Ecotolical Applications. 7(2). 1997. pp. SS2-S63C 1997 by (he Ecoloaica' Society of America

RESPONSE OF RODENTS TO HABITAT FRAGMENTATION INCOASTAL SOUTHERN CALIFORNIA

DoUGLAS T. BOLGER,' ALLISON C. ALBERTS,2 RAYMOND M. SAUVAJOT,J PAULA POTENZA,4CATHERINE McCALVIN,4 DUNG TRAN,4 SABRINA MAZZONI,4 AND MICHAEL E. SOUL~~

IEnvironmental Studies Program. HB 6/82. Dartmouth College. Hanover. New Hampshire. 03755 USA2Center for Reproduction of Endangered Species. Zoological Society of San Diego. P. O. Box 55/.

San Diego. California 92//2 USAJU.S. National Park Service, Santa Monica Mountains National Recreation Area. 3040/ Agoura Road,

Suite /00. Agoura Hills, California 9/30/ USA4Department of Biology, University of California at San Diego. La Jolla. California 92093 USA

'Board of Environmental SlJIdi~s, Ultiv~rsity of California at Saitta Cruz. Santa Cruz. California 95064 USA

Abstract. We employed an island biogeographic approach to determine whether smallfragments of the shrub habitats coastal sage scrub and chaparral, isolated by urbanization,are capable of supporting viable populations of native rodent species. The distribution ofnative rodents in 25 urban habitat fragments was assessed by live-trapping. Over half ofthe fragments surveyed (13 of 25) did not support populations of native rodents. Fragmentssupported fewer species than equivalently sized plots in large expanses of unfragmentedhabitat, and older fragments (fragments that had been isolated for a longer period of time)supported fewer species. Both results implied that local extinctions occurred in the fragmentsfollowing insularization. Stepwise multiple polychotomous logistic regression was used todetermine which biogeographic variables were the best predictors of species Dumber acrossfragments. The area of shrub habitat in each fragment was the most significant predictorof species diversity; age of a fragment was also significant and was negatively correlatedwith species number. but the isolation distance of a fragment had no relationship to speciesdiversity. We found a negative relationship between extinction vulnerability of native rodentspecies and relative abundance: species that were more abundant in unfragmented habitatpersisted in more habitat fragments. Random environmental and demographic ftuctuations(island effects) and edge effects associated with fragmentation are proposed as causes ofthese local extinctions.

Key words: chaparral; coastal sage scrub; extinction; habitat fragmentation; landscape ecology;Neotoma; Peromyscus; Perognathus; persistence; rodents; southern California.

laxation has been documented or inferred for severalgroups of land bridge islands and habitat fragments(Soul~ and Sloan 1966, Brown 1971, Diamond 1972,Willis 1974, 1979, Case 1975, Terborgh 1975, Souleet al. 1979, 1988, Wilcox 1980, Burgess and Sharp1981, Whitcomb et al. 1981, Lynch and Whigham1984, Bolger et al. 1991). In this report, we documentrapid relaxation of rodent faunas in recently isolatedfragments of the shrub habitats, coastal sage scrub andchaparral, in coastal San Diego County, California,USA, and discuss the role of potential causative factors.

The creation of "islands" by habitat fragmentationdiffers from the creation of true land bridge islandsbecause, in addition to the reduction in populationsizes, habitat fragmentation also brings about otherchanges that result from the juxtaposition of naturalhabitat with human development. The tenDS "edge ef-fects" (Lovejoy et al. 1986, Wilcove et al. 1986) or"external threats" (Janzen 1986) have been used torefer to changes brought about by agents that originateon a boundary between two habitats (ecotone specialistplant and animal species) or simply anywhere outsidethe fragment (generalist species, ruderal species, fire-,

INTRODUCTION

The study of the biotic consequences of habitat frag-mentation has often relied conceptually on the relax-ation model of equilibrium island biogeography (Mac-Arthur and Wilson 1967, Brown 1971, Diamond 1972).The equilibrium theory states that species number onislands is determined by an equilibrium between ratesof extinction and immigration. According to equilib-rium theory, land bridge islands (islands created whenrising sea level insularizes an area of continent) shouldbe supersaturated with species and their faunas shouldrelax to some lower equilibrium number of species. InMacArthur and Wilson's model, this relaxation iscaused by increased rates of stochastic extinction re-sulting from the reduction in population sizes causedby the subdivision of the previously contiguous pop-ulations. Many authors have drawn an analogy betweenhabitat fragments and land bridge islands. and pre-dicted relaxation would occur in habitat fragments. Re-

Manuscript received 3 Marcb 1995; revised 5 December1995; accepted 20 December 1995; final version received 29April 1996.

