sign in sign up
<<
Home , Catharus, Passerine, Troglodytes (bird), Wrentit

The Auk 118(3):678-686, 2001

RESPONSE OF TO FOREST EDGE IN COAST REDWOOD FOREST FRAGMENTS

L. ARRIANA BRAND • AND T. LUKE GEORGE Departmentof Fishand Wildlife,Humboldt State University, Arcata, California 95521, USA

ABSTRACT.--Theresponse of passerinebirds to forestedge was examinedin old-growth and mature second-growthcoast redwood (Sequoia sempivirens) forest in northernCalifornia. The study objectiveswere to determinewhich commonpasserine are sensitiveto edgesduring the breedingseason and to estimateedge width for forest interior species. Responseto edgewas measuredalong twelve 100 x 400 m plotsextending from the edge into the forestinterior to obtain relativedensity of birds. Plotswere surveyed4 to 5 times in 1996 and 8 to10 timesin 1997.We found that 14 commonpasserines showed a gradient of edge sensitivity.Steller's ( stelleri) and Swainson'sThrush ( ustulatus) had higher relative densitiesnear edgesthan in the forestinterior (P < 0.05) and were cat- egorizedas edge species. (Certhiaamericana), Winter (Troglodytestrog- Iodytes),Pacific-slope Flycatcher ( difficilis), and VariedThrush (Ixoreus naevius) had lower relative densitiesnear edges(P < 0.05) and were categorizedas interior birds. Based on exponentialregression models, estimated edge widths were 140 m for VariedThrushes, 85 m for Brown Creepers,120 m for Winter ,and 125 m for Pacific-slopeFlycatchers. Creationof edgeswould probablybenefit Steller's Jays (which may be a nestpredator), may not benefit Swainson'sThrushes, and may be detrimental to speciesthat avoid edges.We recommendthat edgeeffects be takeninto considerationwhen planning for theconservation of speciesin the region.Received 18 January2000, accepted 23 January2001.

HABITAT FRAGMENTATION affects the land- ha (including edges) uniformly. Only a few scapeby reducingthe amountand proximity of studieshave defined forest edge versus interior remnant patches of suitable and in- bird speciesby specificallyexamining edge creasing the amount of edge. Avian species avoidance(Kroodsma 1984, King et al. 1997, may respondto one or a combinationof these Germaine et al. 1997). changesin the landscapeas a resultof different Speciesthat requireforest interior may avoid biological mechanisms.However, when ad- edges due to altered microclimate,vegetation dressing the effects of habitat fragmentation, structure,or high densityof predatorsor brood the conceptsof forest-interiorspecies and area- parasites near edges (Temple 1986, Murcia sensitive specieshave often been used inter- 1995).Once a specieshas beenfound to be sen- changeably(Villard 1998).Sensitivity to habitat sitiveto edges,it is importantfor management fragmentationgenerally has been studied by purposesto know the distanceinto the forest examining the responseof a speciesto forest interior that the effect of edge (called edge fragment area rather than distancefrom the width) is observed(Wilcove et al. 1986). A num- forestedge (Formanet al. 1976,Whitcomb et al. ber of authorshave developed patch-area mod- 1981, Blake and Karr 1987, Robinson et al. els based on edge width. Temple (1986) devel- 1995). It has often been assumed that a bird oped the core-area model illustrating the speciesis a forest-interiorspecies if it is less importance of calculating the interior forest abundant or absentfrom small forest patches. area instead of total patch area, but he did not That approachconfounds the effectof edgeand estimatethe edge width empirically.Laurance patch size. For example,a speciesmay require and Yensen(1991) developeda method of es- a minimum patch size of 10 ha due to large timating the total area of forest interior that is home range requirements,but might use all 10 dependent on an empirical measurementof edge width but also did not provide an empir- ical measurement of the distance. Sisk and •Present address: Department of Fishery and Wildlife Biology,Colorado State University, Fort Col- Margules (1993) developedthe effective-area lins, Colorado 80523, USA. E-mail: arriana@cnr. model that goesthe next step of calculatingthe colostate. edu effective area based on measurements of bird

