scales at which habitat fragmentation can occur. groups of plants (Fig. 6b). While this gradient puts fragmentation on a continuous scale, it lacks the biological connection of the species- centered, hierarchical approach advocated by Johnson (1980). The ideal would be a gradient that is continuous and that has a biological con- text. Regardless of how scale is measured, a sit- uational definition should include scale because inferences to population and distributional pro- cesses for a given species are limited to what- ever scale is being examined. Fragmentation tha! affects processes at the home range scale (i.e., individual survival and reproduction) do not necessarily affect processes at a population or range-wide scale (i.e., dispersal between popu- lations of home ranges). For example, fragmen- tation that affects foraging sites within the home range of an individual may not impede the abil- ity of the offspring of that individual to disperse across a wider area. WHAT IS THE EXTENT AND PATTERN OF FRAGMENTATION? Here, we refer to the extent of habitat frag- mentation as the degree to which fragmentation has taken place within a specified spatial scale, whereas the pattern of fragmentation describes patch geometry, e.g., size, shape, distribution, and configuration. Extent describes how much fragmentation has taken place (Fig. 7) whereas geometry describes the pattern of habitat frag- mentation. For example, the patterns of frag- mentation in Figure 8 appear very different even though the total amounts of remaining habitat are the same. Various spatial parameters and sta- tistics (e.g., Turner and Gardner 1991, Mc- Garigal and Marks 1995) can be used to describe the different patterns in Figure 8. A considerable literature exists on how to describe the extent and pattern of habitat fragmentation and we will not review these quantitative methods here. However, a situational definition should include some measure of extent and pattern of fragmen- tation to place it in context. WHAT IS THE MECHANISM CAUSING FRAGMENTATION? Habitat fragmentation often occurs because of some disturbance mechanism. However, habitat fragmentation can be static, such as resulting from topographic differences (Forman 1997: 412). For example, habitat used by Mexican WHAT IS HABITAT FRAGMENTATION?--Franklin et al. 27 as \ I I / ! ! Do FIGURE 6. Example of (a) geographical fragmenta- tion as illustrated by patches of sagebrush and (b) structural fragmentation as illustrated by the distribu- tion of individual sagebrush plants on a plot within one of the patches (after Lord and Norton 1990). Spotted Owls (Strix occidentalis lucida) is dis- tributed on a range-wide scale in a highly frag- mented manner across four states in the U.S. (Keitt et al. 1997; see Fig. 5). This distribution is essentially fixed over an ecological time frame. Dynamic mechanisms occur with some fre- quency within a time frame that is applicable to the ecology of the species and the habitat they use. These mechanisms can be "natural" (fire, wind, etc.) or anthropogenic (logging, agricul- ture, urbanization, etc.; Forman 1997:413). In a given area at a given scale, these mechanisms can simultaneously fragment habitat for some species while creating habitat for others. In con- servation issues, the mechanisms causing habitat fragmentation are often of primary concern, es- pecially when these mechanisms are human-in- duced. A complete description of fragmentation must include an understanding of how the matrix in- fluences the ability of the habitat to support a species. If the matrix differs substantially from the original habitat, the impacts on the species may be more severe than if the matrix differs little. That is, fragmentation is also a function of the degree of contrast in quality between the fo- cal habitat and its neighborhood. For example, both selective logging and building homes may cause tYagmentation of unharvested forest but the consequences may be very different for the species that inhabit the landscape. Most mea- sures of habitat fragmentation do not consider the effects of the matrix on the survival and re- production of individuals or populations within the remaining patches. Understanding what mechanisms are contrib- uting to habitat fragmentation is important for placing habitat fragmentation into the context of either an acceptable ecological process (i.e., re- sulting from natural mechanisms) or a required conservation action (i.e., fragmentation resulting from anthropogenic mechanisms). Current dog- ma on habitat fragmentation is value-biased to- ward a negative connotation (Wiens 1994, Meffe and Carroll 1997); use of the term currently im- plies that the biological effects are negative. However, habitat fragmentation can be value- neutral or positive, depending on the species. FRAGMENTATION--A CONCEPTUAL DEFINITION We propose that the state (or outcome) of hab- itat fragmentation can be defined conceptually as the discontinuity, resulting from a given set of mechanisms, in the spatial distribution of re- sources and conditions present in an area at a given scale that affects occupancy, reproduc- tion, or survival in a particular species. From this, the process of habitat fragmentation can be defined as the set of mechanisms leading to the discontinuity in the spatial distribution of re- sources and conditions present in an area at a given scale that affects occupancy, reproduc- tion, and survival in a particular species. In de- veloping these definitions, we incorporated def- initions proposed by Lord and Norton (1990) and Hall et al. (1997; Table 1) and included three of the four requisites that we previously outlined. The fourth requisite, the extent and pattern of fragmentation, was not included be- cause it hampers the ability of the definition to be general. However, scale and mechanism are included in the definition to avoid, even in gen- eral terms, misleading statements. The term hab- itat fragmentation has acquired a negative con- notation over the years (Wiens 1994). Habitat fragmentation can occur naturally and the term should not be interpreted solely in terms of its potential negative impacts. Our definition re- 28 STUDIES IN AVIAN BIOLOGY NO. 25 None -" High Extent of Fragmentation FIGURE 7. Schematic representation of changes in the extent of fragmentation (after Curtis 1956). moves the value-bias that currently is attached to the phrase "habitat fragmentation." How does our definition differ from previous definitions? We believe our definition is more specific than the definition proposed by Morri- son et al. (1992) and explicitly incorporates the concept of continuity (Lord and Norton 1990) that is lacking in the definitions of Wiens (1989) and Forman (1997) (Tablel). The definition by Faaborg et al. (1993) does not fit the definitions of habitat by Block and Brennan (1993) and Hall et al. (1997), and is more applicable to vegeta- tion type fragmentation than to habitat fragmen- tation. 8ITUATIONAL DEFINITIONS To state that "the habitat is fragmented" is insufficient for understanding the scope of a par- ticular conservation problem or the potential ef- fects on the status of a given species in a given area. When defining fragmentation for a given situation (say, within a particular study, conser- vation plan, or for a given species), statements a FIGURE 8. Examples of different patterns of habitat fragmentation for an area having equal habitat amounts but (a) fewer large patches with higher edge to interior ratio versus (b) greater number of small patches with lower edge to interior ratio. about habitat fragmentation should include the four requisites discussed earlier. The first requi- site, what is being fragmented, requires an un- derstanding of a species' habitat. The second requisite, scale, is essentially a statement as to where inferences are being made and the level of habitat description being considered (e.g., stands of vegetation versus structure of vegeta- tion within stands). The third requisite, extent and pattern of fragmentation, provides a descrip- tion of the magnitude and type of habitat frag- mentation. The fourth requisite, mechanisms, puts habitat fragmentation into a temporal scale (how rapidly changes occur over time) and also into an ecological and conservation context ("natural" versus anthropogenic, or situations in between). A situational definition for habitat fragmen- tation will not necessarily be limited to a com- pact statement as is the conceptual definition. Rather, it should be considered as a series of paragraphs, or even an entire manuscript that in- cludes the four requisites. However, the four req- uisites should be identified and stated clearly to put habitat fragmentation for a particular situa- tion into its appropriate context. CONCLUSIONS By defining habitat fragmentation as we have proposed here, people will have to think more clearly about the characteristic attributes of frag- mentation. While some may consider our at- tempts at defining habitat fragmentation as an over-emphasis on semantics, we agree with Pe- ters (1991) and Hall et al. (1997) that vague and inconsistent terminology in the ecological sci- ences leads to ineffective and misleading com- munication, poor understanding of concepts, and WHAT IS HABITAT FRAGMENTATION?--Franklin et al. 29 generally sloppy science. Habitat is a unifying concept in ecology (Block and Brennan 1993) and central to many of the conservation prob- lems that ecologists face. We believe that de- veloping precise definitions for key concepts at the interface between ecology and conservation is paramount before these concepts become so muddled that ecologists become ineffective in their ability to deal with problems and to com- municate those problems to others. ACKNOWLEDGMENTS We thank R. A. Askins and J. A. Wiens for their thoughtful reviews of this manuscript. We also thank D. Dobkin and J. Rotenberry for their useful comments and for editing this volume. Studies in Avian Biology No. 25:30-48, 2002. HABITAT EDGES AND AVIAN ECOLOGY: GEOGRAPHIC PATTERNS AND INSIGHTS FOR WESTERN LANDSCAPES THOMAS D. $ISK AND JAMES BATTIN Abstract. Habitat edges are an important feature in most terrestrial landscapes, due to increasing rates of habitat loss and fragmentation. A host of hypothesized influences of habitat edges on the distri- bution, abundance, and productivity of landbirds has been suggested over the past 60 years. Never- theless, "edge effects" remains an ill-defined concept that encompasses a plethora of factors thought to influence avian ecology in heterogeneous landscapes. The vast majority of research on edge effects has been conducted in the broad-leafed forests of northeastern and midwestern North America. In general, many western habitats are more heterogeneous and naturally fragmented than their eastern counterparts, and habitat edges are a ubiquitous component of most western landscapes. These dif- ferences in landscape structure suggest that edge effects, and the mechanisms underlying them, may differ markedly in the West. We examined over 200 papers from the peer-reviewed literature on edge effects, focusing our efforts on empirical results and trends in research approaches. The relative dearth of western studies makes geographic comparisons difficult, but it is clear that mechanistic understand- ing of edge effects has lagged behind pattern identification. Bird responses to edge effects tend to vary markedly among species and among different edge types, while no clear pattern emerges re- garding species diversity. In the context of the review, we discuss research and modeling approaches that could move our understanding of edge effects toward a more mechanistic and predictive frame- work. Key Words: core area model; density; edge effects; effective area model; habitat edge; habitat frag- mentation; heterogeneity; species diversity. Habitat fYagmentation increases landscape het- erogeneity as continuous patches of native hab- itats are broken into numerous smaller, isolated patches surrounded by a matrix of different, of- ten heavily disturbed or anthropogenic habitats (Wilcox 1980, Wilcove et al. 1986, Wiens 1994, Franklin et al. this volume). The loss of native habitat cover and the increasing isolation of the resulting patches from one another have been the subject of numerous empirical and theoreti- cal studies and several reviews (e.g., Saunders et al. 1991, Faaborg et al. 1995). Since the early 1970s these two factors have dominated debates about conservation planning in increasingly fragmented landscapes (e.g., Diamond 1976; Simberloff and Abele 1976, 1982; Terborgh 1976). Another result of habitat fragmentation is an increase in the amount of edge habitat, as well as the proliferation of new types of edges, as anthropogenic habitats (e.g., agriculture, logged forest, and urbanized areas) replace native hab- itats and abut the remaining fragments. The in- creasing number of smaller patches, and the lin- ear or irregularly shaped patches that often result from fragmentation (Feinsinger 1997), contrib- ute to the rapid, often exponential increase in the amount of edge habitat in the landscape (Fig. 1). Implications of the proliferation of edge hab- itat for bird populations are numerous, ranging from the alteration of microclimatic conditions to changes in interspecific interactions, such as competition, predation, and nest parasitism. These and other edge effects are often distinct from the effects associated strictly with the loss of habitat and the increasing isolation of the re- maining patches. By influencing the quality of nearby habitat in the remaining fragments, edges may also directly affect the amount of available suitable habitat (Temple 1986, Sisk et al. 1997). Thus, edge effects constitute a class of impacts that are of increasing importance as fragmenta- tion advances and the heterogeneity and struc- tural complexity of the landscape increases. Despite over 60 years of active research, our understanding of edge effects remains diffuse and largely site-specific. Interestingly, the liter- ature on "edge effects" predates research on habitat fragmentation by some 45 years, and be- cause of this long history, a summary of the lit- erature on edge effects parallels the development of avian ecology in general. In fact, edge effects can be viewed as the earliest attempt to study avian ecology at the landscape scale, a perspec- tive that received less attention as the focus of field ecology shifted to population dynamics and community ecology in the 1950s through the 1970s. The conservation imperative that emerged in the seventies, driven by the recog- nition of rapid habitat loss and fragmentation, returned consideration of edge effects to the forefront of avian research, but in a very differ- ent context. Our overview of edge effects traces the de- velopment of conceptual approaches through field studies, experiments, and modeling ap- 30 EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 31 NUMBER OF PATCHES % DEFORMATION FIGURE h Edge habitat proliferates with increasing fragmentation, due both to the increased edge per unit area as the number of patches increases (top), and as individual patches become, on average, more linear or irregularly shaped, as represented here as an increas- ingly flattened patch (bottom). From Sisk and Mar- gules (1993). proaches. The paper focuses on patterns in the literature, particularly the disparity in the level of research in the eastern and western United States and the emphasis upon certain habitat types. We list working hypotheses derived from the literature, and we provide brief summaries of supporting and refuting evidence. Finally, we examine more predictive approaches to the study of edge effects so that the accumulated knowl- edge might be put to work in efforts to predict the impacts of ongoing fragmentation. Our ulti- mate goal is to incorporate a consideration of edge effects into efforts to reverse the negative impacts of fragmentation and improve reserve designs, restoration efforts, and management plans for the conservation of avian biodiversity. EDGE EFFECTSAN ILL-DEFINED "LAW" OF ECOLOGY "Edge effect" is among the oldest surviving concepts (some would say "buzz-words") in avian ecology. In 1933, Leopold referred to "the edge effect" to explain why quail, grouse, and other game species were more abundant in patchy agricultural landscapes than in larger fields and forested areas (Fig. 2). He hypothe- sized that the "desirability of simultaneous ac- cess to more than one (habitat)" and "the great- er richness of (edge) vegetation" supported higher abundances of many species and higher species richness in general (Leopold 1933). This common-sense definition drew on years of ex- perience as a forester and game manager, and reflects the focus of early wildlife managers on game species, many of which utilize early suc- cessional and/or edge habitats preferentially. Lay (1938) provided some of the earliest empir- ical evidence supporting both increased abun- dance and greater species richness at woodland edges. His interpretation of these patterns also began a long tradition of deriving management guidelines from studies of bird abundances and species diversity at edges. His claim that the "maximum development of an area for wildlife requires ... small but numerous clearings" was accepted by many wildlife managers and found its way into many textbooks over a period of several decades, culminating in what has been called the "law of edge effect" (Odum 1958, Harris 1988). General acceptance of the hypoth- esis that diversity and abundance are higher near edges led wildlife biologists to advocate the cre- ation of edge under the assumption that it would benefit biodiversity (e.g., Giles 1978, Yoakum 1980, Dasmann 1981). This understanding of the beneficial nature of edge effects influenced land management practices for decades and served as a de facto prescription for habitat fragmentation in the name of wildliI management. Even to- day, land managers frequently advocate the cre- ation of edges via (for example) forest clearing and prescribed fire, with the intention of increas- ing avian abundance and diversity. More recently, the relationship between forest fragmentation and both nest predation and par- asitism has spawned a different view of edge effects. Edges have been shown to support high- 32 STUDIES IN AVIAN BIOLOGY NO. 25 INTERSPERSION OF TYPES - RELATION TO MOBILITY & DENSITY OF QUAIL A.' Poor Inferspersion (I Cove,/) ... CULTIVATION ,,VEY :,:.....I,. ,.- ...........   ;*,   t tt,.:s 3? :.. :?::':  FIGURE 2. Leopold (1933) coined the term "edge effect" to explain increased abundance of game birds in heterogeneous landscapes with many edges. In this figure, 160 ac (64.7 ha) blocks of 4 habitat types, each 40 ac (16.2 ha), are displayed in the two panels. Panel (a) has 2 mi (3.2 km) of edge, while panel (b) has 10 mi (16 km). Leopold argued that greater bird abundances are associated with the heterogeneous landscapes, such as (b). er rates of nest predation and parasitism (Wil- cove 1985, Paton 1994, Andrdn 1995). Current texts are likely to present evidence that edge ef- fects are "bad" and that the creation of edge habitat by fragmentation leads to the decline of "interior species" that are particularly suscep- tible to nest parasites and predators (e.g., Meffe and Carroll 1997). Again, the focus on certain aspects of edge effects (in this case nest preda- tion and parasitism rates) has led to a widely accepted, general rule of edge effects. However, in this case, the supposedly beneficial effects are often ignored, while the adverse effects, dem- onstrated for a subset of species in particular habitats and in certain geographic areas, are highlighted. Thus, perceptions of the relationship between edge effects and habitat fragmentation are often contradictory, and the reality is almost always more complex than perceptions. In some cases, edges are thought to benefit birds; in others they are seen as the primary threat to bird diversity. And in cases where edges support high bird den- sity but low nest productivity, edge effects on population persistence may be particularly neg- ative (Ratti and Reese 1988). Nevertheless, the term continues to be applied with little discrim- ination, and the assumption that all influences of habitat edges can and should be grouped into a uniform class of ecological impacts persists in the literature. The complexity and diversity of the responses of different species to differing edge types, combined with the lack of an inclu- sive theoretical framework for organizing the plethora of field observations reported in the lit- erature, has turned "edge effects" into a grab- bag term, one that too often is used casually to explain anomalous or inconclusive results. In- deed, the term edge effect has become so widely accepted in the management literature that it is commonly used to explain diametrically op- posed observations. Part of the confusion may result from changes in the scale at which species diversity is as- sessed. Historically, biologists and planners have focused on alpha (local) diversity, which is often high near habitat edges. As conservation plan- ning has shifted to larger areas, and scientists have assessed regional and global patterns in biodiversity, the focus on species diversity has shifted to the gamma (regional) level, which may be lower in fragmented landscapes due to the loss of edge-avoiding species. Until scien- tists and managers are able to adopt a multi- scaled approach to assessing biodiversity (see Noss 1990), confusion over edge effects is likely to persist. HISTORICAL PERSPECTIVES: RESPONSE VARIABLES, FOCAL SPECIES, AND GEOGRAPHIC PATTERNS METHODS We reviewed the literature on edge effects dating back to the mid-1930s in an attempt to synthesize the large and diverse body of published work in arian ecology and wildlife management. Drawing from on- line searches, published abstracts, examination of lit- erature cited in all papers reviewed, and inquiries with colleagues, we created an annotated bibliography to facilitate analysis of patterns from published studies of edge effects. We limited our review to the peer-re- viewed literature after initial attempts to include un- published reports and other "gray literature" demon- EDGE EFFECTS AND AVIAN ECOLOGYSisk and Battin 33 TABLE 1. ANALYSIS OF THE EDGE EFFECTS LITERA- TURE BASED ON PARAMETERS LISTED BELOW, RECORDED FOLLOWING REVIEW OF 215 PAPERS PUBLISHED OVER A 66-YR PEPrOD Study Type-observational, experimental, theoretical, or modeling Location-country, state/province Focal habitat type Adjacent habitat Edge definition (e.g., is the edge treated as a gradient or separate habitat type) Focal species Study design Replication Response variable(s) Explanatory variable(s) measured Results and Conclusions strated a tremendous volume of work of highly vari- able quality. Inclusion of gray literature would have substantially increased our sample size, particularly in the West, but that literature could not be accessed in any consistent manner, and a haphazard sampling of material would have compromised our analyses. In this article we attempt to present an unbiased review of the peer-reviewed literature, and we invite the reader to critically explore the voluminous gray literature for ad- ditional site- and species-specific information on edge effects. A total of 215 publications were examined for this chapter. Of these, we eliminated from further consid- eration any field studies that did not explicitly address avian response to edges (for example, studies that em- ploy edge as one of many possible explanatory vari- ables in multivariate analyses of fragmentation effects; see citations in other chapters in this volume). This left us with 125 studies, providing a comprehensive per- spective on the development of the edge effects con- cept in the primary literature, current understanding of edge effects in the context of habitat fragmentation, and the application of this knowledge in the manage- ment of avian populations. Of the 125 publications re- viewed, 90 presented original research results involv- ing avian subjects (Appendix), and these are included in the analyses presented below. For this subset of the edge literature, we quantified aspects of each study pertaining to the location, focal habitats, species stud- ied, key results, and several related parameters (Table 1). Conceptual and theoretical treatments of edge ef- fects are discussed in subsequent sections of this chap- ter. Unlike the nest predation literature (see recent re- views by Paton 1994, Andrrn 1995, HartIcy and Hunt- er 1998), the literature on patterns of bird density and diversity with respect to habitat edges has not under- gone a recent review. For this reason, we analyze this body of literature in detail. We report the density and species richness response(s) for every treatment con- sidered in each study (Appendix). For multi-year stud- ies, we consider a treatment to show a response if a statistically significant response (increased or de- creased density or species richness at edges) was ob- served in at least one year, and a non-significant trend in the same direction was observed in other years. GEOGRAPHIC PATTERNS AND RESPONSE VARIABLES The majority of published studies of edge ef- fects in avian ecology (88%, N -- 60) are from the eastern half of North America (Figs. 3, 4a). Furthermore, the West has produced less than half as much research on this topic than has FIGURE 3 Map of North America showing number of studies addressing edge effects in landbirds. 34 a) 5 18 53 STUDIES IN AVIAN BIOLOGY c) s 7 B Eastern NA 29 nWestern NA 8/' B Scandinavia 4 '...._._--  Tropics 13 Other 4 9 15 9 ß Agriculture nClearcut ß Road/Trail ß Powedine ß Urban ß Other Induced [3 Natural-Water ß Natural--Other ß Undifferentiated NO. 25 5 b) 6 3 4  d) 5 9b5 [] Forest  Predation: Artificial [] Agriculture 37 9 9 [] Predation: Natural [3 Native Open Habitat Parasitisrn ß Clearcut [3 Wetland ß Density  Species Richness [] Powerline Corridor 40 [] Urban 10 ß Nest Density 7 FIGURE 4. The number of edge studies (a) by region, N = 90; (b) by habitat type, N = 90; (c) by adjacent (matrix) habitat, forest edges only, N = 75; (d) by response variable, N = 112 (some studies involved more than one edge type). Scandinavia, where conditions are, arguably, more similar to eastern North America (Fig. 4a). Clearly, as measured by the number of peer-re- viewed publications, studies in Europe and east- ern North America have had a tremendous influ- ence on our understanding of edge effects. Not surprisingly, since forests are the domi- nant natural habitats in these regions, 73% of all empirical studies focused on forest edges (Fig. 4b), and 33% of these were edges with agricul- tural habitats (Fig. 4c). Again, there is a geo- graphic bias, as conversion of forested habitats to agriculture (and the reverse) has been a pre- dominant land-use trend in the East and Mid- west, whereas edges in western habitats are most often due to timber harvest and a range of fac- tors that degrade, but less often radically trans- form, native habitats. When this distribution of research effort is viewed in the context of the overall habitat diversity of North America, and when the range of natural and anthropogenic factors that modify habitats and create edges is considered, it is apparent that our understanding of edge effects is largely the product of research focused on a small subset of edge types in east- ern, midwestern, and northern European forest edges. Examination of the response variables mea- sured in empirical edge studies reveals a strong tendency to focus on patterns in species abun- dance (44% of all studies) and species richness (17%; Fig. 4d). This work highlights patterns in avian distribution near edges but typically does not examine the factors creating the patterns. Fifty-two per cent of all studies quantified rates of nest predation, but of these only 21% looked at natural nests. The remainder manipulated the placement of artificial nests to estimate relative rates in the wild. Nest parasitism, a topic men- tioned at least parenthetically in most recent publications on edge effects, was quantified in only 7 of the papers that we reviewed (8%; Fig. 4d). Many other potentially important variables, including competitive interactions, pairing suc- cess, movement and dispersal rates, and edge permeability have received scant attention in empirical studies of avian edge effects. EDGES AND NEST PREDATION Three recent reviews that have examined the relationship between forest edges and predation have found that, while evidence exists for higher predation rates at edges, this pattern is far from universal (Paton 1994, Andrn 1995, Hartley and Hunter 1998). These reviews addressed not only the question of how frequently predation edge effects occur, but also looked for explana- tions regarding why some studies found edge ef- fects and others did not. Landscape context was the primary explanatory variable used by all au- EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 35 thors, but they drew markedly different conclu- sions about its importance. Paton (1994) examined edge effects in nest predation on artificial nests and in both preda- tion and parasitism on natural nests. He found that 10 of 14 studies using artificial nests showed evidence of differential nest predation at edges, compared with 4 of 7 studies of natural nests. Of the 14 studies showing differences, most showed higher predation at edges. Just un- der half of the 32 studies examined by Andrdn (1995) showed higher predation rates near edg- es, while only 5 of the 13 North American stud- ies examined by Hartley and Hunter (1988) found a difference in predation rates between habitat edges and interiors. These reviews indi- cate that high nest predation rates occur near edges, but not consistently. Some studies re- viewed by Andrdn (1995) and Paton (1994) even found lower predation near edges. In seeking to explain this variable pattern of edge effects, the three reviews draw strikingly different conclusions, though they consider many of the same papers. Paton (1994) conclud- ed that "significant edge effects were as likely to occur in forested as in unforested habitats." Andrdn (1995) concluded that predation near edges was more likely in agricultural than in for- ested landscapes. Hartley and Hunter (1998), who conducted a substantially more rigorous meta-analysis of the association between forest cover and edge effects, found a marginally sig- nificant (P = 0.095) pattern of higher predation in unforested than in forested landscapes. Un- fortunately, the power of their analysis was lim- ited, as they considered only two studies from unforested landscapes. One possible explanation for the inconsisten- cies in the findings of these different studies is that Andrdn (1995) considered both edge effects and patch size effects in a single analysis, while Paton (1994) and Hartley and Hunter (1998) an- alyzed edge effects and patch size effects sepa- rately. In contrast to their equivocal findings on the relationship between landscape context and the presence of edge effects, both Paton (1994) and Hartley and Hunter (1998) found a very strong relationship between nest predation rate and patch size. This result suggests that Andrdn (1995) may have confounded effects by lumping patch size and edge effects in his analysis, and that the strong pattern that he detected could be due to patch size rather than edge effects per se. Another difficulty in interpreting these results is that most of the studies of edge effects on nest predation have been conducted using artificial nests. Hartley and Hunter (1998) used only ar- tificial nest studies in their analysis, while An- drdn combined artificial and natural nests. Paton considered artificial and natural nest studies sep- arately, but he found only 7 natural nest studies. The use of artificial nests has been questioned repeatedly in recent years (see Willebrand and Marcstrom 1988; Haskell 1995a,b; Major and Kendal 1996, Yahnet 1996), and Haskell (1995a,b) suggested that there is a systematic bias toward increased predation on artificial nests in smaller fragments, a finding that could be especially misleading in studies of predation near edges. While evidence of increased predation rates near edges does exist, it is not clear that this is a widespread phenomenon, or that it is pro- nounced in the West. We found only two studies of nest predation in the West, one that used ar- tificial nests (Ratti and Reese 1988) and one that used natural nests (Tewksbury et al. 1998). Nei- ther study found a significant edge effect in nest predation. PATTERNS IN COMMUNITY ORGANIZATION For several decades, "edge effects" referred almost exclusively to the increase in species di- versity and/or density commonly observed near the edge (Johnston 1947, MacArthur et al. 1962, Giles 1978). A total of 21 studies, with 34 sep- arate treatments, examined density or species richness of the entire bird community (Appen- dix). Of these, 21 treatments reported higher bird densities near edges, while 10 reported no edge response and 3 showed a decrease. The vast majority of these studies (19 studies, ad- dressing 27 treatments) were conducted in for- ested habitats, so we restrict our more detailed analyses to these results. Overall, forest studies showed a strong pattern of higher density at edges but a weaker pattern with regard to species richness. Sixteen treat- ments recorded higher bird abundance near edg- es, with 8 showing no significant response and 3 a negative response. Nine treatments found higher species richness at edges, while 10 found no difference, and 2 found a decrease. While an unequivocal pattern of higher bird density and species richness at edges does not emerge from this analysis, it seems clear that, in the recent literature, negative responses to edges are rela- tively rare and positive responses are common. This could be a manifestation of a general eco- logical principle (i.e., density and species rich- ness increase at most edges) or the result of a bias in the literature (edge responses in areas where studies have been done are different from those in unstudied areas). Because, as we have shown, there is a strong geographical bias in the literature, this second explanation cannot be ruled out. All studies (9 studies, 9 treatments) conducted 36 STUDIES IN AVIAN BIOLOGY NO. 25 in temperate zone forests that examined total bird abundance at edges between native forests and large anthropogenic openings (matrix = ag- riculture, clearcut, clearing, anthropogenic grassland; see Appendix) found higher bird den- sities near the edge. Of the 7 studies that also looked at species richness, 3 found an increase while 4 found no significant pattern. On the oth- er hand, the only study that looked at the dif- ference in overall bird density and species rich- ness along an anthropogenic edge gradient in the tropics found that both decreased near the edge (Lovejoy et al. 1986). Another tropical study, which analyzed edge response by foraging guild, found that two guilds did not differ in abundance and one (insectivores) decreased at the edge (Canaday 1997). These results suggest that even the strongest patterns detected in temperate for- ests may not generalize well to other habitats and geographic regions. The effects of linear drivers of habitat frag- mentation (roads and powerlines) and natural edges appear to be less consistent. While no studies of road or powerline edges found com- munity-level decreases in avian density, 4 of 7 treatments showed increases and 3 of 7 showed increased species richness. Of the studies that examined natural edges (6 studies, 8 treatments), 3 treatments showed increased density, 4 showed no change, and 1 showed a decrease. Four treatments showed increased species rich- ness at natural edges, with 2 showing no change, and one showing a decrease. Aside from the suggestion that edge responses may differ between the tropics and the temperate zone, no clear geographical patterns of edge re- sponse were evident. No studies from eastern North America recorded decreases in total bird abundance (Fig. 5a) or species richness (Fig. fib) at edges, but almost as many treatments showed no response in overall bird density (6) as showed an increase (9). As many treatments showed no response in species richness (7) as showed a positive response near edges (7). The only study from western North America had one treatment that showed increased density and species rich- ness at the forest edge and one that showed no change in either variable (Sisk 1992). Two Scan- dinavian studies showed decreases in density at edges, while I reported no change and 2 found increases. We were surprised at the small num- ber of studies that reported on the entire avian community, especially considering the widely held "rule of thumb" associating edges with higher densities and/or species richness. Many of the studies most commonly cited to support this idea examine only part of the bird commu- nity present at the study site. Many explanations for the reported trends in FIGURE 5. Numbers of treatments from studies con- ducted in eastern and western North America finding positive, negative, or neutral edge responses in total bird density (a) md species richness (b). avian abundance and diversity near edges have been proposed, and few are mutually exclusive. Few studies have attempted to distinguish among them, and many authors have invoked "edge effects" when discussing any of the myr- iad influences of habitat fragmentation on dis- turbance-sensitive species. From this broad range of uses, four general categories of edge effects can be identified: ß Habitat interspersion. Species diversity may increase at habitat edges due solely to the proximity of diflrent habitats (Leopold 1933, Giles 1978). At the habitat edge, each com- munity contributes, on average, more than half of its fauna, resulting in higher species diversity at the edge where the two commu- nities mix (MacArthur and MacArthur 1961, Wiens 1989). ß Resource availability. Many authors have suggested that birds may utilize more than one habitat type during different activities (e.g., nesting and foraging) or during different life stages. Allocating different activities to the most appropriate habitat may allow some species to maintain higher population densi- ties near edges. It also may provide suitable habitat for species that require more than one habitat type (Kendeigh 1944, MacArthur et al. 1962, Yoakum 1980, Dasmann 1981). ß Edge as a unique habitat. Edges may support higher densities of species characteristic of both the adjoining communities, due to in- EDGE EFFECTS AND AVIAN ECOLOGY--SiNk and Battin 37 creased diversity of the vegetation that typi- cally occurs where two habitats intergrade. Many workers have shown correlations be- tween foliage height diversity and bird species diversity (e.g., MacArthur 1958, Cody 1968, Karr and Roth 1971; but see also Willson 1974). Other studies have shown that floriNtic composition and the presence or absence of particular plant species are good predictors of both diversity and density of birds (Wiens 1989). Vegetation structure and floriNtic com- position are generally more diverse at edges, so increases in both species diversity and avi- an density might be expected, even without the addition of edge-dependent species. ß Interspecific interactions and cascading biotic efkcts. Edges, especially those associated with habitat conversion and fragmentation, may permit edge-dependent or habitat-specific species to penetrate some distance into adja- cent habitats where they normally do not oc- cur. Their presence can influence the abun- dance of species in the adjacent habitat, gen- erating cascading effects that penetrate further than the direct environmental changes asso- ciated with the edge (Diamond 1978, 1979; Pulliam and Danielson 1991, Fagan et al. 1999). Such secondary effects, including competition, predation, and nest parasitism, are thought to result in the exclusion of forest species from otherwise suitable habitat near habitat edges (Ambuel and Temple 1983, Wil- cove et al. 1986, Harris 1988). SPECIES-LEVEL RESPONSES UNDERLYING COMMUNITY PATTERNS Each of the definitions of edge effects pre- sented above implies that population densities of some species will change as a function of the distance from the habitat edge. However, few authors have stated explicitly which species they expect to be influenced by habitat edges or how they will respond. In fact, many early studies that support the hypothesis of elevated diversity at edges do not report which species contribute to the diverse assemblages found there. Those that do often show that the increase in species richness is due to the addition of common, cos- mopolitan, or disturbance-tolerant species, which may mask the loss or decline of sensitive species. A better understanding of the dynamics in community organization near edges emerges from studies of the responses of individual spe- cies near habitat edges (Giles 1978, DaNmann 1981, Harris 1988, Reese and Ratti 1988, NoNs 1991, Bolger this volume). Many studies have shown that certain species reach their highest or lowest abundance at particular habitat edges (e.g., Kendeigh 1944, Johnston 1947, Hansson 1983, Kroodsma 1984b, NoNs 1991, Bolger et al. 1997, Germaine et al. 1997, King et al. 1997). Species that are encountered more com- monly near the edge are often termed "edge spe- cies" (e.g., Johnson 1975, Giles 1978, Reese and Ratti 1988), and those whose densities are low near the edge are considered to be habitat- interior species (e.g., Brittingham and Temple 1983, Wilcove et al. 1986, Thompson 1993, Bol- ger et al. 1997). A more quantitative approach to understanding how species respond to habitat edges involves measurement of a species-specif- ic edge response, defined as the pattern of change in population density at incremental dis- tances from the habitat edge (NONs 1991, SiNk and Margules 1993). SiNk and Margules (1993) proposed a classi- fication scheme for population-level edge re- sponses based on changes in density along a transect from one interior habitat, across the edge, and into the adjacent habitat (hereafter the edge gradient). For some species, the edge itself has no effect on population density (null re- sponses), and changes in density are attributable to differences between the two adjoining habi- tats. Other species reach their highest density ("edge exploiters'*) or lowest density ("edge avoiders") near edges (see also Bolger this vol- ume). While classification schemes differ among the published studies reviewed here, it is clear that a diversity of responses is manifest in any particular avian community. Four studies from eastern North America show that edge-exploit- ing responses are generally more common than edge-avoiding responses, with neutral responses (i.e., no edge effect) more common than either in 3 out of 4 studies (Fig. 6a). The small number of Western studies showed similar patterns, ex- cept that edge-exploiting responses outnumbered edge-neutral responses (Fig. 6b). Villard (1998) compared the edge responses of forest-interior neotropical migrants reported in 4 studies from the eastern seaboard stretching from Florida to New Hampshire. He found that there was little consistency in the way that the authors classified responses for the same spe- cies. We extended this analysis to all species that occurred in two or more of the studies (Table 2). While there is considerable variability in the re- sponses reported for these species, some patterns do emerge. Most neotropical migrants are edge avoiders, and all disagreements among authors have to do with whether a species shows a neu- tral response or a positive or negative response; no species is considered an edge-exploiter by one author and an edge-avoider by another. Con- versely, species that are not latitudinal migrants showed neutral or edge-exploiting responses. 