~

May 19'}~ EFFECTS OF ~nAl f}{A(JMj:J'OlAI1U~ U~ KUUb~ 1~

assemblages, containing elements of coastal sagescrub, maritime succulent scrub, mixed chaparral. andchamise chaparral (Munz and Keck 1959, Beauchamp1986, Alberts et aI. 1993). The species composition isdetermined by the slope, aspect, and soil characteristicsof the site. Most fragments were dominated by speciesof the coastal sage scrub assemblage. Portions of thenorth-, east-, and west-facing slopes were often dom-inated by lemonadeberry (Rhus integrifolia) and laurelsumac (Malosma laurina). South-facing slopes andcanyon bottoms were dominated by black sage (Salviamellifera) and California sagebrush (Artemesia cali-fornica). Chaparral shrub species were generally pres-ent on north-facing slopes. These included toyon (Het-eromeles arbutifolia), spiny redberry (Rhamnus cro-cea), and scrub oak (Quercus dumosa) (Albens et al.1993). Two of the fragments, Solana and Montanosa,contained large areas of chamise chaparral dominatedby chamise (Adenostomafasciculatum). In previous pa-pers (Soule et aI. 1988, Bolger et al. 1991) we referredto this mosaic of vegetation types with the generic tenDchaparral. Use of this term is somewhat misleading,however, as coastal sage scrub is the dominant plantassemblage in all but the two chamise chaparral frag-ments. Here we will use the term "shrub habitat" torefer to this mosaic.

wind-. human-caused habitat degradation). The frag-ments we studied are small enough (1-80 ha) that thesedisturbances can often be manifested throughout thefragment. not just along its boundary. The tem1 "islandeffects" has been used to refer to the changes that resultfrom the population subdivision caused by habitat frag-mentation. which may include increased extinction ratedue to increased vulnerability to demographic and en-vironmental stochasticity. inbreeding depression. andgenetic drift.

We have previously described the effects of habitatfragmentation on shrub specialist bird species in 37habitat fragments in coastal San Diego County (Souleet al. 1988. Bolger et al. 1991). The natural landscapein this region is composed of broad coastal mesas dis-sected by extensive dendritic canyon systems. The hab-itat patches we studied are fragments of these canyonsystems. The natural habitat of this region is a mosaic.with chaparral typically dominating the mesa tops.coastal sage scrub on the slopes. and riparian woodlandin the broader canyon bottoms. Extensive developmentof the "mesa tops has isolated variously sized patchesof this habitat mosaic. In this study. we examined thedistribution of native and introduced rodents in 25 ofthese fragments. We had two primary aims in under-taking this study. The first was to determine whetherrodent populations within habitat fragments are viableover time. or if most or all populations will eventuallygo extinct. and to determine which ecological factorsinftuence the ability of a particular fragment to supportrodent populations. Factors expected to affect viabilitywere fragment size. degree of isolation of fragments.habitat quality and quantity. and predation by and/orcompetition with human commensal species. The sec-ond aim was to ascenain whether species were differ-entially vulnerable to extinction. and if so. to determineif vulnerability was correlated with density. Density.or its correlates. trophic position and body size. havebeen shown to be indicators of extinction vulnerability(see Diamond 1984a. b for review and references).

Mainland plots

To compare species diversity in fragments to that inunfragmented habitat, we trapped rodents on three con-trol plots measuring 0.12, 0.5, and 1 ha in continuous,relatively undisturbed habitat. Extending the landbridge island analogy, we refer to these as mainlandplots. These plots were located in Tecolote CanyonPark (0.5 and 1 ha plots) and in an undeveloped areaknown as Del Mar Mesa (located between PenasquitosReserve and Carmel Valley) (0.12-ha plot), and werechosen to approximate the habitat fragments in termsof vegetation and slope. These two sites contain con-tinuous areas of habitat many times the area of ourlargest fragment (-1000 and 4500 ha respectively).