678 July2001] ResponsetoForest Edge 679 densityacross edges. However, other than Sisk nus rubra), California bay (Llmbellulariacalifornica), and Margules' (1993) calculationof bird den- big-leaf maple (Acer macrophyllum),and tan-oak sity across edges, and Gates and Mosher's (Lithocarpusdensifiorus). The understory was domi- (1981) estimationof edge use by birds on the nated by rhododendron (Rhododendronmacrophyl- lure),salal ( Gaultheriashallon), California huckleberry basis of dispersionof nests,no other studies (Vacciniumovaturn), red huckleberry(V. parvifiorum), have empirically estimatededge width based cascara (Rhamnuspurshiana), salmonberry (Rubus on actualuse of edgesby birds. spectabilis),California blackberry (R. ursinus),Hi- Most studiespertaining to the effectof hab- malayanblackberry (R. discolor),and red elderberry itat fragmentationon songbirdshave focused ( Sambucusracemosa). on the eastern and central United States (Am- The redwood region is characterizedby a mari- buel and Temple 1983, Lynch and Whigham time climatewith cool,dry summers(13 to 18øC)and 1984, Askins et al. 1989, Robbins et al. 1989). In wet, mild winters (7 to 13øC)(Harris 1991). Approx- the only study of its kind in the Douglas-fir imately 90% of the precipitationfalls from October through April. Averageannual rainfall rangesfroin (Pseudotsugamensiesii) forests of the Pacific 96 cm in Eureka to 177 cm near Orick. northwest, Rosenberg and Raphael (1986) In California, coast-redwoodforests originally oc- found that few bird speciesshowed sensitivity curred as a wide strip on the west slopeof the coast to patch size, although 12 passerine species range.Only -3.5% of the presettlementdistribution showed avoidanceof edges. Coast redwood remains as original growth (Larsen 1991). Much of (Sequoiasempivirens) forests have likewise un- the current distribution of old-growth and mature dergone drastic fragmentation.Initial studies second-growthcoast-redwood forest habitat is frag- have shown that nest predation increasesnear mented and bordered by natural as well as anthro- redwood forestedges (Brand and George2000) pogenicallyinduced edges (Fox 1997). and that the Varied (Ixoreusnaevius) is We definedforest edge by gaps ->100in in the for- sensitiveto patch size (Hurt 1996). However, est canopyoccurring adjacent to continuousforest patches.Only abrupt edgeswere usedin the study there have been no studiesdesigned to look at design consisting of a number of different types. edge effectson avian communitycomposition Natural edges resulted froin flood plains and prai- in coast redwood forests. ries, and anthropogenicallyinduced edgesresulted The primary purpose of this study was to froin one power-line corridor, as well as residential identify which common passerine speciesin- development and roads. Accessible forest edge habiting old-growth and mature second lengthsof 0.5 to 4 kin were identified froin U.S. Geo- growth coastredwood forestsare sensitiveto logical Survey orthophotoquadmaps and field re- edges during the breeding season.This may connaissance.Acceptable study-site characteristics be important both for forest interior species includeda continuousgap in the forestcanopy >100 that avoid edges because the edge habitat in exterior to an adjacentforest, a sufficientinterior forest area (>5 ha), and a location >1 kin froin other may be unsuitable, as well as for speciesthat edge lengths.Study siteswere within approximately prefer edgesand thus may be suffering high- three hourscombined driving and hiking time froin er rates of predation. The second objective Arcata, California. Once an edge length was identi- was to estimate the edge width for forest-in- fied and measuredwith a map wheel, the locationof terior speciesas a tool to develop manage- the center for each plot was selectedrandomly. A ment recommendations. distance of at least 1 kin between plots is several timeslarger than the territorydiameter of any of the METHODS bird speciesthat were examined. Twelve 4 ha rectangular plots were established. This study was conducted in redwood forest Plots were 100 in wide and extended 400 in into the patchesin Humboldt County, California, froin May forest perpendicularto the forest edge. Plots were to August 1996 and 1997. Study siteswere locatedon surveyedwithin 2 h of sunriseby walking up one public landsmanaged by Humboldt RedwoodsState transect and down a parallel transect located 25 in Park, Redwood National Park, Prairie Creek Red- on either side of the central plot line and stopping woodsState Park, and the Arcata Community Forest. for 5 rain at 50 in intervals(18 locationsper plot). The Study sitesconsisted of old-growth and mature sec- location of each bird seen or heard was recorded as ond-growth (>80 ) coastredwood forest. The x,y-coordinates.Plots were surveyed4 to 5 times in overstoryof all standswas dominatedby redwoods 1996by two observers,and 8 to 10 times in 1997by (>50%), but other speciesfound in those standsin- three observers.Observer visits were rotated sequen- cluded Douglas-fir, Sitka spruce (Piceasitchensis), tially so that eachplot was visited approximatelythe westernhemlock (Tsugaheterophylla), red alder (AI- samenumber of timesby eachobserver. 680 BRANDAND GEORGE [Auk, Vol. 118