38 STUDIES IN AVIAN BIOLOGY NO. 25 FIGURE 6. Numbers of bird species in four studies showing positive, negative, or neutral responses to habitat edges. Eastern studies (a) were conducted in Vermont (Germaine et al. 1997), New Hampshire (King et al., 1997), Florida (Noss, 1991), and Tennes- see (Kroodsma, 1982). Western studies (b) are from California redwood stands (Brand and George this vol- ume) and California oak woodlands (Sisk 1992, Sisk et al. 1997). Again, no species was assigned a positive re- sponse by one author and a negative response by another (Table 2). Unfortunately, there are not enough studies of western birds to make similar comparisons, and there is little overlap in species among the few published studies. Three studies from California do, however, seem to show greater variation in the responses of both neotropical migrants and resident species (Sisk et ai. 1997, Brand and George this volume, Bolger this volume). Ecologists and wildlife managers have often assumed that birds will show consistent, char- acteristic patterns of habitat selection at edges, even when the adjoining habitats differ in veg- etation structure and/or species composition. Im- plicit in this assumption is the idea that edges of all types share some intrinsic qualities, and that their influence on the distribution of organisms and the composition of assemblages is similar. There is little evidence to support these views. Few studies have measured edge responses at more than one type of edge in a given region, and those that have report differences in the con- sistency of arian responses at different edge types. Noss (1991) found considerable variation among species and among sites in longleaf pine (Pinus palustris) bird communities. Sisk et al. (1997) showed that over half of the breeding birds in oak woodland showed different respons- es at edges with grassland versus edges with chaparral, and Kristan et al. (in press) found sig- nificant site-to-site variation in edge response in several southern California coastal sage scrub bird species. Brand and George (this volume) found general consistency at redwood forest edges adjoining habitats as different as logged forest and grassland. In summary, our examination of empirical studies of edge effects did not identify a simple pattern in avian responses, but it did uncover several important points regarding patterns in community organization and population re- sponses to habitat edges: ß "Edge effects" is an ambiguous term in arian ecology and conservation. Its usefulness is limited by widely varying assumptions that permeate its history. ß Edge effects do not contribute to species di- versity in a consistent manner that is easily generalized among sites. ß The abundances of many species change dra- matically near habitat edges. ß Edge responses vary markedly among spe- cies. ß A given species often responds very differ- ently at different types of edges (but a few studies show consistency). MECHANISMS UNDERLYING SPECIES-LEVEL RESPONSES Mechanisms underlying edge effects are many, but few have been adequately investigat- ed (Bolger this volume). Sisk and Haddad (2002) hypothesize that several basic driving factors may underlie the broad range of responses typ- ically grouped together under the term ':edge ef- fects". These include: ß Edges influence movement. Edges may influ- ence behavior, creating barriers to movement even when animals are clearly capable of crossing them (Ries 1998, Haddad 1999). The influence of edges may prevent dispersal through complex landscapes and isolate ani- mals. Sisk and Zook (1996) have shown that "passive accumulation" of migrating birds may generate widely reported increases in density observed near forest edges. © Edges influence mortality. Particularly for habitat interior species, edges may lead to higher mortality in plants and animals. Higher mortality may occur in three different ways. First, edges create greater opportunity for loss EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 39 TABLE 2. VARIATION IN SPECIES-SPECIFIC EDGE RESPONSES REPORTED IN DIFFERENT EMPIRICAL STUDIES FROM THE EASTERN USA New Tennessee Hampshire Vermont (Kroodsma Florida (King et al. (Germaine Common name Scientific name 1984) (Noss 1991) I997) et al. 1997) Neotropical Migrants Yellow-billed Cuckoo Coccyzus americanus 0 + Acadian Flycatcher Empidonax virescens - - Wood Thrash Hylocichla mustelina 0 Hermit Thrash Catharms guttatus 0 - - Red-eyed Vireo Vireo olivaceus 0 - - Black-and-white Warbler Mniotilta varia + 0 0 Black-throated Blue Warbler Dendroica caerulescens 0 + Black-throated Green Warbler Dendroica virens 0 + Hooded Warbler Wilsonia citrina 0 - Ovenbird Seiurus aurocapillus - 0 0 - American Redstart Setophaga ruticilla 0 0 Summer Tanager Piranga rubra + + Scarlet Tanager Piranga olivacea 0 0 0 Temperate Migrants American Robin Turdus migratorius 0 0 + Residents Red-bellied Woodpecker Melanerpes catolinus + 0 Downy Woodpecker Picoides pubescens 0 0 Carolina Chickadee Parus ctirolinensis 0 + Northern Cardinal Cardinalis cardinalis + + Note: Results from four studies allowed the classification of 12 species according to their density responses near edges: '+' for edge-exploiting response; '0' lbr no edge response; '-' for edge-avoiding response; ' ' if not reported (after Viilard 1998). of dispersers into unsuitable habitat. For ex- ample, plants with wind-dispersed seeds that are near the edge will lose more of their prop- agules into unsuitable habitat. Second, edges alter microclimate, including temperature, light, and moisture (Sisk 1992, Chen et al. 1993, Young and Mitchell 1994, Camargo and Kapos 1995). In doing so, edges impact com- petitive interactions between species. Third, edges provide points of entry for predators and parasites, such as the Brown-headed Cowbird (Molothrus ater; Wilcove et al. 1986, Murcia 1995). ß Edges provide feeding or reproductive subsi- dies. From the edge, species may be able to obtain a greater quantity and quality of food resources from each of the habitats that create the edge, leading to positive effects on pop- ulation sizes (MacArthur et al. 1962, Fagan et al. 1999). ß Edges define the boundary between two sep- arate habitats, creating new opportunities for species to mix and interact. By their very na- ture, edges influence species interactions be- cause they bring into close proximity species that would not normally be present in the same habitat. Species that are brought togeth- er at the edge, including predators and prey, new competitors, and mutualists, generate novel interactions and create new communi- ties of species. Despite the diversity of hypothesized and doc- umented mechanisms underlying edge effects, surprisingly few studies have attempted to iden- tify the mechanistic basis for edge response and patterns in community organization reported in the literature. Of the 90 field studies considered in this review, most were observational, typical- ly involving some count of individuals or nests in unmanipulated landscapes. The vast majority of experimental studies involved manipulation of artificial nests for the purposes of examining nest predation and parasitism rates; few involved the experimental manipulation of bird habitats (but see Lovejoy et al. 1986). Forty studies focused on estimates of abun- dance or species richness, but few examined the mechanisms driving the observed patterns. Don- ovan et al. (1997) noted that little work has been devoted to exploring the mechanisms underlying observed patterns of edge effects in nest preda- tion and parasitism. This is even more pro- nounced for studies examining patterns in bird density and species richness. Clearly, the eluci- dation of mechanisms driving edge effects has lagged far behind pattern identification. In- creased attention to the mechanistic drivers un- 40 STUDIES IN AVIAN BIOLOGY NO. 25 derlying edge effects and their relative contri- bution to observed patterns of distribution and abundance is a fruitful area for future research. PREDICTIVE APPROACHES TO MODELING EDGE EFFECTS Despite recent advances in understanding the general consequences of fragmentation, the de- velopment of tools for predicting specific im- pacts has progressed slowly. A growing body of research is demonstrating that edges are often highly influential in determining habitat suit- ability and population persistence in fragmented landscapes (Robinson et al. 1995a, Donovan et al. 1997, Howell et al. 2000). Like the work fo- cusing explicitly on edges, this landscape-scale research is showing that the importance of hab- itat edges varies from species to species and from landscape to landscape. Thus, it is increas- ingly clear that informed habitat management will necessitate the incorporation of our increas- ing understanding of the role of habitat edges in fragmented landscapes into predictive models that will allow assessment of alternative man- agement options in novel landscapes. Most mod- eling efforts addressing birds in fragmented hab- itats have focused on the loss of habitat area and the isolation of remnant patches, typically fo- cusing on a single species (e.g., Thomas 1990, Noon and Sauer 1992, Pulliam et al. 1992). However, models that focus on habitat patches in isolation from matrix and edge effects olen prove to be disappointing in management situ- ations (see Saunders et al. 1991). An integrated approach for assessing edge responses and pre- dicting the impacts of increasing edge habitat is needed before the influence of habitat edges can be incorporated into assessments of the effects of habitat fragmentation. Effective management of habitat edges re- quires knowledge of population-level responses and a conceptual framework for linking this un- derstanding to spatially explicit information about the landscape. Area-based approaches that treat the edge as an area influenced by adjacent habitats, rather than as a separate habitat type, show some promise for guiding management de- cisions. In addition, predictive models oflr a powerful means for advancing our understand- ing of the mechanisms that drive observed pat- terns. The generation of explicit predictions based on empirical measures of species-specific edge responses, followed by field tests and mod- el revision, offer the possibility of more rapid progress in understanding edge effects. Temple (1986) presented a simple, straight- forward approach for including edge effects into a patch-based model of arian abundance. He as- sumed that the effects of nest predators and par- a. Total area 47 ha, core area 20 ha b. Total area 39 ha, core area 0 ha FIGURE 7. Temple's (1986) original core area model of edge effects used sensitivity to edge as a predictor of habitat use by forest-interior birds. The model as- sumed that edge effects, in general, penetrate 100 m into a forested patch, dramatically infuencing the "core area" of suitable habitat within a forest patch (contrast panels a, b). The approach motivated a series of efforts that placed edge effects in landscape context and considered edge effects in predictions of the im- pacts of habitat fragmentation. asites penetrate about 100 m into remnants of midwestern forest and woodland patches, and that the abundances of species that are "sensi- tive to fragmentation" would be low or zero within 100 m of the edge patch. He found that linear regressions of species' abundances against the "core area" of the patch--the area greater than 100 m from the edge--were significantly stronger than regressions against total patch area. This idea provided a conceptual foundation for incorporating the effects of edges and patch shape into patch-based approaches to estimating habitat suitability (Fig. 7). Subsequent work re- laxed some of the assumptions of the core area model, allowing the distance of edge penetration to vary among species (Temple and Cary 1988) and to vary monotonically with distance from the edge (Laurance and Yensen 1991), adding realism to the approach. Extension of the core area approach to ad- dress all species--those with edge-exploiting as well as edge-avoiding responses--and multiple habitat and edge types, led to the effective area model (EAM; Sisk and Margules 1993, Sisk et al. 1997, Sisk and Haddad 2002). EAM ap- proaches predict species abundances (or other variable of interest) in any number, size, or EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 41 Patch-specific population estimate: N = E(A, d,) 150m 50m EDGE 50m 150m Chaparral Oak Wood/and FIGURE 8. Schematic of the effective area model (EAM). Sisk et al. (1997) extended the core area ap- proach to multiple habitat and edge types, using digital habitat maps to describe landscape pattern. The EAM incorporates variation in edge responses among species and at different edge types to estimate the abundances of the breeding bird community in any number of patches of any shape. shape of habitat patches by projecting density estimates from species-specific edge response curves onto digitized maps of all the habitat patches within the focal landscape. The predict- ed density of each species within each patch varies with distance from the edge. In the dis- crete approach illustrated in Figure 8, the patch is divided into sub-regions. These sub-regions correspond to the distance intervals used for field surveys of species abundances, which are used to define species-specific edge responses, illustrated here by the bar graph for Spotted To- whee (Pipilo maculatus). Multiplying the area of each sub-region by the corresponding estimate of population density, and then summing the products for all sub-regions, gives a predicted population size for the species in a particular patch (Fig. 8). The degree to which the predicted density differs from predictions that assume equal abundance throughout the patch reflect the importance of "edge effects." Sisk et al. (1997) reported that the EAM performed significantly better than a null model that ignored edge effects and estimated bird abundances based on patch area alone. Other applications of the EAM are presented in Sisk and Haddad 2002. Several practical considerations influence how the core area and effective area models are ap- plied. First, the spatial resolution of the edge re- sponse measured (i.e., the magnitude of the re- sponse at various distances from the edge) de- termines the spatial resolution of the edge ef- fects modeled. Therefore, the sampling design and survey techniques for measuring the edge response should be scaled to the life history characteristics (e.g., territory size, vagility) of the animals being studied. Logistic and meth- odological limitations often constrain sampling designs somewhat, but the variety of proven methods for sampling avian populations pro- vides flexibility in quantifying edge responses and facilitates the application of these patch- based models to birds operating at different spa- tial scales. In complex, heterogeneous land- scapes, detailed habitat maps reflecting species- specific requirements are needed. Advances in mapping technologies and the application of re- motely sensed data to habitat mapping (e.g., Scott et al. 1993, Imhoff et al. 1997), offer promise for rapid and cost-efficient methods for mapping habitats across large regions. EDGE EFFECTS IN THE WEST: IMPLICATIONS FOR STUDIES OF HABITAT FRAGMENTATION After 60 years of attention and relatively little progress toward articulating general principles pertaining to edge effects, it might be tempting to conclude that the topic is intractable. Indeed, the early adoption of simplistic rules of thumb regarding habitat edges--for example, that more edge leads to higher diversity--may have led to poor habitat management and stalled progress in identifying the mechanisms underlying edge ef- fects. However, slow progress in the past is not a reason to ignore the compelling reasons for expanding mechanistic and management-rele- vant research in the future. Why study edge effects? First, anthropogenic disturbances are rapidly increasing the preva- lence of edges in most terrestrial landscapes. This process is sure to continue, and ignoring edge effects will become increasingly debilitat- ing to conservation efforts. Edge effects may compound the effects of habitat loss and the iso- lation of fragments on the distribution, abun- dance, and persistence of many sensitive bird species. Second, edges are amenable to manage- ment. The area of habitat protected and its lo- cation are often the result of societal decisions based on many factors that often lie outside the purview of conservation biologists. However, management of boundaries often is left to the discretion of the manager. Better understanding of the influences of edges on bird populations will lead to more effective strategies for man- aging habitat fragments. Third, edges are inher- ently dynamic environments and, therefore, they offer opportunities for studying avian responses to changing landscape pattern. What do we know? Not nearly enough, but 42 STUDIES IN AVIAN BIOLOGY NO. 25 the numerous studies from eastern North Amer- ica offer some important lessons tbr those pur- suing studies in western landscapes undergoing fragmentation. ß Our understanding of the many biological phenomena associated with habitat edges is dominated by the description of patterns from eastern forests. ß Western landscapes are, in general, more nat- urally heterogeneous than their eastern coun- terparts, and edges are common components in many landscapes (e.g., riparian corridors). ß The relationship between natural heterogene- ity and avian sensitivity to the increased prev- alence of edge due to habitat fragmentation is not well understood. ß Mechanistic explanations for avian responses near habitat edges are, in general, poorly de- veloped and inadequately tested. Work in the West should pursue mechanistic understand- ing and predictive capabilities of use to hab- itat managers. These lessons, derived from our review of an extensive literature on edge effects and aug- mented by landscape-scale studies of avian re- sponses to habitat fragmentation, argue that edge effects occur commonly in many habitats, that they are of increasing importance as habitats be- come more fragmented, and that we currently know too little about what causes them to pre- dict accurately where and to what degree they will influence bird populations. This knowledge should be sufficient to inspire a more focused, and hopefully more fruitful, effort to understand the many driving factors underlying edge effects and to incorporate this knowledge into strategies for avian conservation. ACKNOWLEDGMENTS We are indebted to L. Ries whose role in designing the edge review was fundamental to our efforts. We also thank P. Paton and J. Faaborg for insightful com- ments on an earlier version of this manuscript. Our work was supported by the Strategic Environmental Research and Development Program (project CS- 1  00). EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 43 I oo+++o+++++oo+++ - I o  + ¸ ¸ ¸ 44 STUDIES IN AVIAN BIOLOGY NO. 25 +++ +++ <<<< . c-i < ++1+ EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 45 + ++ +++ I 46 STUDIES IN AVIAN BIOLOGY NO. 25 I I1 +++1 EDGE EFFECTS AND AVIAN ECOLOGY--Sisk and Battin 47 48 STUDIES IN AVIAN BIOLOGY NO. 25 <<< Studies in Avian Biology No. 25:49-64, 2002. EFFECTS OF FIRE AND POST-FIRE SALVAGE LOGGING ON AVIAN COMMUNITIES IN CONIFER-DOMINATED FORESTS OF THE WESTERN UNITED STATES NATASHA B. KOTLIAR, SALLIE J. HEJL, RICHARD L. HUTTO, VICTORIA A. SAAB, CYNTHIA P. MELCHER, AND MARY m. MCFADZEN Abstract. Historically, fire was one of the most widespread natural disturbances in the western United States. More recently, however, significant anthropogenic activities, especially fire suppression and silvicultural practices, have altered fire regimes; as a result, landscapes and associated communities have changed as well. Herein, we review current knowledge of how fire and post-fire salvaging practices affect avian communities in conifer-dominated forests of the western United States. Specif- ically, we contrast avian communities in (1) burned vs. unburned forest, and (2) unsalvaged vs. salvage-logged burns. We also examine how variation in burn characteristics (e.g., severity, age, size) and salvage logging can alter avian communities in burns. Of the 41 avian species observed in three or more studies comparing early post-fire and adjacent unburned forests, 22% are consistently more abundant in burned forests, 34% are usually more abun- dant in unburned forests, and 44% are equally abundant in burned and unburned forests or have varied responses. In general, woodpeckers and aerial foragers are more abundant in burned forest, whereas most foliage-gleaning species are more abundant in unburned forests. Bird species that are frequently observed in stand-replacement burns are less common in understory burns; similarly, species com- monly observed in unburned forests often decrease in abundance with increasing burn severity. Gran- ivores and species common in open-canopy forests exhibit less consistency among studies. For all species, responses to fire may be influenced by a number of factors including burn severity, fire size and shape, proximity to unburned forests, pre- and post-fire cover types, and time since fire. In addition, post-fire management can alter species' responses to burns. Most cavity-nesting species do not use severely salvaged burns, whereas some cavity-nesters persist in partially salvaged burns. Early post- fire specialists, in particular, appear to prefer unsalvaged burns. We discuss several alternatives to severe salvage-logging that will help provide habitat for cavity nesters. We provide an overview of critical research questions and design considerations crucial for evalu- ating the effects of prescribed fire and other anthropogenic disturbances, such as forest fragmentation. Management of native avifaunas may be most successful if natural disturbance regimes, including fire, are permitted to occur when possible. Natural fires could be augmented with practices, such as pre- scribed fire (including high-severity fire), that mimic inherent disturbance regimes. Key Words: burn severity; cavity-nesters; fire effects; fire suppression; passerine birds; prescribed fire; salvage logging; silviculture; snags; wildland fire; woodpeckers. Understanding the consequences of anthropogen- ic activities that alter natural systems requires a thorough knowledge of the natural disturbance re- gimes that shape communities and landscapes. Often, the ecological consequences of anthropo- genic activities have been evaluated in the con- text of relatively undisturbed, mature forest (e.g., Whitcomb et al. 1977, Mladenoff et al. 1993, King et al. 1997, Morse and Robinson 1999). However, this approach may be inadequate for systems that evolved with major and persistent disturbances, such as fire. In the West, fire has played a dominant role in shaping communities and landscapes. Thus, one of the greatest threats to the ecological integrity of western forest sys- tems may be alteration of natural disturbance re- gimes and landscape structure through livestock grazing, fire suppression, logging in burned for- ests (hereafter "salvaging" or "salvage log- ging"), and other silvicultural activities. Concern that decades of fire suppression may lead to more frequent, larger wildfires has prompted government agencies to expand pre- scribed-burning programs and fire-management policies to diminish the chances of large, severe wildfires (U.S. Dept. of Interior and U.S. Dept. of Agriculture 1998). Unfortunately, our under- standing of historical fire regimes remains rudi- mentary and may be inadequate for setting such goals (Tiedemann et al. 2000). Furthermore, the new government-sanctioned program of prescrip- tion burning focuses on reducing fuel loads, with relatively little consideration given to the efikcts on wildlife (Tiedemann et al. 2000). In part, this problem stems from a paucity of rigorous field studies that have evaluated the eflkcts of fire on wildlife communities. Without a better under- standing of how historical fire regimes influenced communities (Bunnell 1995) and landscapes, as well as how anthropogenic activities have altered fire regimes, programs of prescription burning and other mitigation measures could be as mis- guided as widespread fire suppression. In the review and discussion that follow, we 49 50 STUDIES IN AVIAN BIOLOGY NO. 25 examine avian communities in post-fire forests in conifer-dominated systems of the West, and compare them to those in unburned forests. We focus in particular on the responses of wood- peckers and passefine birds. Because avian re- sponses to fire may vary with burn severity and size, time since fire, ecological contexts of burns, and post-fire salvage logging, these issues are also discussed. We preface our review by providing an overview of historical fire regimes of western forests and how human activities, particularly fire suppression, may have altered those regimes. This background is essential for understanding the patterns observed among avi- an communities using unburned and burned for- ests. We conclude with a discussion of compel- ling management implications that arise from this review, and we identify essential research questions for improving and enlarging our un- derstanding of how fire shapes and perpetuates avian communities in western forests. FIRE REGIMES IN CONIFEROUS FORESTS OF THE WESTERN UNITED STATES Although current knowledge of historical fire regimes in western forests remains somewhat ru- dimentary, it is possible to place those systems into broad fire-regime categories. The regime that characterizes any one system is an interplay between gradients in burn severity and fire fre- quency (i.e., fire-return interval). Generally, burn-severity gradients are divided into three levels, based on vegetation responses to fire: (1) low-severity fires kill or temporarily remove above-ground portions of herbaceous and un- derstory layers and sometimes scorch the lower portions of mature trees, typically without kill- ing them; (2) moderate-severity fires may kill but usually do not consume leaves of canopy trees, although some tree mortality may result; and (3) high-severity fires usually burn the can- opy, killing the majority of trees (Agee 1993). One level of burn-severity may dominate a giv- en burn, but most burns are mosaics of various fire sevefities (Agee 1993, Turner et al. 1994). Furthermore, there is variation among tree spe- cies' responses to fire intensity (e.g., heat). For example, the thick, fire-retardant bark of mature ponderosa pines (Pinus ponderosa) generally provides them protection from understory fires, whereas subalpine firs (Abies lasiocarpa) are of- ten killed by understory fires (Agee 1993). Un- derstory fires also typically kill the above- ground biomass of quaking aspen (Populus tre- rnuloides) stands, although lateral roots readily respond to fire by resprouting vigorously (Agee 1993). Thus, variations in burn severity can have profound effects on the composition and struc- ture of plant communities. For simplicity, most forest systems of the West can be characterized by one of three fire- regimes based on the effects of fire intensity on the dominant tree species: high frequency/low severity, moderate frequency and moderate to high severity, or low frequency/high severity (Agee 1993, 1998). High frequency/low severity fires (i.e., 1- to 40-yr fire-return intervals) are characteristic of many dry, warm forests. The combination of dry conditions and pervasive surface fuels (grasses and duff) allows fire to recur frequently. Many tree species in these sys- tems are adapted to fire (e.g., fire-retardant bark, seedling germination requires bare substrates). Generally, fires in these systems are restricted to herbaceous and understory layers, thereby elim- inating the majority of saplings and perpetuating a discontinuous forest canopy. Examples of such systems include ponderosa pine forests of foot- hills along the Rocky Mountains and Sierra Ne- vada (Arno 1980, Verner and Boss 1980, McKelvey et al. 1996). In forests characterized by intermediate mois- ture and temperatures, fire regimes are generally moderate in severity and frequency, although in many cases severity can be high (Agee 1993, 1998). The mix of burn severities often results in heterogeneous burns and multiple-age struc- tures of dominant trees (Agee 1993, 1998). Fire- return intervals tend to be longer (40-150+ yr) than those in drier sites, but can be quite variable (Agee 1993). Examples of this type of system include red fir (Abies magnifica) and coastal red- wood (Sequoia sernpervirens) in California (Agee 1993, 1998). Low frequency/high severity fire regimes typ- ically result in stand-replacement events. Because of the long fire intervals, trees in these systems often lack the ability to withstand fire (Agee 1993), although some species have reproductive adaptations to fire (e.g., serofinous cones of lod- gepole pine, Pinus contorta; Agee 1993). Typi- cally, climatic conditions (e.g., severe drought and strong winds) necessary for these systems to bum occur only several times per century, and fires spread only if sufficient fuels have accu- mulated (Romme 1982). Once started, fires in these systems often burn vast areas and may last for months (Agee 1993). Regeneration in larger bums can take decades if viable seed sources are distant (Agee 1993). Fire return intervals range from 200-300 years in lodgepole pine forests (Romme 1982, Veblen 2000) to more than 1000 years for some cedar/spruce/hemlock forests of the Pacific Northwest (Agee 1993). Local factors, such as elevation, topography, and climate, can modify the general fire regimes described above. For example, surface fires may occur less frequently in naturally dense systems FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. 51 of ponderosa pine with limited herbaceous cov- er; in turn, canopy fuels may become sufficiently dense to support crown fires (Shinneman and Baker 1997, Brown et al. 1999, Veblen 2000). Especially high probabilities of lighting strikes in mountainous terrain may result in small, fre- quent surface fires that often perpetuate open meadows in moist forests (Agee 1993, Veblen 2000). Overall, the complex mosaic of western forest systems has been shaped by an equally complex mosaic of fire regimes. CHANGES IN FIRE REG1MES Attempts to understand how contemporary human activities have altered natural fire re- gimes are fraught with difficulties. Fire regimes are inherently dynamic, largely due to variations in climate, both long-term (Clark 1988, Romme and Despain 1989, Johnson et al. 1990) and short-term (e.g., E1 Nifio-driven events; Swet- nam and Betancourt 1990, Veblen et al. 2000). In ponderosa pine systems, the degree to which severe fires result from the long-term accumu- lation of fuels due to fire suppression or the short-term accumulation and desiccation of fine fuels following E1 Nifio/Southern Oscillations is poorly understood and can vary among sites (Veblen et al. 2000). Likewise, decades of fire suppression at Yellowstone National Park, which may have delayed the onset of extensive fires, were apparently overshadowed by severe drought and high winds in August 1988 (Rom- me and Despain 1989). Thus, the relative con- tributions of fire suppression and climate on ex- treme fire behavior remains unclear. The relatively ephemeral nature of fire records (e.g., fire scars, stand cohorts) limits our recon- struction of fire histories for most locations (but see Agee 1998). Charcoal deposits in lake-bed sediments have revealed longer histories (Mills- paugh and Whitlock 1995), but they are influ- enced strongly by prevailing winds and water- shed dynamics so that the overall area they rep- resent may be quite limited. Historic accounts of fire behavior and forest conditions during Euro- American settlement can also be biased (Wagner et al. 2000). Furthermore, humans have influ- enced fire regimes in North America for at least 6,000-10,000 years. Native Americans used fire in warfare and for driving game (Stewart 1956), and Euro-American settlers used fire to clear land for mining, logging, and even in land dis- putes (Veblen and Lorenz 1991); settlers also caused many accidental fires (Johnson et al. 1990). Extensive livestock grazing after the mid- 1800s coupled with effective fire suppression (particularly after World War II) led to structural changes in forest stands (Saab et al. 1995), which altered fire regimes further (Madany and West 1983, Covington and Moore 1994; but see Swetnam et al. 1999). Thus, it is difficult to de- termine what constitutes "natural" or "anthro- pogenic" changes to fire regimes. For the pur- poses of this review, we focus on anthropogenic changes that began in the mid 1800s, including grazing, unprecedented fire suppression, and large-scale silvicultural activities (e.g., wide- spread clearcutting, salvage logging). Effects office suppression Given the complexity and limited understand- ing of historical fire regimes, the full ramifica- tions of fire suppression remain unknown. Cer- tainly, the long-term, global-scale effects of fire suppression and their potential interactions with climate changes caused by anthropogenic activ- ities are cause for concern (Leenhouts 1998). On a continental scale, however, it is clear that fire suppression over the last six or seven decades has reduced the number of fires and the total area burned across the U.S. (Ferry et al. 1995). Using satellite imagery, maps of potential natu- ral vegetation, and estimated fire regimes, Leen- houts (1998) concluded that only 8-14% of the area that burned annually in the conterminous United States 200-500 yr ago still burns today. In western forest systems, effects of fire sup- pression vary with forest type and inherent fire regime, as well as accessibility (Romme 1982). In many systems adapted to high-frequency/low-se- verity fire regimes (e.g., ponderosa pine), changes in forest structure since Euro-American settlement have included increased stem densities resulting from decreased mortality of saplings and increased recruitment, and changes in species composition (Gruell 1983, Veblen and Lorenz 1991, Covington and Moore 1994, Swetnam and Baisan 1996, Bel- sky and Blumenthal 1997, Allen 1998). Accumu- lation of fuels may promote more extensive, severe fires than those that occurred prior to Euro-Amer- ican settlement (Barrett 1988, Covington and Moore 1994, Lissoway 1996, Covington et al. 1997, Fule et al. 1997, Veblen et al. 2000). How- ever, wetter climates post-settlement may also con- tribute to a decrease in fire frequency (Veblen et al. 2000, Wagner et al. 2000). The consequences of fire suppression in forests characterized by infrequent fires of high severity (e.g., high-elevation spruce-fir forests of the central Rockies) are less apparent, in part because the lon- ger fire-return intervals may delay, or reduce, the effects of fire suppression (Romme 1982, Romme and Despain 1989, Veblen 2000). Even in regions where the frequency of fires has declined, burn severity may not have changed (Romme and Des- pain 1989). Although the relative contribution of climate and fire suppression is debatable, clearly 52 STUDIES IN AVIAN BIOLOGY NO. 25 the effects of both have influenced fire regimes across western landscapes. Other human activities may amplify or con- found the effects of fire suppression. Overgrazing by livestock or elevated populations of native un- gulates protected from wolf predation may di- minish fire frequency (Hess 1993, Belsky and Blumenthal 1997). For example, during the late 1800s to early 1900s, livestock grazing in many ponderosa pine systems led to decreased surface fuels and increased areas of exposed soil; the re- sult was diminished fire frequencies and in- creased germination and survival of tree seed- lings (Swetnam and Baisan 1996, Veblen 2000). In addition, the combined effects of fire suppres- sion, grazing, and contemporary silvicultural practices in many western forests has promoted the growth of dense, monospecific, even-aged stands (Swetnam et al. 1995, Fule et al. 1997). In turn, this stand structure is believed to present opportunities for more extensive outbreaks of tree-damaging insects than would have occurred prior to the mid-1800s when stands were ofien more open and complex in structure (Swetnam et al. 1995, Veblen 2000, Veblen et al. 2000). Wide- spread tree mortality resulting from insect out- breaks can increase a given stand's susceptibility to fire. Although our current knowledge of the interactive effects of fire suppression and other factors is limited, it has become clear that these factors can alter fire regimes significantly. EFFECTS OF FIRE AND SALVAGE LOGGING ON AVIAN COMMUNITIES Understanding fire regimes in western forests is essential to understanding forest structure, overall landscape patterns, and the responses of bird communities to fire. Fire affects avian nest- ing and foraging activities by generating snags, altering insect communities, eliminating foliage, and altering the size, abundance, and distribution of tree species across the landscape (Finch et al. 1997, Huff and Smith 2000). The degree to which fire affects any of these factors depends, in part, on the severity and ecological context of a par- ticular burn. A thorough understanding of the in- fluence of fire and fire-management activities, such as prescribed burning and post-fire salvage logging, on avian communities is essential to both conservation biology and sound management. Here, we summarize the best current knowl- edge about the influence of fire and salvage log- ging on avian communities in conifer-dominated forests (which often include quaking aspen) of the West. Most of the relatively few published studies were conducted in the northern Rocky Mountains. Because these studies encompassed many cover types and were usually poorly rep- licated, many of our conclusions are prelimi- nary. However, some general patterns, as well as a number of questions, have emerged from four comparisons: (1) avian abundance in burned and unburned forests, (2) avian abun- dance among different fire severities, (3) chang- es in avian-community structures associated with post-fire forest succession, and (4) nesting patterns of cavity-nesting birds in salvaged and unsalvaged, burned forests. AVIAN ABUNDANCE IN RECENTLY BURNED AND UNBURNED FORESTS We summarized the results of 11 studies that compared the abundance of breeding bird spe- cies in early post-fire burns and adjacent mature, unburned forests (Tables la-lc; Fig. 1). Al- though "unburned" forests may have burned previously, these forests were largely mature (i.e., late-successional). All 23 burns surveyed were severe (predominantly stand-replacement) and less than 10 yr old (most were <4 yr old). All but a few burns were greater than 400 ha, and four burns were greater than 1400 ha. Co- nifers, including ponderosa pine/Douglas-fir (Pseudotsuga menziesii), Jeffrey pine (Pinus jef- fryi)/white fir (Abies concolor), lodgepole pine, spruce/fir, and mixed conifers, were the domi- nant cover types. The studies covered seven western states; seven studies were conducted in the northern Rocky Mountains, one was in the southern Rocky Mountains, two were in the Pa- cific Northwest, and one was in the Pacific Southwest (Fig. 1). Studies of post-fire bird communities that were older than 10 yr, were predominantly aspen or riparian, or sampled only burn edges were excluded from analysis. For each species present in >--3 of the 11 stud- ies, we classified abundance patterns into three response classes by study: (1) occurred only in burns or abundance was >-50% higher in burns than in unburned forest; (2) occurred only in un- burned forest or abundance was >-50% higher in unburned than in burned forest; and (3) results varied among samples or there were similar abundances in burned and unburned forest (Ta- bles la-lc). Because only one study (Johnson and Wauer 1996) included both pre- and post- fire surveys, we used this comparison of abun- dance patterns to infer response to fire. Many species showed remarkably consistent patterns, despite the wide geographic area and va- riety of cover types surveyed. Species that com- monly occurred in burns, but were uncommon or absent in unburned forests (Table la), included Black-backed Woodpecker, Three-toed Wood- pecker, Olive-sided Flycatcher, and Mountain Bluebird (see Appendix for species' scientific names). Species that used unburned forests, but rarely occurred in early post-fire forests (Table FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. TABLE 1. SUMMARY OF AVIAN ABUNDANCES IN BURNED AND UNBURNED FORESTS 53 Response categories (number of studies) Similar abundance More abundant or response Species in burns mixed More abundant in unbumed (A) Typically more abundant in burns Three-toed Woodpecker 8 a, b, d, e, g, h, i, j Black-backed Woodpecker 6 b, d, e, i, j, k Olive-sided Flycatcher 8 a, c, d, f, g, h, i, k Mountain Bluebird 9 a, b, c, d, g, h, i,j, k Western Wood-Pewee 7 a, c, d, g, h, i, j Hairy Woodpecker 8 a, b, c, e, f, g, h, j 2 d, i House Wren 5 a, b, d, g, j 1 h Tree Swallow 4 b, h, i, j Northern Flicker 5 a, c, f, i, j 3 b, g, h (B) Typically exhibited mixed or neutral response to burns Mourning Dove 2 d, h Common Nighthawk 2 c, h Cassin's Finch 4 c, h, i, j Pine Siskin 3 c, f, i Chipping Sparrow 2 a, c Dark-eyed Junco 3 c, f, i American Robin 4 a, f, J, k Townsend's Solitaire 1 f Hammond's Flycatcher 1 f Clark's Nutcracker 2J, h Red-naped Sapsucker I h Western Tanager 1 c White-breasted Nuthatch I a Evening Grosbeak lg Pygmy Nuthatch I a Yellow-rumped Warbler WilliamsoWs Sapsucker 1 a Red Crossbill (C) Typically more abundant in unburned forests Steller's Jay 1 g Plumbeous/Cassin's Vireo lg Warbling Vireo lg Gray Jay 2 , h Ruby-crowned Kinglet 2g, h Brown Creeper 2 f, g Red-breasted Nuthatch 1 g 2 h, i Hermit Thrush 1 c Mountain Chickadee Golden-crowned Kinglet Townsend's Warbler Swainson's Thrush Varied Thrush 1 g 3a, d, g 3d, g, h 4g, h, i, j 5a, d, g, h, j 5c, d, g, h, i 5a, c, d, g, h 3d, g, h 2 d, g lg 4d, g, h, lg 1 h 5a, c, g, i, k lg 2g, h 1 ½ 1 c I d l i 2 a, b 2a, J 1 d 1 d 2g , h 3 d, h,j 2 h, i l d 3a, f, h 2 a, h 2 d, h 3 f, i,j 3d, i, j 5a, d, h, i, j 6a, b, d, f, j, k 5a, g, h, i, j 6a, g, h, i, j, k 6a, d, f, h, j, k 3d, f, k 3d, J, k 3d, 1', k Notes: Only species observed in three or more studies were included. More abundant in burns - only occurred in bums or abundance was >50% higher in early post fire forests than unburned forest; similar or mixed - abundance was similar in burned and unburned forest or results varied among samples; more abundant in unburned occurred only in unburned forest or abundance was >50% higher in unburned than early post-fire forests. aBock and Lynch 1970. b Caton 1996. c Davis 1976. d Hairis 1982. e Hoffman 1997. t Huff 1984, Huff et al. 1985. g Johnson and Wauer 1996. h N. Kotliar and C. Melcher, unpubl. data.  Pfister 1980.  Taylor and Bannore 1980. k R. Sallabanks and J. Mclver, unpubl. data. 54 STUDIES IN AVIAN BIOLOGY NO. 25 FIGURE 1. Approximate location of study sites re- ferred to in Table 1. Center location of study area is indicated in cases where multiple burns were surveyed. References (dominant cover type; number of burns; survey years post-fire): A--Bock and Lynch 1970 (Jef- frey pine/white fir; 1 burn; 6-8 yrs); B--Caton 1996 (lodgepole pine; 1 bum; 2-4 yrs); C--Davis 1976 (lodgepole pine; 2 burns; 6 yrs, 9 yrs); D--Harris 1982 (ponderosa pine/Douglas fir; 2 burns; 2-4 yrs, 2 yrs); E-Hoffman 1997 (lodgepole pine; 2 bums; 1-2 yrs); F--Huff 1984, Huff et al. 1985 (w. hemlock/Douglas fir; I burn; 1-3 yrs); G--Johnson and Wauer 1996 (ponderosa pine; 1 bum; 1 yr pre-fire; 3 yrs); H--N. Kotliar and C. Melcher, unpubl. data (ponderosa pine; lodgepole; spruce/fir; mixed conifer; 8 burns; varied from 0-8 yrs); IPfister 1980 (lodgepole pine; 2 burns; 2 yrs, 4 yrs); J Taylor and Barmore 1980 (lodgepole pine; spruce/fir; 2 burns; 1-3 yrs, 5/7 yrs); K--R. Sallabanks and J. Mclver, unpubl. data (mixed conifers; I burn; 1-3 yrs). lc), included Mountain Chickadee, Golden- crowned Kinglet, Hermit Thrush, Varied Thrush, and Townsend's Warbler. Generally, wood drillers and aerial insecfivores were more abundant in early post-fire forests, whereas foliage and bark gleaners were usually more abundant in unburned forests. However, there were several exceptions to this generalization. Overall, these results sug- gest that species with either the strongest affinity for, or aversion to, young burns are responding primarily to the dramatic changes in structural characteristics (e.g., increased availability of snags, decrease in canopy coverage) and/or den- sities of insect prey brought about by burning. Numerous species showed more varied, or ap- parently neutral, responses to bums (Table lb). For example, Townsend's Solitaire, American Robin, Dark-eyed Junco, Chipping Sparrow, and Cassin's Finch were common in both burned and unburned forests, indicating that both types of forests often may provide suitable habitat for these species. Many species, including Red- breasted Nuthatch, Brown Creeper, Yellow-rom- ped Warbler, and Western Tanager, were fre- quently observed in burns, but typically reached their highest abundance levels in unburned for- ests. Many granivores, bark gleaners, and spe- cies that prefer a mixed, open canopy had a var- ied responses. The mixed results may be due, in part, to the influence of site-specific character- istics (see FACTORS THAT AFFECT SPECIES' RE- SPONSES TO BURNS). Several species observed in fewer than three studies exhibited higher abundances in burned compared to unburned forests, including Lewis's Woodpecker (V. Saab, unpubl. data), Rock Wren, Western Bluebird (N. Kotliar and C. Melcher, un- publ. data), Lazuli Bunting (Bock and Lynch 1970), and White-crowned Sparrow (Pfister 1980; N. Kotliar and C. Melcher, unpubl. data). Our personal observations of these species suggest that they readily use bums in certain contexts. Although the generality of these observations is unknown, the apparent suitability of burned for- ests for these species warrants further study. A comparison of bird abundances in more than 30 fires that burned in the northern Rockies in 1988, with bird abundances derived from the lit- erature for nine other major Rocky Mountain for- est cover types (Table 3 in Hutto 1995), generally corresponds to the results of our review. Most of the species that exhibited higher abundances in burned forests (Table l a) were more commonly observed in recently burned forests than in all other mature forest types (Hutto 1995). Likewise, species that exhibited higher abundances in un- burned forests (Table lc) commonly occurred in one or more mature forest types but were infre- quently observed in recently burned forests (Hut- to 1995); however, Mountain Chickadee and Red- breasted Nuthatch occurred in a relatively high percentage (52-74%) of the 1988 bums surveyed by Hutto (1995). Many of the species that showed a mixed or neutral response to bums (Table la) also had a higher frequency of occurrence in ear- ly post-fire forests compared to mature forest types (Table 3 in Hutto 1995). Some of the species that showed mixed pat- terns across studies may use forest edges as well as forest interiors (e.g., Mountain Chickadee, Hermit Thrash; N. Kotliar and C. Melcher, un- publ. data), and because some are rather nomad- ic (e.g., Red Crossbill), the degree to which bums represent suitable habitat cannot be in- ferred easily from surveys that abut the edges of bums. Further research is needed to determine how various factors can alter the relative suit- ability of burned and unburned forests for such species (see next section). FACTORS THAT AFFECT SPECIES' RESPONSES TO BURNS The suitability of bums for birds often will depend on bum characteristics (e.g., severity, FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. 55 time since fire, burn geometry) and landscape context (e.g., forest cover types), as well as re- gional variation (Finch et al. 1997, Huff and Smith 2000). To begin to address these issues, we summarized the results of several studies that evaluated how burn severity and time since fire influenced bird communities. To provide impe- tus for future studies, we also speculate (based on personal observations and a few limited stud- ies) about the ways in which burn characteristics and context may contribute to variation in re- suits among studies. Burn severity Three studies compared avian abundances across various burn severities in reference to un- burned forests. Taylor and Barmore (1980) ex- amined two burn severities (moderate, severe) for the first three yr post-fire in a 1414-ha burn in lodgepole pine and spruce/fir forests in Grand Te- ton and Yellowstone National Parks. Preliminary results (first three yr post-fire) are available from a study of a 9283-ha burn in Oregon in which three burn severities (low, moderate, severe) were examined in mixed coniferous forests (R. Salla- banks, unpubl. data; R. Sallabanks and J. Mclver, unpubl. data). In addition, preliminary results are available for a comparison of two understory-pre- scribed (1 yr post-fire, 200 ha, and 1-3 yr post fire, 1200 ha) and two stand-replacement burns (1 yr post fire, 200 ha, and 3 yr post-fire, 4450 ha) in ponderosa pine/Douglas-fir forests in Col- orado (N. Kotliar and C. Melcher, unpubl. data). The trends observed in the burn-severity studies generally are consistent with the patterns we found in our review of severely burned versus unburned forest, which represented the extremes of the burn-severity gradient (Tables la-c). The general patterns presented here should be viewed as preliminary and in need of further testing, giv- en that two of the studies are unpublished and only six burns were studied. Many bird species whose abundances were consistently higher in burned compared to un- burned forests (Table l a) also appeared to use stand-replacement burns more readily than low- and moderate-severity burns. These species in- cluded Black-backed Woodpecker (R. Sallabanks, unpubl. data), Three-toed Woodpecker and Cas- sin's Finch (Taylor and Barmore 1980; N. Kotliar and C. Melcher, unpubl. data), Olive-sided Fly- catcher (R. Sallabanks and J. McIver, unpubl. data; N. Kotliar and C. Melcher, unpubl. data), Mountain Bluebird (Taylor and Barmore 1980; R. Sallabanks, unpubl. data; N. Kotliar and C. Melcher, unpubl. data), and Western Bluebird (N. Kotliar and C. Melcher, unpubl. data). Dark-eyed Juncos occurred at similar abundances across all burn severities (Taylor and Barmore 1980; N. Ko- tliar and C. Melcher, unpubl. data). Several species reached their highest abun- dances in moderate-severity burns. Brown Creeper and Chipping Sparrow exhibited highest abundances in moderate-severity and severe burns (Taylor and Barmore 1980). Townsend's Solitaire was fairly abundant across all severi- ties, but was most abundant in moderately se- vere burns (N. Kotliar and C. Melcher, unpubl. data). Western Tanager occurred at similar abun- dances in moderately burned and unburned for- ests, but was less abundant in severely burned forests (Taylor and Barmore 1980). Cavity nest- ing species that usually glean the bark of live trees (e.g., nuthatches, Brown Creeper) may re- spond positively to moderate-severity burns that increase availability of snags for nesting, but re- tain live trees for foraging. Species common in open canopy forests (e.g., Townsend's Solitaire, Western Tanager, Chipping Sparrow) may use the mixed open canopy of moderate-severity burns, whereas they may avoid large areas of stand-replacement burns. Thus, the varied re- sults observed for these species in our review of severely burned and unburned forests (Table 1 b) may reflect, in part, the heterogeneity of burn severities within and across studies. Species that were consistently more abundant in unburned than in burned forests (Table 1 c) also decreased in abundance with increasing burn se- verity. These species include Plumbeous Vireo, Steller's Jay, and Hammond's Flycatcher (N. Ko- tliar and C. Melcher, unpubl. data), Gray Jay (Taylor and Barmore 1980); Mountain Chickadee (Taylor and Barmore 1980; R. Sallabanks, un- publ. data; N. Kotliar and C. Melcher, unpubl. data); Ruby-crowned and Golden-crowned king- lets (Taylor and Barmore 1980; R. Sallabanks, unpubl. data); Townsend's Warbler and Varied Thrush (R. Sallabanks, unpubl. data). Many of these species are foliage gleaners; thus their abun- dance patterns probably reflect the incremental loss of foliage area with increasing burn severity. Several species showed slightly different pat- terns across the three studies. Red-breasted Nut- hatch and Yellow-rumped Warbler were least abundant in severe burns across all three studies, but their abundances varied across other severi- ties (Taylor and Barmore 1980; R. Sallabanks, unpubl. data; N. Kotliar and C. Melcher, unpubl. data). Western Wood-pewee increased in abun- dance with burn severity in a lodgepole pine burn (Taylor and Barmore 1980), but was most abun- dant in low-severity ponderosa pine burns (N. Kotliar and C. Melcher, unpubl. data). Again, var- iation in results among studies may be due to the heterogeneity of burn severities both within and among studies. Furthermore, if patches of low- 56 STUDIES IN AVIAN BIOLOGY NO. 25  I Open Canopy I  .................. [Closed Canopy ] e- Low High Fire Severity FIGURE 2. Conceptual model of the interactive ef- fects of burn severity and forest structure on the den- sity of avian species preferring open forest structure. In open-canopy forests (e.g., ponderosa pine) avian densities are high in unburned forests but may be low in severely burned forests. In closed-canopy forests (e.g., lodgepole pine), avian densities are low, but may increase as fire opens up the forest canopy. Thresholds responses to degree of burn severity may result in de- parture from linear relationships depicted here. and moderate-severity burns occur along the burn periphery, as is often the case, it may be difficult to differentiate between the influence of burn se- verity and edge effects (i.e., the juxtaposition of burned and unburned forest). Interactions between burn severity and pre- fire forest structure also may lead to mixed re- sponses to burn severity, particularly for bird species that are sensitive to differences in can- opy coverage (Fig. 2). Some species that occur in open-canopy forests (e.g., Western Wood-pc- wee, Western Tanager) are common in unburned ponderosa pine forests but uncommon in stand- replacement burns in this cover type (N. Kotliar and C. Melcher, unpubl. data). In contrast, these species may be uncommon in dense lodgepole pine (Pinus contorta) forests, but common im- mediately following stand-replacement fires in lodgepole pine forests (N. Kotliar, unpubl. data). Such interactions makes it difficult to predict how a species will respond to burns without a better understanding of how context (e.g., cover type, canopy closure, regional differences, pre- vious silvicultural treatments) can alter suitabil- ity of burned forests for a particular species. Post-fire succession and associated changes in .[orest structure and arian communities No studies have followed bird communities from early through late successional stages after fire (but see Bock and Lynch 1970, Bock et al. 1978, Raphael et al. 1987, Johnson and Wauer 1996); therefore, to examine changes in bird communities from early successional to mature forests we also rely on comparisons of stands that vary in time since fire (e.g., Peterson 1982, Huff et al. 1985). In general, forest structure and avian communities change fairly rapidly after fire, although the rates of change depend, in part, on burn severity as well as pre- and post-fire cover type. Because tree mortality is low, and ground cover often rapidly resprouts, evidence of fire in understory burns may be minimal with- in a few years after fire. In contrast, stand-re- placement burns may persist as a forest of snags for decades. The structure of burned snags typ- ically changes within the first few years. First, needles (if remaining) and smaller branches are shed, then bark and larger branches slough away. Smaller snags typically decay faster than larger snags (Morrison and Raphael 1993, Bull et al. 1997). Factors such as topography, root depth, moisture regime, wind, and tree species can all influence how long snags remain stand- ing, which may exceed a century. Early post-fire forests and associated insect outbreaks attracts cavity-nesting birds due to in- creases in nest sites and food supplies (e.g., Blackford 1955, Koplin 1969, Lowe et al. 1978, Raphael and White 1984, Bock et al. 1978, Saab and Dudley 1998). Duration of occupancy, how- ever, varies among bird species, presumably due to differences in preferred prey availability, as well as the size, distribution, and age of snags. Black-backed and Three-toed woodpeckers rap- idly colonize stand-replacement burns within one to two years of a fire; within five years, however, they become rare, presumably due to declines in bark and wood-boring beetles (Ko- plin 1969, Bock and Lynch 1970, Bock et al. 1978, Bull 1980, Taylor and Barmore 1980, Ap- felbaum and Haney 1985, Dixon and Saab 2000). In contrast, Lewis's Woodpecker is re- ported to be abundant both in recent burns (2-4 yr; Saab and Dudley 1998) and older burns (10- 25 yr; Bock 1970, Linder and Anderson 1998). Hairy Woodpecker and Northern Flicker exhibit more mixed responses, but usually decline with- in the first 25 yr post-fire (Bock and Lynch 1970, Bock et al. 1978, Taylor and Barmore 1980, Huff et al. 1985, Raphael et al. 1987). Mountain and Western bluebirds are secondary- cavity nesters that commonly nest in recently burned forests (e.g., Hutto 1995, Saab and Dud- ley 1998; Table la), but they typically decline in mid-successional stages (Bock and Lynch 1970, Bock et al. 1978, Pfister 1980, Peterson 1982, Raphael et al. 1987). Vegetation regrowth after fire also can lead to increases in flower, seed, and insect abundance, which attracts nectarivores, granivores, and ae- FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. 