SpeciesThe coastal sage scrub and chaparral plant assem-

blages of coastal southern California support a richnative rodent community (M'Closkey 1972, Meserve1972, 19760. b, Price and Waser 1984). Most of ouranalyses are limited to what we term the shrub-inhab-iting rodents (RODENTS), rodent species that areclosely associated with coastal sage scrub and chaparralhabitats-the California mouse (Peromyscus ca/ifor-n;cus) , the cactus mouse (Peromyscus eremicus), theSan Diego pocket mouse (Perognathus tal/ax), thedusky-footed woodrat (Neotoma fuscipes), the desertwoodrat (N. /ep;da) , and the western harvest mouse(Reithrodontomys mega/otis). The only native rodentwe trapped that is not included in this category is theCalifornia vole (Microtus ca/ifornicus). M. ca/ifom;cus

METHODS

Habitat fragments

We surveyed the rodents in a subset of 25 of the 37habitat fragments described in Soul~ et aI. (1988).These 25 are listed in Table I and their approximatelocations noted in Fig. I. This subset spanned the rangein size and time since isolation of the fragments incoastal San Diego County (see Table I). Most of thesites are fragments of canyons. although a few alsocontain mesa-top habitat. These canyons typicallyrange from 15 to 61 m in depth. The boundaries of thefragments are delineated by urban and suburban de-velopment. and all fragments are completely surround-ed by human-modified habitat.

Because of the complex topography of the canyons.the vegetation in each fragment is a mosaic of plant

S54 DOUGLAS T. BOLGER ET AL. EcoqicaJ ApplicationsVol. 7. No.2

TABLE I. Values of biogeographic variables for 2S habitat fragments and Ihree mainland plols. The meaning of the variablesis explained in Mrrhods: S~cirs and Biogrographic variablrs.

RO- AREA CIJ SHRUB DISTX DISTY Ale Trap Cap-DENTS (ha) SHRUB (ha) (m) (m) (yr) nilhts turesCanyon Trapping dates

Florida 4 2S 66 l6.~ 2100 2100 so .57.5 7

Sandmark 84. 90 7'.6 914 914 20 616 1606

34th Street 3 .53.8 7' 40.3 1676 853 34 677 so

Balboa Terrace 5 51.8 75 38.8 243 121 34 498 102

2.5.6 60 .'.3 822 121 499 93Kate Sessions 4 16

Alia La Jolla 32 50 1216 21.9 121 .. 583 119

0 9.7 ,Laurel 0.49 1'S4 1554 79 293 0

Canon 0 8.7 20 .7 1219 1219 58 248 0

ana 0 8.5 30 2.6 286.5 286.5 36 399 0

Baja 0 ,S2 6708.4 4.4 670 31 368 0

0 8 IS .2 365 365 74 260 0Washington

Solana Drive 7.6 90 6.9 .150 75 ,570 294 II

Syracuse 3 7.' 8' 40 18 314 286.4 40

32nd Street South 0 15 304 304 56 300 06.4

60th 0 3'6.1 2.1 2386 33' 37 275 0

0 6 50 3 228 228 23 330 0Juan Street

Acuna 0 .5.1 30 1.5 662 22 224 0110

Edison 904 4.8 4.3 61 61 8 325 106

SpruceOak Crest

I2

4.33.9

10SO

0.431.9

17671(0)

1767400

866

175275

128

0 3.6 so .8 609 609 20 225 054th Street

Titus 0 3.5 7 0.25 33' 280 77 249 0

Montanosa 3 9.5 2 341 12.3 1.3 91 91

EI Mac 0 60 0.66 883 883 32 245 0

32nd Street North

Tec 1

0

3

0...I

23

100

0.095

I

487 4S 77 148

250

03S

Tec 2 4 0..5 100 0.$ 274 55

Del Mar'Mesa 6 0.125 100 0.125 200 21

Itay I~~ t.i"i"t.Li~ Vi" HAliiiA! i"KAU1Vll:.l'lAllU", U.

FIG

is a grassland species that was caught in three habitatfragments adjacent to grassland but not in coastal sageor chaparral. Because of its different habitat require-ments and ability to live in disturbed and exotic grass-land. it is probable that M. californicus would responddifferently to habitat fragmentation than the chaparralspecialist species. For this reason it was excluded fromRODENTS. Its inclusion produces no qualitativechanges in the analyses described below. In additionto the native rodents, two introduced murid rodents,Rattus rattus and Mus musculus. were also trapped insome fragments.