For the analysis,plots were divided into sixteen25 in GEE, was used to obtain robust estimates of stan- x 100 m distanceintervals (bands)from the edge to dard errors. This is a theory developedto give con- interior. The responsevariable for eachspecies was sistent P-values and correct inferential results even the number of detectionswithin eachband per plot when the analysis model is incorrectly specified per (n = 16 bands x 12 plots x 2 years = 384). (Liang and Zeger 1986). This analysiswas done for The total number of detectionsfor a given speciesin the 14 most commonpasserines (those species with eachband-plot combinationwas convertedfrom the >60 detectionsover the two field seasons).The anal- x,y-coordinates of recorded bird locations. Because yses of relative bird density was done with the of short distancesbetween transectsand stops,in- "xtgee" command in Stata 5.0 statistical analysis dividual birds may have been recorded more than software (StataCorp1997). once. As such, point recordingscan not all be con- Yahner (1988) suggestedthat the "functional use sideredindependent and that was takeninto account of edges by wildlife" be used to quantify edge ef- in the analysismethodology in two ways.First, point fects, although he did not recommend a specific recordingsat each visit were summarized by band method for doing so. In this study,edge width was within each plot, and it was assumedthat average operationallydefined for edge sensitivespecies on counts per band were directly proportional to spe- the basis of an exponentialregression model with cies-specificpopulation density and that the propor- one asymptote.First, the speciessensitive to edges tionality factor is independentof distance from the were identified. Interior species were defined as edge within each field season.Consequently, ratios those having significantly lower relative density of averagecounts can be consideredestimates of rel- alongthe edgethan in the forestinterior. The interior ative bird density (see Appendix). Second,possible specieswere first modeledby a "saturated"model, multiple sightingsof a specificbird at a particular implementedby GEE, which providesan estimateof visit could also occur across band boundaries. This the mean observedbird countsin eachband adjusted and other forms of spatial autocorreletionwould in- for the other covariatesin the model. The band-spe- duce local dependencebetween relative density for cific mean observed bird counts were converted to adjacentor nearby bands.This band-to-banddepen- relative density (seeAppendix) then smoothedby an dencewas taken into accountby use of a dispersion exponentialfunction (Y = b•(1- b2X))with one asymp- parameter and robust estimationof standard errors tote and two parameters (b• and b2)where x = dis- (seeAppendix). The distancefrom edgewas takenas tancefrom edge.It was necessaryto use the average the midpoint of the bands (12.5, 37.5, etc.) and treat- relative densityper band to estimatethe exponential ed as a continuous variable. Year was treated as an regressionmodel because the GEE methodologyhas indicator variable (1996 or 1997). not yet been developedto work with count data for An augmentedform of Poissonregression was exponentialmodels. Edgewidth was defined as the used,implemented by GeneralizedEstimating Equa- distancefrom the edge at which 90% of the asymp- tions (GEE) methods, to examine whether distance totic relative density is achieved. from the forest edge and the type of edge were af- Edgespecies were defined as those having signifi- fecting relative density of (McCullagh cantly higher relative density along the edge band and Nelder 1989, Liang and Zeger 1986). The Ap- than in the forest interior. The relative density of pendix includesa detailed descriptionof the model edge specieswas graphed first with the saturated used. To account for differences in the number of model, and then with an exponentialfunction with countingtimes, the number of visits per plot per year one asymptote.The edge width was not determined was log transformedand includedin the modelas an for the edge species. offset (McCullagh and Nelder 1989, see Appendix). Ad ditiona11y, three strategieswere employedto aug- RESULTS ment the Poissonregression approach. First, to ac- count for inherent differencesin bird density from plot to plot, random plot effectswere includedin the Passerinespecies of the coast redwoods in- augmented Poisson regression model. Second, a cluded in this analysisshowed different pat- scaleparameter was includedin the model which,in terns of relative density in relation to distance effect, extendsPoisson regression to allow the vari- from the forest edge. Of the 14 most-common ance to be greater than or less than the mean. That species,most observations were of Pacific-slope is, bird detections within bands were allowed to be Flycatcher (1767 observations) and Winter under- or over-dispersedcompared to a standard Wren (1242 observations),whereas Red-breast- Poisson distribution. Non-Poisson dispersion of ed (85 observations) and American count data could result from aggregationof birds within territories, repulsion of birds between terri- Robin (74 observations) were observed the tories,possible failure to detectindividual birds, or least. multiple detectionsof individual birds. Finally, the Steller'sJay and Swainson'sThrush were cat- Hubur/White/sandwich estimator,as implemented egorizedas edge specieswith approximately July2001] ResponsetoForest Edge 681

TABLE1. Relative density of common passerinesin relation to distancefrom the forest edge in coastred- wood forests of northern California. The average count per band; exp[ln(number of visits) + B0 + B• (distance)+ B2 (year)]•'. A negative coefficientindicates decreased relative density with distance,and a positivecoefficient indicates increased relative densitywith distance.