57 rial and ground insectivores (Lowe et al. 1978, Apfelbaum and Haney 1981, Huff et al. 1985). Olive-sided Flycatcher may appear immediately after fires (Table la; Hutto 1995; N. Kotliar and C. Melcher, unpubl. data) and can persist as long as snags are available and canopy cover remains low (Huff et al. 1985; N. Kotliar and C. Melcher, pers. obs.). Seed-eating birds exhibit a mixed re- sponse to burns, but there is some evidence that several species readily use burns, Clark's Nut- cracker, Pine Siskin, Cassin's Finch, and Red Crossbill in particular (Table lb; Hutto 1995). Whether theses species are responding to in- creased seed availability (e.g., serotinous cones), minerals in the ashes (C. W. Benkman, pers. comm.), or other factors remains unclear. Fur- thermore, these species are rather nomadic, or have large home ranges, and may use burned forests opportunistically. Many species absent or uncommon immediate- ly post-fire begin to increase in mid-successional stages as snags decay or fall, shrubs and saplings become well-developed, and canopy cover in- creases. Although Cordilleran and Dusky fly- catchers may appear at the edges of early post- fire forests (N. Kotliar and C. Melcher, unpubl. data), they sometimes reach peak abundances at mid-successional stages (Peterson 1982, Raphael et al. 1987; N. Kotliar and C. Melcher, unpubl. data). Resprouting aspen stands can attract spe- cies commonly associated with deciduous sys- tems (e.g., Warbling Vireo, Dusky Flycatcher; N. Kotliar and C. Melcher, pers. obs.). Red-naped Sapsucker also has been observed drilling holes in lodgepole pine and aspen saplings within 5-10 years following disturbances (N. Kotliar and C. Melcher, pers. obs.). Lewis's Woodpecker may use burned forests 10-20 yr after fires, presum- ably in response to improved conditions for aerial foraging following a decrease in snag density and an increase in flying arthropods associated with shrub regrowth (c.f., Bock 1970, Linder and An- derson 1998). Species such as Mountain Chick- adee, Ruby-crowned Kinglet, and Swainson's and Varied thrushes reach peak abundance in late-suc- cessional forests (Bock and Lynch 1970, Bock et al. 1978, Peterson 1982, Huff et al. 1985, Raphael et al. 1987). In contrast, species that favor open canopies (e.g., American Robins) begin to decline in mid- to late-successional stages (Peterson 1982, Huff et al. 1985, Raphael et al. 1987). Several species that occur in early post-fire forests also may occur in later successional stag- es. Hammond's Flycatcher occasionally has been detected in young post-fire forests (Harris 1982, Huff et al. 1985, Hutto 1995, Johnson and Wauer 1996; N. Kotliar and C. Melcher, unpubl. data), but they typically reach peak abundance in mature forests (Peterson 1982, Sedgwick 1994; N. Kotliar and C. Melcher, unpubl. data). However, its occasional occurrence immediately after fire suggests that Hammond's Flycatcher may temporarily exhibit site-fidelity. Several species, such as Olive-sided Flycatcher, Brown Creeper, and Dark-eyed Junco, initially may de- cline in mid-successional stages, but may in- crease as canopy gaps and snags are created (Huff et al. 1985, Carey et al. 1991). Fire geometry Although no studies have explicitly examined how birds respond to burn size or shape, one study examined whether bird abundance was af- fected by differing patch sizes created by the extensive fires of 1988. Of the 87 species pres- ent, only Plumbeous Vireo and Townsend's Sol- itaire decreased with increasing patch size (Hut- to 1995). However, the relatively large minimum patch size surveyed (40 ha) may have masked important area effects at lower size ranges. Thus, the response of birds to total burn area needs additional study. Given that area effects have been found to be important in other ecosystems, we should con- sider these effects as they relate to fires as well. For example, post-fire specialists may require a minimum burn size. In contrast, some species may select openings created by small burns and avoid larger burns. Increase in burn size may also lead to increased heterogeneity of bums (e.g., variation in burn severity). The proportion of burn to edge area is also affected by burn size and shape. Thus, species that show positive responses to burns may be attracted to the juxtaposition of burned and un- burned forest. For example, Olive-sided Fly- catcher and Townsend's Solitaire (Table la) reached their highest abundances at burn edges (N. Kotliar and C. Melcher, unpubl. data). In ad- dition, fire damaged trees (not killed outright by fire), which often occur along the periphery of crown fires, are used by several post-fire wood- pecker species (Murphy and Lehnhausen 1998). Many of the species showing mixed response to burns (e.g., American Robin, Townsend's Soli- taire, Western Tanager, Dark-eyed Junco, Chip- ping Sparrow, Pine Siskin, and Cassin's Finch; Table lb) reached their highest abundances within 50 m of the edges of burns (N. Kotliar and C. Melcher, unpubl. data). Many crown fires also contain "peninsulas" and "islands" of unburned forest remnants, which can increase edge habitats or retain un- burned forest well inside of large burns. For ex- ample, the moist microclimate of riparian areas, which may inhibit fire or limit burn severity, can result in riparian remnants. Thus, species not typically associated with early post-fire forests 58 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 2. NUMBER OF CAVITY-NESTING SPECIES IN UNLOGGED AND SALVAGE-LOGGED POST-FIRE FORESTS DURING THE BREEDING SEASON 1N THE NORTHERN ROCKY MOUNTAINS Number of nesting species Partially Severely Forest type Unsalvaged salvaged salvaged Totals Study Mixed conifer/deciduous 16 12 4 17 Caton 1996 a Mixed conifer/deciduous 18 -- 8 18 Hitchcox 1996 b Ponderosa pine/douglas-fir 9 9 -- 10 Saab and Dudley 1998 c Mixed conifer 8 9 -- 9 S. Hejl and M. McFadzen, unpubl. data d a Salvage logging of entire 4000-ha burn included clearcuts (all trees were removed except for a few snags) and partial cuts (individual trees or small groups of trees were logged). b Salvage logging of entire 500-ba burn created an interspersion of harvest treatments with unlogged control plots, In severely salvaged areas, all merchantable (>15 cm dbh, >4.5 m tall) fire-killed trees were harvested. c In salvage-logged units, about 50% of all trees >23 cm dbh, and 70% of trees >53 cm, were harvested. d Salvage logging varied among three bums (bin'ns ranged from 494 3,321 ha). The salvaged portions of burns were partially logged with several areas of severe salvage logging. A portion of each burn was left unbarvested. (e.g., Wilson's Warbler, Lincoln's Sparrow; N. Kotliar and C. Melcher, pers. obs.) may be ob- served in remnant patches immediately post-fire. In burns, detections of birds more typically as- sociated with unburned forest may be artifacts of study design. Few studies explicitly control for distance from survey points in burned habitats to unburned edges and remnant patches. Yet, some species characteristic of unburned forests (e.g., Mountain Chickadee, Ruby-crowned Kinglet, Hermit Thrush) may use live trees along burn edges (N. Kotliar and C. Melcher, unpubl. data). Thus, these species, which also have highly de- tectable songs, may appear to use recently burned forests if survey points are too close to edges. Conclusions: effects of fire on avian communities Although there are relatively few studies that address the effects of fire on avian communities, the consistent presence of many woodpeckers and aerial insectivores in early post-fire forests, and the near absence of many foliage-gleaning species associated with closed-canopy forests, appear to be robust patterns. Many additional species appear to use post-fire forests in certain contexts. For most species, however, we still have a poor understanding of how fire alters habitat suitability. We clearly need more infor- mation about how species' responses to fire can be altered by burn severity (including within- burn heterogeneity), fire geometry, proximity to unburned edges and remnants, pre- and post-fire cover types (e.g., tree species, forest structure, previous silvicultural treatments), and time since fire. Finally, because most burns outside national parks are salvaged, information about the effects of post-fire salvage logging is also critical. EFFECTS OF POST-FIRE SALVAGE LOGGING ON AVIAN COMMUNITIES Salvage logging following stand-replacement fires has occurred since the early 1900s (D. At- kins, pers. comm.). Initially, salvage logging was uncommon due to limited access to burned forests (K. McKelvey, pers. comm.). In the 1950s, however, the demand for lumber in- creased greatly, and subsequent road-building in national forests provided opportunities to har- vest more burns (D. Arkins, pers. comm.). Typ- ically, salvage logging was implemented imme- diately post-fire, leaving few, if any, standing snags. Only within the past two decades have forest managers begun to retain snags within sal- vaged areas to benefit wildlife. The effects of salvaging on avian communities remain poorly understood. Only four studies, all of which were restricted to coniferous and mixed coniferous/deciduous (hereafter "mixed") forests of the northern Rocky Mountains (Montana and Idaho), specifically examined the effects of sal- vage logging on cavity-nesting bird communities (Caton 1996, Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data; Table 2). Two other studies evaluated salvaged burns (Blake 1982, Raphael and White 1984) but did not replicate treatments, thus they were not emphasized in this review. As a result, we focus our discussion on cavity-nesting species in the northern Rocky Mountains. Effects of salvage logging on birds Severely salvaged burns (Table 2) may de- crease the suitability of post-fire forests for most cavity-nesting species. However, the effects of partial salvaging are more equivocal (Table 2). In general, species richness declined only in the most severely salvaged burns, although even partial salvaging altered species composition (Table 2; Raphael and White 1984). Several cavity nesters showed consistent pat- terns of abundance in logged or unlogged con- ditions across studies. Black-backed and Three- toed woodpeckers were most abundant in unsal- vaged burns and rarely nested in salvaged areas FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. 59 of burns (Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). In contrast, nesting Lewis's Woodpeckers were most abundant in partially salvaged burns (Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). Mountain Bluebird and Hairy Woodpecker nested in both unsalvaged and sal- vaged portions of burns, but tended to nest more often in unsalvaged portions (Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. Mc- Fadzen, unpubl. data). The responses of several species to salvage logging varied among studies. Red-breasted Nuthatch and Williamson's Sapsucker nested primarily in partially salvaged burns in conifer- ous forest (S. Hejl and M. McFadzen, unpubl. data), whereas in mixed forest they nested only in the unsalvaged portions of severely salvaged burns (Hitchcox 1996). These mixed responses to salvage logging may be due to differences in salvage severity or cover type. In general, it ap- pears that species most closely tied to early suc- cessional post-fire forests (Table la) may be the most sensitive to salvage logging. The effects of salvage logging on nesting suc- cess also varied among species and studies. In the three studies that examined nesting success (>20 nests per treatment per species), Hairy Woodpecker (Saab and Dudley 1998), Northern Flicker (Hitchcox 1996), and Mountain Bluebird (S. Hejl and M. McFadzen, unpubl. data) expe- rienced significantly higher nesting success in unsalvaged treatments. Three-toed Woodpeck- ers, House Wrens, and Western Bluebirds had similar nesting success among treatments. Variation in characteristics of snags used for nests sites and foraging Salvage-logging practices often call for the harvest of larger, more economically valuable tree species. By altering species composition, sizes, and densities of snags, salvaging may alter resource availability for birds. Therefore, we de- scribe characteristics of post-fire forests required for foraging and nesting cavity-nesting birds and relate those needs to management practices. Although tree species selected for nest sites varied among bird species and studies, some gen- eral patterns were evident. In three studies of mixed forests (both salvaged and unsalvaged) dominated by conifers (95% conifers, 5% Popu- lus spp.), a disproportionate percentage of nests (35-80%) were located in deciduous trees (Hutto 1995, Caton 1996, Hitchcox 1996). Most nests were located in snags. In two other studies of coniferous and mixed conifer forests, birds nested in snags of western larch (Hitchcox 1996; S. Hejl and M. McFadzen, unpubl. data) and ponderosa pine (S. Hejl and M. McFadzen, unpubl. data) Williamson's Sapsucker Lewis's Woodpecker Northern Flicker Brown Creeper White-headed Woodpecker Hairy Woodpecker Western Bluebird Mountain Bluebird Red-breasted Nuthatch Black-backed Woodpecker Three-toed Woodpecker Increasing Snag Density at Nest Sites FIGURE 3. General distribution of cavity-nesting birds in burned forests (unsalvaged and salvage logged) as a function of nest-tree diameter (DBH) and snag density at nest sites (Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). more often than expected. In one study in Idaho and Montana, 45% of all nests were in Douglas- fir (S. Hejl and M. McFadzen, unpubl. data). Such variation in nest-tree selection among stud- ies may result from variation in species compo- sition and the relative availability of preferred trees (S. Hejl and M. McFadzen, unpubl. data). The extent of snag decay influences which snags woodpeckers select for nesting. For ex- ample, strong excavators such as Black-backed, Three-toed, Hairy, and Downy woodpeckers, nested in snags with intact tops (Caton 1996, Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). Weak excava- tors such as Lewis's Woodpecker, White-headed Woodpecker, and Northern Flicker, nested more frequently in broken-topped snags (many broken pre-fire) that were presumably more decayed than intact snags (Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). Because the extent of decay influences nest-tree selection, selective salvaging of less decayed snags likely affects bird species differentially. Cavity nesters also respond to differences in the sizes and spatial distribution of snags (Fig. 3), which, in turn, could be affected by different salvage prescriptions (Saab et al. 2002). In both coniferous and mixed burns, most cavity nesters selected large-diameter trees more often than ex- pected (Caton 1996, Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). Black-backed and Three-toed woodpeck- ers nested in medium-sized snags (Hitchcox 1996, Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). This size class was among the smallest used by any woodpecker species, but is within the size-range targeted for salvaging. In general, cavity nesters selected dense patches of snags more often than dis- 60 STUDIES IN AVIAN BIOLOGY NO. 25 persed or isolated snags (Raphael and White 1984, Saab and Dudley 1998, Saab et al. 2002). Despite the paucity of foraging studies in post- fire forests, some general patterns regarding pref- erences of woodpeckers for certain tree species and sizes emerged from our review. Woodpeckers selectively foraged on large snags in both winter (Kreisel and Stein 1999) and summer (Hutto 1995, Powell 2000; S. Hejl and M. McFadzen, unpubl. data). However, use of tree species in summer varied among studies (Hutto 1995, Caton 1996, Powell 2000; S. Hejl and M. McFadzen, unpubl. data), among habitats within one study (Caton 1996, Powell 2000), and among Picoides woodpeckers within a study (S. Hejl and M. McFadzen, unpubl. data). In northeastern Wash- ington during winter, Downy, Hairy, Three-toed, and Black-backed woodpeckers selectively for- aged on western larch and ponderosa pine, which are also preferentially salvage logged. Thus, by altering the size, distribution, and species com- position of post-fire snags, salvage logging dif- ferentially affects cavity-nesting species. Co-occurring species of woodpeckers some- times select different prey, which could influ- ence avian diversity in post-tire habitats. For ex- ample, in a recent study of an unsalvaged burn in Alaska, Murphy and Lehnhausen (1998) an- alyzed the contents of 33 woodpecker stomachs and found that Three-toed Woodpeckers con- sumed bark beetle larvae (Scolytidae) almost ex- clusively, whereas Black-backed and Hairy woodpeckers primarily consumed wood-boring beetles (Buprestidae and Cerambycidae). In an unsalvaged burn in east-central idaho, Black- backed Woodpeckers were observed feeding their nestlings the larvae and pupae of wood- boring beetles approximately 65% of the time (Powell 2000). Beal (1911), however, reported that 65-75% of the prey consumed by Three- toed and Black-backed woodpeckers were wood-boring beetles. Differences among studies could be due to prey availability (Powell 2000), which in turn is affected by tree species com- position, burn severity, and salvage severity. Conclusions: effects of post-fire salvage logging on cavity-nesting birds Overall, salvage logging in burned forests can have pronounced effects on cavity-nesting species that use post-fire habitats. In conjunction with a substantial reduction in fire-killed trees due to fire suppression, salvage logging has resulted in dra- matic reductions in the availability of snags in these ephemeral habitats. The effects of such re- ductions have serious implications for the viability of Black-backed and Three-toed woodpeckers, which rarely use even partially- logged post-tire forests. Although forest managers have begun to retain some snags (including large snags) in sal- vaged areas, this is not sufficient for species that prefer high densities of snags that characterize un- salvaged bums. Some types of partial salvaging may actually benefit a few species, but historically such species may have been more closely associ- ated with later successional stages of burns after snag densities had decreased naturally, with forests kept open by frequent, low-severity rites, or open post-tire forests. Retention of a diversity of snag species, sizes, and spatial distributions, as well as snags in various stages of decay, in burned forests is essential to the conservation of avian diversity in northern Rocky Mountain forests. The applica- bility of these conclusion across western forests or other avian communities (e.g., open-cup nesting species) requires further research. MANAGEMENT IMPLICATIONS FIRE MANAGEMENT Given the importance of fire to many bird spe- cies, restoration of natural fire regimes may be critical to the ecological integrity of western for- ests. However, the problems associated with re- producing the complexity and diversity of fire processes at multiple scales pose great challeng- es (Baker 1993). The recent emphasis on pre- scribing frequent, low-intensity fires in low-el- evation forests of the Rocky Mountains is a good start toward. reintroducing fire in systems where frequent understory burns maintained open, old-growth stands (but see Covington and Moore 1994, Tiedemann et al. 2000), but this treatment will not be adequate for bird species that associate with stand-replacement burns. For example, prescribed fire may alter the availabil- ity of large snags, depending on fire severity (Horton and Mannan 1988, Tiedemann et al. 2000). In general, the effects of prescribed fire on avian communities are poorly understood (Finch et al. 1997, Tiedemann et al. 2000); the few studies of prescribed fire have been plagued by methodological problems, and thus the con- clusions of these studies are suspect (Finch et al. 1997). Furthermore, incorrectly applied pre- scribed fire can alter landscape structure (Baker 1993). Fire-management practices that include allowing wildland fires of all severities to bum, when and where they are appropriate, may help re-create natural conditions (Hejl et al. 1995). Given the uncertainty about specific, local fire regimes (Baker 1994, Tiedemann et al. 2000, Veblen et al. 2000) and the variation among bird species in response to fire characteristics (Hutto 1995), managers may wish to mimic natural var- iation in fire regimes (e.g., size, severity, fre- quency, timing) that may have occurred within a given cover type and geographic area (Baker FIRE EFFECTS ON AVIAN COMMUNITIES Kotliar et al. 61 1992, Hejl et al. 1995, Veblen et al. 2000). This approach will help to avoid overemphasis on any particular prescription. Post-fire forests can be altered significantly by salvage logging. Although bird species will vary in their responses to different management op- tions, few cavity-nesting species, if any, will ben- efit from severe salvaging (i.e., clearcut, or re- moval of most medium and large snags). Here, we evaluate several alternatives to severe salvage logging based on our knowledge of nesting re- quirements for six cavity-nesting birds in the northern Rocky Mountains: (1) leave the burn un- salvaged; (2) lightly salvage throughout the burn (e.g., leave many of the biggest snags); (3) sal- vage the burn (e.g., light or partial) after a delay of several years (Murphy and Lehnhausen 1998, Kreisel and Stein 1999); (4) salvage part of the burn severely and leave the remainder unsalvaged (Hutto 1995); and (5) apply different salvage treatments across the burn (including variation in tree distributions, sizes, and species left uncut). The species most likely to benefit from unsal- vaged burns, or unsalvaged portions of burns, are those most-closely tied to early post-fire condi- tions. Because Black-backed and Three-toed woodpeckers appear to depend on the short-lived availability of prey resources that quickly invade post-fire habitats, a delay in salvaging may be warranted (Murphy and Lehnhausen 1998). Some species (e.g., American Kestrel, Lewis's Wood- pecker) may tolerate or benefit from partial or light salvage logging provided the large snags and tree species (e.g., deciduous trees, Douglas- fir, ponderosa pine, western larch) they tend to select are left uncut (Saab and Dudley 1998; S. Hejl and M. McFadzen, unpubl. data). Species may inhabit partially salvaged burns (Saab and Dudley 1998; S. Hejl and M. Mc- Fadzen, unpubl. data) because they resemble the later successional stages of burns (when snags begin to thin out naturally) or open forests. Giv- en our limited understanding of the cumulative effects of fire suppression and post-fire salvage logging, and their effects on post-fire habitat availability across western landscapes, allowing succession to proceed naturally in unsalvaged burns may benefit the most species. MIMIC NATURAL DISTURBANCE REGIME Many bird species are adapted to, and may de- pend upon, natural disturbance such as fire. Over the last century, however, logging has supplanted fire as the dominant process shaping coniferous forests in many regions of the West. Yet, the con- sequences of this shift for avian communities is poorly understood (Hansen et al. 1991). It has been suggested that the disturbance created by logging may create adequate habitats for some fire-depen- dent species in areas where severe fires are im- practical (Hutto 1995). Indeed, fire and logging could have similar effects on western landscapes if logging were modified to mimic natural fires more closely (Hunter 1993, Hejl 1994). However, there are profoundly different ways in which past fire and silvicultural activities have affected west- em forest systems. First, they often operate on vastly different spatial and temporal scales (e.g., disturbance size and frequency), which, in turn, will lead to different landscape structure (Hansen et al. 1991, Gluck and Rempel 1996). Second, there are many unique features produced by fire (e.g., a high density of snags and consequent in- creases in wood-boring beetles) that may not be replicated readily by current logging practices (Hansen et al. 1991, Hutto 1995). Finally, selective logging often removes larger trees whereas low- severity fires typically kill smaller trees (Finch et al. 1997). Thus, natural disturbances may provide useful models for developing logging and salvag- ing techniques that would diminish the negative impacts on birds (Hunter 1993, Hejl et al. 1995). Our understanding of how birds respond to silvicultural activities is based primarily on com- parisons of logged versus relatively undisturbed, mature forests (Hejl et al. 1995). However, as- sessments that include comparisons of logged and naturally disturbed forests with similar dis- turbance severities (e.g., thinned forests might be compared to moderate or understory burns) would be valuable. For example, a recent study of 16 burned and 16 logged conifer forests in Colorado found that severely logged forests (i.e., logged areas contained few, if any, live or dead trees) were generally unused by most species as- sociated with stand-replacement burns (N. Kot- liar and C. Melcher, unpubl. data). Overall, avi- an species richness was much higher in burns than in logged forests. The pattern was espe- cially salient when comparing clearcuts (i.e., no retention trees) to unsalvaged burns. Of the spe- cies that did occur in clearcuts, most also oc- curred in burns, whereas the reverse was not ob- served. Hansen et al. (1995b) also found that retaining canopy trees benefits many bird spe- cies in the west Cascades of Oregon. In general, clearcut conifer forests do not function as sub- stitutes for burned forests. In many respects, the effects of logging on avian communities in un- burned forests may be similar to those of salvage logging in stand-replacement burns. The high density of snags in burns is the most obvious distinction between burned and clearcut forests. However, the edges of these disturbanc- es can also differ dramatically. For example, clearcut forests often have well-defined edges with few, if any standing snags. In contrast, burn edges are often a heterogeneous mix of burned 62 STUDIES IN AVIAN BIOLOGY NO. 25 and unburned trees, except along fire breaks where burn edges are usually more abrupt. The juxtaposition of live and dead forests may be important to many species, such as Olive-sided Flycatcher, which generally sings and conducts foraging sallies from dead trees in open areas but nest in nearby live, mature trees (Altman and Sallabanks 2000). In Colorado, Olive-sided Fly- catcher only occurred in cuts that contained both snags and live trees (i.e., not clearcuts; N. Kot- liar and C. Melcher, unpubl. data). The com- plexity of burn edges may also help to diminish deleterious edge effects (e.g., increased nest pre- dation and parasitism) in adjacent undisturbed forests that could result from high-contrast edges of clearcuts. Thus, silvicultural practices that in- corporate structural elements of burns (e.g., re- talning or creating high densities of snags and patches of live trees, and increasing the com- plexity of edges) may improve the suitability of logged forests for many post-fire bird species. There are also important differences between natural and anthropogenic disturbances at larger spatial and temporal scales. For example, in both conifer- and aspen-dominated forests, differenc- es in bird communities among burned and cut forests were most evident in early successional stands, but were still apparent in mid-succes- sional forests (Hutto 1995, Hobson and Schiek 1999). We explore this idea further by compar- ing the fragmenting effects of severe disturbanc- es: stand-replacement fire, silvicultural activities (e.g., post-fire salvage, clearcuts), and forest conversion. Here, we restrict the meaning of for- est fragmentation to the fragmenting effects of anthropogenic disturbance relative to the natural heterogeneity of the landscape. Fragmentation can alter several landscape-scale parameters, in- cluding the number, size, and spatial distribution of forest patches, the degree of contrast between disturbed and adjacent undisturbed forests, and the persistence of the disturbed patches. By def- inition, natural disturbance regimes, such as stand-replacement fires, create and reinforce nat- ural heterogeneity (e.g., spatial configuration of forest patches, variation in successional stages among patches). Because most post-fire forests eventually resemble pre-fire forests (e.g., cover type), persistence and contrast are relatively low compared to the highly persistent patches that result from forest conversion in agricultural or suburban landscapes. The fragmenting effects of silvicultural practices will generally fall some- where between these two extremes, depending on logging severity (e.g., thinning vs. clearcut- ting) and frequency (e.g., cut rotation). For some species, however, the negative effects resulting from alteration of landscape structure and dy- namics through fire suppression may rival the negative consequences of forest fragmentation in some western forests. However, few studies have evaluated the consequences of fire sup- pression or other alterations of fire regimes on avian communities (Lyon et al. 2000). The degree to which anthropogenic disturbance results in forest fragmentation depends on differ- ences between the scale, intensity, and frequency of natural and anthropogenic disturbance, as well as the natural heterogeneity of the landscape. Spe- cies adapted to frequent natural disturbance may tolerate or even prefer the conditions created by disturbance over undisturbed forests. The Pygmy Nuthatch, for example, which is endemic to pon- derosa pine forests (relatively short fire return in- tervals), had higher abundance in prescribed un- derstory bums than in adjacent unburned forests, and was absent in stand-replacement bums (N. Kotliar, unpubl. data). In contrast, species such as the Golden-crowned Kinglet, Varied Thrush, and Townsend's Warbler, which are most often found in association with spruce-fir and cedar-hemlock cover types (relatively long fire return intervals; Hutto 1995), consistently occurred at lower abun- dances in burned forests (Table l c). Although many species may tolerate, or be adapted to, nat- ural disturbance, we expect that most bird species (except for some generalists and introduced spe- cies) will be extremely sensitive to the high degree of persistence and contrast of forest conversion, regardless of inherent disturbance regimes. Super- imposed on these factors are other landscape-scale issues such as local cowbird abundance or the composition of predator communities. Thus, local, landscape, and regional diftErences need to be ad- dressed when basing silvicultural practices on nat- ural disturbance regimes. RESEARCH RECOMMENDATIONS SPECIFIC RESEARCH QUESTIONS Based on our review of past research, we have identified some general patterns regarding the responses of avian communities to fire. How- ever, the studies have raised more questions than they have answered. Thus, applications of man- agement prescriptions involving fire and fire-re- lated silvicultural practices should be considered experimental and be designed to increase our knowledge about fire effects. For example, we need more information about the basic ecology of post-fire forests, including: ß how various fire characteristics (e.g., severity, size, successional stage, and season of burn- ing), landscape contexts, and cover types (both pre- and post-fire types) affect avian communities; ß the extent to which avian use of burns is pred- FIRE EFFECTS ON AVIAN COMMUNITIES--Kotliar et al. 63 icated on the juxtaposition of burned and un- burned forest; ß the effects of fire on life histories (foraging behavior, nest site selection) and demograph- ics, particularly reproductive success, survi- vorship, and recruitment for both breeding and wintering populations (Finch et al. 1997, Lyon et al. 2000); ß variation in avian use of fire-generated snags for nest, foraging, or perch sites compared to use of snags generated by other process (e.g., lightning, disease, insects); ß how avian communities differ in naturally dis- turbed forests compared to managed forests across successional stages; ß the effects of seed-eaters, flycatchers, and oth- er specialists on seed dispersal, forest regen- eration, and overall forest health; ß the manner in which snag characteristics and distributions affect insect prey and, in turn, foraging birds; and ß whether or not there is geographic variation in avian responses to disturbances such as fire. We also need to understand the ways in which fires differ from other natural disturbances (e.g., blowdowns, insect kills) that can be extensive and severe. In Colorado, for example, a recent wind- storm uprooted or damaged trees across 10,000 ha (Flaherty 2000) and, in 1939, an outbreak of spruce beetles (Dendrocotonus rufipennis) killed nearly 290,000 ha of trees (Veblen et al. 1991). Information required for sound management in- cludes: ß improved information on the range of natural variation both within, and among, historic fires; ß the effects of fire suppression on forest and landscape structure and wildlife communities; ß the ecological tradeoffs between wildland, prescribed fire, and mechanical treatments (in- cluding thinning and burning; Tiedemann et al. 2000, Wagner et al. 2000); ß appropriate management of post-fire forests, including how salvage treatments affect spe- cies that require post-fire habitats; ß how wildlife species respond to different stag- es of succession and whether or not those stages are similar across disturbance types; ß the responses of forests and wildlife to re- peated management treatments in the same lo- cation (Andersen et al. 1998); ß the effects of severity in natural compared to anthropogenic disturbances; ß differences and similarities among fire and forest harvesting practices and how these dis- turbances affect avian communities; and ß which management treatments are most likely to conserve the biological integrity of forest systems. RESEARCH DESIGN The inherent nature of fire limits the opportu- nities to conduct well-replicated, controlled ex- periments that evaluate the full spectrum of fire characteristics across all western forest types. Rather, we must rely on several complementary approaches, including: (1) unplanned compari- sons of wildfires (e.g., Finch et al. 1997); (2) meta-analyses that combine data from numerous studies to generate larger datasets and greater sta- tistical power (e.g., Hutto 1995); and (3) con- trolled experiments using prescribed (planned) burns, logged, logged and burned, and unburned controls. The collective results of all approaches should help us develop a greater overall under- standing of how fire affects wildlife. Most studies of fire effects on avian communi- ties have been unplanned comparisons of wildfires (Finch et al. 1997). Although variation among wildfires (e.g., forest type, burn characteristics) and post-fire management strategies, plus the lack of pre- and post-fire treatments, has limited the scope of inference provided by unplanned comparisons, they nonetheless provide unique opportunities to study extensive, severe wildfires. This approach is most useful immediately after years with extensive fire, when researchers can establish numerous, similar-age replicates across regions and in many forest types. Intensive studies of single sites can provide useful information as well, especially in large burns. For example, the effects of burn se- verity could be studied in one large burn by strat- ifying survey points across severities (e.g., R. Sal- labanks and J. Mclver, unpubl. data). In addition, single-site studies can generate data for use in meta-analyses. Meta-analyses provide excellent opportunities for improving the results of multiple studies that have little or no replication (Brett 1997). Even non-statistical compilations of unplanned com- parisons can reveal biologically meaningful trends, as we found in our review that Three- toed and Black-backed woodpeckers were either restricted to, or more abundant in, burned forests (Table la). In order for meta-analyses to be pos- sible, however, researchers must publish detailed study protocols, and they must cooperate with one another to the extent possible to standardize protocols and share data. To complement and expand the existing knowl- edge gained from unplanned comparisons and meta-analyses, we need more experiments that control for and test variations among fire charac- teristics, forest type, and landscape context (e.g., Breininger and Schmalzer 1990). Because annual variation in bird populations can be considerable, 64 STUDIES IN AVIAN BIOLOGY NO. 25 several years of pre- and post-treatment data ide- ally should be collected. Whenever possible, re- searchers should incorporate the full range of fire characteristics provided by natural fire regimes in the systems of interest (Andersen et al. 1998). It will be difficult to find sites for conducting severe bums, but there is increasing support for conduct- ing such studies in national parks (J. Connor, pers. comm.) and wildemess areas. In general, it will be more feasible to conduct experiments of low- to moderate-severity burns in systems that typically experience lower-severity fires. Research programs must also take into ac- count some important design and interpretation problems that are often ignored in fire studies. Because burn edges and burn severity may have pronounced effects on avian use of burns, sur- vey points must be stratified across burn edges, adjacent unburned forest, and distant unburned forest, and over a range of burn severities to control for these sources of variation. In addi- tion, to determine whether avian species use of post-fire habitats immediately after fire repre- sents a preference for burns, or site-tenacity for breeding territories, studies need pre- and post- fire measures of abundance, as well as measures of reproductive success and recruitment over several years. Finally, researchers need to implement long- term studies to develop a full picture of post-fire successional changes and how they affect avian communities. Although habitat loss may be the immediate effect of severe fire on species that typ- ically inhabit mature forests (e.g., Golden-crowned Kinglet, Spotted Owl), the long-term effects (e.g., decades or centuries later) may be habitat improve- ment. Thus, clearing forests of fuels to prevent se- vere fires that could decrease Spotted Owl habitat in the short term could preclude more significant habitat improvements that would benefit Spotted Owls in the future. Overall, researchers will need to consider a wide variety of research approaches, as well as the full spectrum of fire characteristics and forest types, both unmanaged and managed, to understand how proposed management strate- gies may affect the future health and integrity of western-forest systems. ACKNOWLEDGMENTS We thank W. L. Baker, W. H. Romme, and T T. Veblen for their gracious assistance in helping us un- derstand the finer points of fire terminology and ecol- ogy. H. D. Powell's expertise on insects and wood- pecker foraging were instrumental in drafting related sections of the manuscript. W. L. Baker, D. S. Dobkin, T. L. George, and J. E. Roelle all provided valuable suggestions and improvements to earlier drafts of the manuscript. J. Connor contributed to numerous discus- sions regarding the effects of fire management on bird communities in national parks. APPENDIX. SCIENTIFIC NAMES OF BIRD SPECIES Species American Kestrel (Falco sparverius) Spotted Owl (Strix occidentalis) Mourning Dove (Zenaida macroura) Common Nighthawk (Chordeiles minor) Northern Flicker ( Colaptes auratus) Lewis's Woodpecker (Melanerpes lewis) White-headed Woodpecker (Picoides albolarvatus) Black-backed Woodpecker (Picoides arcticus) Downy Woodpecker (Picoides pubescens) Three-toed Woodpecker (Picoides tridactylus) Hairy Woodpecker (Picoides villosus) Red-naped Sapsucker (Sphyrapicus nuchalis) Williamson's Sapsucker (Sphyrapicus thyroideus) Olive-sided Flycatcher ( Contopus cooperi) Western Wood-Pewee (Comopus sordidulus) Hammond's Flycatcher (Empidonax hammondii) Dusky Flycatcher (Empidonax oberholseri) PlumbeDus Vireo (Vireo plumbeus) Cassin's Vireo (Vireo cassinii) Warbling Vireo (Vireo gilvus) Tree Swallow (Tachycineta bicolor) Steller's Jay (Cyanocitta stelleri) Clark's Nutcracker (Nucifraga columbiana) Mountain Chickadee (Poecile gambeli) Red-breasted Nuthatch (Sitta canadensis) White-breasted Nuthatch (Sitta carolinensis) Pygmy Nuthatch (Sitta pygmaea) Brown Creeper ( Certhia americana) Rock Wren (Salpinctes obsoletus) House Wren (Troglodytes aedon) Ruby-crowned Kinglet (Regulus calendula) Golden-crowned Kinglet (Regulus satrapa) Hermit Thrush (Catharus guttatus) Swainson's Thrash ( Catharus ustulatus) Varied Thrush (Ixoreus naevius) Townsend's Solitaire (Myadestes townsendi) Mountain Bluebird (Sialia currucoides) Western Bluebird (Sialia mexicana) American Robin (Turdus migratorius) Yellow-rumped Warbler (Dendroica coronata) Townsend's Warbler (Dendroica townsendi) Wilson's Warbler (Wilsonia pusilia) Western Tanager (Piranga ludoviciana) Lazuli Bunting (Passerina amoena) Dark-eyed Junco (Junco hyemalis) Lincoln's Sparrow (Melospiza lincolnii) Chipping Sparrow (Spizella passerina) White-crowned Sparrow (Zonotrichia leucophrys) Pine Siskin (Carduelis pinus) Cassin's Finch (Carpodacus cassinii) Red Crossbill (Loxia curvirostra) Pine Grosbeak (Pinicola enucleator) Evening Grosbeak (Coccothraustes ve,72ertinus ) Studies in Avian Biology No. 25:65-72, 2002. GEOGRAPHIC VARIATION IN COWBIRD DISTRIBUTION, ABUNDANCE, AND PARASITISM MICHAEL L. MORRISON AND D. CALDWELL HAHN Abstract. We evaluated geographical patterns in the abundance and distribution of Brown-headed Cowbirds (Molothrus ater), and in the frequency of cowbird parasitism, across North America in relation to habitat fragmentation. We found no distinctive parasitism patterns at the national or even regional scales, but the species is most abundant in the Great Plains, the heart of their original range, and least common in the southeastern U.S. This situation is dynamic, because both the Brown-headed and two other cowbird species are actively expanding their ranges in the southern U.S. We focused almost entirely in this paper on the Brown-headed Cowbird, because it is the only endemic North American cowbird, its distribution is much wider, and it has been much more intensively studied. We determined that landscape is the most meaningful unit of scale for comparing cowbird parasitism patterns as, for example, in comparisons of northeastern and central hardwood forests within agricul- tural matrices, and suburbanized areas versus western coniferous forests. We concluded that cowbird parasitism patterns were broadly similar within all landscapes. Even comparisons between prominently dissimilar landscapes, such as hardwoods in agriculture and suburbia versus coniferous forest, display a striking similarity in the responses of cowbirds. Our review clearly indicated that proximity of feeding areas is the key factor influencing presence and parasitism patterns within the landscape. We considered intensity of landscape fragmentation from forest-dominated landscapes altered in a forest management context to fragmentation characterized by mixed suburbanization or agricultural devel- opment. Our review consistently identified an inverse relationship between extent of forest cover across the landscape and cowbird presence. Invariably, the variation seen in parasitism frequencies within a region was at least partially explained as a response to changes in forest coven The most salient geographic aspect of cowbirds' response to landscape fragmentation is the time since fragmentation occurred. Eastern landscapes generally experienced 200 years ago the development and fragmentation that western landscapes experienced less than 75 years ago. Consequently, there is a broad east-west contrast in which more numerous human settlements and smaller unbroken forest stands are found in the East, a difference that permits cowbirds to be more pervasive and ubiquitous. The locality of suitable feeding areas is a hallmark trait of the cowbirds' strategy in exploiting specific forest frag- ments. Host abundance influences parasitism patterns only secondarily at the landscape scale. These two limiting factors come into play differently in different landscapes. For example, cowbird abun- dance in unbroken forested landscapes are limited primarily by the availability of foraging areas rather than by host density, whereas cowbirds are limited primarily by host availability in landscapes that are extensively fragmented with feeding areas. Key Words: Brown-headed Cowbird; cowbird parasitism; fragmentation; geographic variation; host defense; Molothrus ater. The laying of eggs by one species in the nests of another species, allowing the host species to raise their young, is a fascinating evolutionary story (e.g., Rothstein and Robinson 1988, Or- tega 1998:37-63). In North America, the Brown-headed Cowbird (Molothrus ater) is the primary nest parasite, although two other species are expanding their ranges in the southern U.S. (Cruz et al. 1998, Ortega 1998). The trait of par- asitizing nests apparently developed in the Brown-headed Cowbird in the Great Plains. As reviewed below, this cowbird species expanded its range eastward in the 1800s and westward in the 1900s, and now occupies most states and provinces in North America (Rothstein 1994, Peterjohn et al. 2000). Parasitism, along with the cowbird's range expansion, has caused scientists to consider the role that cowbirds might be hav- ing in population declines of certain of their host species. Thus, the goal of our paper is to review cowbird abundance, distribution, and parasitism frequencies across North America so a better un- derstanding of cowbird ecology and its impact on host species can be gained. In this paper we assumed no difference in cowbird parasitism behavior by geographic lo- cation. We reviewed the literature (including un- published manuscripts and reports) in order to characterize the relationship between host and parasite. Given the striking differences in envi- ronmental conditions across North America--in- cluding the distribution of bird species--we can presuppose that one can easily find some amount of difference in the frequencies of cowbird par- asitism just by looking for it. And, in fact, we know this to be the case (see reviews by Ortega 1998, Trine et al. 1998). We were primarily in- terested in examining the process of parasitism. That is, are there fundamental differences in cowbird behavior in different regions that have ecological implications and evolutionary expla- 65 66 STUDIES IN AVIAN BIOLOGY NO. 25 nations? In our review we considered both feed- ing behavior and host selection behavior. VALIDITY OF AN EAST-WEST COMPARISON OF BROOD PARASITISM Our perception of geographic location is based in part on historic context and tradition. It is also difficult to lump large geographic areas under a common descriptor. Where does the East begin and the West end; where does the East becomes the Southeast? These geographical terms are frequently used subjectively and an- thropocentrically in ways that are not supported by ecological characteristics that affect birds. Thus, dividing North America into "East" and "West" is an inappropriate means of examining an ecological relationship such as parasitism and fragmentation. This does not mean, however, that geographic differences do not occur in land- use practices and ecological processes and in the response of animals to these practices and pro- cesses. But establishing a priori boundaries con- strains the analysis to preconceived categories and notions. THE RESPONSE OF COWBIRD HOSTS TO FOREST FRAGMENTATION In this section we set the stage for evaluating regional differences in cowbird parasitism by defining fragmentation and placing this concept into an ecological framework. The emphasis of this volume is on fragmentation, and from the perspective of cowbirds, the most important as- pects of fragmentation are, first, that it affects the abundance and distribution of host species by altering their habitat and, second, that it alters the abundance and distribution of feeding areas associated with developments. These twin themes about the influence of fragmentation on hosts' breeding habitat and on feeding areas of cowbirds associated with human development recur throughout our review. The classic description of fragmentation im- plies extensive landscapes of homogeneous veg- etation, but this conception is an artifact of graphic art framed at a large spatial scale. Ex- amined at finer resolutions, most ecological sys- tems are actually a mosaic of different plant as- sociations. Even changes of a few meters can change soils, slope, and aspect, and thus the as- sociated plants. Further, these mosaics are dy- namic and change, often rapidly, through suc- cession, catastrophic events (e.g., fire, flood, wind), or development activities such as crop plantings or settlements (Meffe and Carroll 1997:274-275; Franklin et al. this volume). The definition of "fragmented" habitat de- pends upon the spatial scale of observation. Our analyses use fragmentation at a scale relevant to selection of habitat by birds, particularly song- birds. Briefly, habitat selection is often viewed as a hierarchical process where individuals first select a broad geographic range, a decision that is largely innate. Within the geographic range the individual then makes a series of decisions based on increasingly refined combinations of vegetation structure, floristics, food resources, and nest sites (Johnson 1980, Hutto 1985). Thus, in an analysis of brood parasitism, frag- mentation is an ambiguous concept unless it is defined in spatial terms relevant to the series of responses a host