One species conspicuously absent from all fragmentssurveyed was the deer mouse (Peromyscus manicula-Ius). Although this species is common in southern Cal-ifornia, it is more typically found in mesa-top habitatrather than on canyon slopes. We did catch a numberof P. maniculatus in one of our control plots, whichwas adjacent to a large area of mesa-top habitat. Be-cause it was absent from all 25 fragments and two ofthe three mainland plots, we assume that P. manicu-latus never occurred in these fragments (rather thanassuming it went extinct in all the fragments).

traps. Different areas within each site were trapped onsuccessive trapping nights to insure that all areas ofeach fragment were trapped and to avoid recapturingthe same individuals. The number of trap nights perfragment ranged from 148 to 677, with a mean of 363trap nights per fragment and a total of 9072 trap nights.Another 724 traps were set in the three mainland plots,each of which was trapped on 2-3 nights. Where pos-sible, traps were set in roughly linear transects fromthe canyon bottom up the sides of the canyon. Trapswere placed -2 m apart, and transect lengths differeddepending on site topography and the spatial distri-bution of shrub habitat. In all but the smallest frag-ments, parallel traplines were a minimum of 100mapart. Traps were baited with bird seed and set beforedusk in the evening and retrieved after dawn the fol-lowing morning. Each animal trapped was identified tospecies and the animal was then released at the site of

capture.An additional 1780 trap nights were set in 12 of the

fragments during the period 26-29 November 1992.The additional trapping in 1992 had two interdependentobjectives: to assess the completeness of our earliersampling, and to investigate the extent of recoloniza-tion of fragments. If many new occurrences were re-corded in 1992 it would indicate that either these spe-cies were present in 1987 and not trapped, or that they

Census techniques

During the period 5 October 1986 to 27 May 1987each canyon fragment was trapped from two to fivetimes depending on its area. using small Sherman live

DOUGLAS T. BOLGER ET AL. Ecologio:al ApplicationsVol. 7. No.2

-R0rtd8Sendm8rk34th SIre.,Baa.,a Twr-Kat. SaNion.801- Drive

These plots should plateau once all the species in afragment have been recorded. Eleven of the 12 frag-ments from which native rodents were trapped reacheda stable species number that did not change after oneto several hundred additional trap nights. In the 12thfragment. Florida. a new species. Neoloma fusc;pes.was recorded in 1992. One additional night of trapping.60 trap nights. did not reveal any additional species.We did not continue trapping this fragment; conse-quently its plateau is shorter than those of the otherfragments.

--+--+--.--.-

6

4

2

00 200 400 600 800

6c~80~

"i0~jE~

u

Biogeographic variables

The values of the biogeographic variables (Table 1)are the same as those used in the earlier bird study(Soule et al. 1988). with the exception of those forFlorida Canyon. Only a portion of this fragment. de-limited from the rest of the canyon by surface streets.was trapped for rodents. so the area values for thiscanyon are correspondingly smaller than those in Souleet aI. (1988).

The age of a canyon (AGE) is the time elapsed sincethe fragment was isolated by development. Ages weredetermined from records of the City of San DiegoBuilding Department and corroborated when possiblewith dated aerial photographs. The total area of eachcanyon (AREA) was measured from aerial'photo con-tour maps (orthomaps) using an Apple computer dig-itizing tablet. The percentage shrub cover in each can-yon (%SHRUB) was estimated by a visual on-site in-spection or from recent aerial photos when available.Since this is our most subjectively measured variable.it is worth noting that it was estimated in 1985 alongwith all the other biogeographic variables. before thetrapping commenced. and so was not subject to bias.The area of each canyon still retaining native shrubvegetation (SHRUB) was calculated by multiplyingAREA by %SHRUB. Two isolation measures were alsomeasured on orthomaps. DISTX is the distance to thenearest large (>100 ha) "source canyon." DISTY isthe distance to the nearest canyon fragment of equalor greater size.

4

2

00 200 400 600 800

AnalysisStatistical analyses were performed with BMDP

(BMDP Statistical Software. Incorporated. Los Ange.les. California) and Statview (BrainPower. Incorporat.ed. Calabasas. California) statistical software. In orderto achieve normality. most variables were log-trans-formed prior to statistical analyses. The variable%SHRUB was normal in the untransformed state andso was not transformed. The independent variable RO.DENTS was not normally distributed (Kolmogorov.Smirnov. D = 0.314. P < 0.01) and natural log trans.formation did not normalize the distribution. The dis-tribution is non-normal because 13 fragments did notsupport native rodents and so have zero values for RO-DENTS (see Results. Species-area relationship).