Robust Relative Densityb Species Coef. (B•) S.E. P-value 0 m 200 m 400 m Pacific-slopeFlycatcher Empidonaxdifficilis 0.0011 0.0003 0.001 1.00 1.26 1.58 Steller'sJay Cyanocittastelleri -0.0018 0.0007 0.008 1.00 0.69 0.48 corax -0.0001 0.0017 0.931 1.00 0.97 0.94 Chestnut-backedChickadee Parusrufescens -0.0001 0.0006 0.886 1.00 0.98 0.97 Red-breasted Nuthatch Sitta canadensis 0.0024 0.0012 0.052 1.00 1.63 2.64 Brown Creeper Certhiaamericana 0.0015 0.0005 0.003 1.00 1.36 1.84 Troglodytestroglodytes 0.0016 0.0006 0.004 1.00 1.38 1.91 Golden-crownedKinglet Regulussatrapa 0.0002 0.0006 0.703 1.00 1.05 1.10 Swainson's Thrush Catharus ustulatus -0.0017 0.0008 0.030 1.00 0.71 0.51 Turdusmigrat•riu• -0 0099 00018 0.2!6 !.00 0.64 0.42 Ixoreus naevius 0.0019 0.0005 0.000 1.00 1.47 2.18 Chamaeafasciata 0.0014 0.0025 0.571 1.00 1.32 1.74 Hermit Dendroica occidentalis 0.0014 0.0009 0.128 1.00 1.31 1.72 Wilson'sWarbler Wilsoniapusilla 0.0002 0.0004 0.582 1.00 1.05 1.10 • Seeappendix for a detaileddescription of the augmentedPoisson regression model. bThe relativedensity of eachspecies at 0, 200,and 400 m fromthe edge was obtained by takingthe value of e raisedto thecoefficient multiplied by the distancefrom edge.For example,the relative densityfor the Varied Thrush at 400 m from the edge = exp[0.0019(400)]= 2.14 timesthe relative density at the edge. (The slight discrepancyis due to rounding of coefficientsin this table). Pleasesee appendix for a descriptionof the generalmethod of interpretingPoisson regression model parameters. half of their relative density at 400 m as at the of the relative densityper band, the estimated forest edge (Table 1). American Robin showed edge width was 140 m for the Varied Thrush, a similar tendencybut the relationshipwas not 85 m for the BrownCreeper, 120 m for the Win- significant(P = 0.21). Five speciesappeared ter Wren, and 125 m for the Pacific-slopeFly- unaffectedby distancefrom the forestedge (P catcher(Fig. 1). The relative densityof the two -> 0.571)with a relative density at 400 m from edge speciesdecreased in relation to distance the edge ranging from 0.94 to 1.74 times the from the forestedge (Fig. 2). relative density at the forest edge (Common Raven, Chestnut-backed Chickadee, Golden- DISCUSSION crowned , Wrentit, and Wilson'sWar- bler). and Red-breasted Nut- hatch showed nonsignificant (P -< 0.128) Estimatingedge sensitivityis onemethod of althoughsubstantial tendencies for lowerrela- identifyingspecies that may be affectedby for- tive densityclose to the forestedge, with a rel- est fragmentation.Biotic and abioticdifferenc- ative density at 400 m from the edge ranging es betweenedges and forest-interiorlocations, from 1.72 to 2.64x that at the forestedge. Four suchas differences in predation(Gates and Gy- species were categorized as forest interior sel 1978), microclimate(Chen et al. 1993), veg- birds: Brown Creeper, Winter Wren, Pacific- etation structure (Ranney et al. 1981), and in- slopeFlycatcher, and VariedThrush with a rel- sect composition (Shure and Phillips 1991), ative density at 400 m from the edge ranging probably affect different speciesin different from 1.58 to 2.18x the relative density at the ways dependingon their breeding and forag- forest edge (Table1). ing requirements.Patch size has often been Relativedensity was plotted for the six spe- usedto measuresensitivity to forestfragmen- ciesthat showeda significanteffect of distance tation. Although patch-size sensitivity and from the forestedge. For the forest-interiorspe- edgesensitivity are probablycorrelated (due to cies, there were very few detectionsof birds edge influencesthat operate on a larger pro- from 0 to 25 m from the edge. Detectionsin- portion of the area in smaller patchesthan in creasedfrom 25 to 100 m from the edge, and larger patches),patch size is not a precisemea- then a leveled off from approximately100 to sure of edge sensitivitynor vice versa(Villard 400 m. Basedon exponentialmodel smoothing 1998). 682 BRANDAND GEORGE [Auk, Vol. 118

Varied Th n,.qh ø Paeifie-slope Flycatcher

e

e ß e ß ß

o

ß

o

140 m i 125m

i i , i i i i i , , , , 0 50 100 150 200 250 3• 3• 4• 0 50 100 150 200 250 300 350 400

Winter Wren. e Brown C!•el)er

o o

ß ß 4 e

e ...... ß 3 i ß ß ß ß

2

I I I t I I I , [

0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400

Distancefrom edge(m) FiG.1. Relativedensity with respectto distancefrom the forestedge and estimatededge width for the VariedThrush, Pacific-slope Flycatcher, Winter Wren, and BrownCreeper. The points represent the band- specificrelative density. The smooth curve represents the relative density based on an exponential regression modelwith oneasymptote. The dottedline showsthe asymptoticrelative density and the dash-dotline illustratesthe edgewidth definedas the distance from edge at which90% of theasymptotic relative density has been achieved.