makes. There are changes that take place in the environment at several scales of resolution (see also Angelstam 1996). Such descriptions of the environment and habitat se- lection are not restricted geographically, but should apply across eastern and western envi- rons. Consequently, we would not expect diffEr- ent behavioral processes in either host species or cowbirds to be operating geographically. The proportion of birds that show a particular re- sponse to fragmentation (e.g., area sensitive, en- hanced by edge) may differ geographically de- pending on the historic factors that formed the initial bird assemblage (e.g., Morrison et al. 1998:16-26). For example, fewer Dendroica warblers occur in the West than in central and eastern locations. This is apparently the result of Pleistocene and post-Pleistocene events (Mengel 1964, Morrison et al. 1998:18-21). Thus, there is simply a greater opportunity for fragmentation to cause negative impacts on these warblers in more eastern locations, and perhaps a propor- tionally more apparent impact to the bird assem- blage due to fragmentation. Fragmentation in managed forests can be con- sidered dynamic in that stands are cut and re- forested; stands are not retained in early succes- sional conditions. This means the songbird com- munities that cowbirds parasitize continue to have extensive natural breeding habitat although the vegetation communities are less stable than they would be in unmanaged forest. In contrast, disturbances due to human development activi- ties result in permanent or static fragmentation (McGarigal and McComb 1995). This eradicates some host-breeding habitat, leaving disjunct fragments separated by patches that have food for cowbirds. They concluded that it is unlikely that the empirical findings on forest lYagmenta- tion from urban and agricultural landscapes ex- tend to the dynamic forest landscapes of the Pa- cific Northwest and elsewhere. Likewise, Keller and Anderson (1992) concluded that fragmen- tation in Wyoming could not be directly com- pared with fragmentation occurring in the Pacif- ic Northwest. Freemark et al. (1995) also noted that most studies in the West have been con- GEOGRAPHICAL VARIATION IN COWBIRDS Morrison and Hahn 67 TABLE 1. COMPARISON OF THE EFFECTS OF LANDSCAPE STRUCTURE ON NEOTROPICAL MIGRATORY SPECIES BREED- ING IN NORTHEASTERN AND CENTRAL HARDWOOD FORESTS WITHIN AGRICULTURE AND SUBURBANIZED LANDSCAPES VERSUS WESTERN FORESTS Landscape structure Northeastern and central vs. western comparison Landscape composition Forest type Same Forest cover Same; less severe in west Habitat proportion Same Landscape configuration Patch size Same; perhaps less severe in west Patch shape N/A a Interpatch distance Same Nonforest edge N/A Habitat juxtaposition Same Note: Information summarized from Freemark et al. (1995). a N/A - comparison not iliade or conlparable. ducted in forested landscapes fragmented by sil- vicultural activities--which usually do not have rich food sources for cowbirds--rather than in agricultural and urban landscapes as in the East, which do include sources of food (see also Hejl et al. this volume). Yet, Freemark et al's. (1995) extensive liter- ature review of the response of breeding com- munities of neotropical migrants to landscape structure across much of North America does show similarities in songbird responses. A sub- jective comparison of communities nesting in northeastern and central hardwood forests within agricultural and suburbanized areas with com- munities nesting in western coniferous forests revealed similar responses of birds to broad measures of landscape structure (Table 1). Par- ticularly because this is a comparison among very dissimilar landscape settings (i.e., hard- woods within agriculture and suburbia versus managed coniferous forest), the similarity in re- sponse by breeding birds is striking. Although there are similarities in the respons- es of host communities in different regions to fragmentation, Freemark et al. (1995) concluded that birds in western (coniferous) forests have not shown as strong a negative response to frag- mentation as have birds in northeastern and cen- tral hardwood forests. They attributed this sev- eral factors: fragmentation is a more recent oc- currence in the West; fragmentation has rarely resulted in habitat isolation; and western forests are naturally fragmented and human-induced fragmentation has not had time to negatively im- pact birds. The key insight here is that there are not inherent differences in the response of bird communities to forest fragmentation. The earlier stage of fragmentation typical of western forest means that many western coniferous forests are actually "perforated" rather than fragmented (Forman and Collinge 1996), or, as Freeman et al. (1995) described them, "punctuated" by clearcuts. Of course there are also numerous ex- amples of both extensively forested areas and forests perforated by logging and agriculture outside of western environs (e.g., Robinson et al. 1995a, Robinson and Robinson 1999). McGarigal and McComb (1995), working in the Oregon Coast Range, found that landscape structure (composition and configuration) ex- plained <50% of the variation in each species' abundance among the landscapes. Species' abundances were generally greater in areas with a relatively fragmented distribution of habitat. Note that from the cowbird's perspective this means host abundance increases as fragmenta- tion progresses. They cautioned, however, that species sensitive to fragmentation at the scale of their study may have been rare already and therefore not subject to the approach they used. Again from the cowbird's perspective the spe- cies that drop out do not reduce the number of host individuals available to cowbirds. They concluded, however, that their results were gen- erally similar to studies conducted in forest- dominated landscapes in New Hampshire, Mis- souri, Maine, and Wyoming. Thus, when com- parisons are made between similar vegetation types, birds respond in a similar manner across broad geographic regions. They noted that ef- fects of fragmentation in forest-dominated land- scapes altered in a forest management context is not comparable with fragmentation caused by urbanization or agricultural development, which is typically how eastern and western regions have been compared in the literature. In conclusion, the same ecological processes associated with fragmentation seem to operate regardless of geographic region. It is the longev- ity of those land-use changes that precipitated fragmentation that causes any geographic differ- ences in current responses by birds. Verner 68 STUDIES IN AVIAN BIOLOGY NO. 25 (1986) concluded that in western forests frag- mentation was in the early stage and tended to produce two-dimensional islands (clearcuts) in three-dimensional seas (forests), while in eastern forests (as in European forests) the later stages of fragmentation have resulted in three-dimen- sional islands (forest fragments) in two-dimen- sional seas (e.g., agricultural lands). Askins et al. (1990) likewise concluded that the longer his- tory of fragmentation in Europe has resulted in the extirpation of most area-sensitive species, a situation now in progress in North America. The localized abundance, breeding success, and sur- vival of birds is related primarily to factors of habitat quality such as resource availability and predator-competitor activity, but these factors can be overridden when patches becomes very small (<10-20 ha) and isolated. In summary, landscape fragmentation affects the songbird communities that cowbirds parasit- ize. At one level of intensity, fragmentation re- fers to the transformation of extensive forests into smaller stands, with the consequence for cowbird hosts of smaller, often shifting, breed- ing areas, and habitats with a greater edge to interior ratio. As fragmentation progresses, it evolves to a heterogeneous landscape composed of a mix of patches of breeding habitat with patches of development activities such as agri- culture and settlements. With these twin aspects of fragmentation--smaller forest stands and in- creasing food sources associated with develop- ment--an increase in cowbird abundance and parasitism is likely. HISTORIC DISTRIBUTION OF BROWN-HEADED COWBIRD AND POPULATION TRENDS Peterjohn et al. (2000) described the continen- tal decline in cowbird numbers in North Amer- ica since the mid-1960s. Maximum cowbird abundance occurs in the northern Great Plains. Regionally, numbers are declining in the south- ern plains and throughout most of the East. The decline in the East is attributed to substantial increases in forest cover. There appears to be an overall steady abundance of cowbirds in the West. Within the region there is perhaps a slight decrease in the Pacific Northwest, while the Central Valley of California showed perhaps the greatest proportional increase in cowbird num- bers in North America. While there is consensus that the ancestral range of cowbirds in the Great Plains is still the area of their greatest abundance, other aspects of the extent and timing of their range expan- sions both eastward and westward are less cer- tain. Rothstein (1994) suggested that cowbirds have been in the East in small numbers since at least the 1700s, the earliest era of European col- onization. In the West, cowbirds may not be re- cent additions to the avifauna. While their col- onization up the Pacific Coast from southern California to Oregon and Washington has been well documented over the course of the 20th century, there is also evidence of earlier popu- lations in the northwest (Rothstein 1994). They apparently occurred historically, however, across the Great Basin to the eastern edge of the Sierra Nevada (Rothstein 1994). Thus, contrary to pop- ular belief, the cowbird did occur historically in western North America. The Sierra Nevada-Cas- cade mountain ranges may have served as a bar- rier to widespread expansion onto the Pacific slope. There is also fossil evidence that cowbirds (of unknown breeding behavior) occurred along the edges of the species' current range in Cali- fornia, Oregon, and Florida in the late Pleisto- cene (Lowther 1993). Chace and Cruz (1999) suggested that cowbirds formerly ranged to near timberline in the Rocky Mountains because of the historic presence of bison (Bison bison). Cowbirds retreated from these elevations with the extirpation of bison from these mountains. The addition of cattle to former bison range is now allowing cowbirds to return to the moun- tains. If this is the case, we would expect that birds in at least some regions of the Rocky Mountains have had a longer exposure to cow- birds than our recent data indicate, and they may still express behavioral traits that evolved during the bison-cowbird period. SUBSPECIES DIFFERENCES Differences among the three subspecies of the Brown-headed Cowbird have been little studied. Rothstein (1994) speculated that the smaller southwestern subspecies, the "dwarf" cowbird, M. a. obscurus, might be more vagile or more competitive than M. a. artemisia, forrod to the north, east of the Rockies, because the westward range expansion of the species to the Pacific md up the west coast seems to have been driven by obscurus. At some point later artemisia appears also to have crossed the Rockies into northern California such that the two have subsequently intermixed as cowbirds moved north into Oregon and Washington. Recent evidence of the range expansion of the eastern subspecies M. a. ater into the Florida peninsula makes it feasible that ater may be as successful as obscurus was in colonizing the Pa- cific west coast. Cruz et al. (1998) noted that ater has spread rapidly since the 1950s and now has breeding records confirmed halfway down the peninsula, with non-breeding sightings re- ported throughout the state. The expansion of the Brown-headed Cowbird into Florida is ex- GEOGRAPHICAL VARIATION IN COWBIRDSMorrison and Hahn 69 pected to have significant negative consequences for the indigenous breeding passerines, many of which are patchily distributed and breeding in small populations. The character of natural hab- itats and human settlements in Florida consists of mangrove on the west coast and dunes and beach on the east coast, with relentless human settlement along both coasts. The central section of the peninsula is higher and drier and agricul- tural and livestock developments are pervasive. Two mangrove-obligate species, the Black-whis- kered Vireo (Vireo altiloquus) and the Florida subspecies of Prairie Warbler (Dendroica dis- color), are already reflecting local population extirpation due to parasitism (W. Pranty, pers. comm.). OTHER COWBIRD SPECIES: RECENT NORTH AMERICAN INVADERS While it is only speculative to compare the invasive character of Shiny (Molothrus bonar- iensis) and Bronzed (M. aeneus) cowbirds to Brown-headed Cowbirds at this stage, recent de- velopments in their respective range expansions suggest that both may be successful and increas- ingly widespread in the United States. Both are also host generalists, although perhaps not as ex- treme as the Brown-headed Cowbird (Rothstein et al. 2002). The rapid and impressive northward range expansion of the Shiny Cowbird across the Caribbean and into North America makes it a likely candidate to become established in the southeastern U.S. in the next few decades. While no breeding records have yet been recorded in Florida, the Shiny Cowbird is expected to be- come established there with little difficulty (Ste- venson and Anderson 1994; W. Pranty, pers. comm.). Nothing is known about the extent of habitat specialization for either Brown-headed or Shiny cowbird within Florida. The Bronzed Cowbird has only recently shown marked range expansion, apparently in association with loss of songbird breeding hab- itat in lower Rio Grande Valley in Texas. How- ever, it has expanded both eastward and west- ward and could thus become a factor in regions of the U.S. (Cruz et al. 1998). In Texas, the Bronzed Cowbird parasitizes over 23 species, and at this stage it appears to prefer larger host species than does M. ater. The bronzed is thought to have contributed to the extirpation of Audubon's Oriole (lcterus graduacuada) in por- tions of lower Rio Grande Valley. Together with the brown-headed, the bronzed may also have contributed to declines of the Orchard (1. spur- ius), Hooded (I. cucullatus), and Northern (1. galbula) orioles in south Texas (Cruz et al. 1998). HOST BEHAVIOR AND GEOGRAPHY Much interest has focused on the question why most host species of the Brown-headed Cowbird do not show effective anti-parasite be- havior. Rothstein's (1975) early experimental study of twelve eastern species used artificial eggs colored to resemble cowbird eggs and showed that only a few species regularly ejected the parasite eggs. Since then a large number of studies have been conducted in a variety of sites both east and west, showing that parasitism de- fenses (i.e., egg ejection, egg burial, or nest de- section) occur occasionally and unpredictably among species. Some western-residing species and subspecies show eftkctive anti-parasite behaviors that pre- vent or minimize deleterious eftcts of parasit- ism, which may have developed after contact with cowbirds, or which may have been present as pre-adaptation. For example, the Black- throated Gray Warbler (Dendroica nigrescens) regularly buried cowbird eggs in its nests in the Inyo-White mountains of eastern-central Cali- fornia (J. Keane and M. Morrison, unpubl. data), and Rich and Rothstein (1985) showed that Sage Thrashers regularly rejected cowbird eggs throughout their western range. Egg-ejection behavior is one of the best-stud- ied anti-parasite behaviors, yet a thorough sum- mary of the proportion of acceptor and rejecter species by geographic region is still lacking be- cause that would require systematic comparative studies of dirtbrent populations of a large num- ber of host species. Although evidence for egg rejection exists for many species, the quantita- tive estimates of frequency of this behavior can usually only be confirmed through experimen- tation, usually with artificial eggs (Ortega 1998: 19). Of the >225 species known to be parasit- ized by Brown-headed Cowbirds, fewer than 20 are known to regularly eject parasitic eggs (Or- tega 1998:19-20). Despite the obvious advan- tages to hosts of removing cowbird eggs, there are also many reasons why birds accept them (Ortega 1998:23-27). The most prominent rea- son is that parents risk breaking their own eggs when they try to move the cowbird egg. Little is known about the degree to which egg-ejection behavior is genetically based or learned. Briskie et al. (1992) concluded that some anti brood-parasitic defenses are probably genetically determined. Robertson and Norman (1977) thought that the presence and intensity of aggression should vary widely geographically depending on the length of exposure to brood parasitism. For example, they compared aggres- sion in an area of long-term host-cowbird sym- patry (Manitoba) with an area (Ontario) of more 70 STUDIES IN AVIAN BIOLOGY NO. 25 recent sympatry. They found that the Manitoba host populations showed more aggression to- wards a model cowbird, and concluded that this was because of the longer history of sympatry. Hobson and Villard (1998) studied the response of American Redstarts to model cowbirds in western Canada and found that they exhibited more vigorous nest defense in fragmented for- ests where cowbirds are more common than in extensively forested landscapes. There is a widespread assumption that all hosts would evolve measurable anti-parasite be- haviors given long enough sympatry with cow- birds. According to this hypothesis, some spe- cies along the Pacific slope may not have had adequate exposure to parasitism to evolve reg- ular ejection behavior (Rothstein 1975). As dis- cussed above, however, additional evidence must be gathered before any analysis of geo- graphic trends in egg-rejection behavior. We suggest that the variability and relative rarity of anti-cowbird defenses reflects the inconsistent selection pressure exerted by cowbird parasitism in those landscapes where parasitism is relative- ly low and where the level of parasitism on in- dividual species and communities varies from year to year. In several areas where long-term studies of cowbird parasitism have been con- ducted and where parasitism pressure is both high and consistent on particular species in the community (such as central Illinois, the Edwards Plateau in Texas and Oklahoma, and southern California), the study populations should be tracked for the emergence of anti-parasite be- haviors. Similarly, the evolution of defenses by forest interior birds should be watched in the context of fragmentation in both east and west. COWBIRD PARASITISM AND GEOGRAPHY We present a summary of patterns of cowbird parasitism in relation to vegetation structure, host community, and degree of landscape de- velopment based on studies conducted across North American a variety of vegetation types in different geographic regions (Table 2). Our review indicates that proximity of feed- ings areas is the key factor influencing which host community a local cowbird population will parasitize. Although Payne (1973, 1977) dis- cussed the importance of temporal mismatch of breeding seasons (i.e., differing lengths of ex- posure, sensu Mayfield 1965) and documented the phenomenon for the birds of northern Cali- fornia, temporal mismatch is often overlooked. It is a notable phenomenon in eastern and west- ern locations. The local abundance of cowbirds resulting from fragmentation and feeding oppor- tunities further correlated with parasitization (Payne 1973, 1977). It is commonly stated that the heavily para- sitized riparian communities in the western and southwestern United States are physiographical- ly unique because of the often abrupt change from the relatively roesic riparian vegetation and the xeric surrounding landscape (Ortega 1998: 267, Farmer 1999). However, cowbirds fre- quently use riparian areas in eastern and central, as well as western regions for passage, nesting, and foraging. Riparian corridors allow passage by cowbirds into an otherwise less suitable land- scape matrix, including both eastern and western forests. The primary development impact to western riparian areas is loss of area and frag- mentation (isolation), which is the same pattern seen in eastern deciduous forests (i.e., isolated patches of forest in a matrix of different vege- tation). Several riparian obligate species in the West and Southwest have been nearly extirpated because of habitat loss. The isolation of these species into small patches exacerbated the effect of cowbird parasitism on their host populations. This situation, however, is not restricted to ri- parian vegetation of the West and Southwest. In three eastern regions where small and restricted species or subspecies occur in conjunction with a unique and limited habitat, development has created the classic situation in which cowbird parasitism (and nest predation) accelerate the de- cline of the resident species. In northern Mich- igan, in jack pine (Pinus banksiana) habitat, the species at risk is the Kirtland's Warbler (Den- droica kirtlandii). In the coastal mangrove for- ests of Florida, the species at risk are Black- whiskered Vireo and Prairie Warbler (Cruz et al. 1998, Stevenson and Anderson 1994). In Central Texas and Oklahoma, on the Edwards Plateau, the species at risk are the Golden-cheek Warbler (Dendroica chrysoparia) and Black-capped Vir- eo (Vireo atricapillus). VALIDITY OF GEOGRAPHICAL COMPARISONS OF COWBIRD PARASITISM One of the most important aspects of geog- raphy in analyzing the impact of cowbirds is the use of different spatial scales. Robinson (1999) noted that cowbird ecology can be analyzed at continental, regional, and landscape scales as much as at a local scale in relation to factors such as distances from edges. In this section, we discuss the findings of investigators who ana- lyzed patterns at different scales. Hochachka et al. (1999) emphasized that investigators must define the scale they are using when predicting cowbird abundance and parasitism level. Several investigators have considered whether aspects of cowbird parasitism vary on a conti- GEOGRAPHICAL VARIATION IN COWBIRDS--Morrison and Hahn TABLE 2. FACTORS CORRELATED WITH INCREASED COWBIRD PRESENCE, ABUNDANCE, OR PARASITISM 71 Factor Location Source Temporal mismatch Proximity of feeding Local stand factors a Presence of riparian corridor Host density Species richness Fragmentation Original range E. Washington 1 Arizona/California 11 E. Washington 1 N. Rockies 2, 3, 5 Sierra Nevada 6, 10 N. Michigan 7 Midwest 8, 13, 14, 15 Vermont 9 Florida 16, 17 New Mexico 18 Texas 19 Pennsylvania 20 Virginia 21 N. Michigan 7 New York 22 N. Rockies 2, 3, 5 Coastal California 4 Southern California 23 Sierra Nevada 10 Missouri 12 N. Rockies 2, 5 Midwest 13 Nationally 24 Sierra Nevada 6 Illinois 8 Arizona/California 11 a Florida 17 Tennessee 25 Nationally 26 Northeast 27 Sources: 1: Vander Haegen and Walker (1999); 2: Young and Hutto (1999); 3: Hejl and Young (1999); 4: Farmer (1999); 5: Tewksbury et al. (1999); 6: Purcell and Verner (1999); 7: Striblcy and Haultier (1999); 8: Robinson et al. (1995a); 9: Coker and Capen (1995. 1999); 10: Lynn ctal. (1998); 11: Rosenberg et al. (1991:265, 335); 1 la: Rosenberg et al. (1991:282-283); 12: Thompson et al. (1992); 13: Donowm et al. (1997); 14: Thompson (1994); 15: Trine et al. (1998); 16: Cruz et al. (1998); 17: W. Pranty, pets. comm.; 18: Gogucn and Mathews (1999); 19: Eckrich et aL (1999); 20: E. Morton, pets. COlrim.; 21: J. Kan; pets. comm.; 22: Hahn and Hatfield (1995); 23: Kus (1999); 24: Hahn and O'Connor (2(X)2); 25: Miles and Buehler (1999); 26: Smith and Myers-Smith (1998); 27: Hoover and Brittingham (1993). a When in close proximity to feeding areas. nental scale (Smith and Myers-Smith 1998, Robinson 1999). At a national scale, Hahn and O'Connor (2002) found that the most important factor predicting cowbird abundance is the pres- ence of their preferred mix of host species (i.e., the seventeen most common hosts identified by Friedmann [1963]; Table 2). Landscapes in which host communities are found in close prox- imity to feeding areas typically occur where considerable habitat fragmentation occurs, that is, intrusion of agricultural activities, including concentrated livestock grazing, into a formerly undisturbed area. When they examined ancestral versus invaded ranges separately, they found that the predictive value of these host species actually operated only in the invaded ranges. Robinson et al. (1995b) suggested that because some western coniferous forests are more open than eastern forests, it was unclear whether or not western and eastern cowbirds differed in their preferences for forests, or if host distribu- tion or some other factors influenced habitat oc- cupancy by cowbirds. Our review indicates that the relationship between the openness of forests and cowbird abundance holds regardless of re- gion. In fact, the variation seen in parasitism rates within a region was at least partially ex- plained as a response to changes in forest cover. Further, many western forests have interlocking canopies with dense understories (e.g., Pacific Northwest, many western riparian forests). Again, sweeping generalizations regarding East and West seem unwarranted. Hochachka et al. (1999) evaluated the rela- tionship between torest coverage and parasitism among eastern, central, and western regions of the United States to provide a biological expla- nation for differences in the relationship tween forest coverage and rates of cowbird par- asitization across the continent. They also ex- amined if variation in forest coverage was as- sociated with the presence or absence of 72 STUDIES IN AVIAN BIOLOGY NO. 25 cowbird parasitization in a study area, and, where cowbirds were present, if the frequency at which nests were parasitized was associated with forest coverage. They obtained data on par- asitization rates of forest birds from the Breed- ing Biology Research and Monitoring Database (BBIRD), with data from 23,448 individual nests being analyzed. There were 26 study sites on which the nesting success of forest-nesting birds was monitored. Hochachka et al. (1999) reported that the con- clusions of previous research suggested that larger proportions of forest cover will result in a lower impact of Brown-headed Cowbirds on their hosts. They further suggested that the re- lationship between forest coverage and parasiti- zation might differ away from the Midwest for a number of reasons. They offered that variation in cowbird abundance may not only affect ab- solute rates of parasitization, but also the pattern of variation in parasitization rate with varying forest coverage. Cowbirds in different parts of the continent encounter communities of hosts with different lengths of exposure (e.g., May- field 1965) and responses (e.g., Briskie et al. 1992) to parasitization, and host species with longer exposure to cowbirds may be resistant to parasitization regardless of the proportion of for- est in a landscape. This appears true, but we do not see any evidence of this varying predictably by region in our review--all host responses are seen across the country, and all responses were seen within different localities within a region. Hochachka et al. (1999) continued that the re- lationship between cowbird parasitization and forest coverage may also vary as a function of the local area over which forests were measured. Within local areas, forest coverage varied in its power to predict parasitization, depending on the size of the area over which forest coverage was measured (Tewksbury et al. 1998, Donovan et al. 2000). It is clear that vegetated patches sur- rounded by agriculture are different than those surrounded by more forest; this holds regardless of region. Hochachka et al. (1999) failed to find any substantial differences in the behavior and hab- itat requirements among the races of Brown- headed Cowbirds (Lowther 1993). They con- cluded that although cowbird abundance de- clined westward--away from the center of the cowbird's range--the lower abundance of cow- birds in the West should result in a lower rate of parasitization, but not in a complete reversal of the relationship between parasitization rate and forest coverage. In the analyses by Ho- chachka et al. (1999), we see the importance of examining parasitization in a spatially explicit manner. Local factors, such as presence of ag- riculture and patch size, will usually override relatively region-wide factors, such as absolute forest coverage and host density, in determining parasitization rates. Our review shows that the major factors determining the impacts of cow- birds on hosts operate continent-wide (Table 2). Fragmentation increases the degree of local sympatry between cowbird and host. Peterjohn et al. (2000) found no evidence to suggest that changes in cowbird populations differentially in- fluenced population changes in cowbird hosts and rejecter species. Trends from BBS data showed that both cowbird host species and spe- cies rarely parasitized showed the same pattern of direct association with trends in cowbird abundance, and all of the correlations were low. The general direct relationship between cowbird trends and trends of neotropical migrants reflect- ed the broad regional patterns of increasing bird populations in western North America and de- clines in the southern United States. They con- cluded that large-scale changes in weather pat- terns, land use practices, and habitat availability were primarily responsible for the direct asso- ciations they found between population trends in cowbirds and their host species. The strong in- fluence of weather was also used by Johnson (1994) to explain the numerous range expan- sions of western birds. Lowther (1993) concluded that fragmentation of eastern deciduous forest leads to increased parasitism by cowbirds. Further, he summarized that similar patterns were becoming evident in western montane areas as human settlement ex- pand. We agree, and conclude that geographic differences in the response of birds to fragmen- tation-and thus our characterizations of the as- semblage of birds in different locations (e.g., species richness)--are largely determined by the time since fragmentation occurred, rather than any inherent differences in the response. Cow- birds respond in distribution to fragmentation first by the location of suitable feeding areas, and secondarily to host abundance. As aptly summarized by Robinson et al. (1995a), cow- birds in heavily forested landscapes appear lim- ited primarily by the availability of foraging ar- eas rather than by host density. In fragmented landscapes, however, cowbirds appear limited primarily by host availability because feeding areas are readily available as a result of the frag- mentation. ACKNOWLEDGMENTS We thank the editors of this volume for inviting our presentation and for critical reviews of several drafts. We also thank additional comments provided by sev- eral anonymous referees. Studies in Avian Biology No. 25:73-80, 2002. EFFECTS OF FOREST FRAGMENTATION ON BROOD PARASITISM AND NEST PREDATION IN EASTERN AND WESTERN LANDSCAPES JOHN E CAVITT AND THOMAS E. MARTIN Abstract. The fragmentation of North American forests by agriculture and other human activities may negatively impact the demographic processes of birds through increases in nest predation and brood parasitism. In fact, the effects of fragmentation on demographic processes are thought to be a major underlying cause of long-term population declines of many bird species. However, much of our understanding of the demographic consequences of fragmentation has come from research conducted in North America east of the Rocky Mountains. Thus, results obtained from these studies may not be applicable to western landscapes, where habitats are often naturally heterogeneous due to topographic variation and periodic fire. We utilized data from a large database of nest records (> 10,000) collected at sites both east and west of the Rocky Mountains to determine if the effects of fragmentation are consistent across broad geographic regions. We found that forest fragmentation tended to increase the frequency of brood parasitism by Brown-headed Cowbirds (Molothrus ater) east of the Rockies but we were unable to detect a significant difference in the West. Within the eastern United States, nest predation rates were consistently higher within fragmented sites relative to unfragmented sites. Yet, in the West, fragmentation resulted in a decrease in nest predation relative to unfragmented sites. This is perhaps accounted for by differential responses of the local predator community to fragmentation. Our results suggest that the effects of fragmentation may not be consistent across broad geographic regions and that the effects of fragmentation may depend on dynamics within local landscapes. Key Words: brood parasitism; forest fragmentation; nest predation; Western North America. Forest fragmentation occurs when large, contin- uous, forested tracts are converted to other veg- etation types or land uses so that only a few scattered fragments remain (Faaborg et al. 1995). Fragmentation is a characteristic feature of most human dominated landscapes (Burgess and Sharpe 1981) and is particularly evident in portions of northern Europe and eastern North America (east of the Rocky Mountains) where agricultural production and urban development have reduced once contiguous forests into small, and often isolated patches (Andrn 1992, Don- ovan et al. 1995b, Robinson et al. 1995a). For the past several decades considerable at- tention has been given to the effects of forest fragmentation on avian populations within North America because of widespread population de- clines (Gates and Gysel 1978, Ambuel and Tem- ple 1983, Wilcove 1985, Askins et al. 1990, Robinson et al. 1995a). The fragmentation of once continuous forests may result in both a quantitative and qualitative loss of habitat for species (Faaborg et al. 1995). Fragmentation can negatively influence avian populations by reduc- ing the total area of native vegetation resulting in the extinction of some species. In addition, as an area is fragmented into increasingly smaller patches, the amount of edge relative to interior area increases. This exposes populations to the conditions of a different surrounding ecosystem and consequently to what are known as "edge effects" (Murcia 1995). Research conducted to date suggests several characteristics of forest fragments that may negatively affect avian pop- ulations. Small forest patches with a high edge to interior ratio have: (1) High rates of nest pre- dation. The abundance of avian and mammalian nest predators (avian and mammalian) often are higher along forest edges than within the forest interior (e.g., Gates and Gysel 1978, Chasko and Gates 1982, Hanski et al. 1996). (2) High rates and intensities of brood parasitism. The Brown- headed Cowbird (Molothrus ater) is often more abundant along forest edges, and nests adjacent to edges typically have higher rates of parasitism (Donovan et al. 1995b, Robinson et al. 1995a, Young and Hutto 1999). (3) Reductions in pair- ing success. Several species within forest frag- ments and near forest edges have a reduced chance of attracting mates than when in large continuous forests and within the forest interior (Wander 1985, Gibbs and Faaborg 1990, Villard et al. 1993, Burke and Nol 1998). (4) Lower food availability for breeding birds. Burke and Nol (1998) demonstrated that invertebrate bio- mass was lower within forest fragments than large continuous forests. These fragmentation effects are thought to be a major underlying influence of long term pop- ulation declines of many birds, particularly for- est-interior species within eastern North Ameri- ca (Whitcomb et al. 1981, Robbins et al. 1989b, Sauer and Droege 1992, Ball et al. 1994). Con- sequently, many small forest fragments in east- ern North America support few if any forest- 73 74 STUDIES IN AVIAN BIOLOGY NO. 25 interior species (Robbins et al. 1989b, Freemark and Collins 1992). Concern over avian population declines and the potential demographic consequences to frag- mentation have led to numerous studies de- signed to examine the potential effects of forest fragmentation on avian productivity. Previous studies have suffered from two major problems. First, studies of fragmentation effects have often depended on data from artificial nests, which of- ten do not reflect rates or patterns of predation on real nests (e.g., Major and Kendal 1996). Studies using artificial nests also cannot provide information on the rates and patterns of cowbird parasitism. Second, much of our current under- standing of the demographic consequences of fragmentation has come from research conduct- ed east of the Rocky Mountains (George and Dobkin this volume). Because most fragmenta- tion studies are conducted over a relatively small geographical area (but see Donovan et al. 1995b, Robinson et al. 1995a), often with no replica- tion, the results cannot be generalized to other locations or regions. The effects of forest frag- mentation within eastern North America may not automatically be applied to the West for sev- eral reasons. Unlike once contiguous eastern for- ests, forests west of the Rocky Mountains have a naturally heterogeneous pattern due to topo- graphic variation, periodic fire, flooding and oth- er climatic events (Franklin et al. this volume, Hejl et al. this volume). Thus, human induced fragmentation in the West (e.g., logging) may not have yet created sufficiently different land- scape patterns to affect avian populations (Hejl 1992, Freemark et al. 1995, Hejl et al. this vol- ume). Unlike fragmentation in eastern North America, fragmentation in the West is a rela- tively recent phenomenon and thus there may not have been sufficient time for birds to re- spond (Rosenberg and Raphael 1986). Addition- ally, the pattern of nest predation may not be comparable between regions because local pred- ator communities likely diffen Large predators found in western North America, but largely ab- sent in the East, may keep mesopredator popu- lations in check (Soulfi 1988, Rogers and Caro 1998). Thus, the effects of fragmentation on avi- an demographic processes in the East may not apply to western North America. In this paper, we utilized data from 20 repli- cated study sites to examine the effects of forest fragmentation on the reproductive success and nest predation rates of a suite of forest nesting species breeding at sites east and west of the Rocky Mountains. We also examined if forest fragmentation affects the frequency (number of nests parasitized) and intensity (number of par- asite eggs laid per nest) of brood parasitism dif- ferently in eastern versus western sites. Finally, we review the available literature on the effects of fragmentation on nest predation by geograph- ic region (east vs. west). METHODS We used nesting data from 10,446 nests (103,855 days of exposure) of 23 species of open nesting pas- serines (Table 1). The data used in these analyses come from the Breeding Biology Research and Monitoring Database, a collaborative effort in which researchers monitor avian breeding productivity and habitat con- ditions using standardized sampling protocols (Martin et al. 1997) at sites located throughout the continental U.S. Data were utilized from 20 study sites located east and west of the Rocky Mountains (Fig. 1). Examina- tion of Figure 1 illustrates that sites were not evenly distributed across North America and include a group- ing centered along the Mississippi River and a group- ing along the western side of the Rocky Mountains. For simplicity we refer to sites east of the Rocky Mountains as eastern sites and those along the western side of the Rockies as western sites. Each site utilized was replicated and composed of 4 to 30 separate study plots. Sites were chosen from the database for this analysis if the principal investigator designated them as either largely fragmented by human activities (ag- riculture or logging), or unfragmented. Because our classification of sites is subjective, we also calculated the proportion of forest within a 10-km radius of each study plot from a GIS layer produced by the USDA Forest Service covering the entire United States. A 10- km radius was chosen because this area relates well to distances most cowbirds commute between breeding and feeding areas (Thompson 1994, Thompson and Dijak 2000), and previous studies have used this area as a simple measure of forest fragmentation (Robinson et al. 1995a, Donovan et al. 1995b, Hochachka et al. 1999, Thompson et al. this volume). Forest coverage was calculated using FRAGSTATS (McGarigal and Marks 1995). Three unfragmented sites in the east and three in the west were paired with a nearby fragmented site to ex- amine local landscape-level effects of fragmentation on daily mortality rates (Table 2). Species were chosen for the analysis if they satisfied all three of the follow- ing criteria: (l) they are open nesting passerines that primarily nest in forest habitats, (2) the total number of nests available for each species was greater than 50, and (3) the species were recorded breeding at more than one site. All statistical analyses were conducted using PC-SAS (SAS Institute 1998). Tests were para- metric unless transformations of the data could not meet assumptions of normality and homogeneous var- iances. Results from statistical tests are referred to as significant when P < 0.05. Values reported in the RE- SULTS section are means + SE. REPRODUCTIVE SUCCESS We examined the effects of fragmentation on com- ponents of reproductive success by performing paired t-tests on mean clutch size and mean number of off- spring fledged per nest, blocking by species and testing for habitat diftbrences. Because cowbirds often remove host eggs before parasitizing nests (Nolan 1978), we PARASITISM, PREDATION, AND FRAGMENTATION--Cavitt and Martin TABLE 1. FOCAL SPECIES USED IN ANALYSES 75 Colllnlon nanle Scientific name Nest placement Number of nests Eastern Wood-pewee Contopus virens Tree 169 Western Wood-pewee Contopus sordidulus Tree 264 Acadian Flycatcher Empidonax virescens Tree 1624 Blue-gray Gnatcatcher Polioptila caerulea Shrub 210 Wood Thrush Hylocichla mustelina Shrub 814 Swainson's Thrush Catharus ustulatus Shrub 162 Veery Catharus fuscescens Shrub 100 American Robin Turdus migratorius Shrub 1461 Cedar Waxwing Bombycilla cedrorum Tree 163 Warbling Vireo Vireo gilvus Tree 468 Red-eyed Vireo Vireo olivaceus Shrub 673 Yellow Warbler Dendroica petechia Tree 1276 Kentucky Warbler Oporornis formosus Ground 115 Hooded Warbler Wilsonia citrina Shrub 363 Worm-eating Warbler Helmitheros vermivorus Ground 286 Ovenbird Seiurus aurocapillus Ground 411 American Redstart Setophaga ruticilla Tree 335 Northern Cardinal Cardinalis cardinalis Shrub 307 Indigo Bunting Passerina cyanea Shrub 492 Black-headed Grosbeak Pheucticus melanocephalus Tree 180 Song Sparrow Melospiza melodia Shrub 218 Northern Oriole Icterus galbula Tree 65 Western Tanager Piranga ludoviciana Tree 291 included only unparasitized nests in the analysis of clutch size. BROOD PARASITISM The frequency of brood parasitism was calculated by determining the number of nests containing cow- bird eggs or young for a species within each study site. We calculated parasitism frequency for a species only when evidence of cowbird parasitism could be found within the database. The intensity of cowbird parasit- ism was calculated by determining the mean number of cowbird eggs laid within each species' nest, within each study site. Each species was classified according to nest placement as either a ground, shrub, or tree nester (Table 1 ) to determine if nest placement affected a species' response to forest fragmentation. The clas- sification of nest placement was based on Ehrlich et al. (1988) and Baicich and Harrison (1997). Differenc- es in the frequency of cowbird parasitism between fragmented and unfragmented sites were examined us- ing Friedman's nonparametric analysis of variance (ANOVA) for randomized blocks (Sokal and Rohlf 1981) and differences in intensity of cowbird parasit- ism were examined by using parametric ANOVAs. For each analysis we blocked by species and tested for habitat affects. Nonparametric Wilcoxon 2-sample tests (Sokal and Rohlf 1981) were performed on the arcsine transformed proportion of nests parasitized for each nesting classification to determine if nest place- ment affected a species' response to fragmentation. NEST PREDATION The daily mortality rate of nests and their associated standard errors were estimated using the Mayfield (1961, 1975) method as modified by Johnson (1979) and Hensler and Nichols (1981). We calculated the dai- ly mortality rate for nests of each species as the total number of failures divided by the total number of days nests were observed, pooled across all nests within each study site. Differences in daily mortality rates be- tween fragmented and unfragmented sites were ex- amined using analysis of variance blocking by species and testing for habitat affects. We also partitioned daily mortality rates into cause-specific components (pre- dation and parasitism) to determine the mechanisms that may influence reproductive success in fragmented versus contiguous sites. As in the parasitism analyses, we classified each species according to its nest place- ment. Differences in predation rates between paired fragmented and unfragmented sites were examined us- ing the program CONTRAST (Hines and Sauer 1989). This program uses chi-square statistics to test for ho- mogeneity of mortality rates by creating a linear con- trast of the rate estimate (Sauer and Williams 1989). LITERATURE REVIEW We also reviewed the available literature to sum- marize the effects of forest fragmentation and edge ef- fects on nest predation rates between sites east and west of the Rocky Mountains. We limited our review to studies conducted in forested systems and to those that examined the effects of anthropogenic fragmen- tation (e.g., agriculture and forestry practices). Be- cause most nest predation studies have used artificial nests, we have included them in our review, but rec- ognize that there are inherent weaknesses in their use (Haskell 1995a, Ortega et al. 1998). RESULTS Sites classified by investigators as fragmented had significantly lower proportion of forest cov- 76 STUDIES IN AVIAN BIOLOGY NO. 25 FIGURE 1. Locations of study sites used in analyses. Squares indicate sites designated as "eastern" and circles as "western." Open symbols indicate fragmented sites and closed unfragmented. Each site plotted on the map is composed of several independent study plots. er within a 10 km radius (0.45 _+ 0.10) relative to unfragmented sites (0.90 _+ 0.04, t = -4.199, df = 6.2, P = 0.005). REPRODUCTIVE SUCCESS We found no difference in clutch size of un- parasitized nests between fragmented and un- fragmented sites (East 0.01 +_ 0.10, t = -0.091, df = 1, P = 0.930; West 0.10 _+ 0.07, t = 1.46, df = 1, P - 0.194). Yet, the mean number of offspring fledged per nest attempted was significantly greater in unfragmented rela- tive to fragmented sites in the east (-0.23 _+ 0.08, t = -2.72, df = 1, P = 0.02), but we found no difference between fragmented and unfrag- mented sites west of the Rocky Mountains (0.09 _+ 0.08, t - 1.06, df = 1, P = 0.314). BROOD PARASITISM The frequency of parasitism by Brown-head- ed Cowbirds was significantly higher in eastern fragmented sites relative to unfragmented sites (X 2 = 317.34, df = 1, P < 0.001) but there were no significant differences among western sites (X 2 = 2.29, df = 1, P > 0.1; Fig. 2). In addition, fragmentation resulted in a significantly higher frequency of brood parasitism for all eastern TABLE 2. LOCATIONS OF PAIRED FRAGMENTED AND UNFRAGMENTED SITES Site Landscape Latitudelongitude Location Columbia Frag 38.95-92.11 Mofep Unfrag 37.04-91.12 SE Forest I Frag 43.61-91.25 SE Forest 2 Unfrag 43.61-91.25 St. Croix Frag 45.36-82.72 Cheque. NF Unfrag 46.06-91.11 Bitterroot I Frag 46.10 114.23 Bitterroot 2 Unfrag 46.10-114.23 South Fork 1 Frag 43.62-111.63 South Fork 2 Unfrag 43.62-111.63 PNFF Frag 44.67-116.20 PNFU Unfrag 44.67-116.20 Columbia, MO Ozarks, MO Southeastern MN Southeastern MN Eastern MN, Western WI Chequemegon NE WI Bitterroot Valley, MT Bitterroot Valley, MT South Fork of Snake River, ID South Fork of Snake River, ID Payette National Forest, ID Payette National Forest, ID PARASITISM, PREDATION, AND FRAGMENTATION--Cavitt and Martin 77 0.4 E o.3 , 0.2 ._ o P o.1 n i Fragmented Unfragmented East FIGURE 2. Mean frequency of brood parasitism (_+ SE) by Brown-headed Cowbirds in fragmented and un- fragmented eastern and western sites. A * indicates P < 0.05. West nest placement classifications, but no differences were found among western sites (Table 3). The intensity of brood parasitism was not af- fected by forest fragmentation east (F = 0.07, df = 1, 10, P = 0.80) or west (F = 0.14, df = 1, 2, P = 0.75) of the Rockies. Within nest place- ment classifications, shrub nesters at fragmented western sites had a significantly higher intensity of cowbird parasitism relative to unfragmented sites (Table 3). There were no other differences in parasitism intensity by nest placement clas- sification (Table 3). DAILY MORTALITY Eastern fragmented sites tended to have high- er daily mortality rates than unfragmented sites (F = 3.03, df = 1, 47, P = 0.08) but the differ- ence was not significant (Fig. 3). However, west- ern unfragmented sites had significantly higher daily mortality rates relative to fragmented sites (F = 3.87, df = 1, 30, P = 0.05; Fig. 3). Eastern shrub nesting birds suffered significantly higher daily mortality rates on fragmented than on un- fragmented sites, but no other differences by nest placement classification were observed (Ta- ble 3). The daily mortality rate due to nest predation was not significantly different between eastern fragmented (0.031 -+ 0.002) and unfragmented sites (0.030 _+ 0.003; F = 0.10, df = 1, 31, P = 0.76), but was significantly higher in western unfragmented sites (0.038 -+ 0.003; F = 4.04, df = 1, 30, P = 0.05) relative to fragmented loca- tions (0.029 _+ 0.003). The daily mortality rate due to parasitism was significantly greater in TABLE 3. EFFECTS OF HABITAT FRAGMENTATION ON THE FREQUENCY (PERCENTAGE OF NESTS) AND INTENSITY (NUMBER OF EGGS) OF BROWN-HEADED COWBIRD PARASITISM AND DAILY MORTALITY RATES WITHIN NEST PLACE- MENT CLASSIFICATIONS FOR SITES EAST AND WEST OF THE ROCKY