FIG. 2. Plots of cumulative species counts vs. cumulativenumber of trap nights for each fragment in which native ro-dents were caught. The dashed ponions of the lines representthe 1992 trapping.

were missing in 1987 and recolonized some time in theintervening years.

Because an effort was made to trap all areas of eachcanyon. larger canyons were trapped more heavily onan absolute basis. but less heavily per unit area. thanwere smaller canyons. In Fig. 2, the cumulative num-bers of species recorded from each fragment are plottedas a function of the cumulative numbers of traps set.

Two statistical techniques were used to analyze thenon linear relationships between species number andthe biogeographic variables. age. area. and isolation.A stepwise logistic regression (BMDP) was performedwith the dependent variable being the number of spe-cies of the total pool of six species that were presentand absent in each fragment. Stepwise polychotomouslogistic regression (BMDP) was also used. Polycho-tomous logistic regression is similar to the more com-monly used logistic regression, with the difference thatthe dependent variable may assume more than two dif-ferent states. The underlying mode.) in either case islogistic and so does not assume a linear relationshipbetween dependent and independent variables. Inde-pendent variables may be continuous. nominal. or or-dinal. The dependent variable. in the logistic analysis.was nominal. the presence or absence of a particularspecies in a particular fragment. In the polychotomouslogistic analysis the dependent variable used was or-dinal, the number of species present in a fragment.ranging from zero to six. The results of the two analysesdid not differ qualitatively. so only the logistic regres-sion results are presented.

(Table 4). The area variable entered was log SHRUB(the area of shrub vegetation within a canyon fragment)which was a better predictor of the number of nativerodents in these sites than was total fragment area(AREA). This suggests that the native species do notextensively use the exotic herbaceous vegetation thataccounts for the nonshrub area in each fragment (Al-bens et al. 1993). The second and last variable addedto the regression equation was log AGE. The sign ofthe coefficient was negative. indicating that older frag-ments suppon fewer species than younger fragments.If local extinctions have occurred following fragmen-tation. then older fragments should suppon fewer spe-cies than younger fragments. After controlling for theeffect of habitat area. there is a significant negativerelationship between species number and fragment age.This is correlational evidence for local extinctions fol-lowing fragmentation.

Comparison to mainland plots

Fig. 3 shows the number of rodent species trappedin the three mainland plots superimposed on the com-parable data for the canyon fragments. Surprisingly,there is a suggestion of a negative relationship betweenarea and species number among the three plots. Thisis almost certainly a product of the small number ofplots sampled, and we expect that surveys from a largernumber of plots would reveal the expected positivespecies-area relationship. Nevertheless, the mainlandplots support more species of native rodents than mostcanyon fragments with a similar area of shrub vege-tation, strongly suggesting that extinctions have oc-curred following insularization.

RESULTS

Correlation between variables

Table 3 shows the simple product-moment correla-tions between the variables. Inspection of the table re-veals many significant correlations. The pattern of cor-relation between the dependent and independent vari-ables can be summarized as follows: high native rodentspecies number is associated with young fragmentshaving large areas of native shrub vegetation with highpercent cover, which are not highly isolated, and whichdo not support R. rattus. The two isolation measures,DISTX and DISTY. are correlated positively with AGE;in other words, older fragments are more isolated. Thisis because older neighborhoods (which contain olderfragments) have fewer remaining fragments; conse-quently they are farther apart. Because of this func-tional correlation between the age and isolation of frag-ments, disentangling their effects on species numberwith regression methods may not be possible.

Species-iJrea relationship

Of the 2S fragments surveyed, 13 did not supportpopulations of native rodents (Table 2, Fig. 3). Frag-ments without native rodents were in general smallerthan fragments which supported rodent populations.

Habitat degradation

Scrub vegetation is easily damaged by physical dis-turbance such as trampling. and recovers slowly. Con-sequently. the vegetation in older fragments should bein poorer condition than that in younger fragments.Percent cover of shrub habitat decreases as a functionof fragment age (Fig. 4). presumably due to humandisturbance. This loss of habitat area should result inloss of species. but there is also a negative effect ofage on species number above and beyond that causedby the loss of habitat area. as demonstrated by thesignificant partial regression of species on age.