Edge-sensitivespecies.--The edge-sensitive ests. Winter Wrens breed in moist coniferous speciesfound in thisstudy concurred relatively forestsand nestin densebrush especially along well with other studies that measured sensitiv- stream banks (Ehrlich et al. 1988). Barrows ity to forestfragmentation. Winter Wrens have (1986) found that Winter Wrens in California beenfound to be sensitiveto edges(Rosenberg had broadhabitat preferences in fall and win- and Raphael1986), patch size (Rosenbergand ter, but that habitat selection shifted in the Raphael1986), and clearcutsembedded in old- breeding seasonalmost exclusivelyto old- growth (Hejl and Paige1994). Varied Thrush growth forest characterizedby dense,moist distributionwas positively related to forest- understory.Likewise, McGarigal and McComb fragmentsize in an earlierstudy in coastred- (1992) found that the Winter Wren was associ- woods (Hurt 1996). Pacific-slopeFlycatcher, atedwith ripariansystems in Oregon.The Var- Brown Creeper,Red-breasted Nuthatch, and ied Thrush breeds in moist coniferous forest Hermit Warbler were also found to be sensitive with a denseunderstory and is associatedwith to edgesby Rosenbergand Raphael(1986). streams(Beck 1997). The Pacific-slope Flycatch- There are many possibleexplanations for er breedsin forestsespecially near water (Ehr- edge sensitivity.Some speciesthat avoided lich et al. 1988).Edges receive higher levels of edgesshow habitat preference for moistfor- incident radiation, and thus microclimate near July2001] ResponsetoForest Edge 683

Reese1988). Brand and George(2000) used ar- tificial nestsand foundpredation to be signifi- cantly higher near edgesthan in the forest in- terior in coast-redwood forest stands. If 06 ,••o Steller'sJay Swainson'sThrushes have higher relative den-

0.4 sity closerto edgesbecause they are selecting edgesfor nesting, then they may also be suf-

02 fering higher levelsof nestpredation near edg- f , , , 0 50 100 150 200 250 300 350 400 es. Though edge-preferringspecies may also have mechanismsto reduce predation, it is

Swainson's Thrush questionable whether the active creation of 10 edges would be beneficial for Swainson's Thrush, and that is an area which warrants fur- O8 ther study. 06 The averageedge width of the four forest-in- terior speciesis --115 m. The distance of 115 m O4 from the forestedge alsocorresponds with the