MOUNTAINS Nest Region placement Statistics Fragmented Unfragmented Frequency of Cowbird Parasitism Intensity of Cowbird Parasitism Daily Mortality Rate Z, dr, P Median, Median, Upper-Lower Upper-Lower Quartiles Quartiles East Ground 2.33, 1, 0.02 28.5, 50.0-20.2 2.0, 13.3-0.9 Shrub 2.01, 1, 0.05 30.1, 62.7-26.5 6.4, 34.8-0.0 Tree 1.94, 1, 0.05 9.6, 17.0-8.0 1.6, 5.6-0.07 West Shrub 0.44, 1, 0.66 35.3, 52.6-0.0 3.1, 19.0-1.4 Tree 0.0, 1, 1.00 8.6, 21.1-0.0 1.8, 32.0-0.0 F, dfmode I error, P Mean -+ SE Mean -+ SE East Ground 1.16, 1, 1, 0.48 1.7 + 0.4 1.0 +- 0.5 Shrub 1.25, 1, 6, 0.31 1.7 -+ 0.2 1.4 -+ 0.3 Tree 1.07, 1, 2, 0.41 1.1 _+ 0.05 1.0 _+ 0.1 West Shrub 21.36, 1, 2, 0.04 1.6 + 0.06 1.2 _+ 0.06 Tree 2.2, 1, 2, 0.27 1.2 +_ 0.08 1.4 _+ 0.08 Z, df, P Median, Median, Upper-Lower Upper-Lower Quartiles Quartiles East Ground 1.3, 1, 0.20 0.4, 0.06-0.04 0.03, 0.05-0.03 Shrub -2.1, 1, 0.04 0.043, 0.05-0.04 0.037, 0.04-0.03 Tree 0.14, 1, 0.89 0.039, 0.04-0.03 0.034, 0.04-0.03 West Shrub 0, 1, 1.0 0.041, 0.04-0.03 0.04, 0.06-0.03 Tree -0.3, 1, 0.77 0.031, 0.04-0.03 0.037, 0.06-0.03 78 STUDIES IN AVIAN BIOLOGY NO. 25 0.06 656 Fragmented Unfragmented 0.05 0.04 - 0.03 0.02 0.01 0.00 1463 East 1122 West FIGURE 3. Mean daily mortality rate (2 SE) of nests in fragmented and unfragmented eastern and western sites. The total number of nests used in analyses are given above each bar. A * indicates P < 0.05. eastern fragmented sites (X 2 = 29.04, df = 1, P < 0.001; median, upper-lower quartiles of frag- mented sites 0.005, 0.01-0; unfragmented sites 0, 0.003-0) but not among western sites (X 2 = 0.278, df = 1, P > 0.5). In two of the three paired eastern sites, daily mortality rates were significantly higher on frag- mented relative to unfragmented plots (Fig. 4). This pattern was reversed in the west where two of the three paired sites had significantly higher daily mortality rates on unfragmented plots rel- ative to fragmented ones. LITERATURE REVIEW Our review consisted of 39 studies; the vast majority (33) were located east of the Rockies, with only six studies in the West (Table 4). The results of eastern studies were based on 53 field seasons with a mean duration of 1.6 field sea- sons per study. Western studies were based on only 11 field seasons with a mean of 1.8 field seasons per study. Of the studies that have tested for edge effects, 56% of 16 studies detected an eflct in the East, whereas only one of four stud- ies observed an edge eflct in the West. Eastern studies that examined the eflct of fragmentation on nest predation rates typically found negative relationships. A negative relationship between fragmentation and nest predation was found in 68% of 19 studies, no relationship in 21%, and two studies (--10%)reported a positive relation- ship. Only three western studies reviewed tested for fragmentation eflcts; two of three studies found a positive relationship between nest pre- dation rates and fragment size with the third demonstrating no relationship. MO wI MN East MT Unfragmented N-ID S-ID West FIGURE 4. Comparison of fragmentation effects on the daily mortality rates (+ SE) of paired local sites east and west of the Rocky Mountains. The number of nest records utilized in each comparison is indicated above each bar. A * indicates P < 0.001; other com- parison P > 0.05. It has been suggested that forest fragments embedded in different matrices may differen- tially affect patterns of nest predation (Andr6n 1995, Bayne and Hobson 1997). According to the "Eastern Paradigm," birds nesting in forest patches imbedded in an agriculture or urban/sub- urban matrix are expected to have lower repro- ductive success relative to those nesting in more natural settings (Thompson et al. this volume). Thus, we classified studies according to the ma- trix of the surrounding landscape (e.g., agricul- ture and forest dominated). Six studies in the East tested for edge effects within an agricultur- ally dominated matrix and nine within a forested matrix. Five of the six forest-agricultural edge studies demonstrated an increase in nest preda- tion, whereas only four of eight found edge ef- fects within a forested matrix. We were unable to review any western studies that tested for edge effects within an agricultural matrix. Brand and George (2000), however, compared preda- tion rates on artificial nests between sites with different types of adjoining habitat. In contrast to predictions of the "Eastern Paradigm," Brand and George (2000) found predation rates were lower in patches adjacent to urban/suburban ar- eas than those adjacent to natural grasslands. Three of four western studies within forest-dom- inated landscapes failed to demonstrate an edge eflct. Ten of the eastern studies reviewed tested for fragmentation effects within an agricultural ma- trix. Two of ten found no relationship between forest area and nest predation rates but the re- maining eight reported significant and negative relationships. The results of eastern studies con- ducted within a logging matrix are not as ap- parent; of the nine studies reviewed, five report- PARASITISM, PREDATION, AND FRAGMENTATION--Cavitt and Martin 79 TABLE 4. SUMMARY OF STUDIES EXAMINING THE EFFECTS OF EDGE AND FOREST FRAGMENTATION ON NEST PRE- DATION RATES EAST AND WEST OF THE ROCKY MOUNTAINS Nest Duration Edge Fragnlentation Reference Location type a Matrix of stud?' effect effect c Eastern Studies Bayne and Hobson 1997 SK A Agriculture 2 no Burger 1988 MO A Agriculture I yes Donovan et al. 1995 Midwest R Agriculture 3 - Donovan et al. 1997 Midwest A Agriculture I - Fauth 2000 IN R Agriculture 3 no 0 Gates and Gysel 1978 MI R Agriculture 2 yes Haskell 1995 NY A Agriculture I 0 Hobson and Baynes 2000 SK R Agriculture 4 - Hoover et al. 1995 PA R Agriculture 2 - Linder and Bollinger 1995 IL A Agriculture 1 yes Marini et al. 1995 IL A Agriculture 1 yes Robinson et al. 1995 Midwest R Agriculture 5 - Saracco and Callazo 1999 NC A Agriculture 1 yes Sargent et al. 1998 SC A Agriculture 1 - Seitz and Zegers 1993 PA A Agriculture 1 Weinberg and Roth 1998 DE R Agriculture 2 - Wilcove 1985 MD, TN A Agriculture I - Bayne and Hobson 1997 SK A Forested 2 no DeGraaf and Angelstam 1993 NH A Forested 1 0 Fenske-Crawford and Niemi 1999 MN A Forested 1 no Gale et al. 1997 CT R Forested 2 0 Hanski et al. 1996 MN R Forested 1 no King et al. 1996 NH R Forested 2 yes King et al. 1998 NH A Forested 1 yes Niemuth and Boyce 1997 WI A Forested 2 yes Rudnicky and Hunter 1993 ME A Forested 2 no 0, + Small and Hunter 1988 ME A Forested 1 no - Vander Haegen and DeGraaf 1996 ME A Forested 1 yes - Vander Haegen and DeGraaf 1996 ME A Forested 1 + Yahner and Mahan 1996 PA A Forested 1 - Yahner and Scott 1988 PA A Forested 1 - Yahner and Wright 1985 PA A Forested 1 no Keyser et al. 1998 AL A Residential I - Western Studies Hannon and Cotterill 1998 AB A Agriculture 2 0, + Tewksbury et al. 1998 MT R Agriculture 2 + Brand and George 2000 CA A Forested I yes Cotterill and Hannon 1999 AB A Forested 3 no 0 Ratti and Reese 1988 ID A Forested 1 no Song and Hannon 1999 AB A Forested 2 no a R - study monitoring the effect on real nests, A - study monitoring effect on artfficial nests. b Number of field seasons on which results are based. c This column indicates the direction of the relationship between forest area and nest predation rates. A "0" indicates no relationship, a .... indicates a negative relationship and a "+" indicates a positive relationship. Studies with more than one symbol represent annual variation in response. ed negative relationships between forest area and nest predation rate, two reported a positive relationship, and two reported no relationship. Only two western studies reviewed tested for fragmentation effects within an agricultural ma- trix, and both of these studies found a positive relationship between nest predation rates and fragment size. DISCUSSION We found that the patterns of brood parasitism were not consistent between sites east and west of the Rocky Mountains. The frequency of brood parasitism was significantly higher in eastern fragmented sites relative to unfragment- ed sites, but not in the West. In addition, all nest placement classifications within fragmented eastern sites had a higher frequency of parasit- ism relative to unfragmented sites, but we were unable to detect a difference in the West. It ap- pears this differential response may, in part, be due to greater variation in the frequency of par- asitism among western sites. For example, some fragmented western sites reported no cowbird 80 STUDIES IN AVIAN BIOLOGY NO. 25 parasitism for shrub and tree nesting species and others reported rates as high as 52%, a rate com- parable to the most severely affected eastern fragmented sites. This higher variability among western sites in their response to brood parasit- ism may be attributed to lower cowbird abun- dance in the West as compared to the East (Sauer et al. 2000). Morrison and Hahn (this vol- ume), in an extensive review of the literature, did not find evidence to suggest that cowbird parasitism varied by region. Rather, they suggest that the major factors determining the impacts of cowbirds on their hosts operate continent- wide. The frequency and intensity of cowbird parasitism may be difficult to predict across large geographic regions and may depend pri- marily on local factors such as the presence of agriculture and patch size (Hahn and Hatfield 1995, Hochachka et al. 1999, Morrison and Hahn this volume). It is clear from this study that the effects of forest fragmentation on nest predation rates are not necessarily consistent across the continent. We found that eastern fragmented sites had few- er offspring fledged per nest attempted, and tended to have higher daily mortality rates rel- ative to unfragmented sites. These results are in agreement with the "Eastern Paradigm" (e.g., Thompson et al. this volume). In contrast, west- ern unfragmented sites had significantly higher daily mortality rates due to nest predation rela- tive to fragmented ones. Paired sites east and west of the Rockies also tended to follow this same general pattern, higher daily mortality rates in fragmented eastern sites and unfrag- mented western sites (see Fig. 4). Studies reviewed for this paper also suggest that forest fragmentation may not be generalized between sites east and west of the Rockies. East- ern studies typically reported a negative rela- tionship between forest area and nest predation rates (68%). This generality is improved when only studies conducted within an agricultural matrix are examined (80%). Unfortunately, only two western studies could be located, and thus any conclusions regarding the effects of frag- mentation on nest predation in the West are speculative. However, both of these studies re- ported a positive relationship between forest area and nest predation rates and both studies explained their results on the basis of a differ- ential response of nest predators. Tewksbury et al. (1998) demonstrated that nest predation was higher on unfragmented sites relative to sites fragmented by agriculture and human develop- ment within the Bitterroot Valley of Montana. They suggested this pattern was due to the re- sponse of nest predators to fragmentation. Red squirrels (Tamiasciurus hudsonicus), important nest predators in their system, were more abun- dant in forested landscapes and declined with increasing forest cover (but see Bayne and Hob- son 2000). Similarly, an artificial nest study con- ducted in woodlots surrounding agricultural land in Alberta, Canada, found higher rates of nest predation within larger woodlots during one breeding season and no difference during anoth- er (Hannon and Cotterill 1998). They suggested that forest interior predators, such as small mam- mals, were important in driving this response. Any attempt to uncover patterns associated with nest predation is difficult because predation is an inherently complex phenomenon. Each study site will have a particular suite of reptilian, mammalian, and avian predators (e.g., Miller and Knight 1993, Fenske-Crawford and Niemi 1997, Thompson et al. 1999; Cavitt 1999, 2000) and these predators will either take nests inci- dentally (Vickery et al. 1992) or deliberately for- age for nests (Sonerud and Fjeld 1987). Fur- thermore, this suite of nest predators will vary from site to site across North America and will likely respond to fragmentation differently (Bayne and Hobson 1998, 2000). Unfortunately, few studies have been con- ducted within the western U.S. that examine the effects of forest fragmentation on nest predation rates. Our analyses and literature review are based on only a handful of western sites in com- parison to the numerous studies conducted in the East. Consequently, we are not certain of the generality of our results throughout the West. However, these results do suggest that (1) suf- ficient evidence exists to question the applica- tion of patterns observed in the eastern U.S. across broad geographic regions, (2) more stud- ies on the effects of fragmentation are needed throughout the western U.S., particularly studies that simultaneously monitor both the fates of real nests and the response of the predator com- munities, and (3) long-term studies are needed to separate real effects from stochastic process- es. ACKNOWLEDGMENTS The results presented here are the products of an extensive list of investigators and their field assistants. Without their collective efforts and generous contri- butions of data to the BBIRD (Breeding Biology Re- search and Monitoring Database) project this research would not have been possible. We also wish to thank T. L. George, D. Dobkin, and an anonymous reviewer for comments and suggestions on drafts of this man- uscript, and N. Summers for editorial assistance. This research was supported by the BBIRD program under the Global Change Research Program of the USGS Biological Resources Division. Studies in Avian Biology No. 25:81-91, 2002. EFFECTS OF FOREST FRAGMENTATION ON TANAGER AND THRUSH SPECIES IN EASTERN AND WESTERN NORTH AMERICA RALPH S. HAMES, KENNETH V. ROSENBERG, JAMES D. LOWE, SARA E. BARKER, AND ANDRI A. DHONDT Abstract. It is likely that selective forces on forest-specialist birds differ by region across the North American continent, and closely related species that evolved under presumably differing selective regimes may show markedly different responses to human-caused habitat fragmentation. We report the results of research by the Cornell Laboratory of Ornithology that used volunteers to gather data on the effects of habitat fragmentation on forest tanager and thrush species across their ranges and the continent. This large-scale approach permits the comparison of effects between regions within species as well as between species. Although forested landscapes in western North America are often naturally fragmented compared to historically contiguous forests in eastern North America, an identical set of principal components described forest fragmentation in both regions. Response by the Western Tanager (Piranga ludoviciana) to overall fragmentation was very similar to that of the Scarlet Tanager (P. olivacea) in eastern regions; probability of breeding dropped significantly for both species in highly fragmented landscapes. The Hermit Thrush (Catharus guttatus), with both eastern and western pop- ulations, is highly affected by fragmentation, with no geographic variation. Additionally, both the Swainson's Thrush (C. ustulatus) in the West and the Veery (C. Jhscescens) in the East showed similar strong effects of fragmentation. Predation and parasitism pressures as estimated by detections of mam- malian and avian predators or of Brown-headed Cowbirds (Molothrus ater) differed between eastern and western study sites, as did the response by cowbirds to fragmentation gradients in different regions. Overall, however, we found that closely related species and populations showed similar responses to habitat fragmentation, regardless of the historic configuration of the forests in which they occurred. Key Words: Catharus fuscescens; Catharus guttams; Catharus ustulatus; geographic variation; Hy- locichla rnustelina; Molothrus ater; Piranga ludoviciana; Piranga olivacea; predators; principal com- ponents analysis. Selective forces on forest-specialist birds differ by region across the North American continent, with differing levels of disturbance, nest para- sitism, and of predation by a variable suite of predators. Further, closely related species, or populations within widely distributed species, that have evolved under differing selective re- gimes may show markedly different responses to human-caused habitat fragmentation. How- ever, testing whether presumably different selec- tive regimes have indeed led to different re- sponses to fragmentation in western and eastern North America is not a trivial matter. It requires several things that, heretofore, have not been combined in one research project (or even in a series of research projects); these include a large geographic extent, a large sample size, standard- ized data collection, and a widely applicable measure of fragmentation. Further, the species to be studied must have continent-wide distribu- tions, or comparisons must be made between closely related species with primarily eastern or western geographic ranges. We report the results to date from the Cornell Lab of Ornithology's Birds in Forested Landscapes (BFL) project, which used volunteers to gather data on the ef- fects of habitat fragmentation on forest tanager and thrush species across their ranges and across North America. Several authors have pointed out the differ- ences between western, often coniferous, forest and eastern deciduous forest landscapes as se- lective environments for obligate forest-nesting birds (Hejl 1992, Freemark et al. 1995, Tewks- bury et al. 1998). For example, western and eastern forests differ both in their original con- figuration and in their subsequent use by humans (Hejl 1992). Western forests are naturally patchy, and in many areas are confined by mois- ture regimes to riparian zones or to topographic "islands" (Tewksbury et al. 1998). Further, hu- man-caused fragmentation in western North America has often been due to logging (Hejl 1992), and is of fairly recent origin. In contrast, the formerly contiguous eastern hardwood for- ests have been cleared for agriculture as long as 200 years before present (Smith et al. 1993, Yahner 1997), and are now increasing from his- torical lows as abandoned farms revert to forest. In addition to disturbances caused by humans, naturally occurring disturbances also play a large role in shaping the selective environment in which forest bird species evolve, and it is clear that the type, scale, and frequency of dis- turbance are different in the two regions. In 81 82 STUDIES IN AVIAN BIOLOGY NO. 25 western North America, the rainiest months oc- cur in winter and spring, with relatively little rain occurring during the summer and fall (Perry 1994). There are also extensive stands of early- successional, serotinous tree species (lodgepole pine, Pinus contorta; jack pine, P. banksiana; and black spruce, Picea mariana) in boreal and temperate montane forest (Perry 1994). Further, dryer forests throughout the West are dominated by the equally fire-adapted ponderosa pine (Pi- nus ponderosa; Perry 1994). This combination of seasonal droughts and fire-adapted vegetation is reflected in frequent disturbance by fire (Free- mark et al. 1995). In contrast, eastern deciduous forests are relatively free of fire because of fre- quent rains during the summer and fall, and be- cause the combination of warmth and high mois- ture levels leads to rapid decomposition of fallen trees and other potential fuels (Perry 1994). Other selective forces such as predation and rates of nest parasitism also appear to differ be- tween western and eastern North America. Both the suites of predator species present, nest par- asites, and their abundance (Donovan et al. 1995a), appear to combine to alter the selection regimes in the two regions (Tewksbury et al. 1998, Rosenberg et al. 1999). For example, red squirrels (Tamiasciurus hudsonicus), which are the most common nest predator in some western landscapes (Bayne and Hobson 1997, Darveau et al. 1997, Tewksbury et al. 1998), are relative- ly rare in the East where avian predators such as corvid species are much more common (Ho- grefe et al. 1998). Moreover, rates of nest para- sitism by the Brown-headed Cowbird (Moloth- rus ater) also vary with region, with highest rates in the Midwest region (42.1% of Wood Thrush, Hylocichla mustelina, nests) and lower rates in the Mid-Atlantic (26.5%) and Northeast (14.7%) (Hoover and Brittingham 1993). Final- ly, the responses of both nest predators and par- asites to fragmentation has also been shown to vary across physiographic regions (Robinson et al. 1995b, Trine 1998, Rosenberg et al. 1999). These large differences between eastern and western forest vegetation, historical land uses, disturbance, and between parasitism and preda- tion regimes provide ample grounds to suspect differences in responses to fragmentation be- tween eastern and western landscapes (Freemark et al. 1995). The question becomes how to test for these hypothesized differences. The first re- quirement is for a measure of fragmentation that is applicable across the continent, and in land- scapes with differing conformations of habitat. Habitat fragmentation implies loss of habitat, a reduction in mean habitat patch size, increases in the mean isolation of patches, and increases in the mean amount of forest/non-forest edge (Andr6n 1994). Most workers agree that loss of habitat is one of the primary mechanisms by which human-caused habitat fragmentation af- fects populations of birds; some even suggest that habitat loss is the primary (Trzcinski et al. 1999) or only (Fahrig 1997, 1998) mechanism. Others have cited the effects of increased edge (Paton 1994, Hoover et al. 1995, Donovan et al. 1997) and isolation (Robbins et al. 1989a, Vil- lard and Taylor 1994, Villard et al. 1995, Des- rochers and Hannon 1997), or of decreased patch size (Schieck et al. 1995, Bellamy et al. 1996a, Keyser et al. 1998, Trine 1998) as also playing an important role. However, it seems most likely that both habitat abundance and con- figuration (McGarigal and McComb 1995, Vil- lard et al. 1999) play important roles, with the effect of configuration increasing in importance below a critical threshold in abundance (Turner 1989, Andrn 1994, Andrn et al. 1997, With et al. 1997, Andrn 1999). What is needed is a composite measure of habitat fragmentation that captures a large proportion of the information contained within these variables. Such a com- posite measure should include information cap- tured at the level of the surrounding landscape, as well as at the patch (Freemark et al. 1995), to afford a more complete understanding of the factors affecting the distribution of sensitive spe- cies (Hinsley et al. 1995). The Cornell Labora- tory of Ornithology's BFL project provides both the fragmentation data needed to calculate such a composite measure, as well as data on species occurrence from across the continent that are necessary to test the hypothesis of different re- sponses to fragmentation in eastern and western landscapes. BFL is a natural continuation of the Cornell Lab of Ornithology's Project Tanager, which be- gan as a National Science Foundation (NSF) Na- tional Science Experiment. Project Tanager used volunteers across North America (north of Mex- ico) to study the effects of forest fragmentation on four species of tanagers (Rosenberg et al. 1999). BFL uses the same methodology to study the effects of fragmentation on seven species of forest thrushes and two species of Accipiter hawks. BFL was undertaken during the 1997 and 1998 breeding season in cooperation with Partners in Flight, an umbrella organization of government agencies, conservation organiza- tions, and industry working together to promote the conservation of birds in the Americas. Birds in Forested Landscapes was continued during the 1999 and 2000 field seasons in cooperation with the United States Department of Agricul- ture (USDA) Forest Service. For simplicity's sake, we will refer to both Project Tanager and TANAGERS AND THRUSHES IN EAST AND WEST--Hames et al. 83 Birds in Forested Landscapes as BFL hereinaf- ter. METHODS DATA COLLECTION The data-collection protocol for both Project Tana- ger (Rosenberg et al. 1999) and BFL were essentially identical. Each protocol consisted of four stages: the unbiased selection of one or more study sites; repeated visits to the study sites with the playback of conspe- cific vocalizations to elicit responses from territorial birds so that they could be counted; the estimation of a number of patch- and landscape-scale measures of fragmentation; and the coding of data onto computer- readable bubble-forms, which were returned to the Lab of Ornithology for collation and analysis. In both studies, the volunteer participants selected study sites in suitable wooded habitat (e.g., trees >6 m tall, canopy coverage 30%). The instructions stressed that almost any patch of relatively mature for- est or woodland was acceptable, and participants were urged to find a range of patch sizes in similar habitat. To avoid bias, participants were cautioned to select their study sites based only on apparent habitat suit- ability and to not select sites where the species of in- terest was known to nest (Rosenberg et al. 1999). Each study site was defined as a circle of 150-m radius; point-counts and playbacks were conducted at the cen- ter of each study site. Participants made two visits to each site to census for territorial males of the focal species. During a ten-minute point count on each visit, participants looked and listened for territorial individ- uals of the species of interest within the study site. Participants also recorded the presence of avian and mammalian predators, as well as any detections of Brown-headed Cowbirds during the two point-counts. The two required visits were timed to coincide with pair bonding or nest building, and with the nestling/ fledgling stages of the breeding cycle. If no individuals of the species of interest were detected within the point count period, participants used playback of conspecific territorial vocalizations to elicit a response from any previously silent birds in order to verify that no terri- torial males were present (Viilard et al. 1995, Rosen- berg et al. 1999). Based on the behavior of birds that were detected, each site was scored as missing, pres- ent, possible, probable, or confirmed breeding using breeding atlas codes (Anonymous 1986, Butcher and Smith 1986, Rosenberg et al. 1999). To avoid counting birds passing through on migration, we scored study sites as "possible" breeding sites only if a singing male of the focal species was detected on both visits. While in the field, participants also used simple techniques to estimate canopy height and amount of canopy closure and noted other site characteristics such as the forest type (coniferous, deciduous or mixed), three most common tree species, and presence or absence of surface water (streams or ponds) at each site. After completion of the fieldwork, participants used USGS topographic maps in conjunction with a clear acetate grid overlay to estimate a number of mea- sures of fragmentation for each site. (The grid was intended for use with 1:24000 maps or aerial photos, and was divided into 1 ha squares at that scale.) Esti- mated fragmentation measures included the size of the forest patch surrounding the study site, the isolation of that patch from other patches, and the proportion of forest and edge density (amount of forest/non-forest edge corrected for the amount of forest) in the sur- rounding 1000 ha block. The site's elevation above mean sea level (MSL) was also recorded, as was an estimate of the canopy height. A number of other data were also collected at each site, but were not used in this analysis. For further details on the development of this protocol see Rosenberg et al. (1999). Participants then coded these data onto computer-readable forms and returned the forms to the Lab of Ornithology. At the Lab, we edited each form by hand to ensure it had been correctly completed; simple checks were also performed when the SAS (SAS Institute 1989) dataset was constructed to ensure that each datum was within possible ranges. We excluded all sites with missing data from subsequent analyses. ANALYSES At each site participants collected a number of data, including measures of forest fragmentation. We checked the distributions of all fragmentation variables on normal probability plots and transformed variables as needed before analysis began. Many of the mea- sures of fragmentation are highly significantly inter- correlated (Hames et al. 2001). To avoid multicolli- nearity and the fitting of complicated models with dif- ficult-to-interpret interaction terms, we used principal component analysis (PCA) on the transformed data to simplify the dataset by yielding fewer uncorrelated factors (principal components), which explained a high proportion of the variance in the original dataset (John- son and Wichern 1982, Villard et al. 1995, Rosenberg et al. 1999). We then used multiple logistic regression to model the probability that territorial birds would be found, based on the principal component values at each site. We also used logistic regression to model the probability of occurrence of the Brown-headed Cowbird. To test the hypothesis that the effects of frag- mentation varied between eastern and western land- scapes, we compared the magnitude of the fragmen- tation coefficients derived from logistic regression for each region. Principal components analysis To conduct the PCA we combined all unique study points from the 1995, 1996, 1997, and 1998 field sea- sons of BFL into one dataset. We then used PROC FACTOR (SAS Institute 1989) with the orthogonal varimax rotation option to ensure that there was max- imal separation (Johnson and Wichern 1982) and no intercorrelation between the resulting principal com- ponents. These rotated factors were then standardized to a mean of zero and a standard deviation of one (SAS Institute 1989) to facilitate comparison of estimated coefficients, before they were used as predictor vari- ables in the logistic regression. We included a number of transformed variables from each study site in the PCA. These variables were the natural log of the forest patch size (Ln Size), edge density (Ln Edge Density), elevation above msl (Ln Elevation) and canopy height (Ln Canopy Height), as well as the arcsine square-root transformed proportion of forest (Asqrt %Forest; Table 1). The natural log of 84 STUDIES IN AVIAN BIOLOGY TABLE 1. CORRELATION MATRIX FOR VARIABLES INCLUDED IN PRINCIPAL COMPONENTS ANALYSIS NO. 25 Ln(size) Asqrt(%forest) Ln(edge density) Ln(eIevatiol) Ln(canopy height) Ln(Size) 1.000 0.556** -0.386** 0.114** 0.064** Asqrt(%Forest) 1.000 -0.740** 0.151 ** 0.029 Ln(Edge Density) 1.000 -0.156** -0.033 Ln(Elevation) 1.000 -0.031 Ln(Canopy Height) 1.000 Notes: Ln(Size) is the natural log of the patch size; Asqrt(% Forest) is the arcsine square-root transformed % forest in the surrounding 10130 ha; Ln(Edge Density) is the linear measure of forest/non-forest in nffha; Ln(Elevation) is the natural log of distance above Mean Sea Level, in m; L(Canopy Height) is the natural log of canopy height, in m. * P < 0.01, ** P < 0.001. isolation, measured as distance to the nearest forest patch of 40 or 200 ha, was not included in the PCA because these data were missing from a substantial number of records. As this variable was highly signif- icantly correlated with Ln Size (r = -0.228, P --< 0.001), Ln Edge Density (r = 0.413, P < 0.001), and Asqrt %Forest (r = -0.567, P < 0.00l), we felt that the increase in sample size gained by omitting this variable more than compensated for any loss of ex- planatory power caused by its omission. Logistic regression analysis We used PROC LOGISTIC (SAS Institute 1996) to model the probability that a singing male of the species of interest would be detected on the two required vis- its, either vocalizing spontaneously or in response to playback of conspecific territorial calls, based on the level of fragmentation at each site. We fit multiple lo- gistic regressions using all of the calculated predictor variables (Principal Components), and used manual backward elimination of non-significant (Wald chi- square P > 0.1 ) variables to fit the best model. Models were compared using the G 2 statistic (difference in -2 log-likelihood between two nested models; Agresti 1996) and Akaike Information Criterion (AIC; Agresti 1996). The model chosen in each case was the most parsimonious one that minimized the AIC and had a G 2 that was not significant at the P < 0.05 level. Comparison of fragmentation effects To compare the effects of fragmentation in eastern and western landscapes, we first subset our data into two parts at the 98th meridian, a natural break in the dataset that coincides roughly with the Great Plains. We focused our analyses on widespread species that had both eastern and western populations (e.g., Hermit and Swainson's, Catharms ustulatus, thrushes and Vee- ry, C. fuscescens) or congeneric species pairs (e.g., Western, Piranga ludoviciana, and Scarlet, P. oliva- cea, tanagers) with one eastern and one western mem- ber. In addition to these focal species, we compared the effects of fragmentation on the presence of Brown- headed Cowbird across North America. Additionally, we used contingency table analysis to test for differ- ences in the frequency of occurrence of several species of predators in eastern and western landscapes. We fit separate regression models for each member of species pairs, and tested for differences in the strength of regression coefficients between the pair us- ing a large sample t-test. We rejected the null hypoth- esis of no differences if P < 0.05. However, because we had very large sample sizes for several species, we also compared 95% confidence intervals for the frag- mentation coefficient in each model, to avoid rejecting the null hypothesis based on differences that were sta- tistically, but not biologically, significant. We accepted the null hypothesis of no difference in the effects of fragmentation between species pairs if the 95% con- fidence intervals for the mean estimated effect of frag- mentation overlapped substantially. For single species, we fit regression models that included an east/west dummy or indicator variable, and region by factor in- teraction terms. We rejected the null hypothesis of no difference in effects of fragmentation for widespread species if P < 0.05 (Wald chi-square) for the region by fragmentation interaction term. Comparison of predator and nest parasite pressure To characterize differences in predation and nest parasitism pressures between eastern and western land- scapes, we used contingency table analysis to test for differences in frequency of occurrence for the Brown- headed Cowbird and for several species of predator. Predator species included nest predators such as squir- rels, chipmunks, and corvid species, as well as pred- ators of fledglings and adult birds such as Accipiter hawks. RESULTS DATA COLLECTION Volunteers collected data at a total of 1840 sites during the 1995 and 1996 field seasons (tanager species) and at an additional 1298 sites during the 1997 and 1998 field seasons (thrush species), for a total of 3138 sites (Fig. 1). These sites spanned North America, covering 50 states and provinces, and 55 physiographic regions (Robbins et al. 1986). However, many sites were missing required data, and we based subsequent analyses only on sites for which complete data were available. The proportion of sites which contained a territorial male of the focal species on both visits varied from 0.15 for the Swain- son's Thrush to 0.325 for the Scarlet Tanager. ANALYSES Principal component analysis Our principal component analysis was based on 2515 unique study sites. These sites included 1933 sites with complete data east of the 98th meridian (East), and 582 west of the 98th me- TANAGERS AND THRUSHES IN EAST AND WEST--Hames et al. 85 FIGURE 1. Locations of the approximately 2500 study sites on which this analysis is based. Because of the large size of the symbols representing study sites relative to distances on the map, one study site may cover several others. ridian (West). The correlation matrix for the in- cluded variables showed highly significant cor- relations between patch size, proportion of for- est, and edge density (Table 1), which was re- moved by the orthogonal varimax rotation, thus yielding uncorrelated and easily interpretable principal components (Table 2). The first three principal components ex- plained 83% of the variance in the data set (Ta- ble 2). The first principal component (PC1) had high positive loadings (coefficients >0.5) for patch size and proportion of forest, and high negative loading for edge density in the sur- rounding landscape; we interpreted this principal component as an overall measure of fragmen- tation. PC1 varies from negative values for small patches in a landscape with little forest and a large amount of forest/non-forest edge, to posi- TABLE 2. FACTOR LOADINGS DERIVED FROM PRINCIPAL COMPONENTS ANALYSIS OF 2515 FORESTED SITES Variable PC 1 PC2 PC3 Ln(Size) 0.742 0.029 0.082 Asqrt(%Forest) 0.921 0.067 -0.041 Ln(Edge Density) -0.850 -0.090 0.178 Ln(Elevation) 0.093 0.995 -0.016 Ln(Canopy Height) 0.032 -0.016 0.997 Eigenvalue 2.187 1.027 0.923 Cumulative variance explained 0.437 0.643 0.827 86 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 3. COMPARISONS OF FRAGMENTATION VALUES (PC1) AT SITES SAMPLED IN EASTERN AND WESTERN NORTH AMERICA FOR BIRDS IN FORESTED LANDSCAPES PROJECT Geographic region N Mean SD Minimum Maximum Range East 1933 0.036 1.0069 -2.440 2.819 5.259 West 582 -0.127 0.9820 -2.591 2.393 4.984 Notes: Data were included from both the 1997 and 1998 field season of BFL. The mean fragmentation values from eastern and western landscapes were not significantly different (pooled test of H0: iz I - Iz2 = 0, z - 0.366, df - 2513, P - 0.373). tive values for large patches in a landscape with high proportions of forest and little edge. Inter- pretations of the second and third principal com- ponents were straightforward: PC2 had a high loading only for elevation and PC3 had a high loading only for canopy height. PC3 was re- tained in the PCA despite an eigenvalue (0.92), which was less than the commonly accepted cut- off of 1.0, because other studies have suggested that the height of the canopy plays an important role in habitat selection by forest-obligate birds (Cody 1985, Hames 2001). Hereinafter PC1, PC2, and PC3 will be referred to by their inter- pretations as overall fragmentation, elevation, and canopy height, respectively. Overall, there was little difference between western and eastern sites in the PCA-derived overall fragmentation values (PC1). The mean overall fragmentation values were not signifi- cantly different for western and eastern sites (z = -0.158, df = 2347, P = 0.874) and the min- ima, maxima, and ranges were very similar (Ta- ble 3). Logistic regression analysis The effect of fragmentation was a very similar decrease in the probability of detection with in- creasing habitat fragmentation for both tanager species. Both the Scarlet Tanager in the East, and the Western Tanager in the West, showed a strong, highly significant increase in probability of "possible" breeding as the fragmentation measure PC1 increased (Table 4). This resulted in an approximately five-fold decrease in the es- timated probability of occurrence from the least to the most fragmented site. The probability of detection also increased with increasing eleva- tion in the Scarlet (Fig. 2), but not the Western, tanager (Fig. 3). Sample sizes for the eastern populations of the Swainson's Thrash, and for western popu- lations of the Veery, were insufficient to make within-species comparisons for these species. We therefore treated these as a species pair and restricted the regression analyses to eastern sites for the Veery and to western sites for the Swain- son's Thrush. Both of these thrushes displayed similar highly significant increases in the prob- ability of "possible" breeding as PC1 increased, and fragmentation decreased (Table 4). As in the tanager species, this resulted in an approximate- ly five-fold decrease in probability from the least to most fragmented sites. In both species the probability of detection also decreased with in- creasing elevation (Figs. 4, 5). The Hermit Thrush (Table 5) also showed a highly signifi- cant negative response to fragmentation of ap- proximately the same magnitude as that dis- TABLE 4. STRENGTH OF THE EFFECTS OF FRAGMENTATION (PC1), ELEVATION (PC2), AND CANOPY HEIGHT (PC3) ON THE PROBABILITY OF DETECTING TERRITORIAL BIRDS, SHOWN AS ESTIMATED COEFFICIENTS DERIVED FROM MUL- TIPLE LOGISTIC REGRESSION ScarIet Tanager Western Tanager Veery Swainson's Thrush Wood Thrush Intercept -0.6174'** - 1.4866'** - 1.5899'** -0.8660*** -0.7545*** PC 1/east 0.3648 a*** -- 0.5755 b** -- -0.1668** PC1/west -- 0.5954 a*** -- 0.7315 b** -- 95% CI low 0.2304 0.2765 0.3747 0.3054 -0.3089 95% CI high 0.5016 0.9299 0.7835 1.1902 -0.0453 PC2/east 0.3148'* -- -0.2061' -- ns PC2/west -- ns -- -0.3184* -- PC3/east ns -- ns -- 0.3245*** PC3/west -- ns -- ns -- Notes: The PCA was calculated using all data from across North America; the notations "east" and "west" refer to the region in which each species was studied; denotes that the corresponding coefficient was not calculated; ns indicates that the coefficient was not significant at the P <- 0.05 level. a Test of H0: no difference between coefficients, z - 1.2812, P - 0.176, ns. b Test of H0: no difference between coefficients, z - 0.5968, P = 0.334, ns. * P <- 0.10, ** p < 0.01, *** P <- 0.001. TANAGERS AND THRUSHES IN EAST AND WEST Hames et al. 87 . 1.00 1 n = 934 0.05 .... O= 0.00_7.1fi 1 PC2 Higher Fragmention Elevation FIGURE 2. The effects of fragmentation (PC1) and elevation (PC2) on the probability of detecting a sing- ing or ca]ling male Scaler Tanager on both required visits. Probability of occuence increases as agmen- tation decreases and elevation increases. Model is highly significant ( 2 log-likelihood = 39.876, df = 2, P (0.001). n = 659 õ 1.00 '/ 0.50 L .... :' 0.25 '  F;mentation - 0. ..... PC1 -1.87 Lower - ' 0 94 Elev:;i' Higher PC2 Higher Fragmentation Elevation FIGU 4. The effects of fragmentation (PC1) and elevation (PC2) on the probability of detecting a sing- ing or ca]ling male Veery on both required visits. Prob- ability of occuence increases as fragmentation and elevation decrease. Model is highly significant (-2 log-likelihood 35.932, df - 2, P < 0.001). played by the other thrushes. In addition, the Hermit Thrush showed a highly significant in- crease in the probability of "possible" breeding with increases in elevation. The best model also contained a significant region by canopy height interaction term, so that we can conclude that the effect of canopy height differed between eastern and western populations. The uniform response to fragmentation across species and also across genera was striking, and somewhat troubling. To test if this trend was universal and potentially an artifact of our analytic design, we also fit a logistic regression model to data for the Wood Thrush, a purely eastern species. The Wood Thrush showed the opposite trend (Table 4), a somewhat weaker but still significant in- crease in probability of "possible" breeding with increases in fragmentation. The Wood Thrush was also more likely to be detected in forests with higher canopies (Fig. 6). In both the East and the West, the Brown- headed Cowbird likewise showed an increase in the probability of occurrence with increases in fragmentation (Table 6). The best model for the cowbird also contained a significant effect of year, an indicator variable used to partition var- iance due to slight differences in the Project Tanager and BFL protocols as to when cowbirds could be counted. In addition, there was a highly significant year by region interaction term, g 1.00  n = 339 0.75 0.50¾  0.25 2.23 a. 0.00_ p2.1'/'; 1 PC2 Higher Fragmentation FIGURE 3. The effect of fragmentation (PC 1) on the probability of detecting a singing or calling male West- em Tanager on both required visits. Probability of oc- currence increases as fragmentation decreases. Model is highly significant (-2 log-likelihood = 13.757, df = 1, P < 0.001). Note there is no significant effect of elevation; elevation is only included for comparison between graphs. n = 140 ._(2  0.75 '5 0.50 Lower  Fragmenlation  0.25 ' 2.0s O 73 a. 0.00 -O.9O Elevation Higher PC2 Higher Fragmentation Elevation FIGURE 5. The effects of fragmentation (PC1) and elevation (PC2) on the probability of detecting a sing- ing or calling male Swainsoh's Thrush on both re- quired visits. Probability of occurrence increases as fragmentation and elevation decrease. Model is highly significant ( 2 log-likelihood = 12.588, df = 2, P = 0.002). 88 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 5. RESULTS OF LOGISTIC REGRESSION OF GEOGRAPHIC REGION, FRAGMENTATION, ELEVATION, VEGETATION STRUCTURE, AND THEIR INTERACTIONS, ON THE PRESENCE OF THE HERMIT THRUSH Variable Parameter estimate df SE Wald X2 p Intercept - 1.7284 1 0.1724 100.5300 <0.001 West -0.5680 1 0.3253 3.0498 0.081 PC 1 0.6793 I 0.1270 28.6187 <0.001 PC2 0.4099 1 0.1517 7.3037 0.007 PC3 -0.3452 I 0.1432 5.8130 0.016 West*PC3 0.3968 1 0.2245 3.1245 0.077 Notes: Regression based on data from 617 study sites censused for BFL from 1997 to 1998. "West" is an indicator variable: West - 0 east of the Great Plains and West - I west of the Gmat Plains. Overall model X 2 = 57.879, df - 5, P < 0.001. Concordant pairs = 71.8%. which showed that fewer cowbirds were detect- ed in the West during the 1997 and 1998 BFL field seasons. Further, there were other highly significant region by fragmentation, and region by elevation interactions, as well as a region by elevation by year three-way interaction. East/West comparisons Because we used a standardized measure of fragmentation that included both patch size and the landscape measures proportion of forest and of edge, and because this variable had similar distributions in the East and West, we were able to directly compare fragmentation coefficients from logistic regressions. We tested for differ- ences in the strength of fragmentation between species using a large sample, two-tailed t-test, or between populations within species by using the Wald chi-square for the region by fragmentation interaction term from the logistic regression. We also directly compared the relative strengths of the effects of fragmentation across species by comparing 95% confidence intervals for the es-  1.00  n = 1089  0.50  ._.., Higher  0'2511 I I  -F2r.:: mentatiøn a. 000 ' . 0. PC1 1.95 -058 Higher " Lower Canopy PC3 Lower Fragmentation Canopy FIGURE 6. The effects of fragmentation (PCI) and canopy height (PC3) on the probability of detecting a singing or calling male Wo Thrash on both required visits. (Note that fragmentation axis is reversed from other graphs.) Probability of occuence increases as fragmentation increases and canopy height increases. Mode] is highly significant (-2 log-likelihood = 26.381, df = 2, P < 0.001). timated fragmentation coefficients. We found that the strength of the negative effects was not significantly different (z = - 1.281, P = 0.176) in the Scarlet and the Western tanagers (Table 4) and that their 95% confidence intervals showed considerable overlap. Likewise, there was no significant difference in the strength of fragmentation effects (z = 0.597, P = 0.334) between the Veery in the East and the Swain- son's Thrush in the West (Table 4), and the 95% confidence interval for the Veery was complete- ly contained within that of the Swainson's Thrush. Logistic regression for the Hermit Thrush (Table 5) did not yield a significant re- gion by fragmentation interaction term (Wald X 2 = 0.099, P = 0.753), indicating that there was no significant difference in the strength of frag- mentation effects between eastern and western populations. Thus, neither objective hypothesis testing, nor a more subjective examination of the degree of overlap in confidence intervals, pro- vided strong evidence to reject the null hypoth- esis of no difference in eastern and western re- sponses to fragmentation, at least in these tana- ger and thrush species. Conversely, although the Brown-headed Cow- bird, like the Wood Thrush, showed an overall increase in probability of detection with increas- es in fragmentation, a significant region by frag- mentation interaction term showed that the re- sponse to fragmentation was stronger in the West than in the East (Table 6). Additionally, contingency table analysis of the number of sites at which the Brown-headed Cowbird was de- tected (Table 7) showed a somewhat higher fre- quency of occurrence in the East than the West, although this difference was not significant (P  0.058). For predators, however, the picture is more straightforward. Overall the East had a sig- nificantly higher proportion of sites with at least one mammalian (49.3%) or at least one avian (64.4%) predator, than did the West (39.3% and 25.4%, respectively.) In fact, the West signifi- cantly surpassed the East only in the frequency of occurrence for the red or Douglas (Tamias- TANAGERS AND THRUSHES IN EAST AND WEST Hames et al. 89 TABLE 6. RESULTS OF LOGISTIC REGRESSION OF PROTOCOL, GEOGRAPHIC REGION, FRAGMENTATION, ELEVATION AND THEIR INTERACTIONS, ON THE PRESENCE OF BROWN-HEADED COWBIRDS Variable Parameter estimate df SE Wald X 2 P Intercept -0.8756 I 0.0747 137.37 <0.001 Year -0.2883 I 0.1259 5.24 0.022 West -0.3474 1 0.2210 2.47 0.116 PC1 -0.1998 1 0.0571 12.22 <0.001 PC2 0.1398 1 0.0999 1.96 0.162 Year*West -0.7813 1 0.3562 4.81 0.028 Year*PC2 -0.2965 1 0.1516 3.83 0.050 West*PC1 -0.5577 1 0.1604 12.09 <0.001 West*PC2 -0.7654 1 0.1798 18.12 <0.001 West*PC2*Year 0.8415 1 0.2944 7.65 0.006 Notes: Regression based on data from 2068 study sites certsused for Project Tanager and BFL from 1995 to 1998. "Year" is an indicator variable that partitions variation due to differences in the protocols of the two projects. "West" is an indicator variable: West = 0 east of the Great Plains and West = 1 west of the Great Plains. Overall model X 2 = 101.25, df = 9, P < 0.001. Concordant pairs = 63.0%. ciurus douglasii) squirrels and for Accipiter spe- cies. For all other predators the proportion of sites with detections was significantly higher in the East than in the West (Table 7). DISCUSSION Despite regional differences in topology, veg- etation structure, suites of predators, and land uses past and present, compounded by differ- ences in phylogeny, there is a surprising unifor- mity in the strength and direction of the respons- es to fragmentation across the regions and the species studied. This is particularly surprising because Rosenberg, et al. (1999) showed clear regional differences in the strength of responses to fragmentation in the Scarlet Tanager. This lack of regional effects in the present study may be due to the "lumping" of variation occurring at smaller scales, due to the extremely large re- gions defined for the current study. However, as measured by presence/absence of singing males, for at least the tanager and thrush species we studied, increasing fragmentation is strongly correlated with decreasing probability of detec- tion. What is perhaps not intuitively clear is the correct interpretation of our results. Our study measured the distribution (presence or absence) of the focal species in relation to fragmentation, not the demographic consequenc- es of that fragmentation. However, demonstrated sensitivity to fragmentation alone (shown as changes in distribution of sensitive species) is sufficient to infer that the tanager and thrush species studied are adversely affected by frag- mentation (Winter and Faaborg 1999). For ex- ample, in a recent study of fragmentation effects on grassland birds, Winter and Faaborg (1999) make a clear distinction between the distribu- tional consequences (lower densities, lower probability of occurrence) and the demographic consequences (lower nesting success) of frag- mentation. Further, their results demonstrate that some area-sensitive species may show distribu- tional effects such as absence from small patch- es (Robbins et al. 1989a), while other species may show demographic effects such as lower nesting success in fragments (Donovan et al. 1995a, Winter and Faaborg 1999). This useful partitioning of the adverse effects of fragmen- tation can equally well be applied to forest- dwelling species. This is important because di- rectly determining the demographic consequenc- es of fragmentation requires a skilled field crew and is extremely labor-intensive, making it im- TABLE 7. PERCENTAGES OF SITES, BY REGION, AT WHICH NEST PREDATORS OR BROWN-HEADED COWBIRDS WERE DETECTED DURING THE 1995, 1996, 1997 OR 1998 FIELD SEASON Brown- Chipmunk Red or Gray Crow Jay Accipiter headed (any Douglas or fox (any (any (any Mammalian Avtan Cowbird species) squirrel squirrel species) species) species) predator predator N 644 783 388 685 1148 1222 161 1429 1838 East % 21.88 29.94 9.86 26.83 44.54 45.59 4.38 49.28 64.40 West % 18.66 12.87 20.98 9.40 16.99 23.42 6.82 39.25 25.41 A 3.22 17.07 --11.12 17.43 27.55 22.17 -2.44 10.03 38.99 X2 3.603 88.687 64.622 101.356 187.665 118.740 5.058 23.390 69.124 P --< 0.058 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 Notes: N = Total number of sites at which predators or Brown-headed Cowbird detected; 6, = difference in percentage detected (East - West). 