Species vulnerability

Do species-specific attributes make some rodent spe-cies more vulnerable to extinction than others? For birdspecies. we found a significant positive correlation be-tween density and the number of fragments in whicha species persists (Soule et al. 1988. Bolger et al. 1991).Here we perform a similar analysis on rodents. As anindex of density in undisturbed habitat. we used thetotal number of captures of each species that were madein the three mainland plots. This will be a non biasedindex of relative abundance provided species are equal-

Stepwise logistic regression analysis

Stepwise logistic regression was used to determinethe predictive power of the biogeographic factors. age.area and isolation on species number. Log AREA. logSHRUB. log AGE. and log DISTX. log DISTY. werethe independent variables in the analysis. As expected.area is the primary determinant of species diversity

SS8 DOUGLAS T. BOLGER ET AL. EcOloall:aI ApplicMlonsVol. 7. No.2

TABLE 2. Number of individuals of seven native and two introduced rodent species trapped in 2S habitat fragments andthree mainland control plots. In the Del Mar Mesa plot P~romys,'us maniculalus and P~r()gnalhu.f californicus were alsocaught.

35736536348000001*

390000

78

2t

16

9

30

16

8

0

0

0

0

0

16

12

0

0

0

0

14

0

0

0

0

8

0

0

II

9

I

08000

120000000000000000000

000

020

1802000007t00000400002t002

03030200000000000000000

-000

13I650It

21t350000000000000It0000000

I92t~

II1200000~4t0000

1000002t00

010

442

II16.29343

II40200

000

003003200

2700000

222910

300,000

00000 0

FloridaSand mark34th StreetBalboa TerraceKate SessionsAlta La JollaLaurelCanonZenaBajaWashingtonSolana DriveSyracuse32nd Street South60thJuanAcunaEdisonSpruceOak CrestS4th StreetTitusMontanosaEI Mac32nd Street North

Tec ITec 2Del Mar Mesa

t Denotes a species trapped in 1986-1978 but not in 1992.* DeDotes a species trapped in 1992 but not in 1986-1987,

ly trappable. These values were plotted against thenumber of fragments in which each species occurred(Fig. S). There is a significant nonparametric correla-tion between occurrence and abundance (Spearmanrank. n = 6. r-corrected = 0.94. P = 0.02). suggestingthat the more abundant species are more persistent.

The way that abundance affects occurrence in frag-ments of different size is demonstrated with incidencefunctions (Fig. 6. sensu Diamond 1975). Incidenceamong the three most abundant species is low in thesmaller fragments. but increases steadily with areaclass until these three species occur in virtually all ofthe fragments in the largest area class (Fig. 6a). Overallthe less abundant species achieve a lower proportionof incidence (Fig. 6b). but the greatest divergence be-tween the incidence functions of the three rarer speciesand those of the more abundant species is in the twolarger area classes where the rarer species generallyachieve a much lower incidence. These incidence func-tions support our conclusions about relative abundanceand extinction vulnerability because they demonstratethat even in the largest fragments (those which are mostlikely to have initially contained all the species and inwhich more traps were set) the more abundant speciesappear to be more persistent.

DISCUSSION

Evidence of extinctions

The strongest evidence for frequent extinctions inthese fragments comes from the comparison of speciesrichness in fragments to that in plots of similar sizewithin continuous habitat (Fig. 3). In a time span ofonly 20-80 yr. it appears that all native rodents havedisappeared in 13 of the 2S canyon fragments. Thatextinctions have occurred is perhaps not surprisingwhen the size of these populations is considered. Sto-chastic demography predicts that vulnerability to ex-tinction should be inversely related to population size(MacArthur and Wilson 1967. Richter-Dyn and Goel1972. Leigh 1981. Goodman 1987) and many of thepopulations in the smaller fragments (0-10 ha) wereprobably in the range of 10-100 individuals at the out-set. Also. habitat area (and thus total population size)decreases as canyons age. The number of missing spe-cies in fragments and the lack of a negative partialregression between isolation and species number sug-gests that recolonization or rescue (Brown and Kodric-Brown 1977) of populations is not frequent enough tomaintain rodent populations in most of these fragments.

TABLE 3. Simple product-moment correlations among the dependent and independent variables. Below the diagonal thecorrelations are between the variables as labeled. The line above the diagonal contains the correlations between the naturallog transform of RODENTS and the independent variables as labeled.