02 distanceat which the probability of predation , , , r 0 50 100 150 200 250 300 350 400 on artificial nests declinesby half (Brand and George2000). An estimateof 115 m edgewidth Distancefrom edge (m) and a 115 m diameter core area suggeststhat a FIG. 2. Relative density with respectto distance circular forest patch should be a minimum of from the forest edge for the Steller'sJay and Swain- 9.3 ha in to have 1 ha of interior habitat. son's Thrush. However, this size may not be adequate de- pendingon the territory size of a givenspecies. The edgewidths estimatedin this studycan be edges(such as temperatureand humidity) re- usedto predictpatch sizes that, in effect,would quired by thosespecies may be altered. Micro- be all edgefor particularforest interior species. climate changes,in turn, could affectvegeta- For example, a 140 m edge width for Varied tion compositionand structure as well as the Thrushessuggests that a circularpatch size of prey basenear edges.It is alsopossible that the 6.2 ha consistsof only edge habitat. However, speciesthat avoid edgesmay be respondingto selectionpressure of increasedpredation over Hurt (1996) found Varied Thrushesrequired a ecologicaltime. minimum patch size in coast-redwood forests Edge-neutralspecies.--The Chestnut-backed of 16 ha, thus suggestingthat a minimum of 2.3 Chickadee,Golden-crowned Kinglet, Wrentit, ha of forestinterior are required. Wilson's Warbler, and Common Raven showed no evidencefor avoidingor preferringedges in CONCLUSION coastredwoods. This contrastswith a study in Douglas-fir forests in which The Chestnut- Recommendationsfor the common species backedChickadee and Golden-crownedKing- that breed in the redwood forestregion can be let were found to avoid edges(Rosenberg and madefrom the resultsof this study.(1) The cre- Raphael 1986). ation of edge would probably benefit Steller's Edge-preferringspecies.--The American Robin Jays,but that speciesmay be a nest predator was found to be an edge bird in the eastern and is not a bird of conservationconcern. (2) It United States (Johnstonand Odum 1956). The is questionablewhether the active creation of AmericanRobin also showedthat tendencyin edges would be benefit Swainson'sThrushes. this study,but small samplesize probablyre- Further study is neededto investigatewhether sultedin inconclusiveresults. Steller's Jays had Swainson'sThrushes are sufferinghigher rates higher relative abundancenear edgesand were of nest predation near edgesin coast-redwood observedtaking eggsout of artificial nests(L. forest similar to that which has been found us- A. Brandpers. obs.). In contrastto Steller'sJays, ing artificial nests(Brand and George2000). (3) Swainson'sThrushes may be caught in an eco- Creation of edge may be detrimental to Pacific logical trap (Gates and Gysel 1978, Ratti and Slope Flycatcher,Varied Thrush, Winter Wren, 684 BRANDAND GEORGE [Auk, Vol. 118 and Brown Creeper, as well as Red-breasted GATES,J. E., AND L. GYSEL.1978. Avian nest disper- Nuthatch and Hermit Warbler. sion and fledgling successin field-forest eco- tones.Ecology 59:871-883. GATES, J. E., AND J. A. MOSHER. 1981. A functional ACKNOWLEDGMENTS approachto estimatinghabitat edge width for birds. American Midland Naturalist 105:189- Specialthanks go to C. Campbell, K. Melody, M. 192. Wuestehube, and J. Powell who helped collect data. GERMAINE, S.S., S. H. VESSEY, AND D. E. CAPEN. T. W. Arnold, D. G. Hankin, J. S. Marks, A. N. Powell, 1997. Effects of small forest openings on the K. G. Smith, D. Twedt, and one anonymousreviewer breeding bird community in a Vermont hard- provided helpful review comments. Public land wood forest. Condor 99:708-718. managers of Prairie Creek Redwoods State Park, HARRIS, S. W. 1991. Northwestern California Birds. Redwood National Park, Humboldt Redwoods State Humboldt State University Press, Arcata, Park, and the Arcata Community Forestwere partic- California. ularly helpful in grantingpermission to conductthis HEJL,S. J., AND L. C. PAIGE.1994. A preliminary as- research.This study was funded by the Humboldt sessmentof birds in continuousand fragmented Area Foundation. forests of western redcedar and western hem- lock in northern Idaho. Pages189-197 in Sym- LITERATURE CITED posium Proceedings,Interior Cedar-Hemlock- White Pine Forests:Ecology and Management. AMBUEL,B., AND S. A. TEMPLE.1983. Area-depen- (D. M. Baumgartner,J. E. Lotan, and J. R. Tonn, dent changesin the bird communitiesand veg- Eds.). WashingtonState University, Pullman. etation of southernWisconsin forests. Ecology HURT, M. 1996. Breeding distribution of Varied 64:1057-1068. Thrushes (Ixoreus naevius) in redwood forest ASKINS, R. A., J. F. LYNCH, AND R. GREENBERG.1989. patches.M.S. thesis,Humboldt StateUniversity, Populationdeclines in migratory birds in east- Arcata, California. ern .Current Ornithology7:1-57. JOHNSTON,D. W., AND E. P. ODUM. 1956. Breeding BARROWS,C. W. 1986.Habitat relationshipsof Winter bird populationsin relation to plant succession Wrens in northern California. Western Birds 17: on the Piedmontof Georgia.Ecology 37:50-62. 17-20. KING, D. I., C. R. GRIFFIN, AND R. M. DEGRAAF. 1997. BECK,M. 1997.Songpost and foraginglocation char- Effect of clearcut borders on distribution and acteristicsof breedingVaried Thrushes(Ixoreus abundance of forest birds in northern New naevius)in coastal redwood forests of northwest Hampshire. Wilson Bulletin 109:239-245. California. M.S. thesis, Humboldt State Univer- KROODSMA,R. L. 1984. Effect of edge on breeding sity, Arcata, California. forest bird species.Wilson Bulletin 96:426-436. BLAKE,J. G., ANDJ. R. KARR.1987. Breedingbirds of LARSEN,C. J. 1991. A status review of the Marbled isolated woodlots: Area and habitat relation- Murrelet (Brachyramphusmarmoratus) in Califor- ships. Ecology68:1724-1734. nia. Department of CandidateSpecies Status Re- BRAND, L. A., AND T. L. GEORGE.2000. Predation risk port 91-1. California Department of Fish and of nesting birds in coast redwood forest frag- Game, Sacramento. ments. Journal of Wildlife Management 64:42- LAURANCE,W. F., AND E. YENSEN.1991. Predicting 5l. the impactsof edge effectsin fragmentedhabi- CHEN, J., J. F. FRANKLIN, AND T. A. SPIES.1993. Con- tats. BiologicalConservation 55:77-92. trasting microclimates among clearcut, edge LIANG,K.-Y. 1987.Estimating functions and approx- and interior old growth Douglas-fir forest.Ag- imate conditional likelihood. Biometrika 4:695- ricultural and ForestMeteorology 63:219-237. 702. EHRI.ICH, P. R., D. S. DOBKIN, AND D. WHEYE. 1988. LIANG, K.-Y., AND S. L. ZEGER.1986. Longitudinal The Birder's Handbook: A Field Guide to the data analysisusing generalizedlinear models. Natural History of North American Birds. Si- Biometrika 73:13-22. mon and Schuster, Fireside, New York. LYNCH, J. F., AND D. F. WHIGHAM. 1984. Effects of for- FORMAN, R. T. T., A. E. GALLI, AND C. F. LECK. 1976. est fragmentationon breedingbird communities Forest size and avian diversity in New Jersey in Maryland, USA. BiologicalConservation 28: woodlotswith someland-use implications. Oec- 287-324. ologia 26:1-8. MCCULLAGH, P., AND J. A. NEI.DER. 1989. General- Fox, L. 1997.Klamath bioregional assessment project ized Linear Models, 2nd ed. Chapman and Hall, GIS maps. Department of Natural Resources, London. Planning and Interpretation. Humboldt State MCGARIGAL, K., AND W. C. MCCOMB. 1992. Stream- University, Arcata, California. side versusupslope breeding bird communities July2001] ResponsetoForest Edge 685