90 STUDIES IN AVIAN BIOLOGY NO. 25 practical for an extensive, volunteer-based study such as this one. In the future, repeated sampling over several breeding seasons will allow us to use rates of site occupation or turnover (Villard et al. 1992, Winker et al. 1995, Bellamy et al. 1996b), rather than simple presence/absence in one season, as a measure of the effects of fragmentation. For example, Hames et al. (2001) have demonstrated that the proportion of breeding seasons a site is occupied by a territorial male Scarlet Tanager over several years is inversely proportional to the degree of fragmentation. Thus, we eagerly await analyses of multiple-year BFL data, which will allow us to make stronger inferences by us- ing rates of territory occupancy as a currency for the effects of fragmentation on habitat qual- ity and reproductive success. However, although direct demographic data are necessary for a complete understanding of these effects, already documented changes in distribution due to frag- mentation are sufficient to demonstrate adverse effects on sensitive species. Our requirement that singing males be de- tected on both visits to the study sites reduces the probability that migrant males would be counted as "possible" breeders; that is, as resi- dent males displaying territoriality. However, al- though the participants were charged to find as many nests of focal species as possible, and to monitor any nests found to determine reproduc- tive success, very few nests were in fact found (Rosenberg et al. 1999). This lack of direct mea- sures of reproductive success, per se, limits our ability to determine the processes that lead to the observed patterns, and hence our ability to make inferences about population effects of fragmen- tation. In particular, Van Horne (1983) pointed out that the use of density alone as a measure of habitat quality could give rise to misleading results, especially where territorial behavior lim- its access to high quality habitat. Others (Maurer 1986, Hobbs and Hanley 1990, Winker et al. 1995) have supported her conclusion, and still others (Vickery et al. 1992, Donovan et al. 1995b, Winter and Faaborg 1999) have also pointed out that density does not necessarily track reproductive success. However, although density (measured as number of birds per area) is mathematically equivalent to probability of occurrence (with the same units), probability of occurrence based on presence/absence data is a special case of density measures with density bounded by zero and one. In fact, the only re- liable evidence of the effects of fragmentation available from census data is arguably based on the presence or absence of a species (Freemark et al. 1995, Winter and Faaborg 1999). Further, Boyce and McDonald (1999) point out that hab- itat usage involves both active habitat selection and passive persistence in a habitat. The fitness consequences of utilizing that habitat, expressed as selection on survival or reproduction (South- wood 1977), is what gives rise to the perceived patterns of distribution (Boyce and McDonald 1999). Thus, in most cases, extent of habitat use (or presence/absence) reflects fitness in those habitats (Fretwell and Lucas 1970). The patterns described above (Van Horne 1983, and others) may be exceptions to this generalization (Boyce and McDonald 1999). Thus, at first glance, it is somewhat surprising that the high sensitivity to fragmentation shown in most of the species studied was not correlated with population trends as measured by the Breeding Bird Survey. For example, while the Veery showed a significant decline survey-wide between 1966 and 1996 (trend = -1.4%, P < 0.01; Sauer et al. 1997), the Hermit Thrush, which displayed approximately the same level of fragmentation sensitivity as the Veery, showed a survey-wide significant increase over the same period (trend = +1.4%, P = 0.01; Sauer et al. 1997). The equally sensitive Swainson's Thrush displayed no significant trend at all survey-wide. Finally, the Wood Thrush, whose probability of occurrence increased with increasing fragmen- tation, has shown a strong and highly significant negative trend (trend = -1.8%, P < 0.01) over the same 30 years (Sauer et al. 1997). However, this is perhaps not a total surprise. As migrant species, the thrushes' and tana- gers' population trends reflect influences on the birds on their breeding grounds, during migra- tion, and on their wintering grounds. In the case of the Wood Thrush, population decreases co- incide with deforestation in their tropical win- tering grounds (Morton 1989) and a decrease in the survival of non-territorial "floaters" while over-wintering (Rappole et al. 1989). In contrast, the Hermit Thrush, which is the only thrush ex- hibiting an increasing population trend, is also the only species we studied that does not winter in the tropics. Thus, demonstrated sensitivity to fragmentation on the breeding grounds alone may not be sufficient for prediction of popula- tion trends of these neotropical migrant species. Data from all portions of the annual cycle are important to understand changes in migratory bird demography (Danielson et al. 1997). How- ever, at least in the case of the Wood Thrush, the preponderance of recent evidence suggests that declining trends are due, in large part, to poor reproductive success in fragmented land- scapes on the breeding grounds (Robinson and Wilcove 1994, Hoover et al. 1995, Trine 1998). Another surprising result of our analysis was the uniformly negative correlation between the TANAGERS AND THRUSHES IN EAST AND WEST--Hames et al. 91 degree of fragmentation and presence of our fo- cal species (with the exception of the Wood Thrush), which held across western and eastern regions. As recently as five years ago, Freemark et al. (1995) pointed to differences between the landscape contexts of studies of fragmentation in the East and the West as a means to explain the clear differences in levels of response be- tween the regions. They pointed to the fact that most western studies had taken place in forested regions fragmented by silviculture, as opposed to most eastern studies that took place in land- scapes where forests were fragmented by agri- culture and urbanization (Freemark et al. 1995). Our current analysis did not take the nature and extent of adjacent habitat into account because these data were not always available, but instead made comparison based on patch and landscape configuration alone. Further, Freemark et al. (1995) cited an earlier study by Rosenberg and Raphael (1986), which suggested that the lack of strong reaction by western species may also be due to relatively recent fragmentation com- bined with a time-lag in response by sensitive birds, as well as a lack of truly isolated forest patches. It is possible that the intervening 16 years was a sufficient time for a time-lagged re- sponse to become apparent, or for levels of par- asitism by Brown-headed Cowbirds to increase with increasing human populations throughout the West (Tewksbury et al. 1998). It seems just as likely, however, that our study was simply the first that undertook a large scale comparison of fragmentation effects in the East and West using the same methodology and measures of frag- mentation in both regions, and that the nature of adjacent habitat has a far from negligible effect on sensitive species' response to fragmentation. In summary, it is clear that the trends in prob- ability of detecting tanager or thrush species in landscapes with varying proportions of fragmen- tation are the same, in both direction and strength, in both western and eastern landscapes. Further, this similarity in response to fragmen- tation occurs despite differences in both the suites and abundances of predators and of nest parasites, and despite significant regional differ- ences shown in other analyses (Rosenberg et al. ! 999). The Brown-headed Cowbird increased in all landscapes with increases in the level of frag- mentation, and this effect was stronger in the West. However, all the focal species except the Wood Thrush showed strong negative effects of fragmentation on possible breeding, whatever their distribution and whatever the history of landuse in their ranges. Finally, this study dem- onstrates that the use of volunteer citizen sci- entists in conjunction with explicit, rigorous pro- tocols using playback to verify the absence of the species of interest, can be effective at ad- dressing a large-scale question such as this by gathering detailed distributional data about spe- cies of interest across North America. ACKNOWLEDGMENTS This research was conducted with funding from the National Science Foundation, the National Fish and Wildlife Foundation, the USDA Forest Service, Archie and Grace Berry Charitable Foundation, Florence and John Schumann Foundation, and the Packard Foun- dation. We also gratefully acknowledge helpful statis- tical advice from C. E. McCullough, W. M. Hochach- ka, and fieldwork by hundreds of dedicated volunteers. We also thank the editors, S. T. Knick, and an anon- ymous reviewer for comments that improved the paper. Studies in Avian Biology No. 25:92-102, 2002. THE EFFECTS OF HABITAT FRAGMENTATION ON BIRDS IN COAST REDWOOD FORESTS r. LUKE GEORGE AND L. ARRIANA BRAND Abstract. Human activities in the redwood (Sequoia sempervirens) region over the last 150 years have changed what was once a relatively continuous old-growth forest ecosystem into a highly frag- mented mosaic of young, mature, and old-growth forest patches, agricultural land, and human settle- ments. We summarize recent studies on the eflcts of forest fragmentation on diurnal landbirds in redwood forests and present new analyses of the effects of forest patch size on the distribution and abundance of breeding birds. Analyses of the relative abundance of 31 bird species in 38 patches of mature and old-growth redwood forest indicate that six species were positively correlated with forest patch area and may be sensitive to fragmentation: Pileated Woodpecker (Dryocopus pileams), Pacific- slope Flycatcher (Empidonax difficilis), Steller's Jay (Cyanocitta stelleri), Brown Creeper (Certhis americana), Winter Wren (Troglodytes troglodytes), and Varied Thrush (lxoreus naevius). These spe- cies (except the Steller's Jay) have been identified as sensitive to forest fragmentation in other studies of wet coniferous forests in the western U.S. The American Robin (Turdus migratorius), Orange- crowned Warbler (Vermivora celata), Dark-eyed Junco (Junco hyemalis), and Song Sparrow (Melos- piza melodia) were negatively correlated with patch area. Song Sparrows and Orange-crowned War- blers are more abundant in young second-growth than mature redwood forests, and American Robins and Dark-eyed Juncos are generally associated with forest openings. Thus, these four species are associated with and likely responding to habitats surrounding forest patches. Previous analyses have shown that four of the species that were positively associated with patch area, Pacific-slope Flycatch- ers, Brown Creepers, Winter Wrens, and Varied Thrushes, were less abundant at forest edges than the forest interior, suggesting that edge avoidance may be responsible for their sensitivity to fragmentation. Two species, Steller's Jay and Swalnson's Thrush (Catharus ustulatus), were more abundant along forest edges. In a previous study, we found that predation on artificial nests increased with proximity to forest edge and that Steller's Jays were observed preying on some of the nests. These and other studies suggest that several bird species are sensitive to fragmentation of old-growth and mature second-growth coast redwoods possibly due to changes in microclimate along forest edges or to increased nest predation and subsequent avoidance of forest edges. Implementation of forest practices that reduce the amount of forest edge on the landscape may reduce the potential impacts of fragmen- tation on bird species in redwood forests. Key Words: area effects; artificial nests; diurnal landbirds; edge effects; forest fragmentation; nesting success; redwoods; Sequoia sempervirens. Numerous studies have documented the negative effects of forest loss and fragmentation on birds breeding in forests of the midwestern and east- ern United States (Ambuel and Temple 1982, Askins et al. 1990, Robinson and Wilcove 1994, Walters 1998, Thompson et al. this volume) and Europe (Andrdn 1992, 1994). Furthermore, a consensus is emerging among scientists working in these regions that habitat fragmentation re- suits in increased nest predation and parasitism, thereby reducing breeding productivity and pos- sibly leading to population declines. Thompson et al. (this volume) have proposed a "top-down" hierarchical model that includes regional, land- scape-level, and local effects to explain variation in nesting success across the landscape. How- ever, there is substantial variation among studies and some results in western forests seem to con- tradict the general pattern (e.g., Tewksbury et al. 1998). This has led to suggestions that the "Eastern Paradigm" may not be applicable to western forests. Over the last 150 years, Westside forests (for- ests west of the Sierra Nevada/Cascade crest) have been extensively logged, resulting in a fragmented pattern of late-seral stage forest in a sea of younger forest (Garmen et al. 1999). Be- cause forest fragmentation has had such a dra- matic impact on birds in other regions, it has been suggested that similar effects may be oc- curring in Westside forests. However, while some species such as the Northern Spotted Owl (Strix occidentalis caurina) and Marbled Mur- relet (Brachyramphus marmoratus) show strong negative responses to forest fragmentation, stud- ies of passerines and other small bird species in Westside forests have documented few effects of forest fragmentation (Rosenberg and Raphael 1986, Lehmkuhl et al. 1991, McGarigal and Mc- Comb 1995). A number of hypotheses have been suggested to explain the lack of response of birds to forest fragmentation in Westside forests, including: (1) insufficient time for species to respond (Rosen- berg and Raphael 1986, Lehmkuhl et al. 1991), (2) limited extent of forest loss (Rosenberg and 92 FRAGMENTATION EFFECTS IN REDWOOD FORESTS---George and Brand 93 Current and Historical Distribution Of Redwood Forests l 1 FIGURE 1. Original distribution of coast redwood (Sequoia sempervirens) forests and current distribution of old-growth and mature second-growth coast redwood forest north of Point Reyes National Seashore. Current distributions are based on Landsat satellite imagery (Fox 1997). Raphael 1986, Lehmkuhl et al. 1991), (3) the matrix (generally young forest) is less detrimen- tal to nesting birds (McGarigal and McComb 1995), and (4) the species are adapted to hetero- geneous landscapes and thus to the kinds of changes that logging has produced on the land- scape (McGarigal and McComb 1995, Hejl et al. this volume). The first two hypotheses do not role out fragmentation effects but suggest that effects may only be evident in forests that have been logged extensively in the past. The latter two hypotheses imply that forest fragmentation due to logging will have little effect even in heavily logged regions of the western United States. Coast redwood (Sequoia sempervirens) for- ests have been heavily logged since the mid 1800s. Only about 3.5% of the pre-settlement distribution remains as original growth, and the current distribution of mature and old-growth redwood forest habitat is highly fragmented (Fig. 1; Larsen 1991). Logging began earlier and has occurred more extensively in redwood than in other Westside forests (Sawyer et al. 2000). Thus, the effects of fragmentation may be more evident in redwood than in other Westside for- ests. The birds of the redwood forest have not been extensively studied. However, over the past sev- eral years there have been a number of studies that have examined the effects of forest frag- mentation on the birds of the region. Our objec- tives in this paper are to: (1) present new anal- yses of bird response to patch size and nesfing success of Winter Wrens and Swainson's Thrushes (see Table 1 for scientific names of bird species studied) with respect to distance from forest edge, (2) summarize published and unpublished studies on the effects of forest frag- mentation on birds in redwood forests, and (3) compare the effects of forest fragmentation on 94 STUDIES IN AVIAN BIOLOGY NO. 25 birds in redwood forests to those found in the Midwest and the eastern United States. METHODS We describe the methods for the analysis of bird response to patch size and nesting success of Winter Wrens and Swainson's Thrushes in detail, as these analyses have not been published. Methods for esti- mates of relative bird abundance with respect to dis- tance from forest edge and the artificial nest experi- ments have been published elsewhere (Brand 1998; Brand and George 2000, 2001). STUDY AREA We conducted our studies in redwood forest patches in Humboldt County, California. Point counts that we used for analysis of bird response to patch size were conducted from I May to 15 July, 1994. Monitoring of Winter Wren and Swainson's Thrush nests took place during May-August 1998-1999. Study sites con- sisted of old-growth as well as mature second-growth (>80 years) coast redwood forests. The overstory of all stands was dominated by redwoods (>50%), but other tree species found in these stands included Doug- las-fir (Pseudotsuga menziesii), Sitka spruce (Picea sitchensis), western hemlock (Tsuga heterophylla), grand fir (Abies grandis), red alder (Alnus rubra), Cal- ifornia bay (Umbellularia californica), big-leaf maple (Acer macrophyllum), and tan-oak (Lithocarpus den- sifiorus). The understory was dominated by rhododen- dron (Rhododendron rnacrophyllum), sword fern (Po- lystichurn munitum), salal (Gaultheria shallon), Cali- fornia huckleberry (Vacciniurn ovaturn), red huckle- berry (Vaccinium parvifiorum), cascara (Rhamnus purshiana), salmonben'y (Rubus spectabilis), Califor- nia blackberry (Rubus ursinus), Himalayan blackberry (Rubus discolor), and red elderberry (Sambucus race- rnosa). The edge of each patch was defined by gaps ->100 m in the forest canopy occurring adjacent to several features such as rivers, grasslands, young forest (<30 years), residential development, and roads. Study sites were located on public lands managed by Humboldt Redwoods State Park, Redwood National Park, Prairie Creek Redwoods State Park, the City of Arcata (Arcata Community Forest), Humboldt State University Wildlife Department (Wright Wildlife Ref- uge), the City of Eureka (Sequoia Park), and Grizzly Creek State Park. Study sites were also located on Simpson Timber Company property and other private lands. Stands on privately owned land have been in- tensively managed in the past 100 years. Most of the sites on public lands have never been logged; some were logged once and are now mature stands (>100 years). For the patch size study, we used orthophotographic quadrangles of the region to identify potential forest patches characterized by >50% redwood canopy and a stand age of >80 years. From approximately 90 el- igible patches, we randomly chose 38 forest patches to survey. The size of patches ranged from 0.89 ha to 4252 ha. However, 35 of the 38 patches were <160 ha. The study sites were distributed over approximate- ly 700 km 2, all within 50 km of the Pacific Ocean. The fate of Winter Wren and Swainson's Thrush nests was studied at the Wright Wildlife Refuge, the Arcata Community Forest, and Redwood National Park. Plots were established along forest edge (edge plots) and in forest interior (interior plots, >400 m from forest edge). One edge plot was established at the Wright Wildlife Refuge, two interior and one edge plot were established in the Arcata Community Forest, and two interior plots were established in Redwood Na- tional Park. Both the Wright Wildlife Refuge and the Arcata Community forest bordered on suburban areas. BIRD RESPONSE TO PATCH SIZE To examine which passedfie bird species are sensi- tive to forest patch size and shape during the avian breeding season, we investigated the distribution and relative abundance of birds in redwood forest patches using point counts (Verner 1985). The location of the first point in a patch was randomly selected. From that point, a direction was randomly chosen to establish the succeeding points placed 200 m apart, until no further points could be placed within the patch or we had es- tablished 4 points. Most points were > 100 m from the edge of the patch. In some cases the size and shape of the patch made this impossible, but in all cases points were placed no closer than 50 m from the edge of the patch. Each patch was surveyed four times (twice by each of two observers), approximately once every two weeks. Point counts lasted 8 min, and were conducted at least 5 min apart. Some patches were too small to contain four points. In these patches, we established fewer points but maintained equal sampling effort by conducting additional counts at the points. If one point was established in a patch, then four, 8-min point counts spaced 5 min apart were conducted at one point. If a patch contained two points, two point counts were conducted 5 min apart at each point. If a patch contained 3 points, two point counts were done at a randomly chosen point, then one point count was con- ducted at the two remaining points. If four points were established in a patch, one point count was conducted at each point. All point-counts were conducted within four hours after sunrise. Data were recorded separately for each 8-min point count even if occurring 5 min apart in the same loca- tion. During an 8-min point count, birds were not counted twice unless there was a high certainty that it was a different individual of the same species. The number of birds counted at each point in each patch across all visits to each patch was summed to get an index of relative abundance for that patch. To quantify the landscape variables of habitat patch size and patch shape, we used a planimeter and ortho- photoquads to measure the area (ha) of each forest patch and a map wheel to measure the total perimeter (m) of each patch. Because perimeter length is corre- lated with area, we computed an index of patch shape using the ratio of the perimeter (m) of a given forest patch to the perimeter (m) of a circular forest patch of equal area. Both patch area and patch shape were log transformed for analysis. Because the bird data are counts, we used Poisson regression (McCullagh and Nelder 1989) to examine the effect of patch area and shape on bird abundance. Only species that were observed in at least 20% of the patches were included in the analysis. We used the FRAGMENTATION EFFECTS IN REDWOOD FORESTS--George and Brand 95 natural log of patch area to deal with wide disparity in patch areas. The natural log of patch area and patch shape were correlated (r 2 = 0.34, df = 36, P = 0.037) and therefore we used only log patch area in the anal- yses because it explained a higher proportion of the variation in bird abundances than log patch shape, and patch area is generally a better predictor of bird abun- dance than patch shape (Galli et al. 1976. Blake and Karr 1987, Askins et al. 1990). A scale parameter was included in the model, which allows the variance to be greater than the mean to allow for over-dispersion of bird detections within patches compared to a standard Poisson distribution (McCullagh and Nelder 1989). Species that were positively associated with area were considered sensitive to fragmentation. All analyses were conducted using SAS statistical software (SAS Institute 1999). NATURAL NESTS In 1998 and 1999 nests of Swainson's Thrushes and Winter Wrens were monitored in plots established along forest/suburban edges and at locations distant (>400 m) from suburban edges (J. Kranz and T L. George, unpubl, data). Nests were monitored at 3-4 day intervals until the nest failed or the young fledged. Daily Survival Rate (DSR) was computed for edge (<100 m from suburban edge) and interior (>100 m from suburban edge) nests using the Mayfield method (Hensler and Nichols 1981 ) and comparisons were per- formed using program CONTRAST (Hines and Sauer 1989, Sauer and Williams 1989). Because of small sample sizes of nests, we used a = 0.10 to reduce the chance of a Type II error. LITERATURE SURVEY We surveyed the literature for studies of the re- sponse of diurnal landbirds to forest fragmentation in wet coniferous forests of the Pacific Northwest. We classified a species as area sensitive if its abundance increased with patch size (Schieck et al. 1995; this study) or with the amount of mature or old-growth forest within a surrounding buffer. Buffers differed in extent fi'om 100 ha (Manuwal and Manuwal this vol- ume) to 250-300 ha (McGarigal and McComb 1995). Rosenberg and Raphael (1986) examined both patch size and the amount of mature or old-growth in a 1,000-ha buffer surrounding the stand. Lehmkuhl et al. (1991) examined three scales: patch size, the area ad- jacent to the patch (within 400 m of the boundary), and the landscape (circular 2,025 ha area centered on the patch). Hejl and Paige (1994) compared bird rel- ative abundance between a continuous stand of old- growth forest, an old-growth forest with 1-8 year-old clearcuts, and a selectively logged forest. A species was classified as edge sensitive if its abundance de- clined with proximity to edge (Brand and George 2001) or declined in abundance as the amount of edge increased in a surrounding buffer area. Buffer areas varied from 10 ha (Rosenberg and Raphael 1986), to 100 ha around each patch (Manuwal and Manuwal this volume), to 400 m surrounding the patch (Lehmkuhl et al. 1991). Thus there were seven studies that ex- amined area effects and four that examined edge ef- fects. We included fewer studies in our analysis than Manuwal and Manuwal (this volume, Table 1) because we only included studies that specifically addressed area or edge sensitivity. Life history characteristics (nest type, migratory status, and foraging mode) of each species were obtained from the studies included in the summary and from the literature (Ehrlich et al. 1988). Species that showed evidence of area effects in two or more studies are included in Table 3. RESULTS Thirty-one species were included in the anal- ysis of bird abundance and patch size (Table 1). Three species, the Golden-crowned Kinglet, Pa- cific-slope Flycatcher, and Wilson's Warbler, were detected in all of the patches. The abun- dances of six species, the Pileated Woodpecker, Pacific-slope Flycatcher, Brown Creeper, Stell- er's Jay, Winter Wren, and Varied Thrush, were positively correlated with log forest patch size (Table 2, Fig. 2). These species spanned the whole range of frequency values, from species that were detected in all of the patches (Pacific- slope Flycatcher) to those that were detected in a small proportion of the patches (Pileated Woodpecker). American Robins, Orange- crowned Warblers, Dark-eyed Juncos, and Song Sparrows were negatively correlated with patch size (Table 2, Fig. 2). Varied Thrushes and Pileated Woodpeckers showed a threshold response to patch area. Var- ied Thrushes were detected in only 1 out of 17 patches below and 20 out of 21 patches above 16 ha. Pileated Woodpeckers were detected in 2 of 29 patches below and 6 of 9 patches above 48 ha. None of the other species showed evi- dence of a threshold response (Fig. 2). Twenty-three Swainson's Thrush and 48 Win- ter Wren nests were monitored in the two years. Nest success for both years combined was low for Swainson's Thrushes (25%; DSR _+ sE = 0.940 _+ 0.016), whereas Winter Wrens had high nest success (65%; 0.986 + 0.016). Daily sur- vival rate of Swainson's Thrush nests close (<100m) to forest edges was lower than interior nests (0.92 -+ 0.023 vs. 0.974 _+ 0.018, respec- tively; P = 0.065) but nest success of Winter Wrens did not differ between edge and interior locations (0.991 +_ 0.0053 vs. 0.977 _+ 0.009, respectively; P = 0.17). None of the nests were parasitized by Brown-headed Cowbirds (Mol- othrus ater). LITERATURE SURVEY We found eight studies that had examined the effects of forest fragmentation on diurnal land- birds in Westside forests (Table 3). Because each study used different methods to examine these relationships and species composition varied among sites, the results must be interpreted cau- tiously. However, we felt this comparison was 96 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 1. B1RD SPEC1ES INCLUDED IN ANALYSES OF PATCH CHARACTERISTICS AND B1RD ABUNDANCE IN COASTAL REDWOOD FORESTS Proportion of patches Species occupied (N - 38) Golden-crowned Kinglet (Regulus satrapa) Pacific-slope Flycatcher (Empidonax difficilis) Wilson's Warbler ( Wilsonia pusilia ) Chestnut-backed Chickadee (Poecile rufescens) Winter Wren (Troglodytes troglodytes) Swainson's Thrush ( Catharus ustulatus) Brown Creeper ( Certhia americana) Steller's Jay (Cyanocitta stelleri) American Robin (Turdus migratorius) Hermit Warbler (Dendroica occidentalis) Dark-eyed Junco (Junco hyemalis) Song Sparrow (Melospiza melodia) Orange-crowned Warbler (Vermivora celata) Common Raven (Corvus corax) Purple Finch ( Carpodacus purpureus) Pine Siskin (Carduelis pinus) Vaux's Swift (Chaetura vauxi) Varied Thrush (Ixoreus naevius) Hutton's Vireo (Vireo huttoni) Band-tailed Pigeon (Columba fasciata) Northern Flicker (Colapies auratus) Red-breasted Nuthatch (Sitta canadensis) Western Tanager (Piranga ludoviciana) Cassin's Vireo (Vireo cassinii) Hermit Thrush (Catharus guttatus) Pileated Woodpecker (Dryocopus pileatus) 1.00 1.00 1.00 0.97 0.95 0.92 0.89 0.89 0.84 0.82 0.74 0.68 0.66 0.63 0.63 0.58 0.53 0.47 0.42 0.37 0.32 0.32 0.32 0.29 0.26 0.24 an important first step in identifying species that consistently show evidence of sensitivity to frag- mentation. Out of seven studies that examined area sen- sitivity, ten species were identified as being sen- sitive to fragmentation in two or more and seven in three or more studies (Table 3). There was no tendency for species with particular nest types or foraging modes to predominate, but the ma- jority of the species were residents. Eight of the ten species that were identified as area sensitive also showed evidence of edge sensitivity in one or more studies (Table 3). Thus, there is high concordance between area sensitive and edge sensitive species in these studies. The association between edge sensitivity and area sensitivity that we found, however, must be viewed with caution. Only one of the studies (Brand and George 2001) was specifi- cally designed to examine response to forest edge; the others were based on point counts, which may be a poor indicator of edge effects (Villard 1998). DISCUSSION Six of the 31 bird species we examined in the forest patch size analysis showed a positive as- sociation with forest patch area, suggesting that a substantial portion of the avifauna is sensitive to the effects of forest fragmentation in this re- gion. Four species, American Robins, Orange- crowned Warblers, Dark-eyed Juncos, and Song Sparrows, were more abundant in small than in large forest patches. This is consistent with the habitat associations of these species. Song Spar- rows and Orange-crowned Warblers are more abundant in young second-growth than mature redwood forests (Hazard and George 1999) and therefore are likely to be associated with the edges of mature stands. American Robins and Dark-eyed Juncos are generally associated with forest openings (Ehrlich et al. 1988) and there- fore it is not surprising that they are more abun- dant in smaller patches. Because of the extensive loss and fragmentation of mature and old-growth forest in this region, we will focus our discus- sion on those species that may be negatively af- fected by loss and fragmentation of mature and old-growth forests. Other studies in Westside forests have failed to detect strong evidence for edge or area sen- sitivity among diurnal landbirds (Rosenberg and Raphael 1986, Lehmkuhl et al. 1991, McGarigal and McComb 1995, Schieck et al. 1995). The lack of evidence in other studies may have been due to the landscapes studied and the approaches FRAGMENTATION EFFECTS IN REDWOOD FORESTS--George and Brand 97 TABLE 2. POISSON REGRESSION RELATIONSHIPS BETWEEN BIRD RELATIVE ABUNDANCE AND PATCH AREA IN 38 REDWOOD FOREST PATCHES SURVEYED IN NORTHERN CALIFORNIA IN 1994 Species response to fragmentation Slope -+ SE P Negative Pileated Woodpecker 0.62 -+ 0.25 0.015 Pacific-slope Flycatcher 0.08 -+ 0.04 0.037 Steller's Jay 0.17 -+ 0.07 0.029 Winter Wren 0.30 -+ 0.05 <0.001 Brown Creeper 1.67 -+ 0.09 0.055 Varied Thrush 0.71 -+ 0.06 <0.001 Positive American Robin -0.44 -+ 0.11 <0.001 Orange-crowned Warbler -0.76 -+ 0.26 0.004 Dark-eyed Junco -0.52 -+ 0.24 0.029 Song Sparrow -0.55 -+ 0.27 0.043 Notes: Species that were positively related to area were classified as showing a negative response to fragmentation. Those showing the opposite trend were classified as being positively associated with fragmentation. Only thoc species that occurred in al least 20% of the patches were included in the analysis. that were used. Lehmkuhl et al. (1991) and Ro- senberg and Raphael (1986) studied landscapes that were far less fragmented than the redwood forests we examined. The smallest stand exam- ined by Lehmkuhl et al. (1991) was 51 ha, and most of the area around the stands (2,025 ha) consisted of less than 50% clearcut. Few (4/46) of the stands that Rosenberg and Raphael (1986) studied were true islands (isolated from other mature stands by clearcuts or hardwood forest), and the amount of clearcut forest in the sur- rounding 1000 ha block varied from 0 to 44%. Thus the lack of evidence for sensitivity to frag- mentation in these studies may be because the landscapes were not sufficiently fragmented to affect the bird species they examined. Mc- Garigal and McComb (1995) specifically ex- amined landscapes (250-300 ha) encompassing a wide range of landscape structure based on the proportion of late-seral forest and the spatial configuration of the forest. However, they did not use a patch-centered approach, but rather ex- amined the relationship between landscape char- acteristics and average bird abundance in all ser- al stages within those landscapes. Thus, the scale of their analysis was much larger than our study. Schiek et al. (1995) used a similar ap- proach to ours but their sample of patches was small (21), and therefore their ability to detect eftcts of fragmentation may have been limited. We found no association between sensitivity to fragmentation and life history characteristics. However, most of the species were residents, which contrasts sharply with similar summaries of birds in the midwestern and eastern United States where species that have been identified as sensitive to fragmentation are more often long- distance migrants (Robbins et al. 1989b, Free- mark et al. 1995). Thus, there does not appear to be any suite of life history traits that makes a species more likely to be negatively affected by fragmentation in these forests. This suggests that attempts to classify sensitivity to fragmen- tation based on life history traits are likely to be problematical (Hansen and Urban 1992, Hansen et al. 1993). Two species, Pileated Woodpeckers and Stell- er's Jays, showed evidence of area sensitivity but not edge sensitivity. Pileated Woodpeckers have large territories (>300 ha) in western co- niferous forests (Bull and Holthausen 1993), and therefore small isolated forest patches may be less suitable for nesting and foraging. Hejl (1992) also found that Pileated Woodpeckers showed a threshold response to forest patch area in the northern Rockies and suggested that large stands or aggregates of small stands of late-seral forests are necessary to maintain suitable habitat for this species. Brand and George (2001) found that Steller's Jay abundance declined with dis- tance from edge in redwood forests, which is inconsistent with their area sensitivity. Rosen- berg and Raphael (1986) also found that Steller's Jays were more abundant along edges, and that they were weakly negatively associated with an index of insularity. Thus, the evidence for area sensitivity in Steller's Jays is weak in both stud- ies (Rosenberg and Raphael 1986; this study), and therefore their designation as area sensitive may be a statistical artifact. Eight of ten species that showed sensitivity to fragmentation also showed evidence of edge sensitivity. This suggests that area sensitivity may be related to edge avoidance in these spe- cies. Although edge sensitivity is often assumed to be associated with area sensitivity (Whitcomb et al. 1981, Askins et al. 1990, Freemark and Collins 1992), Villard (1998) found a poor cor- 98 STUDIES IN AVIAN BIOLOGY NO. 25 14 12 : 10 o u 8  6 6 4 z 2 0 0.1 American Robin Orange-crowned Warbler 9 ß 6 eeß ß ee 3 2 1 10 100 1000 10000 0.1 1 10 100 1000 10000 16 14 : 12 o lO  8 Q 6 6 4 z 2 0 0.1 Dark-eyed Junco ß 18 16 14 12 10 8 Song Sparrow ß ß ß ß ß ß ß ß ß 4 ß ß 2 0 1 10 100 1000 10000 0.1 1 10 100 1000 10000 Pileated Woodpecker Pacific-slope Flycatcher 3O ß 3.5  3 25 ß eee ee ß õ 2.5 20 ..  2 ß 15 ß ß ß ß ß  1.5 6 1 # ee ß 10 ß z 0.5  #--e 5 ß 0 0.1 1 10 100 1000 10000 0.1 1 10 100 1000 10000 FIGURE 2. Relationship between relative density and patch area for bird species in redwood (Sequoia sem- pervirens) forest patches in northern California. Species that show a positive COlTelation between patch area and relative abundance are considered area sensitive. Fitted Kine is best fit Poisson regression with log link function. relation between edge- and area-sensitive spe- cies in studies conducted in the eastern United States. There are many factors that change between forest edges and interior locations that may in- fluence bird abundance, such as differences in predation (Paton 1994), microclimate (Chen et al. 1993), vegetation structure (Ranney et al. 1981), and insect composition (Shure and Phil- lips 1991). These factors may act singly or in combination to make forest edges more or less suitable to particular species. For instance, mois- ture gradients may influence the abundance of ground-dwelling arthropods, which in turn could affect the distribution of ground foraging bird species, as has been suggested for Ovenbirds (Seiurus aurocapillus; Gibbs and Faaborg 1990). Reduced moisture along forest edges may play an important role in the edge avoidance for several of the species. Winter Wrens breed in moist coniferous forests and nest in dense brush, especially along stream banks (Ehrlich et al. 1988). Barrows (1986) found that Winter Wrens in California have broad habitat preferences in fall and winter, but that habitat selection shifts in the breeding season almost exclusively to old- growth forest characterized by a dense, moist understory. Likewise, McGarigal and McComb (1995) found that Winter Wrens are associated with riparian systems in Oregon. The Varied Thrush breeds in moist coniferous forest (George 2000) and song post locations are as- sociated with large diameter trees, on steep slopes, surrounded by a high density of trees 14 12  lO o , 8 c,, 6 4 2 0 0.1 FRAGMENTATION EFFECTS IN REDWOOD FORESTS--George and Brand 99 Steller's Jay Brown Creeper 1 10 100 1000 10000 10 8 7 6 5 4 3 2 1 0 0.1 1 10 100 1000 10000 30 .o 20  15 ß 10 o z 5 0 0.1 FIGURE 2. Winter Wren 1 10 100 1000 10000 Patch Area (ha) Continued. Varied Thrush 35 3O ß 25 ß ß  20 ß ß ß Jß 15 lO  ß o.1 1 lO lOO looo 10000 Patch Area (ha) near streams (Beck and George 2000). Thus, male thrushes prefer moist, shady locations for song posts. The Pacific-slope Flycatcher breeds in forests, especially near water (Ehrlich et al. 1988). Edges receive higher levels of incident radiation (Chen et al. 1993), and thus the micro- climate near edges may be unsuitable for these species. Microclimate changes, in turn, could af- fect vegetation composition and structure as well as prey availability near edges. Another factor that may cause bird species to avoid edges is predation (Brittingham and Tem- ple 1983). The mechanism is less clear in this case but it could either be a direct response to the presence of potential predators along edges or occur indirectly as unsuccessful nesters move TABLE 3. BIRD SPECIES IDENTIFIED IN Two OR MORE STUDIES AS SHOWING EVIDENCE OF SENSITIVITY TO FOREST FRAGMENTATION IN WET CONIFEROUS FORESTS OF THE PACIFIC NORTHWEST Nest Migratory Foraging Area Edge Species type a status b mode c sensitive d sensitive d Pileated Woodpecker Cavity R Drill 1, 3, 7 Pacific-slope Flycatcher Cup L Flycatch 6, 7 1, 8 Steller's Jay Cup R Omnivore 1, 7 Chestnut-backed Chickadee Cavity R Foliage 1, 3, 5, 6 1, 6 Red-breasted Nuthatch Cavity R Bark 3, 5, 6 1, 8 Brown Creeper Crevice R Bark 3, 4, 7 1, 8 Winter Wren Crevice R Ground 1, 2, 3, 4, 6, 7 1, 2, 8 Golden-crowned Kinglet Cup R Foliage 4, 6, 7 1 Varied Thrush Cup S Ground 3, 5, 6 8 Hermit/Townsend's Warbler Cup L/S Foliage 1, 6 1 a Cavity-nest in tree cavities; Crevice-nest in niches and behind bark; Cup-open cup nesters. b L-long-distance migrant; R-resident; S-short distance migrant. c Bark-bark gleaner; Drill-excavates insects from dead wood; Flycatch-sallies for insects from a perch; Foliage-gleans insects from foliage; Ground- gleans insects from ground; Omnivore-leds on a variety of food types. d Studies included: l-Rosenberg and Raphael (1986); 2-Lehmkuhl et al. (1991); 3-McGarigal and McComb (1995); 4-Hejl and Paige (1994); 5- Schieck et al. (1995); 6-Manuwal and Manuwal (this volume, Table 6); 7-this study; 8-Brand and George (2001). 