in the central Oregon coast range. Journal of avifauna of the eastern deciduousforest. Pages Wildlife Management 56:10-23. 125-205 in ForestIsland Dynamics in Man-dom- MURCIA,C. 1995.Edge effectsin fragmentedforests: inated Landscapes(R. L. Burgessand D. M Shar- Implicationsfor conservation.Trend s in Ecology pe, Eds.). Springer-Verlag,New York. and 10:58-62. W1LCOVE, D. S., C. H. MCLELLAN, AND A. P. DOBSON. NELDER, J. A., AND R. W. M. WEDDERBURN. 1972. 1986. Habitat fragmentationin the temperate Generalized linear models.Journal of the Royal zone. Pages 237-256 in Conservation Biology: Statistical Society SeriesA 135:370-384. The Scienceof Scarcityand Diversity (M. E. Sou- RANNEY, J. W., M. C. BRUNER, AND J. B. LEVENSON. 16, Ed.). Sinauer Associates, Sunderland, 1981. The importance of edge in the structure Massachusetts. and dynamics of forest islands. Pages67-95 in YAHNER,R. H. 1988. Changesin wildlife communi- Forest island Dynamics in Man-dominated ties near edges.Conservation Biology 2:333-339. Landscapes(R. L. Burgessand D. M. Sharpe, Eds.). Springer-Verlag,New York. Associate Editor: A. Powell RATTI, J. T., AND K. P. REESE.1988. Preliminary test of the ecological trap hypothesis.Journal of APPENDIX: Descriptionof themodel used for augmented Wildlife Management 52:484-491. poissonregression analysis. This study of relativebird ROBBINS,C. S., J. R. SAUER, R. S. GREENBERG,AND S. densityas a functionof distancefrom the forestedge DI•OEGE.1989. Population declines in North was basedon the total observedbird countsper band American birds that migrate to the Neotropics. per plot per year. For eachband, the total observed Proceedingsof the National Academyof Scienc- bird count was the sum of the counts observed over es USA 86:7658-7662. all visits to the plot during eachof two field seasons. ROBINSON, S. K., 17.R. THOMPSON III, T. M. DONOVAN, The year was treated as a binary predictor variable D. R. , AND J. FAABORG. 1995. Re- that allowed for a systematicshift in averagecounts gional forest fragmentationand the nestingsuc- acrossall plot-band combinations.Because the num- cessof migratory birds. Science267:1987-1990. ber of visits varied somewhatfrom plot to plot, that ROSENBERG,K. V., AND M. G. RAPHAEL.1986. Effects varying amount of observationtime had to be ac- of forest fragmentation in Douglas-fir forests. counted for in the analysis model. Pages263-272 in Wildlife 2000: Modeling Hab- Let Cpovrepresent the countobserved in plot p in itat Relationshipsof Terrestrial (J. band b for visit v for a given bird species.As the pre- Verner, M. L. Morrison, and C. J. Ralph, Eds.). liminary step in modeling, those counts were as- University of WisconsinPress, Madison. sumed to be Poisson distributed observations with SHURE, D. J., AND D. L. PHILLIPS. 1991. Patch size of an averagekpmg, > 0 accordingto the probability forest openings and arthropod populations. model: Oecologia86:325-334. SISK,T. D., AND C. R. MARGULES.1993. Habitat edges e-;•pbv )kpbvCpbv P(Cpbv)-- whereCp•,v = 0, 1, 2, and restoration:Methods for quatifying edge ef- Cpsc,! .... fectsand predictingthe resultsof restorationef- Acrossvisits for a specificplot and band combi- forts. Pages57-69 in Nature Conservation3: Re- nation, the variability of countsaround the mean are constructionof Fragmented Ecosystems(D. S. Saunders,R. J. Hobbs, and R. Ehrlich, Eds.). Sur- assumedto be independent.For example,bird activ- ity levels, and thus the likelihood of detection,may rey Beatty and Sons,Chipping Norton, United vary around the averagefor a particular band due to Kingdom. STATACORP. 1997. STATA Statistical Software, re- weather conditionsthat can changefrom visit to visit lease 5-0. Stata Corporation, College Station, within a single field season.From the additive prop- Texas. erty for independentPoisson counts, it thereforefol- TEMPLE,$. A. 1986. Predicting impacts of habitat lowsthat the total count Cp• for theband over Vp total fragmentationon forestbirds: A comparisonof number of visits over both years is distributed ac- two models. Pages 301-304 in Wildlife 2000: cordingto the Poissonprobability model: Modeling Habitat Relationshipsof Terrestrial e VpXpbpp)kpbCpb Vertebrates (J.Verner, M. L. Morrison, and C. J. P(C•,•,)•- , whereC•,• = 0,1,2.... Cp•! Ralph., Eds.). University of Wisconsin Press, Madison. with averagecount )kp•= averageof )kpbv across Vp VILL^RD, M.-A. 1998. On forest-interior species, visits. edge avoidance,area sensitivity,and dogmasin The basic Poisson model can be extended to the avian conservation. Auk 115:801-805. PoissonRegression Model (PRM) by letting: WHITCOMB, R. E, C. S. ROBBINS,J. E LYNCH, B. L. WHITCOMB, M. K. KLIMKIEWICZ, AND D. BYS- kp•= exp[B0 + Bt(xt) + B•(x•) + ... + B•(x•)] TRAK. 1981. Effects of forest fragmentation on such that 686 BRANDAND GEORGE [Auk, Vol. 118