100 STUDIES IN AVIAN BIOLOGY NO. 25 7 6 asym Relative 4 Density 3 2 1 Varied Thrush ] ø 510 160 1}0 260 2}0 360 3}0 460 Distance from Edge (meters) FIGURE 3. Relative density with respect to distance from the forest edge and estimated edge width for the Varied Thrush. The points represent the band-specific relative density. The smooth curve represents the relative density based on an exponential regression model with one asymptote. The dash-dot line illustrates the edge width, defined as the distance from edge at which 90% of the asymptotic interior relative density has been achieved. to new locations (Villard 1998). Brand and George (2000) found that predation on artificial nests that mimicked Varied Thrush and Winter Wren nests declined with distance from edge in redwood forest patches and that Steller's Jays were observed preying on the nests on several occasions. These results are consistent with the hypothesis that Winter Wrens and Varied Thrushes avoid forest edges because of higher nest predation. Steller's Jays are also more com- mon on forest edges than forest interior loca- tions (Brand and George 2001) and thus their presence could provide a proximate cue to nest- ing birds. Other studies of artificial and natural nests have shown similar patterns with respect to dis- tance from forest edge but there are many ex- ceptions as well (Brand and George 2000, Sisk and Battin this volume). In addition, some stud- ies suggest that predation rates on artificial nests may not reflect predation on real nests (Nour et al. 1993, Haskell 1995a, Willebrand and Marc- strom 1988, Wilson et al. 1998, Ortega et al. 1998, King et al. 1999). We found no difference in nesting success between edge (< 100 m from forest edge) and interior (>100 m) nests for Winter Wrens, but nesting success of Swainson's Thrushes was lower on edges. Thus the pattern of decreasing nesting success with proximity to forest edge appears to be species-specific and more studies are needed to document the gen- erality of this pattern. Swainson's Thrush populations may be partic- ularly vulnerable to increased predation along edges because thrushes are more abundant along edges in redwood forest patches (Brand and George 2001). Thus, thrushes may be experi- encing an ecological trap (Gates and Gysel 1978) in this region, which could have severe effects on recruitment and population growth (Donovan and Lamberson 2001). Swainson's Thrush populations may be suffering poor re- cruitment in other parts of their range. Bednarz et al. (1998) found that Swainson's Thrushes are experiencing low nesting success in central Ida- ho, which they attributed to high levels of forest fragmentation in the region. Swainson's Thrush- es have also been included in a draft list of spe- cies of special concern in California because of declines and a shrinkage of their breeding range in the Sierra Nevada mountains (T. Gardali, pers. comm.). Regardless of the mechanism, edge avoidance has important implications for forest manage- ment. Information on the distance over which edge effects occur could provide important man- agement guidelines for minimum widths of for- est stands. Brand and George (2001) found that the distance to 90% of asymptotic interior rela- tive density varied from 85 m for the Brown Creeper to 140 m for the Varied Thrush (Fig. 3). The average distance to 90% asymptotic density of the four forest interior species is approxi- mately 115 m. The distance of 115 m from the forest edge also corresponds with the distance at which the probability of predation on artificial FRAGMENTATION EFFECTS IN REDWOOD FORESTS--George and Brand 101 nests declines by half (Brand and George 2001). The edge widths estimated in Brand and George (2001) can be used to predict the patch sizes that may be suitable for particular forest interior spe- cies. For example, assuming that the average ter- ritory size for Varied Thrushes is 4 ha (George 2000), a circular patch of 19.6 ha would provide a 4 ha core with a 140 m buffer. Breeding Varied Thrushes were found to require a minimum patch size of approximately 16 hectares in coast redwood forests (Hurt 1996), close to the pre- dicted size. Another pattern that has been observed in studies in the eastern U.S. (Wilcove 1985) and Europe (Andrdn et al. 1985, Angelstam 1986, Andrdn and Angelstam 1988, Andrdn 1992), is an increase in nest predation in forest fragments embedded within urban or agricultural land- scapes as compared to regenerating forest or other more natural habitats. This may be due to an increase in generalist predators in landscapes that are dominated by agricultural or urban areas (Thompson et al. this volume). In redwood forest fragments, however, Brand and George (2000) found that rates of predation on artificial nests adjacent to rural (grassland) edge were signifi- cantly higher than nests located adjacent to sub- urbs, rivers, young forests, or roads. Thus, our results suggest that landscape context has a very different eflbct on rates of nest predation in the redwood region than in the eastern U.S. and Eu- rope. Our results are consistent with those of Tewksbury et al. (1998) who found that rates of nest predation in riparian forests in Montana were higher in sites adjacent to undisturbed co- nifer forests than those adjacent to agricultural areas. Thus landscape context may not exert a predictable influence on rates of nest predation in western forests as it does in the eastern U.S. and Europe, perhaps due to the diversity of hab- itats and associated nest predators in the West. It is also possible that the various landscapes examined by Brand and George (2000) and Tewksbury et al. (1998) were not sufficiently different at the regional level to influence the predator community (Thompson et al. this vol- ume). Predation on artificial nests appears to be sub- stantially lower in redwood forests than other forests. Approximately 69% of the artificial ground nests and 55% of the arboreal nests were intact after 14 days, which is substantially higher than has been found for most other studies con- ducted in fragmented forests of the eastern U.S. (Wilcove 1985, Yahner and Cypher 1987, Rud- nicky and Hunter 1993, Whelan et al. 1994, Fen- ske-Crawford and Niemi 1997, Yahner and Ma- han 1997). This difference may reflect lower overall avian abundance as well as lower pred- ator activity in mature and old-growth redwood forests than in eastern deciduous forests. Each of the species that showed evidence of area sensitivity in our survey also has been iden- tified as an old-growth associate in one or more regions of the Pacific Northwest (Manuwal and Manuwal this volume, Table 2). This suggests that there may be an association between area sensitivity and dependence on old-growth forest habitat among the birds in this region. If this is the case, loss and fragmentation of old-growth forests may have a more severe impact on these species than predictions based on the area of old-growth forest alone. EAST VS. WEST The proportion of species showing evidence of sensitivity to habitat fragmentation in red- wood forests (6/31 or 19%) is lower than the proportion that has been reported for studies in the eastern U.S. For example, Freemark and Collins (1992) reported that 34/70 or 49% of the species they examined showed evidence of area sensitivity, which is significantly higher than the proportion we observed (X 2 = 7.67, df = 1, P = 0.006). The proportions may change depending on what bird orders are included and the studies considered, but the pattern of a higher propor- tion of area sensitive species in forests of the eastern and midwestern U.S. relative to redwood forests is unlikely to change. In addition, given the overlap in species identified as area sensitive in the studies we examined, it is likely that this pattern holds for all Westside forests. We also found few long-distance migrants among the species that are area sensitive, which is very dif- ferent from the eastern and midwestern U.S. where long-distance migrants predominate. Our studies also suggest that the ecological processes that are responsible for area sensitivity among redwood forest birds may differ from those in the eastern U.S. Thompson et al. (this volume) have proposed a "top-down" hierarchi- cal model where higher agricultural and human habitation at the regional scale results in in- creased predator and parasite numbers which in turn reduces the nesting success of birds in these landscapes. Contrary to the predictions of this model, we found that predation on artificial nests was significantly higher along natural grassland edges than suburban edges or roads. In addition, although predation on artificial nests declined with distance from forest edge, this pattern dif- fered among species when we examined natural nests. Parasitism also was not a factor as none of the nests we monitored were parasitized by Brown-headed Cowbirds. Our studies suggest that area sensitivity in some species may be a result of edge avoidance and subsequent decline 102 STUDIES IN AVIAN BIOLOGY NO. 25 1800 1600 1400 1200 1000 800 600 400 200 0 0-10 10-20 20-30 3040 40-50 50-60 60-70 70-60 80-90 90-100 >100 Area (ha) FIGURE 4. Size distribution of mature and old-growth redwood (Sequoia sempervirens) forest patches north of Point Reyes National Seashore. Based on Landsat satellite images (Fox 1997). in small forest patches. This suggests a "bottom- up" mechanism where behavioral responses to edge result in changes in abundance in different sized patches. MANAGEMENT IMPLICATIONS Several bird species that breed in coast red- wood forests are negatively affected by forest fragmentation. This means that the regional abundance of these species will be affected not only by the amount of mature and old-growth forest but also its distribution across the land- scape. Most redwood forests are privately owned and are intensively managed for timber production, and it is unlikely that large amounts of land will be added to parks and reserves (Thornburgh et al. 2000). Thus, the abundance of these species in the region will be greatly in- fluenced by how forest practices affect the dis- tribution of mature forests across the landscape. Presently, 79% of the mature and old-growth redwood forest patches north of Point Reyes Na- tional Seashore are less than 10 ha (Fig. 4). This is below the threshold for breeding occupancy by Varied Thrushes, and many of these patches may be poor or unsuitable habitat for the other species that are sensitive to fragmentation. Changes in forest practice rules that result in larger patches of mature forest on the landscape would greatly benefit these species and should be encouraged. ACKNOWLEDGMENTS Special thanks go to M. Hurt, C. Campbell, K. Mel- ody, M. Wuestehube, J. Powell, and D. Kwasny who helped collect data. J. Kranz kindly provided data on nesting success of Swainson's Thrushes and Winter Wrens. S. Elliot helped with the preparation of the manuscript. Public land managers of Prairie Creek Redwoods State Park, Redwood National Park, Hum- boldt Redwoods State Park, and the Arcata Commu- nity Forest, as well as private landowners, were par- ticularly helpful in granting permission to conduct this research. This study was funded by the Humboldt Area Foundation. Support for A. Brand during preparation of this paper was provided by SERDP project CS- 1100. Studies in Avian Biology No. 25:103-112, 2002. EFFECTS OF HABITAT FRAGMENTATION ON BIRDS IN THE COASTAL CONIFEROUS FORESTS OF THE PACIFIC NORTHWEST DAVID A. MANUWAL AND NAOMI J. MANUWAL Abstract. Few studies have been done in the Pacific Northwest on the effects of habitat fragmentation on birds. Comparisons among studies is difficult because of different study designs and possible regional variation in bird response. Timber harvesting and human settlements have greatly fragmented the once vast amounts of old-growth forests. Forest patches of the Pacific Northwest are typically surrounded by forests of different ages rather than agricultural lands, as is found in much of eastern North America. In Washington, one three-year study showed that overall bird species richness and abundance varied little in a managed coniferous forest despite differing degrees of fragmentation. Some individual species, however, increased or decreased with the amount of clearcut area and other landscape variables. Species associated with open habitats or edges increased, while those associated with forests having a well-developed canopy decreased. There is substantial variation in avian response to landscape variables that characterize watersheds. At the stand level, canopy dwellers and cavity nesting species show the most negative response to increasing levels of canopy reduction, whereas species associated with the ground or shrub layer are least affected. Cowbird parasitism is negligible in the mountains of the Pacific Northwest, but apparently is more widespread in the large valleys such as the Puget Sound lowlands and Oregon's Willamette Valley where more farmland and urban, non- forest environments exist. More studies are needed on fragmentation effects on birds and cowbird parasitism in the region. Key Words: birds; habitat fragmentation; Pacific Northwest. Natural forces such as fire, floods, and volcanic eruptions have always created natural heteroge- neity, but humans have accelerated fragmenta- tion and caused reductions in suitable habitat in some biomes. In the early days of wildlife man- agement, managers were encouraged to create fragmentation and edges since game species thrived in this environment (Leopold 1933, Al- len 1962). With more knowledge of the biology of non-game species, we now know that there are edge-sensitive species that often decline in highly fragmented landscapes (Whitcomb et al. 1981, Ambuel and Temple 1983, Wilcove and Whitcomb 1983). The increased concern over the fate of neotropical migrant passerines has re- sulted in numerous studies in eastern North America (e.g., Howe 1984, Temple and Cary 1988, Robbins et al. 1989a, Terborgh 1989, Wil- cove and Robinson 1990, Freemark and Collins 1992, Robinson 1992, Faaborg et al. 1995, King et al. 1998, Friesen et al. 1999, Rosenberg et al. 1999). Thus, most of the published information on this topic for the United States derives from research done east of the Rocky Mountains. Based on the many studies of birds in the eastern portions of North America, the principal effects of forest fragmentation on birds are: (1) reduction in patch size and change of patch shape appear to negatively affect area-sensitive species, (2) species especially adapted to living in edge habitats increase, and (3) depending on landscape context, the increase in the amount of edge results in elevated predation rates and in- creased brood parasitism by the Brown-headed Cowbird (Molothrus ater). Few studies have been conducted on the effects of forest fragmen- tation on birds in the Pacific states. Until re- cently the emphasis has been on relating bird populations to forest age and structural charac- teristics (e.g., Manuwal and Huff 1987, Carey et al. 1991, Gilbert and Allwine 1991, Hansen et al. 1991, Manuwal 1991, Ralph et al. 1991; Han- sen et al. 1995a,b). Our approach in this paper is to evaluate the effects of forest fragmentation on birds by reviewing published as well as un- published studies of birds in the coniferous for- ests of western Washington, western Oregon, and northwestern California, and to present new information from three studies in Washington and Oregon. RESULTS CHARACTERISTICS OF FOREST FRAGMENTATION IN THE PACIFIC NORTHWEST Until Euro-American settlement of the area about 150 years ago, forests in the Pacific North- west were heterogeneous due to natural events such as wildfires. Approximately 50-60% of the forest land base was old-growth forest at the time of settlement (Franklin and Spies 1984, Booth 1991). Due to timber harvesting and other land use activities, only about 20% of the pre- settlement old-growth Douglas-fir (Pseudotsuga menaiesii) forests remain (FEMAT 1993). Due to different management goals, the remaining forest is fragmented in a variety of ways (Figs. 1 and 2). The forests of this region are under federal, 103 104 STUDIES IN AVIAN BIOLOGY NO. 25 FIGURE 1. Typical forest fragmentation in the Oregon and Washington Cascades, Willamette National Forest, Oregon. Photo courtesy of U.S. Forest Service. Photo taken on 12 July 1987. state, or private management. Private manage- ment, which includes forests managed by timber companies, forests owned by private ownership, and forests on Indian lands, traditionally have been harvested for profit as the major objective. This has resulted in large clearcuts, some over 1,000 ha. These large clearcuts are in various stages of regeneration, and some have been con- vcrted into plantations, which typically have a rotation time of 40-60 years (Garmcn ct al. in press). This does not allow for development of structure associated with mature or old-growth forests (>200 years; FEMAT 1993). These lands are regulated by state laws that mandate a ripar- ian zone buffer, but this is generally narrow and susceptible to edge effects such as windfall and increased insect infestation due to stress on the trees. The federal lands are managed by agencies with different mandates. The lands administered by the National Park Service, and those desig- nated as wilderness (which in this region are managed by the Forest Service) have a policy of no forest harvesting. Thus, they serve as a refuge for large (>1,000 ha) patches of old- growth forest. These protected forests are often at high elevation, or are bordered by forests that have undergone extensive cutting. The majority of the lands managed by the Forest Service have been harvested by cutting of small patches of FRAGMENTATION AND BIRDS IN COASTAL FORESTS--Manuwal and Manuwal 105 FIGURE 2. Digitized satellite image of western Washington in the Mount Rainier National Park area. Arrow denotes park boundary. Courtesy of C. Grue and K. Dvornich, Washington Gap Analysis. forest within the old-growth matrix, which has resulted in a checkerboard effect (Franklin and Forman 1987). With time, further cuts between these areas have resulted in different-aged seral forests within the old-growth matrix, causing a loss of large (>1,000 ha) continuous old-growth areas. This technique also results in more edge area than the harvesting practices of the private sector (Spies et al. 1994). The Bureau of Land Management harvesting policy results in mid- sized patches. The study conducted by Chen et al. (1992) provides insights into the effect of clearcuts on adjacent old-growth forests. They report that these eflcts include: (1) reduced canopy cover, (2) increased growth rates of Douglas-fir and western hemlock (Tsuga heterophylla), (3) ele- vated rates of tree mortality, and (4) more Doug- las-fir and western hemlock seedlings but fewer of Pacific silver fir (Abies amabilis; Chen et al. 1992). The eftcts of clear-cutting on vegetation characteristics of old-growth Douglas-fir Ibrests ranged from 16 to 137 m for variables related to distance from the edge. Thus, some forest patch- es, especially those less than 10 ha, may be too small to preserve an interior forest environment (Chen 1991 ). In Washington, approximately half of the 9,971,625 ha classed as forest lands are admin- istered by federal agencies (McGinnis et al. 1997). Of this, about 11% is wilderness. In Oregon, Spies et al. (1994) clarified the diftring rates of harvest in private and public ownership on a 2,589-km 2 study area. Between 1972 and 1988 the closed forest canopy declined from 71% to 58%. In those areas under private own- ership, the decrease was from 50% to 28%, for a net loss of 45%. The non-wilderness lands un- FRAGMENTATION AND BIRDS IN COASTAL FORESTS Manuwal and Manuwal 107 TABLE 2. FRAGSTATS INDICES USED IN LANDSCAPE ANALYSIS OF BIRD SPECIES ABUNDANCE AND COMMUNITY CHARACTERISTICS Index name (units) Description a CCAREA (ha) CCED (m/ha) MAT_AREA (ha) PATCHES ED (m/ha) MNN (m) SHDI IJI (percent) CONTAC (percent) Total area of clearcuts (3-8 yrs old) Total amount of clearcut edge Total area of mature forest (50-80 yrs old) Number of patches Edge density Mean nearest neighbor index Shannon's Diversity Index Interspersion and juxtaposition index Contagion index a See McGarigal and Marks (1995) for a complete description and definition of each index. dividual species abundance and six of the nine FRAGSTAT indices. Nine bird species had a positive and eight species had a negative rela- tionship with total clearcut area (CCAREA; Ta- ble 3). Virtually all species with a positive re- sponse (Table 3) are known to be associated with open, shrubby habitats, so even at the land- scape level, these species tend to be most com- mon in a landscape with a large amount of land in clearcuts. All nine bird species typically for- age or nest either on the ground or in shrubs and small trees. These species are known as pioneer species and typically are the first ones to colo- nize recent clearcuts and fire sites. On the other hand, species having negative responses, such as the Winter Wren (Troglodytes troglodytes), Golden-crowned Kinglet (Regulus satrapa) and Chestnut-backed Chickadee (Poecile rufescens), are most often associated with forests with a well-developed canopy, so their response is somewhat predictable. Eight species were positively correlated with total area of mature forest (MAT. AREA; Table 3). The Pacific-slope Flycatcher (Empidonax dif- ficilis), Wilson's Warbler (Wilsonia pusilia), Hermit-Townsend's Warbler (either Dendroica occidentalis or D. townsendi or their hybrids; see Rohwer and Wood 1998), Red-breasted Nut- hatch (Sitta canadensis), Hairy Woodpecker (Pi- comes villosus), and Evening Grosbeak (Coc- cothraustes vespertinus) all had significant pos- itive responses to the amount of mature forest in the 100 ha circle. The Varied Thrush (Ixoreus naevius) and Winter Wren also had negative re- sponses to clearcuts, so these two species may be attracted at the landscape level to more ex- tensive stands of mature forests away from clearcuts. The Orange-crowned Warbler (Vermivora ce- lata) was the only species associated with the amount of clearcut edge. Chestnut-backed Chickadees had a negative association with edge density, indicating that this bird may be an area- sensitive species. The Swainson's Thrush (Ca- tharus ustulatus) was negatively associated with an increasing number of habitat patches. Alter- natively, the Dark-eyed Junco (Junco hyemalis), White-crowned Sparrow (Zonotrichia leuco- phrys), and Spotted Towhee (Pipilo maculatus) were positively associated with interspersion and juxtaposition. This seems to suggest that these species are attracted to habitat patchiness. At the community level, no significant rela- tionships were found between bird species rich- ness and area of clearcuts or area of mature for- ests in any of the three years of the study. Sim- ilarly, no significant relationships were found between the number of bird detections and area of clearcuts or area of mature forests. Oregon McGarigal and McComb (1995) investigated bird community response to landscape structure in the central Oregon Coast Range. They sam- pled 10 landscapes (250-300 ha) in three basins. Each landscape was characterized by the amount of late-seral forest condition and relative frag- mentation. Among the many bird species de- tected, 12 species were strongly associated with late seral forest condition but were also found in other forest conditions. Three species, the Olive- sided Flycatcher (Contopus borealis), Red-tailed Hawk (Buteo jamaicensis), and Western Wood- Pewee (Contopus sordidulus) were associated with habitats where there was a sharp edge be- tween late-seral and early seral forests. Five spe- cies were positively associated with patch size: Gray Jay (Perisoreus canadensis), Brown Creeper, Winter Wren, Varied Thrush, and Chestnut-backed Chickadee. The following spe- cies were more abundant in fragmented land- scapes: Red-breasted Sapsucker (Sphyrapicus ruber), Western Wood-Pewee, Olive-sided Fly- catcher, and Red-tailed Hawk. The Winter Wren showed the most aversion to fragmented land- scapes. Meyer et al. (1998) and Franklin and Gutierrez (this volume) examine the relationship 108 STUDIES IN AVIAN BIOLOGY NO. 25 ++++++++ I I I I I +++ FRAGMENTATION AND BIRDS IN COASTAL FORESTS--Manuwal and Manuwal 109 between habitat fragmentation and Spotted Owls (Strix occidentalis). In general, McGarigal and McComb (1995) found a large amount of variation in response to a wide variety of landscape variables. Part of the difficulty in assessing species responses to hab- itat variables is the scale at which the compari- sons was made. Bird abundance was generally greater in more fragmented landscapes. As is true for many other studies, uncommon species or those with large territories such as the Pile- ated Woodpecker (Dryocopus pileatus), are gen- erally undersampled and their relationship with landscape variables could not be determined. California Raphael (1984) and Rosenberg and Raphael (1986) assessed the effects of forest fragmenta- tion in Douglas-fir forests of northwestern Cal- ifornia by examining point count survey data relative to 10 fragmentation measures at the plot (N = 136), stand (N = 46), and landscape lev- els. In general, bird species richness increased in fragmented stands. They also found that bird species richness at the plot and stand levels in- creased with proximity and extent of adjacent clearcut. They found 20 species associated with edges and 20 other species that avoided edges. Among the common species, only the Olive-sid- ed Flycatcher was detected more often on the edge than in the forest interion Birds showing the most negative responses to forest fragmen- tation were the Spotted Owl and Pileated Wood- pecker, whereas the Sharp-shinned Hawk (Ac- cipiter striatus) and Blue Grouse (Dendragapus obscurus) showed less population declines in fragmented areas. LOCAL AND STAND-LEVEL EFFECTS Washington riparian zones In an attempt to determine the response of birds to harvest with two different riparian zone buffer widths, eighteen riparian areas within co- niferous forests in the western Washington Cas- cades were studied in 1993, 1995, and 1996 (Pearson and Manuwal 2001). The clear-cuts created adjacent to the sampled riparian zones caused forest fragmentation and created large amounts of edge along the streams. Ten point count stations were visited where birds were counted for 6 min to determine avian relative abundance. Each study site was visited 5-6 times during the nesting season. All sites were studied for one year before harvest and sampled for two years after harvest to evaluate bird re- sponse to the buffer widths. Species richness was higher after harvest in the uplands compared with unharvested con- trols. Wider buffer widths had higher species richness than did unharvested sites. Predictably, species considered to be edge species, for ex- ample Dark-eyed Junco, Song Sparrow (Melos- piza melodia), and Warbling Vireo (Vireo gil- vus), increased after harvest. Some species, tably the Golden-crowned Kinglet, decreased significantly after harvest. Washington and Oregon green tree retention An experimental on-going study initiated in 1992 in the Pacific Northwest, called Demon- stration of Ecosystem Management Options (DEMO), is designed to examine the effects of stand-level green tree retention on ecological at- tributes of the forest. This was a daunting task because of the scale of the study and public con- cern over continued cutting on National Forest lands. Details of the study design are given by Aubry et al. (1999). In general, it consists of a randomized block design of six treatments rep- resenting varying levels of green-tree retention. Each treatment unit is 13 ha in size and leave- trees (trees remaining after harvest) were either clumped (aggregated) or dispersed through the harvested area. Study sites were only in upland areas. There are four blocks in Oregon and four in Washington. There is substantial variation in el- evation between blocks (210-1,710 m), but usu- ally only about 200-300 m variation within a block (Aubry et al. 1999). Birds were surveyed for two years before experimental retention har- vests were made and only two blocks in Wash- ington were surveyed after harvest since the oth- er two blocks had not yet been harvested. An overview of this project and preliminary results of pre-treatment sampling is in Lehmkuhl et al. (1999). We report here some preliminary and geographically limited results of the responses of the following groups of birds: cavity-nesters, forest floor-dwellers, and canopy-dwellers (Ta- ble 4). Birds were surveyed by both point counts (4 points, 160 m apart, 6 visits) and territory- mapping (11 species only). Among the three groups of species, forest floor-dwellers appeared to be less impacted by green-tree removal than the other two groups. Bird populations declined in virtually all con- ditions after harvest, even the control (100% re- tention) sites. The spring of 1998 was cold and wet in the Washington Cascades and several spe- cies of birds either failed in their first nesting attempt or nested late in the season (D. Manu- wal, pers. obs.; M. Leu, pers. obs.). This may account for the lower than expected numbers of birds in control sites. Forest floor birds appar- ently recognize 75% retention sites as little dif- ferent from untreated (100%) retention sites since there was no change in populations (Table 110 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 4. PERCENT CHANGE IN NUMBER OF BIRO TERRITORIES TO GREEN-TREE RETENTION LEVELS AFFER HAR- VEST IN WASHINGTON IN 1998 Cavity nesters a Canopy-dwellers a Forest floor-dwellers a Level of retention Butte Paradise Hills Butte Paradise Hillq Butte Paradise Hills 100% Retention (--0%) b -67 47 -30 -48 -48 23 75% Aggregated (25%) -73 -73 -76 -73 +29 -29 40% Dispersed (-60%) -64 91 -66 -95 -26 47 40% Aggregated (-60%) -48 54 -79 -53 -24 61 15% Dispersed (-85%) 80 -82 -93 -89 48 - 18 15% Aggregated( 85%) -79 -85 -87 91 -51 -50 a Cavity-nesters included: Brown Creeper, Chestnut backed Chickadee and Red-breasted Nuthatch; canopy-dwellers included: Chestnut backed Chick- adee, Hermit Warbler, and Pacific-slope Flycatcher; forest floor-dwellers were: Dark eyed Junco, Winter Wren, Varied Thrush.  Amount of canopy reduction. 4). It seems clear that both dispersed and aggre- gated 15% retention offers little habitat for cav- ity-nesters and canopy-dwellers. The declines in number were close to the decline in green-tree canopy levels. These results and interpretations are preliminary and additional post-treatment sampling may show more definitive trends in bird community and individual species respons- es. The adjustment of bird territory placement relative to retention level and dispersion is an especially interesting aspect of the study. Two examples of how birds adjusted their territories are the Dark-eyed Junco and the Hermit Warbler. The junco was a common bird on the study site, having 3 whole territories and 5 partial territo- ries on a single 40% aggregated retention treat- ment site (Butte) before harvest. After harvest, there were 3 whole territories and 3 partial ter- ritories. Each junco territory contained portions of the retention circles as well as cleared area. This fits with the anticipated response of an edge species. Before harvest, the Hermit Warbler was the most abundant species on the study site; there were 12 complete territories and 5 partial territories on the site. After harvest all but 5 ter- ritories disappeared and each of those were lo- cated such that there was one territory per cir- cular retention patch. Apparently, the patch con- tained a sufficient amount of canopy and asso- ciated insect prey to allow nesting to occur. We have no data on breeding success but all five males were paired. With additional post-harvest sampling in both Oregon and Washington, stron- ger conclusions can be drawn from this inves- tigation on the response of birds to fragmenta- tion at the stand level. OTHER INDIVIDUAL SPECIES STUDIES There are some studies of the effects of frag- mentation on species of conservation concern in the Pacific Northwest such as the Spotted Owl (Meyer et al. 1998, Franklin and Gutierrez this volume), which is strongly positively associated with several landscape attributes of late succes- sional forests. There are on-going studies of fragmentation effects on the Marbled Murrelet (Brachyramphus marmoratus; Raphael et al. this volume). As with studies of eastern bird com- munities, some species such as the Gray Jay, Brown Creeper, Winter Wren, Varied Thrush, and Chestnut-backed Chickadee tend to decrease with fragmentation and are often associated with late successional forests (Rosenberg and Rapha- el 1986, Manuwal 1991). A long-term study of Northern Goshawk (Ac- cipiter gentilis) demography, breeding behavior, and habitat selection for foraging and nesting on Washington's Olympic Peninsula was initiated in 1995 by Dan Varland and John Marzluff. To- gether with graduate students Sean Finn and Tom Bloxton, they are investigating the effects of the local- (forest stand) and landscape-level structure, composition, and spatial arrangement of forests on goshawks. The emphasis of the study is to understand how goshawks respond to habitat loss and fragmentation resulting from timber harvest. The first three years of study concentrated on surveying all known occupied nest areas on the Olympic Peninsula (N = 30) to determine if past habitat modification was correlated with current occupancy. Occupied stands differed from unoccupied ones primarily in having greater canopy closure, although the percentage of the surrounding landscape cur- rently comprised of regenerating forest also was negatively correlated with occupancy. There- fore, fragmentation of the mature forest land- scape may reduce occupancy of historical nest sites. However, their current research on the for- aging and ranging habits of goshawks in frag- mented forests suggest that individual pairs are extremely resilient to forest loss and fragmen- tation. Goshawks forage primarily in mature for- ests, but make use of regenerating forests and riparian gaps. They are notably unaffected by habitat loss and fragmentation that occurs while they are occupying an area. The working hy- FRAGMENTATION AND BIRDS IN COASTAL FORESTS--Manuwal and Manuwal 111 TABLE 5. ABUND^NCE OF BROWN-HEADED COWBIRDS IN LOWLAND HABITAT OF WESTERN WASHINGTON FROM BREEDING BJRD SURVEYS (BBS) Population BBS route Name Years Mean/year trend a Sea level 89907 Vashon Island 2 89905 Deception Pass 5 89072 Mukilteo 4 89034 Everett 15 Mean Lowlands, Cascade Foothills 89111 Carnation 9 89066 Bayview 4 89133 Montesano 11 89078 Pe Ell 3 89059 Raymond 2 Mean Cascades-Low Elevation 89904 Verlot 6 89902 Cascade River 9 89043 Packwood 19 Mean 13.0 ? 22.6 - 20.5 0 10.5 16.7 19.3 15.8 - 0.4 0 13.7 0 7.5 ? 11.3 0.8 1.2 3.0 1.7 a ? indicates insufficient data; 0 no trend, decreasing. pothesis that links these apparently contradictory observations is that specific pairs acclimate and adjust to forest fragmentation in and around their breeding territories, but when these accli- mated pairs die, new pairs are less likely to se- lect the formerly occupied habitat for breeding. Lack of continued selection of fragmented hab- itat by goshawks produces the negative corre- lation between occupancy and fragmentation, while acclimation to fragmentation allows cur- rent territory owners to be unaffected by frag- mentation. BROWN-HEADED COWBIRD PARASITISM The Brown-headed Cowbird is a relatively re- cent immigrant to the coastal regions of the Pa- cific States. It became established in portions of this region only since the 1950s (Rothstein 1994, Morrison and Caldwell this volume). In western Washington it may not have become established until a little later since Jewett et al. (1953:592) reported that the cowbird was (referring to the 1940s and 1950s) a "rare migrant and casual winter visitant in western Washington." Since the 1950s, the cowbird has become established as a breeding bird in western Washington but its distribution is clearly restricted to the Puget Trough lowlands. A review of 12 Breeding Bird Survey (BBS) routes in the Puget Sound area indicates that this species is relatively common in the highly fragmented open habitats from sea level up to the foothills of the Cascade Moun- tains (Table 5). Cowbird abundance decreases with elevation, or at least with a landscape in- creasingly dominated by coniferous forests. Point count bird surveys in coniferous forests conducted from 1983 to 1998 in the Cascade Mountains at elevations ranging from 300 to 1500 m show that the Brown-headed Cowbird is virtually absent (7 detections out of a total 56,290 bird detections; Table 6) in this land- scape even though it is fragmented (Figs. 1 and 2). The cowbirds we detected were in recent clearcuts adjacent to Douglas-fir forests. Factors preventing cowbird colonization of the frag- mented coniferous forests in the Washington Cascades are unknown, but it is apparent that cowbird parasitism is not currently impacting potential hosts in the fragmented landscape of the Washington Cascades. Cowbirds are very rare there now but they could become a problem in the future. Cowbirds are relatively common in the Puget Sound Lowlands so parasitism is undoubtedly occurring there, but its extent has not been investigated. The proximity of the pres- ently occupied areas to mountain habitat makes it possible that cowbirds may eventually occupy some of the Cascade and Coast Range montane forests. The effects of predation on songbird communities of the Pacific Northwest is poorly known. A current study by R. Sallabanks is ex- ploring this aspect in managed forests of the Washington Cascades. CONCLUSIONS Fragmentation in the mountains of the Pacific Northwest consists of open areas created by clearcut or seed-tree logging in a matrix of for- 112 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 6. NUMBERS OF BROWN-HEADED COWBIRDS DETECTED IN CONIFEROUS FORESTS OF THE CASCADE MOUN- TAINS OF WASHINGTON AND OREGON Data source a N Years Cowbirds detected Total bird detections OGWHP 46 2 0 21,962 TFW-RMZ 18 3 0 6,032 TFW-Landscape 24 3 7 20,373 USFS-DEMO-WA 24 2 0 4,446 USFS-DEMO-OR 24 2 0 3,477 Total 7 56,290 aData from point counts within 50 m of points except TFW-RMZ (within 15 m of points). Abbreviations: OGWHP (Manuwal 1991): 12 points, 6 visits, 8 min count duration; 1984, 1985. TFW-RMZ (S. Pearson and D.A. Manuwal, unpubl. data): 10 points, 6 visits, 6 min count duration; 1993, 1995, 1996. TFW-Landscape (Aubry et al. 1997): 12 Points, 6 visits, 8 min count duration, 1993, 1994, 1995. USFS-DEMO-WA (D.A. Manuwal unpubh data): 4 stations, 6 visits, 8 min count duration; 1995, 1996. USFS-DEMO-OR (D.A. Manuwal unpubl. data): 4 stations, 6 visits, 8 min count duration; 1995, 1996. ests of various ages. This pattern differs from many areas of eastern North America where for- ests are located near or adjacent to agricultural lands or human settlements. In the Pacific North- west, fragmentation appears to be most exten- sive on private commercial timberlands com- pared with national forests. The Puget Sound Lowlands have some areas of agriculture, mixed with patches of forests, but this region has not been adequately studied. The effects of forest fragmentation are not well documented in the Pacific Northwest com- pared with the many studies in eastern North America [e.g., those cited in Hagan and John- ston (1992) and Martin and Finch (1995)]. Nev- ertheless, some patterns seem to be emerging from recent studies. Species richness seems to increase in highly fragmented landscapes, chief- ly because of the colonization of edge species, which often nest or forage in open, shrubby hab- itats. However, interior forest birds may be de- clining under these conditions. The identification of specific landscape variables responsible for this has been difficult to determine, perhaps be- cause birds such as the Winter Wren and Hermit Warbler, which have small territories, respond to stand-level factors rather than large scale ones. There are no long term studies in the Pacific Northwest so we have no information on how fragmentation affects bird abundance. Short- term investigations indicate that some species increase while others decrease with fragmenta- tion, a pattern also observed in the eastern Unit- ed States. Brood parasitism and predation have been shown to be a major concern in the fragmented environments of eastern North America (e.g., Robinson et al. 1995b), but there is no evidence that parasitism is an important factor in the coastal mountains of the Pacific Northwest. However, this could become a problem as more forested land is cleared and converted to more open habitat. Coniferous forests in the Pacific Northwest are naturally heterogeneous because of the ef- fects of fire, wind-throw, floods, and volcanic eruptions. Compared with habitat fragmentation in much of eastern North America, fragmenta- tion in the mountains of the Pacific Northwest is fundamentally different in that forest patches are not surrounded by agricultural land or areas dominated by human development. Instead, for- est patches are surrounded by other forest patch- es of different ages. Late successional forest patches remaining after timber harvesting have become smaller in recent decades and are less suitable for area-sensitive bird species than larg- er patches. Cowbird brood parasitism is not common in the mountains but does occur in low- land habitats. It is clear that much more research is needed in the Pacific Northwest to determine relationships between birds and forest fragmen- tation. ACKNOWLEDGMENTS We thank S. Garman, T Spies, and J. Franklin for sharing their information on Pacific Northwest vege- tation. C. Grue and K. Dvornich, Washington Coop- erative Fish and Wildlife Research Unit, Washington Gap Analysis, provided us with digital maps. S Reu- tebush provided the aerial photograph of Willamette National Forest. We are grateful to the Washington De- partment of Natural Resources (Timber, Fish and Wild- life Agreement) for funding the riparian management zone and landscape studies in Washington, and the U.S. Forest Service, Pacific Northwest Forest Experi- ment Station, Portland, OR, for funding the DEMO project. The efforts of many field ornithologists asso- ciated with these projects are gratefully acknowledged. Studies in Avian Biology No. 25:113-129, 2002. BIRDS AND CHANGING LANDSCAPE PATTERNS IN CONIFER FORESTS OF THE NORTH-CENTRAL ROCKY MOUNTAINS SALLIE J. HEJL, DIANE EVANS MACK, JOCK S. YOUNG, JAMES C. BEDNARZ, AND RICHARD L. HUTTO Abstract. We describe historical and current landscape patterns for the north-central Rocky Moun- tains, speculate on the expected consequences of human-induced changes in coniferous forest patterns for birds, and examine the evidence related to the expected consequences. The Rocky Mountain region has one of the most heterogeneous landscapes in North America, combining high complexity in abiotic gradients with fire as a major disturbance factor. In recent decades fire suppression has limited this disturbance, resulting in altered stand structures and relatively homogeneous expanses of mid-succes- sional forest where there were once mosaics of different-aged post-fire stands. Elsewhere, historically homogeneous landscapes that rarely burned have become more heterogeneous due to logging. Many torest types are less common than they were historically due to current management. Land conversion to agriculture and development has primarily occurred in low elevations. We speculate that the con- sequences of these changes include: (1) bird species adapted to historically homogeneous forest land- scapes would be negatively affected by landscape heterogeneity created by timber harvest openings; (2) bird species specialized for forest types that were once prevalent but are now uncommon may be negatively affected by decreasing patch size and increasing isolation; and (3) birds that breed in close proximity to human-added landscape features may be negatively affected by brood parasites or nest predators. Brown Creeper (Certhia americana) and Golden-crowned Kinglet (Regulus satrapa) had the strongest trends of species sensitive to fragmentation indices. Pine Siskin (Carduelis pinus), Chip- ping Sparrow (Spizella passerina) and Dark-eyed Junco (Junco hyemalis) were positively associated with fragmentation across most studies. Nesting success varied among landscape configurations, and some trends paralleled abundance patterns. Brown-headed Cowbird (Molothrus ater) parasitism rates were extremely low (0-3%) where nest success has been studied in coniferous forests of the north- central Rockies. Across extensive and intensive studies, distance to agricultural lands was the strongest predictor of cowbird presence. Therefore, we found evidence for the ideas that birds adapted to homogeneous forest landscapes have been negatively affected by heterogeneity caused by timber harvesting, that patch size is important for some birds in one vanishing habitat (old-growth ponderosa pine, Pinus ponderosa), and that cowbirds are more abundant in conifer forests near human-added landscape features. The effects of changes in landscape patterns on birds in the north-central Rockies seem to be less dramatic than in eastern and midwestern North America, and different landscape measures are more relevant to western conifer forests. We need additional research on most aspects of breeding, nonbreeding, and dispersal ecology in relation to landscape patterns and within-stand changes. We offer our proposed consequences as hypotheses upon which to base future tests. Key Words: birds; fire; fire regimes; fire suppression; forest fragmentation; north-central Rockies; landscape; landscape patterns; wildfire. Forest fragmentation has clearly afl,ected birds in some landscape configurations in the East and Midwest (Porneluzi et al. 1993, Donovan et al. 1995a, Robinson et al. 1995a). In landscapes where forests are fragmented by agriculture and urbanization, resulting in discrete measurable patches, species richness has been shown to in- crease with patch area and decrease as patches become more isolated (Whitcomb et al. 1981, Ambuel and Temple 1983, Freemark and Mer- riam 1986, Blake and Karr 1987). The presence or absence of a species across patches of differ- ent sizes suggested minimum area requirements (Temple 1986, Askins et al. 1987, Robbins et al. 1989a). Nesting success declined (Villard et al. 1993, Donovan et al. 1995b), and edge effects (as indicated by nest predation and parasitism) were particularly strong where the landscape matrix had been highly modified (Robinson 1992). These studies identified long-distance mi- grants as particularly sensitive to area efl,ects. The effects of landscape changes on bird pop- ulations in conifer forests in the West seem to be less dramatic (Rosenberg and Raphael 1986, McGarigal and McComb 1995). Historical and current landscape patterns are quite difl,erent in the West than in the East and the Midwest, es- pecially in the mountainous and sparsely popu- lated north-central Rocky Mountains. Conifer forests dominate the mountain slopes of this re- gion, and conversion of lands to agriculture and urban development generally has been restricted to valley bottoms. While the natural heteroge- neity of these conifer forests was variable, fire suppression and timber harvest have created landscape patterns with different kinds and lev- els of heterogeneity. Nonetheless, they remain forested ecosystems that may not present barri- 113 114 STUDIES IN AVIAN BIOLOGY NO. 25 ers to many native species (Mcintyre and Barrett 1992). The response of avian species to this dy- namic mosaic may be species-specific and pro- cess-specific (Haila 1999). Edge effects may also be substantially different in forest-dominat- ed landscapes than in agricultural ones (Hanski et al. 1996, Bayne and Hobson 1997). Different measures of landscape patterns are more relevant to landscapes in western conifer forests than those used in the East and Midwest. For example, size and isolation of an individual forest patch is almost impossible to measure in conifer forests of the north-central Rockies be- cause the forest is the matrix rather than the patch, with most stands connected in some way to other conifer forests that may or may not be similar in age, species composition, and struc- ture. The exceptions include rarer forest types, such as old-growth ponderosa pine (Pinus pon- derosa) or patches of recent fire disturbance. Measures of fragmentation in western conifer forests are thus better achieved by characterizing patterns within a defined landscape, based on relative amounts of forest and amounts and types of edges. More complex variables may be necessary, such as measures of connectivity (Taylor et al. 1993). When patch size is used, patch boundaries often are created somewhat ar- tificially when a user-defined landscape outline is imposed onto the forest matrix for analysis. Because of these constraints, studies in western coniferous forests usually describe the structure of the landscape mosaic in which the forest is embedded (see Wiens 1989) and then relate that structure to avian populations (Rosenberg and Raphael 1986, van Dorp and Opdam 1987, McGarigal and McComb 1995, Schieck et al. 1995). We investigated whether bird populations are related to landscape changes in north-central Rocky Mountain conifer forests and whether these relationships are similar to what has been reported for other regions. We define the north- central Rockies as that area from eastern Oregon and Washington east through Idaho and western Montana to Wyoming (Fig. 1). We include aspen (Populus spp.) in our discussion of conifer for- ests because it is an integral part of many conifer landscapes. To look at the relationships between birds and landscape patterns, we (1) describe historical landscape patterns and the processes responsible for them; (2) describe current land- scape patterns and their causes; (3) discuss im- plications and potential consequences of human- induced changes between historical and current patterns for coniferous forest birds; (4) examine the current evidence surrounding the expected consequences; and (5) compare our findings for the north-central Rockies to other regions. / FIGURE 1. The north-central Rocky Mountain geo- graphic area. Rocky Mountain forest type boundaries from Bailey's (1995) ecoregions of the United States, including portions of northern, middle, and southern Rocky Mountain steppe provinces. HISTORICAL LANDSCAPE PATTERNS Natural landscape heterogeneity results from the superposition of a disturbance regime onto vegetation patterns created by abiotic gradients (Turner and Romme 1994). Historically, the north-central Rocky Mountain region had one of the most heterogeneous landscapes of any area in North America due to a dry climate and fre- quent lightning-caused fires, and this disturbance regime was superimposed on complex vegeta- tion patterns resulting from moisture gradients and finely dissected topography. Characterizing natural or presettlement land- scapes can be a very difficult task (Noss 1985, Sprugel 1991). The evidence is scattered and subject to many potential biases (Noss 1985). In the recent bioregional assessment of the interior Columbia River Basin, Hann et al. (1997) used scattered evidence, expert opinion, and simula- tion models to estimate broad-scale landscape patterns across the region for the 1850-1900 time period. The mid-scale assessment associ- ated with that project (Hessburg et al. 1999) used historical aerial photographs to characterize landscape conditions in sampled watersheds, but historical photos could be found for only the "recent historical" period of the 1930s to 1960s. Even if accurate historical data could be re- covered for one point in time, the dynamic na- ture of the disturbance regimes diminishes the usefulness of that information. Fire size and se- verity depend on previous disturbance history (e.g., fuel buildup) as well as cyclic weather pat- terns (Bessie and Johnson 1995). There is grow- ing evidence that fire disturbance was extremely variable historically and probably not in equilib- rium across the landscape (Sprugel 1991, Turner and Romme 1994, Brown et al. 1999). In addi- FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTS--Heft et al. 115 tion, native Americans altered fire regimes for hundreds of years before Euro-American settle- ment (Barrett and Arno 1982). Therefore, any characterizations of historical landscape patterns must be considered generalizations and take into account the highly variable nature of the land- scape. ABIOTIC FACTORS The north-central Rockies are composed of many mountain ranges of varying raggedness and orientation. Moisture varies with elevation and topography, and there is also a regional gra- dient in rainfall due to continental climate pat- terns (Habeck and Mutch 1973, Peet 1988). Finely dissected topography interweaves land units of very different slopes, soils, moisture re- tention properties, and exposures, and these pat- terns occur at several spatial scales. Local land- scape vegetation patterns are strongly influenced by these abiotic gradients. Higher elevations have lower temperatures and receive more precipitation. Annual precipi- tation in the north-central Rockies ranges from less than 380 mm in intermontane valleys to more than 1500 mm at higher elevations (Ha- beck and Mutch 1973). These local temperature and moisture patterns create zones of forest hab- itat types based on the physiological require- ments and competitive abilities of the various tree species (Daubenmire 1956). For example, in much of the north-central Rockies, the driest and lowest-elevation forests historically were domi- nated by ponderosa pine, which remains an im- portant early-seral species up into the mid-ele- vation zone, where Douglas-fir (Pseudotsuga menziesii) was typically the major tree species in climax vegetation. The less drought-resistant Engelmann spruce (Picea engelmanni) and sub- alpine fir (Abies lasiocarpa) compete for climax status only in the more moist, upper-elevation zones. Each of these zones had different fire re- gimes (Arno 1980). Fire in many of these re- gimes maintained large areas dominated by shade-intolerant (early-seral) tree species, in- cluding ponderosa pine, lodgepole pine (Pinus contorta), western larch (Larix occidentalis), sometimes grand fir (Abies grandis) and Doug- las-fir, and, historically, western white pine (Pi- nus rnonticola). Local topography and soils can drastically al- ter available nutrients, solar radiation, tempera- ture, and water retention (Peet 1988, Swanson et al. 1988). South-facing slopes and ridge tops are much warmer and drier and may support vegetation typical of lower elevations, if soils allow. Sheltered valley bottoms have lower solar radiation and may collect water and cold-air pockets that support vegetation more character- istic of that nearly 500 m higher on open slopes (Peet 1988). Naturally treeless areas occur wher- ever slopes are too steep or rocky, or where there is prolonged summer soil drought (Dau- benmire 1968). Areas on the east side of the Continental Divide especially have widespread occurrence of forest-grassland-sagebrush mosa- ics, probably regulated by the availability of moisture (Patten 1963) and the frequency of fire (Arno and Gruell 1983). In contrast, moist Pacific air reaches a limited area in southeastern British Columbia, north- eastern Washington, northern Idaho, and north- western Montana. The resulting luxuriant forests in this region appear similar to forests in the Cascade Mountains (Peet 1988), with tree spe- cies including western hemlock (Tsuga hetero- phylla), western redcedar (Thuja plicata), and grand fir (Habeck 1987). The combination of greater precipitation and gentler topography re- sults in relatively continuous forests in this re- gion, including the valley bottoms where there is often no well-defined lower timberline. DISTURBANCE Disturbance imposes further heterogeneity on the landscape, at several spatial scales, by pro- ducing a mosaic of age classes and successional communities. Fire was historically the most prevalent natural disturbance in the northern Rocky Mountains (Gruell 1983). The extent and severity of fires in the north- central Rockies depended on the moisture gra- dient, which varied temporally as well as spa- tially (Arno 1980). Forests in more mesic areas burned less often (every 50-300 years; Table 1), so they were more likely to reach later succes- sional stages and to accumulate larger amounts of