ln0•,b)= Bo+ B•(xt) + B2(x2)+ '" + BL(XL) lows for statisticallydependent data from band to bandwithin plotsand alsoallows for plot to plot var- where Bo,B•, B2,ß ß ß, BLare the regressionmodel co- iation (Liang and Zeger 1986). The GEE method uses efficients to be fitted to the data and x•, x2, ..., xLare a "working correlation matrix" (Liang and Zeger the predictors of the averagetotal countsper band 1986) to representplot to plot variation and provide per plot. The exponentialform is generallyused be- an approximatemodel for the correlationbetween causeit forcesthe averagecount parameter )t,b to be countsfrom differentbands in the sameplot. Forthis >0. Note that the natural log transformation will analysisthe "exchangeable"working correlationma- causeln0•,• ) to be a linearmodel. trix was used. In addition, a "robust" method for es- In the PRM the averagetotal count for band b in timating standard errors of regressioncoefficients plot p is: and P-valuesis provided that accountsfor the fact 2•,•= V, exp[Bo+ B•(x•)+ B2(x2)+ '" + BL(xL)] that the working correlationmatrix usedfor the anal- ysis may not fully agree with the true situation = exp[ln(Vp)+ Bo+ B•(x•) (Liang and Zeger 1986). The robust and correlation (exchangeable)options were used with the Stata + B2(x2)+ ... + BL(XL)] xtgee commandto do the analysis. where the term ln(V,) servesas a predictorwith a Interpretationof model coefficients.--Withthe bird known regressioncoefficient equal to one. A term of countsobtained in this project,it is only reasonable this type is called an offset (McCullagh and Nelder to make the working assumptionthat the averageof 1989).The offsetallows the PRM to deal with vary- observedcounts in a particular plot-band (25 x 100 ing numbersof visits and providesa way to study m area) per year are proportional to the true average systematicdifferences in )•,•depending on distance bird densityin the correspondingplot-band area per from the edge. year. Assumingthat the proportionalityfactor relat- ing the averageof observedcounts to underlying However, the PRM has two important limitations bird densityis independentof distancefrom edge,it relative to the analysisrequirement for this study. is possibleto estimatethe relativedensity (ratios of First, standard Poissonregression requires that the averagebird counts)from the outermost(edge) band varianceof observedcounts equal the averageof ob- relativeto bandsat varying distancesfrom the edge. servedcounts. This property seldomholds in biolog- A generalprocedure for interpretationof PRM co- ical applications (McCullagh and Nelder 1989) and efficientsis basedon ratiosof averagecounts at two is unlikely to be appropriatehere. Secondly, the PRM predictor points; x•m, x?], ..., x•m versusx• [2], X2 [2], requiresthat the total countoutcomes for the differ- ß.., xrt2•.Let RDl•i[21= relative density at predictor ent band and plot combinationsbe statisticallyin- point [1] versuspredictor point [2]. Then, dependent.As discussedin the methodssection, the )tp•,atx•m, X211] • . . . , XL[1] countdata, althoughindependent from plot to plot, RDt•i•21= are not likely to be independentfrom band to band )kpbyat x1[2], x212] , . . . , XL[2] within a plot. Fortunately,both of the restrictiveas- sumptionsof the PRM can be relaxed. = (exp{ln(%)+ Bo+ The generalized linear model (GLM) (McCullagh + B2(x2••) + ... + BL(x,?)}) and Nelder 1989, Nelder and Wedderburn 1972), which can be implemented by the Stata glm and -' (exp{ln(Vp)+ B0 + B•(x?) xtgee commands(StataCorp 1997), allow the count varianceto be proportional,rather than necessarily + B2(x?) + ... + B,•(x?•)}) equal, to the averagecount. This is achievedwith a = exp{B•(x?l - x?) proportionality (scale)parameter that can be higher than or lessthan 1 correspondingto over- or under- d- B2(x2[•1 -- X2[21) d- o.' dispersionrelative to )t,•,.The scaleparameter is es- timated from the data. For this analysis, the + B•(x•l•l - x?)} scale(dev)option with xtgee was used, in which the For example,if x2through x• are the same at both scaleparameter was set to the deviancedivided by predictorpoints and x• is a continuouspredictor for the residual degreesof freedom. the distancefrom edgecoded x? = 0 at the edgeand The PRM has been generalizedfurther by the gen- x? = 1 for I m awayfrom the edge,then exp(B•(Ad)) eralized estimating equation (GEE) extensionof is theratio of relativedensity at Admeters away from GLM (Liang and Zeger 1986, Liang 1987). GEE al- the edge versusthe edge.