woody fuels, not burning until sufficient fuels and weather conditions produced a stand-replac- ing crown fire. Forests in drier areas would burn more often (every 5-50 years; Table 1), before sufficient fuels could accumulate to result in a crown fire. These frequent underburns destroyed seedlings of shade-tolerant tree species while causing minimal harm to fire-resistant early-ser- al trees, thus maintaining non-climax stands of old-growth ponderosa pine and western larch (Arno et al. 1997). Historically, old-growth ponderosa pine and western latch dominated millions of acres on drier valley bottoms and south facing slopes throughout much of the north-central Rockies (Arno et al. 1997). Although these "fire-depen- dent" (Habeck 1988) forests could be extensive, complex topography and moisture gradients usu- ally made these forests less homogeneous than in the Southwest (Arno 2000). Heterogeneity could occur at several scales, with grassland-for- 116 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 1 . REPORTED MEAN AND RANGE OF HISTORICAL FIRE INTERVALS IN GENERAL CONIFEROUS FOREST CLASSES Mean fire General habitat intervals b Fire interval Predominant class a Description (yrs) range b (yrs) fire regime c Limber pine Mostly small stands mixed with grass and 74 variable Nonlethal? shrubs on dry or rocky sites Warm, dry ponderosa Open stands, with grass understory main- 5-30 2-55 Nonlethal pine or Douglas-fir tained by frequent fire Warm, moist ponderosa Typically ponderosa pine dominant with 10 49 3-97 Nonlethal pine an understory of Douglas-fir in the ab- sence of fire Cool, dry Douglas-fir Generally open stands of Douglas-fir with 35-40 variable? Nonlethal sparse understory Moist Douglas-fir Douglas-fir often dominates; closed-cano- 25-30 8-66 Mixed py ponderosa pine, larch, and lodgepole pine common in seral stages Grand tir/mixed conifer Diverse closed-canopy forest; often devel- 13 120 5 150 ops into mixed species stand Cool lodgepole pine Pure stands of lodgepole pine or mixed 24-50 1-88 with grand fir and whitebark pine Subalpine fir and codomi- Spruce and other firs common in seral 57-153 50-300 nant species stages; stand-replacement fires common Moist redcedar and west- Closed-canopy stands of redcedar and 70-120 25-200 ern hemlock western hemlock Mixed-Lethal Lethal-Mixed Lethal Lethal a General classes of forest habitat types employed by U.S. Forest Service (Steele et al. 1981), arranged approximately on a dry to moist gradient. b Fire-interval estimates from Arno 1980, Arno and Gruell 1983; Arno et aL 1995, 1997: Crane and Fischer 1986, Gruell et aL 1982, Gruell 1983. c Historical fire regime thought to occur over most acreage; all habitat types could have all fire types. est mosaics at the drier extremes and with denser forests created by stand-replacing fires at the wetter extremes. East of the Continental Divide, where it is too dry for larch and too cold for ponderosa pine, Douglas-fir forests often had similar fire regimes (Arno and Gruell 1983). In very dry years, stand-replacement fires may have occurred in any of these areas (Bessie and Johnson 1995, Brown et al. 1999). In the more roesic areas of the north-central Rockies (maritime-influenced forests, north-fac- ing slopes, and mid- to high-elevation forest types), the predominant fire regime was one of infrequent, stand-replacement fires (Arno and Davis 1980, Romme 1982, Fischer and Bradley 1987, Barrett et al. 1991). In fact, the origin of most Rocky Mountain forest stands can be traced to stand-replacement fires (Arno 1980). Historically, most individual fires were small (<1 ha; Strauss et al. 1989), because fuels were too moist or sparse to spread the fire. However, most of the area burned by stand-replacement fires was due to a few large fires in dry years (Strauss et al. 1989, Bessie and Johnson 1995), so it was the large fires that created the vegeta- tion mosaic that dominated the landscape until the next extensive fire (Turner and Romme 1994). Large crown fires rarely consumed an en- tire forest because of local variations in wind, topography, vegetation type, natural fire breaks, and fuel loads (Turner et al. 1994). These factors produced a heterogeneous pattern of burn sever- ities, as well as islands of unburned vegetation (Eberhart and Woodard 1987, DeLong and Tan- ner 1996). The degree of patchiness depended on the dryness of fuels in the year of the fire (Turner et al. 1994, Turner and Romme 1994). Data on natural fire intervals in different for- est cover types suggest that fire severity and fre- quency were highly variable prior to current fire suppression activities (Table 1). Frequent non- lethal fires and infrequent stand-replacement fires could occur in the same region depending on weather and fuel accumulations, or individual fires may have been of "mixed severity," with many trees dying and many surviving (Brown 1995, Arno 2000). Mixed-severity fire regimes occurred especially in mid-elevation, mixed-co- nifer forests, where moisture regimes and topog- raphy were variable, and fire-resistant tree spe- cies (especially larch and ponderosa pine) oc- curred. Mixed-severity fires produced heteroge- neity at several scales, killing variable amounts of trees within a forest stand and affecting var- iable numbers of stands within a landscape. The moisture regime influenced this variability in size, with drier areas tending to have smaller patches of lethal burns because fires burned of- ten enough to prevent sufficient fuel accumula- tion for extensive crown fires (Barrett et al. 1991). This typically left a patchy, erratic pattern on the landscape that fostered development of highly diverse communities (Barrett et al. 1991, Arno 2000, Lyon et al. 2000). FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTS--Heft et al. 117 CURRENT TRENDS In the north-central Rockies, most changes in landscape patterns from historical to current are the result of changes in the disturbance regimes due to fire suppression and timber harvesting. The resulting forests may differ in age, structure, species composition, or landscape pattern, but they remain conifer forests. The little land con- version that has occurred is focused within the lower elevations where forest or grassland has been converted to agricultural land, rural resi- dences, or urban areas. FIRE SUPPRESSION Fire suppression has become increasingly ef- fective since the 1930s (Arno 1980, Barrett et al. 1991). Through much of the low and mid- elevation landscapes, fire suppression has altered stand structures and landscape patterns through- out the north-central Rockies (Tande 1979, Arno 1980, Barrett et al. 1991). Because dry, lowland areas had fire-return intervals of 5-50 years, the suppression of low-intensity fires for up to 70 yrs has resulted in abnormal fuel accumulations that make the historically resistant old-growth pine and latch more susceptible to stand-replace- ment fire. Harm et al. (1997) estimated that 19% of the interior Columbia River Basin has changed to a lethal fire regime from mixed or non-lethal over the last century. Complex, uneven-aged stands containing fire-resistant trees are being replaced by even-aged post-fire stands that cover large areas of the landscape (Hann et al. 1997). Future fires that burn in this simplified landscape may be larger and more homogeneous, so the ho- mogeneity may be self-perpetuating (Arno 1980, 2000; Barrett et al. 1991). Many areas naturally had heterogeneous land- scapes due to a mosaic of successional stages following stand-replacement fires. Here, fire suppression is converting this mosaic of forest stands from a variety of age classes into a more homogeneous expanse of mid-successional ma- ture forest (Hann et al. 1997). Because succes- sion changes forest structure most rapidly in the earliest age classes, it has taken only a few de- cades for fire suppression to allow large expans- es of closed-canopy, continuous forest to form on the landscape (Tande 1979). However, in ar- eas with stand-replacement fire regimes, the time period of successful fire suppression may not yet be long enough to greatly affect the historical fire-return intervals of 140 to 400 years (Romme 1982, Barrett et al. 1991). Fire suppression also reduces many unique post-fire habitats on the landscape. Early post- fire patches of standing dead trees are much re- duced throughout the region. There also has been a loss of shade-intolerant tree species, such as ponderosa pine, larch, and aspen, as succes- sion advances in the absence of fires (Hann et al. 1997). Fire-maintained old-growth ponderosa pine stands are an obvious example, but western larch also formed large, open stands of fire- maintained old growth (Arno et al. 1997). Larch is restricted to relatively more mesic areas than ponderosa pine, but it is the most shade-intol- erant and fire-resistant conifer species in the north-central Rockies (Arno and Fischer 1995), so it is an important early-seral species as well as being an important older-aged component of forests in mixed-severity fire regimes. Aspen is another early-seral tree species that regenerates following fire. In the Centennial Mountains of Idaho, aspen cover has been reduced 80% since 1850, while mature conifer forest increased in area, patch size, and connectivity (Hansen and Rotella 2000). Increasing isolation may be an- other landscape factor affecting stands of these tree species. TIMBER HARVEST With the suppression of fires, timber harvest- ing is now the most important disturbance re- turning conifer forests to early successional stag- es. It is unclear whether the total area involved is similar, however. Hann et al. (1997) estimated that the current areal extent of early-successional stands in moist forests (20%) is at the low end of the historical (pre-1900) range (19-29%), is about the same (18%) as historical in low-ele- vation dry forests (8-20%), and higher (33%) than historical (23-25%) in upper-elevation cold forests. There are great differences, however, in the landscape patterns and stand structures pro- duced by timber harvest compared with fire (Hann et al. 1997). Whether timber harvest in- creases or decreases landscape heterogeneity de- pends on the natural heterogeneity of the area (i.e., fire regime and topography), the harvest methods used, and the spatial scale at which analyses are done. Timber harvest has greatly reduced the acces- sible, low-elevation dry forests that historically had non-lethal fire regimes and were dominated by old-growth ponderosa pine or western larch. Accessible forests were preferentially logged first, with more distant ones harvested as tech- nology improved and road systems were created (Hejl 1994). Few old-growth stands remain. In the national forests of eastern Oregon and Wash- ington, where the original low- and mid-eleva- tion ponderosa pine forests may have been about 90% old growth, nearly three-quarters of this old growth had been logged by 1970 (Henjum et al. 1994). In addition, 82% of the remaining old- 118 STUDIES IN AVIAN BIOLOGY NO. 25 growth patches are smaller than 100 acres, with only 7 patches over 5,000 acres (Henjum et al. 1994). Fire suppression has resulted in further danger to these patches by allowing the buildup of fuels and converting patches to denser forests with more shade-tolerant tree species. Hann et al. (1997) estimated that the ponderosa pine cov- er type decreased by 26% throughout the interior Columbia River Basin since 1900. Open-canopy old growth has diminished even more (Henjum et al. 1994). Timber harvesting in combination with fire suppression has also reduced old- growth larch on the landscape. Hann et al. (1997) estimated that the western larch cover type (all ages) has decreased by 36% throughout the interior Columbia River Basin since 1900. Mid-elevation forest with mixed-severity fire regimes historically had a diversity of stand structures and landscape patterns. Timber har- vesting returns some patch heterogeneity to these forests, but generally on a coarser-grained scale than produced by natural fires, with a more regular pattern (Reed et al. 1996a). Clearcuts do not retain the remnant trees or snag structure typical of post-fire forests, nor do they create an environment that could maintain the historical complexity of community composition and structure. Consequently, most of the early-seral forest stands within this type are very different in composition and structure relative to the na- tive conditions (Hann et al. 1997). Harvest methods that retain green trees (e.g., Lehmkuhl et al. 1999) may better mimic some mixed-se- verity fires, but still lack the snag structure or large, downed woody debris. If the same pre- scription is always used for this type of cutting, it will produce a relatively simplified and ho- mogeneous landscape. The most productive forests in this region were the "Cascadian" forests around northern Idaho, where fires were rare and, therefore, large blocks of old growth likely developed. Once fairly homogeneous landscapes have been rid- dled with clearcuts and other logged conditions. In these and other forests with stand-replace- ment fire regimes, (e.g., high-elevation lodge- pole pine), the creation of many small clearcuts is a departure from the pattern of disturbance created by the natural fire regime (Brown 1995). Similarly, in boreal forests in Canada, DeLong and Tanner (1996) found that wildfires created a more complex landscape pattern than clearcut harvesting practices do, with a greater diversity of patch sizes, more irregular shapes and bound- aries, and more patches of mature forest inter- mixed. These patches may be critical for bird species that require heterogeneity in patch struc- ture. They also provide sources of large trees and snags (legacies) within the young post-dis- turbance forest (DeLong and Tanner 1996). No cutting method can create the dense snag struc- ture that is produced by a stand-replacement fire. It is unclear if timber harvest has created more fragmentation than natural disturbance re- gimes. Reed et al. (1996a) found a substantial increase in patchiness created by clearcutting and roads from 1950 to 1993 in high elevation forests in the Medicine Bow Mountains of southern Wyoming. Quantitative landscape in- dices suggested a level of fragmentation greater than that found in the Oregon Cascades. How- ever, the disturbance patterns in Wyoming were superimposed on a landscape of natural hetero- geneity, and it is unknown what the landscape in either 1950 or 1993 would have been like under a natural fire regime. Tinker et al. (1998) found similar results in the Bighorn Mountains of north-central Wyoming. Old-growth forest patches produced by natural disturbances in western coniferous forests were typically much larger and more continuous than are the remnant patches created by timber harvesting and road building (Tinker et al. 1998). However, they found that roads contributed more to this change in landscape indices than did clearcuts. It is not known if roads are wide enough to cause harm- ful fragmentation effects for most Rocky Moun- tain bird species, especially in open forests, but roads are certainly a more permanent distur- bance than clearcuts (Reed et al. 1996b). However atypical the landscape pattern pro- duced by timber harvesting may be, it still leads to forest succession and the retention of natural vegetation. A potentially more serious impact on the forested landscape is the permanent conver- sion of native habitat to agriculture or residential and urban development (including roads). In the north-central Rockies, this conversion has been concentrated in the valley bottoms. While this limits the amount of fragmentation in the overall landscape, these low elevation areas are also the most productive ecosystems for birds (Hansen and Rotella 1999). As rural development accel- erates in the inland west (Knight 1997), we may see much more serious fragmentation and edge effects on birds due to added human features on the landscape (e.g., Friesen et al. 1995). PROPOSED CONSEQUENCES OF LANDSCAPE CHANGES ON BIRDS Based on our knowledge of the historical landscape patterns of the region and the changes that have occurred, we speculate about which birds we would expect to be most affected by landscape changes in the past 100 years in north- central Rockies conifer forests. We offer these speculations as a framework from which to ex- amine the data that exist on bird trends and bird FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTS--Heft et al. 119 relationships with landscape patterns. The pro- posed consequences of landscape changes on birds are: (1) species that are adapted to moist forest types that historically formed the most ho- mogeneous landscapes (e.g., old-growth cedar/ hemlock) would be negatively affected by in- creased landscape heterogeneity created by tim- ber harvest openings; (2) species specialized for forest types that were once prevalent but are now uncommon or rare (i.e., vanishing habitats: aspen, early post-fire forests, old-growth pon- derosa pine, and old-growth larch) may be neg- atively affected by decreasing patch size and in- creasing isolation over and above the general loss of habitat; and (3) birds that breed in close proximity to human-added landscape features (such as cows, horses, bird feeders, agricultural land, or residential development) may be nega- tively affected by brood parasites or nest pred- ators that are attracted to these features. More than one of these consequences could be occur- ring in any one particular landscape. HAS FOREST FRAGMENTATION AFFECTED BIRDS OF THE NORTH-CENTRAL ROCKIES? To evaluate whether and how coniferous for- est birds are affected by changes in landscape patterns, we looked for evidence from each of three sources: (1) regional population trends based on the North American Breeding Bird Survey; (2) studies concerning relationships be- tween bird abundance and specific landscape characteristics, including the effects of logging; and (3) studies concerning relationships between nesting success and human-caused landscape modification. BBS TRENDS We assumed that if populations of some bird species are declining as the result of changing conditions brought on by fire suppression and intense timber harvesting activities, then the re- cent 33 years of Breeding Bird Survey (BBS) data (1966-1998) collected from within the re- gion should reflect that fact, although there may be other reasons for any observed declines. Thus, we determined how many conifer forest bird species breed in the north-central Rocky Mountains, which ones are adequately covered by the BBS, and what the BBS data indicated about their recent population trends. We focused our analysis on the Central Rockies region, as defined by Robbins et al. (1986), and the conifer forest habitats within that region. By our own estimate, there are 87 bird species that breed in conifer forest habitats within the region (Table 2), and 39 of those (45%) were abundant enough (>1.0 bird per route) and detected frequently enough (on more than 14 routes within the re- gion) to obtain reasonably reliable models of their population trends (Sauer et al. 1999). The bird species for which data are too few, and for which we cannot expect the BBS to provide meaningful results in the future, include those that are rarely detected (e.g., diurnal raptors, grouse), those that occur in habitats that are un- common and poorly sampled by the BBS (e.g., burned forests), and those that are primarily noc- turnal (owls). Only one of the 39 forest bird species for which the BBS provides adequate coverage ap- pears to be declining significantly in the Central Rockies Region--the Olive-sided Flycatcher (Table 2; see table for scientific names of bird species mentioned throughout text). This species is associated with forest openings (natural and human-created) and edges (Altman and Salla- banks 2000), and was most common in harvest- ed and recently burned conifer forest at sites across northern Idaho and western Montana (Hutto and Young 1999). Of these forest types, burned forests have become rarer within the past century. Because several of the species that were not covered well by the BBS are also relatively common in burned forests (woodpeckers), there is even more reason to focus management atten- tion toward the effects of fire suppression and post-fire salvage logging, both of which have undoubtedly affected the more fire-dependent species negatively (Hutto 1995, Kotliar et al. this volume). BIRD ABUNDANCE AND LANDSCAPE FEATURES Very few studies have been conducted that look specifically at the relationships between changing landscape patterns and birds in forests of the north-central Rockies. We identified five data sets that addressed the relationships be- tween the abundances of bird species and some aspect of landscape configuration. These studies were conducted in different forest types, eleva- tion, and climatic regimes as follows: (1) a re- gion-wide correlational analysis based on 312- ha landscapes centered on bird count points across western Montana and northern Idaho, where conifer forest was defined as one category that included all major conifer types and a wide range of canopy closures within those types (i.e., closed canopy, seed tree, shelterwood, and group selection harvested sites; R. Hutto and J. Young, unpubl. report); (2) a correlational anal- ysis of spatial patterns within 300-ha landscapes in mid-elevation closed-canopy mixed-conifer forest, dominated by grand fir/Douglas-fir/pon- derosa pine in west-central Idaho (Evans 1995); (3) a comparison of a continuous 240-ha old- growth landscape with two similar-sized old- 120 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 2. RECENT POPULATION TRENDS OF CONIFER FOREST BIRD SPECIES IN THE CENTRAL ROCKIES REGION AS DETERMINED FROM BREEDING BIRD SURVEY DATA, 1966--1998 Species No. routes BBS trend Turkey Vulture, Cathartes aura Sharp-shinned Hawk, Accipiter striatus Cooper's Hawk, Accipiter cooperii Northern Goshawk, Accipiter gentilis Swainson's Hawk, Buteo swainsoni Red-tailed Hawk, Buteo jamaicensis American Kestrel, Falco sparverius Ruffed Grouse, Bonasa umbellus Spruce Grouse, Falcipennis canadensis Blue Grouse, Dendragapus obscurus Wild Turkey, Meleagris gallopavo Flammulated Owl, Otus fiammeolus Great Horned Owl, Bubo virginianus Northern Pygmy-Owl, Glaucidium gnoma Barred Owl, Strix varia Great Gray Owl, Strix nebulosa Boreal Owl, Aegolius funereus Northern Saw-whet Owl, Aegolius acadicus Vaux's Swift, Chaetura vauxi White-throated Swift, Aeronautes saxatalis Black-chinned Hummingbird, Archilochus alexandri Calliope Hummingbird, Stellula calliope Broad-tailed Hummingbird, Selasphorus platycercus Rufous Hummingbird, Selasphorus rufus Lewis' Woodpecker, Melanerpes lewis Williamson's Sapsucker, Sphyrapicus thyroideus Red-naped Sapsucker, Sphyrapicus nuchalis Hairy Woodpecker, Picoides villosus White-headed Woodpecker, Picoides albolarvatus Three-Toed Woodpecker, Picoides tridactylus Black-backed Woodpecker, Picoides arcticus Northern (Red-shafted) Flicker, Colaptes aurams Pileated Woodpecker, Dryocopus pileatus Olive-sided Flycatcher, Contopus cooperi Western Wood-Pewee, Contopus sordidulus Hammond's Flycatcher, Empidonax hammondii Dusky Flycatcher, Empidonax oberholseri Cordilleran Flycatcher, Empidonax occidentalis Plumbeous Vireo, Vireo plumbeus Cassin's Vireo, Vireo cassinii Warbling Vireo, Vireo gilvus Gray Jay, Perisoreus canadensis Steller's Jay, Cyanocitta stelleri Clark's Nutcracker, Nucifraga columbiana Common Raven, Corvus corax Tree Swallow, Tachycineta bicolor Violet-green Swallow, Tachycineta thalassina Northern Rough-winged Swallow, Stelgidopte¸,x serripennis Black-capped Chickadee, Poecile atricapillus Mountain Chickadee, Poecile gambeli Chestnut-backed Chickadee, Poecile rufescens Red-breasted Nuthatch, Sitta canadensis White-breasted Nuthatch, Sitta carolinensis Pygmy Nuthatch, Sitta pygmaea Brown Creeper, Certhia americana Rock Wren, Salpinctes obsoletus House Wren, Troglodytes aedon Winter Wren, Troglodytes troglodytes Golden-crowned Kinglet, Regulus satrapa Ruby-crowned Kinglet, Regulus calendula Mountain Bluebird, Sialia currucoides 81 3.4* 58 -0.7 48 -5.4* 13 -10.2' 39 0.2 57 2.0 11 0.8 84 1.0 74 2.3 106 0.0 57 5.4* 81 -4.0* 90 -0.3 81 1.7 91 -2.0 57 2.1 9 -9.9* 72 1.5' 103 2.2* 67 -0.3 59 5.4* 63 4.6* 105 2.0 79 1.7 84 4.0 64 1.3 94 0.7 91 0.1 25 2.4 104 3.1' 32 1.1 15 1.0 62 3.7 63 3.0* 87 0.8 92 -1.2 64 1.6 FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTS--Heft et al. TABLE 2. CONTINUED. 121 Species No. routes BBS trend Townsend's Solitaire, Myadestes townsendi Swainson's Thrush, Catharus ustulatus Hermit Thrush, Catharus guttams American Robin, Turdus migratorius Varied Thrush, Ixoreus naevius Orange-crowned Warbler, Vermivora celata Nashville Warbler, Vermivora ruficapilla Yellow-rumped (Audubon's) Warbler, Dendroica coronata Townsend's Warbler, Dendroica townsendi MacGillivray's Warbler, Oporornis tolmiei Wilson's Warbler, Wilsonia pusilla Western Tanager, Piranga ludoviciana Green-tailed Towhee, Pipilo chlorurus Spotted Towhee, Pipilo maculatus Chipping Sparrow, Spizella passerina Fox Sparrow, Passerella iliaca Lincoln's Sparrow, Melospiza lincolnii Dark-eyed (Oregon) Junco, Junco hyemalis Black-headed Grosbeak, Pheucticus melanocephalus Lazuli Bunting, Passerina amoena Brown-headed Cowbird, Molothrus ater Pine Grosbeak, Pinicola enucleator Cassin's Finch, Carpodacus cassinii Red Crossbill, Loxia curvirostra Pine Siskin, Carduelis pinus Evening Grosbeak, Coccothraustes vespertinus 80 - 0.5 103 0.8 72 1.2 111 0.5 68 1.4 83 1.0 109 -0.5 70 1.2 96 0.9 77 - 1.0 97 0.8 13 -2.9 55 4.5* 111 0.1 65 6.8* 111 -0.4 59 8.9* 61 3.4* 86 - 1.1 72 - 0.2 79 0.7 109 0.3 62 2.2* Note: Species without trend information were either too rare (<0.1 bird per route) or detected too infi'equently (on fewer than 5 routes) to provide estimates; those without bolded data have either deficient regional abundance (< 1.0 birds per route) or route sample size (fewer than 14 routes). Species showing significant declines or increases (P < 0.05) are noted with an asterisk next to the trend value. growth and selectively-harvested landscapes, each with embedded clearcuts in western red- cedar/western hemlock forests in northern Idaho (Hejl and Paige 1994); (4) a comparison of har- vested and unharvested 20-100 ha stands of spruce/fir in southeastern Wyoming (Keller and Anderson 1992); and (5) a patch-based study of old-growth ponderosa pine/Douglas-fir/western larch in western Montana (Aney 1984). Not all of the landscape metrics were evaluated in all studies, and two studies (Keller and Anderson 1992, Hejl and Paige 1994) focused more on the overall comparison of landscapes modified by timber harvesting to unmodified areas (see Ef- fects of logging patterns). Bird abundances were based on point counts; point locations usually were within conifer forest and encompassed the natural variability in forest cover around points, and analyses generally included only the most common bird species detected. Thus, informa- tion is primarily limited to passerines, because other species are not well-sampled by point counts. Amount of forest The amount of forest covering a landscape is a frequently-reported measure of the degree of fragmentation of that landscape (e.g., Robinson et al. 1995a). It is one metric that can be mea- sured easily in forested landscapes where the forest remains highly interconnected and occurs as the matrix, not as a patch, although it gives no information on the spatial configuration of the remaining habitat. It also is a measure that can be used over large regions when the reso- lution of the map used to measure forest cover is too coarse to adequately capture other spatial parameters such as patch shape and edge. In the three landscape studies we considered, forest cover was measured at similar extents (within 200-312 ha areas) and at similar resolutions (at the scale of an aerial photograph or 30 m X 30 m pixel). The forest cover of interest ranged from 3-100% across all sampled landscapes, al- though these measures are not entirely compa- rable among studies due to different definitions of "forest." A total of 10 species (five residents, three long-distance migrants, and two short-distance migrants) were consistently positively associated with the amount of forest cover in at least one study (Table 3). The probability of occurrence of seven species increased with increasing amounts of conifer forest in the study in which forest was defined most broadly ("all conifer;" R. Hutto and J. Young, unpubl. report). In 122 STUDIES IN AVIAN BIOLOGY NO. 25 TABLE 3. RELATIONSHIPS BETWEEN CONIFEROUS FOREST BIRD SPECIES AND LANDSCAPE METRICS IN THE NORTH- CENTRAL ROCKY MOUNTAINS Proximity Amount of forest Patch size Edge density to edge All Mixed Cedar/ All Mixed Ponderosa All Spruce Mixed conifer a conifer b hemlock c conifer conifer pine d conifer rit e conifer Positively associated with elements of continuous landscapes Vaux's Swift (LDM, CN f) + Gray Jay (R, OCN) (+) Chestnut-backed Chickadee (R, CN) + Red-breasted Nuthatch (R, CN) + + + Brown Creeper (SDM, EN) + + + Winter Wren (R, EN) + + + Golden-crowned Kinglet (R, OCN) + + + + + Swainson's Thrash (LDM, OCN) + (+) Hermit Thrash (SDM, OCN) (+) Varied Thrush (R, OCN) + + Yellow-rumped Warbler (SDM, OCN) + + Townsend's Warbler (LDM, OCN) + + + Black-headed Grosbeak (LDM, OCN)g + Pine Grosbeak (R, OCN) + Mixed associations with fragmentation Cassin's Vireo (LDM, OCN) - Clark's Nutcracker (R, OCN) - Western Tanager (LDM, OCN) - + - Negatively associated with elements of continuous landscapes Hammond's Flycatcher (LDM, OCN) - Dusky Flycatcher (LDM, OCN)g - Common Raven (R, OCN) - (-) Mountain Chickadee (R, CN) - - Ruby-crowned Kinglet (SDM, OCN) - Townsend's Solitaire (SDM, OCN) MacGillivray's Warbler (LDM, OCN)g - Chipping Sparrow (LDM, OCN) - - Dark-eyed Junco (SDM, OCN) - - Cassin's Finch (R, OCN) - - Red Crossbill (R, OCN) - Pine Siskin (R, OCN) - + + (+) + + (+) + + + + + + + Notex: Forest types described in text. Not all landscape metrics evaluated in all five forest types. Positive association (increased abundance) denoted by +; negative association by . Responses in parentheses significant at 0.05 < P < 0.10. All others significant at P < 0.05. a R. Hutto and J. Young, unpubl. report. "All Conifer" forest includes seedtree, shelterwood, and group selection harvested sites. b Evans 1995. Mixed conifer is closed canopy mature mixed conifen "Hejl and Paige 1994. cl Aney 1984. e Keller and Anderson 1992. tLDM long distance migrant, SDM - short distance migrant, R - resident (as defined by Partners in Flight); EN - enclosed nest, OCN open cup nest, CN - cavity nest. g Black-headed Grosbeak and Dusky Flycatcher classified as riparian by Hutto and Young 1999; MacGillivray's Warbler excluded from "All Conifer" analyses--not restricted to conifen closed-canopy mixed conifer forest, five species increased in relative abundance as amount of forest increased (Evans 1995). Three species were more abundant in unharvested cedar/hem- lock landscapes than in harvested landscapes, and were less abundant than expected in har- vested areas based on the amount of forest re- maining (Hejl and Paige 1994). Across these studies, Golden-crowned Kinglet was most fre- quently associated with forest cover; Brown Creeper and Winter Wren associations appeared in two studies. The relationship between abun- dance and amount of forest was not tested di- rectly in spruce/fir (Keller and Anderson 1992), but five species were more abundant in contin- uous forest than in areas interspersed with clear- cuts (see Effects of logging patterns). A similar number (9) of species had the op- posite association, decreasing in abundance with increasing amounts of forest, suggesting that they would have a positive response to fragmen- tation. However, this negative association with forest area was examined directly in only two studies, and there was less correspondence be- FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTSHejl et al. 123 tween these studies. Dark-eyed Junco was the only species that was negatively associated with forest cover in both studies; Western Tanager had opposing associations. More resident spe- cies (five) were negatively associated with in- creased amount of forest than long- (two) or short-distance (two) migrants. Patch size Relationships of abundance with patch size (the area of a continuous block of similar habi- tat) were tested directly in three studies (Table 3). Most of the species positively associated with larger patch size in the two landscape stud- ies (Evans 1995; R. Hutto and J. Young, unpubl. report) also were associated with amount of for- est. The two variables were strongly correlated (r = 0.69) in Hutto and Young's study, as they probably are in many western studies. However, Vaux's Swift, Gray Jay, Hermit Thrush, and Pine Grosbeak were associated with patch size but not to amount of forest in these studies (Evans 1995; R. Hutto and J. Young, unpubl. report). Red-breasted Nuthatch, Golden-crowned King- let, and Townsend's Warbler had the most con- sistent positive associations with patch size be- tween the two studies. Interpreting Aney's (1984) study in old growth ponderosa pine, we identified two spe- cies (Solitary Vireo [now Cassin's Vireo] and Brown Creeper) with possible minimum patch size requirements. These species were absent from stands below a certain size, even though those stands might have been large enough to accommodate at least one territory. Cassin's Vir- eo (reported territory size of 0.5 ha/pair; Aney 1984) was not detected in stands less than 5 ha (9 of 19 stands examined), but was consistently detected in larger stands. Brown Creeper (terri- tory size ranges from < 1 to 6.4 ha/pair; Hejl et al. 2002) was absent from stands less than 4.5 ha (8 of 19 stands). Many species in this study were not detected frequently enough for a pat- tern of area sensitivity to emerge. In addition, Aney (1984) did not consider annual turnover in assessing presence or absence within patches (Freemark et al. 1995). Most species (7 of 9) negatively associated with amount of "all conifer" forest across west- ern Montana and northern Idaho (R. Hutto and J. Young, unpubl. report) also were negatively associated with increasing patch size, with the exception of Clark's Nutcracker and Dark-eyed Junco (Table 3). Hammond's Flycatcher and Red Crossbill also were negatively associated with patch size in this study. Only three species de- creased in abundance as patch size increased in west-central Idaho (Evans 1995). Edge Relationships between birds and edge density or distance I¾om edge were evaluated in three studies. In "all conifer" forests (R. Hutto and J. Young, unpubl. report), all seven of the species that were positively associated with the amount of forest were also negatively associated with edge density (see Table 3; r = -0.048 between these two predictor variables, demonstrating low correlation, and thus reasonable independence, between them). In this instance, edge was defined as the boundary between patches of dissimilar cover, with 15 possible cover classifications (5 forest types, 4 open land types, 3 riparian types, and 3 other classes) within 312-ha landscapes. Two species (Brown Creeper and Hermit Thrash) had a negative association with edge density in spruce/fir (Keller and Anderson 1992). Evans (1995) measured sensitivity to edge directly by comparing abundance across three distances to edge (<50 m, 50-100 m, >100 m). Edges were defined by openings in closed-canopy forest and the juxtaposition of forests of different ages and canopy closure. Red-breasted Nuthatch, Golden- crowned Kinglet, and Townsend's Warbler were significantly more abundant as distance from edge increased (Table 3). Across studies, 10 species increased in abun- dance as edge density increased or distance from edge decreased (Table 3). Chipping Sparrow and Pine Siskin were most frequently positively as- sociated with edge across studies. Effects of logging patterns Two studies in the north-central Rockies (Kel- ler and Anderson 1992, Hejl and Paige 1994) compared the numbers of birds in landscapes modified by timber harvesting to unmodified ar- eas. In both studies, the modified areas were cre- ated by logging (stripcuts, spot cuts, and clear- cuts) interspersed within previously unlogged or partially-logged forest. (Partially-logged forest remained as continuous forest, but some trees had been selectively removed previously.) The two studies differed in habitat and methodology. In the high elevation Engelmann spruce/subal- pine fir study, Keller and Anderson did not sam- ple clearcut areas because they did not want stand comparisons to reflect avian use of unfor- ested areas compared to forested areas. In the low elevation western redcedar/western hemlock study, Hejl and Paige sampled the complete landscapes, allowing points to fall in clearcuts, on edges, or in forest interior, to see how birds responded to clearcut/forest landscapes as a whole. Of 16 species detected in spruce/fir and 38 species in cedar/hemlock, 9 species were com- 124 STUDIES IN AVIAN BIOLOGY NO. 25 mon to both studies. Of these nine species, three had the same results: Brown Creepers were more abundant in unlogged landscapes, Red-breasted Nuthatches were similarly abundant in logged and unlogged landscapes, and Pine Siskins were more abundant in logged landscapes. Hermit Thrush, American Robin, and Yellow-rumped Warbler had opposite trends in the two studies. Of those species found only in one study but with significant associations, three species were more abundant in unlogged landscapes (Moun- tain Chickadee, Winter Wren, Swainson's Thrush) and nine in logged landscapes (Northern Flicker, Olive-sided Flycatcher, Townsend's Sol- itaire, Cassin's Vireo, Warbling Vireo, Orange- crowned Warbler, MacGillivray's Warbler, West- ern Tanager, and Chipping Sparrow). In both of these studies, it was difficult to as- certain whether the associations with logged or unlogged landscapes were caused by a simple decrease or increase in suitable habitat caused by logging or by the changes in landscape con- ditions (i.e., decreased patch size, increased edge). The fact that three species (Brown Creep- er, Winter Wren, and Golden-crowned Kinglet) were less abundant in harvested cedar/hemlock landscapes than would be expected based on the amount of forest remaining (see above under Amount of forest) suggested that landscape changes could at least be a partial cause of lower numbers in those landscapes. In addition, while most of the species identified in the two studies have similar trends to those resulting from log- ging in stand-level studies throughout the West (as summarized by Hejl et al. 1995), Gray Jay, Red-breasted Nuthatch, and Pine Siskin do not, indicating potential landscape effects. Synopsis Given that there was virtually no replication of any of the conditions among the studies that we summarized, we suggest that the species most or least sensitive to fragmentation, based on their patterns of abundance, are those that show a consistent response in several forest types and geographic regions. Based on this as- sumption, Brown Creeper clearly had the stron- gest trend of species sensitive to changes in landscape patterns, as it was associated with at least one variable indicating landscape change (and usually more than one) in four of the five studies examined (Table 3). Golden-crowned Kinglet, Red-breasted Nuthatch, Winter Wren, Hermit Thrush, and Townsend's Warbler also showed consistent results across studies. These species appear as sensitive to disruptions in the pattern of forest cover on the landscape else- where in the West. Brown Creeper, Winter Wren, and Red-breasted Nuthatch were correlated with the amount of forest and/or patch size in coastal Douglas-fir or cedar/hemlock forests (Rosenberg and Raphael 1986, McGarigal and McComb 1995, Schieck et al. 1995), and Red-breasted Nuthatches and Townsend's Warblers avoided edges (Rosenberg and Raphael 1986). Fewer species had consistent positive associ- ations with elements of more fragmented land- scapes in the north-central Rockies. Several spe- cies had consistent associations with more than one landscape element within a study, but only three species (Pine Siskin, Chipping Sparrow and Dark-eyed Junco) were consistent across studies. These three species were also more abundant in logged landscapes (Keller and An- derson 1992, Hejl and Paige 1994). Our results were somewhat inconsistent with other western studies. Chipping Sparrow was associated with edges in Douglas-fir forests in California (Ro- senberg and Raphael 1986), but Pine Siskin and Dark-eyed Junco were positively associated with larger patches of old-growth Douglas-fir and hemlock forests on Vancouver Island (Schieck et al. 1995). While the five studies we reviewed differed in methods, particularly in how forest cover was defined, none attempted to define "fragmented" based on a minimum patch size. Thus, inconsis- tent results among these studies are not attribut- ed to one study considering a 200-ha patch to be a fragment and another considering it contin- uous forest. Two studies (Evans 1995; R. Hutto and J. Young, unpubl. report) measured frag- mentation indices as continuous variables across 300-ha landscapes and related bird abundances in correlation or regression tests. One logging study also based landscape descriptions on 240- ha landscapes (Hejl and Paige 1994). The other logging study used some small (20-40 ha) patches as unmanaged controls (Keller and An- derson 1992), but we used only an edge measure from this study. The old-growth ponderosa pine patch-based study included very small patches (<4 ha) but the only variable discussed from that study was patch size; we used a species' presence or absence across the range of patches as an indication of sensitivity to patch size. DEMOGRAPHIC RELATIONSHIPS WITH LANDSCAPE FEATURES Several studies have suggested that the num- ber of individual birds can temporarily increase in areas adjacent to recent cuts due to displace- ment of birds into the nearest suitable habitat (Schmeigelow et al. 1997, Walters 1998). Over the long term, high abundances can be main- tained from source habitats and a population trend would not be apparent (Van Horne 1983, Vickery et al. 1992). Increased densities could FRAGMENTATION IN ROCKY MOUNTAIN CONIFER FORESTS--Heft et al. 125 have a negative impact on reproductive rates through reduced pairing success, competition for resources, and reproductive failure (Hagen et al. 1996), issues for which demographic studies are needed. Data on the effects of landscape patterns on bird demography are seriously lacking from conifer forest habitats in the north-central Rock- ies. Several recent studies are beginning to pro- vide information to address this gap. S. Hejl (unpubl. data) studied nesting success of cavity-nesting and enclosed-nesting species in a continuous old-growth (>170 yr) cedar/hem- lock forest landscape (240 ha) and compared re- sults to nesting success from a landscape com- posed of recent clearcuts in a matrix of old- growth cedar/hemlock in northern Idaho. Nest- ing success did not differ between landscapes for any of the five focal species (Red-naped Sap- sucker, Chestnut-backed Chickadee, Red-breast- ed Nuthatch, Brown Creeper, and Winter Wren) in 1992-1994, but four species (all but Winter Wren) had trends of lower nesting success in the logged landscape. The sample of nests was lim- ited, and the numbers for some species-land- scape combinations may be too low to compute reliable Mayfield estimates (Hensler and Nichols 1981). D. Evans and colleagues (unpubl. report) studied nesting success of Swainson's Thrushes and Western Tanagers in mixed-conifer forests in west-central Idaho. Data were obtained from 10 separate study plots, four of which were clas- sified as located within relatively continuous for- est areas and six of which were classified as rel- atively fragmented. Stands were classified based on a multivariate analysis of landscape cover within 1 km of the avian demography study plots. Nesting success of neither Swainson's Thrush nor Western Tanager differed between landscape classes, although there was a trend for lower success of Swainson's Thrushes, and high- er success of Western Tanagers, in fragmented landscapes. However, overall nest success esti- mates for both species in either landscape class were substantially below the minimum nest suc- cess thresholds suggested as needed to support self-sustaining populations (Martin et al. 1996). Evans et al. (unpubl. report) also found no re- lationship between nest success and distance to edge for either species. Using survival (recap- ture and resighting of color-marked individuals) and productivity data collected from mixed-co- nifer habitats in Idaho, they modeled continu- ous-landscape and fragmented-landscape popu- lations of Swainson's Thrushes and Western Tanagers. Population trajectories did not differ between continuous and fragmented landscapes for either species, and all populations declined rapidly. Because overall estimates of annual sur- vivorship were relatively high (0.67-0.68 annual survivorship), the authors concluded that the de- clines in simulated populations were mostly tied to relatively low nesting success. Sallabanks et al. (1999) initiated a regional study examining the effects of landscape com- position on avian nesting success. They moni- tored replicate plots in managed forest land- scapes with both silviculture and agriculture, managed forest landscapes with active silvicul- ture only, and unmanaged forest landscapes with neither agriculture nor silviculture. Although statistical analyses have yet to be conducted, a preliminary examination of the data (2,847 nests of 66 bird species) suggests a mix of results: several species tend to have increasing rates of nest success along a spectrum from managed landscapes with both silviculture and agriculture to unmanaged landscapes (e.g., Warbling Vireo), others appear to be unaffected by landscape composition (e.g., Dusky Flycatcher), and still others have their highest success in the most heavily managed landscapes (e.g., Mac- Gillivray's Warbler; R. Sallabanks, pers. comm.). The primary cause of landbird nest failures within the north-central Rockies region is pre- dation, as reported elsewhere (Martin 1993). In Idaho, predators destroyed 31-35% of all nests monitored, depending on species and landscape classification (D. Evans et al., unpubl. report). Based on opportunistic observations, these au- thors recorded evidence of red squirrel (Tamias- ciurus hudsonicus) predation and speculated that avian predators, such as jays, accounted for some losses. In addition, only one of 202 nests had evidence of cowbird parasitism. Based on one year of data, R. Sallabanks et al. (unpubl. report) reported that 43% of total nests (76% of failures) were destroyed by predators in three regions in Idaho and Montana. In a companion study in west-central Idaho using artificial nests baited with clay eggs, Warner (2000) identified deer mouse (Peromyscus maniculatus), yellow- pine chipmunk (Tamias amoenus), red squirrel, and northern flying squirrel (Glaucomys sabri- nus) as the primary predators of nests placed on the ground and in shrubs. Predator assemblages were similar between managed (i.e., with agri- culture and/or silviculture) and unmanaged (i.e., without agriculture or silviculture) forest land- scapes. Warner (2000) also documented attempt- ed predation on clay eggs by deer, sheep, do- mestic cattle, coyotes, ground squirrels, beaver, and other songbirds. Demography data show some consistency with results based on abundance. Abundance data indicated that 14 species are potentially negatively affected by landscape changes caused 126 STUDIES IN AVIAN BIOLOGY NO. 25 by timber harvesting (i.e., numbers for these 14 species are either positively correlated with more of larger forests or negatively correlated with edge density or distance to edge; Table 3). For the four of these 14 species for which we have preliminary nesting success data, three (Brown Creeper, Red-breasted Nuthatch, Swain- son's Thrush) had lower nesting success trends in logged landscapes. The other species (Winter Wren) had inconsistent nesting success trends. One of the species with a mixed association with landscape changes according to abundance data (Western Tanager) had a trend of greater nesting success in fragmented landscapes. This latter re- sult was consistent with findings by Davis (1999) that Western Tanagers in Idaho were most closely affiliated with relatively open stands of primarily Douglas-fir trees. Brown-headed Cowbird occurrence Given that nest parasitism has been shown to be a problem in some fragmented landscapes, we summarized the response of Brown-headed Cowbirds to landscape changes. Studies in the north-central Rockies that have examined cow- bird abundance within a landscape context con- sistently show that proximity to agricultural ar- eas is a strong, if not the strongest, predictor of cowbird occurrence (Hejl and Young 1999, Young and Hutto 1999, Tewksbury et al. 1999). Within conifer forest sites across western Mon- tana and northern Idaho, cowbirds were more likely to be found in xeric forests (especially ponderosa pine), in areas with an abundance of cowbird hosts, close to developed, agricultural, and riparian areas, and less likely to be found in subalpine forests (Young and Hutto 1999). In the Bitterroot Valley, Montana, Brown-headed Cow- bird abundances were greatest in riparian areas, less in xeric conifer forest, and least in riparian conifer Ibrests (Tewksbury et al. 1999). Within 518-ha landscapes in xeric ponderosa pine/ Douglas-fir forests, landscape context was more important than stand attributes in determining cowbird numbers (Hejl and Young 1999). Cow- birds were more abundant in landscapes with more open land (agricultural land and grass- land), deciduous riparian habitat, mature forest (70-120 yr), and less old growth. Forest cover, logged openings, human residences, and eleva- tion were not important predictors of cowbird numbers in these xeric forests. All of these stud- ies suggest that cowbird distribution is limited by the presence and distribution of largely sup- plemental food supplied by human activities. In addition, cowbirds may be more abundant in co- nifer stands near riparian areas (but not in can- yons or riparian conifer forests) because they are attracted to riparian habitats that are dense with potential hosts, and venture into adjacent conifer forests secondarily. Fewer data are available to assess the impact of cowbirds on nest success. From BBIRD sites across the West, forest coverage correlated in- ver