RODENTSIn AREAIn SHRUB%SHRUBIn AGEIn DISTXIn DISTYRA7TUSMUS

-0.315 -0.304 -0.'1.0.877***0.249

-0.0060.0490.159

-0.2790.42*

0.654----0.339--0.089-0.028-0.218

0.324

-o.~-0.1-O.~-0..1

0

0.468*0.458*

-0.069

0.335

0.776***0.216

-0.063

I0.253

-0.04I

-0.356

* P = 0.05. .. P = 0.01. ... P = 0.001

taintops. In contrast. previous studies of rodent pop-ulation dynamics in fragmented landscapes have doc-umented dynamic patterns of local extinction and re-colonization rather that faunal collapse. Many of thesestudies were conducted in forest patches in agriculturallandscapes. and a number have documented dispersalbetween patches across agricultural fields or alongfencerows (Middleton and Merriam 1981. Hendersonet al. 1985. Merriam and Lanoue 1990. van Apeldoornet al. 1992). In the study most similar to ours Dickmanand Doncaster (1987. 1989) documented the distribu-tion of four small mammal species in a heterogenoussample of habitat and garden patches in Oxford City.England. Unlike ours. their system displayed extinc-tion-recolonization dynamics. They demonstrated re-colonization of patches from which they had experi-mentally removed all individuals of two focal rodentspecies. In contrast to these studies. the rodent speciesin our system appear incapable of frequent recoloni-zation across the urban landscape matrix in San DiegoCounty. Species capable of crossing the modified hab-itat matrix can potentially persist in fragmented land-scapes. while species that cannot. such as those westudied. are unlikely to persist. For example. Laurence(1994) found that rodents and other small mammalsthat were less tolerant of human-modified habitat de-clined or disappeared in Australian rain forest frag-ments. while those that were found in the modifiedmatrix remained stable or increased.

Meaning of new records in 1992

Two new species occurrences were recorded in the1992 sampling. One species, Peromyscus califomicusin the Solana fragment. was represented by one indi-vidual in 295 trap nights in 1992. The second newoccurrence. Neotoma fuscipes in Florida canyon. wastrapped twice in 210 trap nights in 1992. These resultssuggest that either these populations were at densitieslow enough to avoid detection in 1986-1987 and arestill at low densities. or these species were absent fromthose fragments in 1986-1987. but have recolonizedsince without achieving high population density. Therewere 46 populations missing from these fragments in1986-1987 (of 150 possible populations. six speciestimes 25 fragments) and only two of those 46 werefound to be present in 1992. so the great majority ofthose missing remain absent. Therefore. the conclusionthat recolonization is infrequent is supported. In ad-dition. there were II species that were recorded in1986-1987 that were not recorded in 1992 (see Table2). Because the sampling effort was not as extensivein 1992 as in 1986-1987, it is not warranted to concludethat these represent additional extinctions.

Our evidence suggests that the lack of immigrationand recolonization leads to a faunal collapse similar toBrown's (1971) study of mammals on isolated moun-

-+f!?zw

~

Causes of extinction

Habitat fragmentation can affect a broad spectrumof ecological processes (Robinson et al. 1992). Deter-mining which of those potential changes causes theextinction of native rodents in these fragments is notpossible from our data. Habitat attrition and degrada-tion which result from trampling of vegetation. trailformation. intentional clearing of vegetation for fire-breaks. increased fire frequency. and the introductionof invasive plant species may reduce rodent fitness.Effects of fragmentation on rodent predators may alsobe involved. Housecats (Felis cattus). coyotes (Canislatrans). striped skunks (Mephitis mephitis). raccoons(Procyon lotor). opossums (Didelphis v;rg;niana).

FIG. 3. The log-log species area relationship for the sixspecies of native rodents in 25 habitat fragments. The opensymbols are species numbers for fragments: the closed sym-bols are species numbers for the three mainland control plots.

r34***199*169178

500 -' &ologll:OII Applil:atlOflS

Vol 7. No.2TABLE 4. Results of stepwise logistic regression analysis of native rodent species number in each fragment. The dependent

variable is the proportion of species present in the fragment of the total pool of six species. Independent variables includedlog(DISTX). log(DISTY). log(AGE). log(AREAI. and log(SHRUB). Stepping stopped after two variables. log(SHRUB)and log(AGE). were added to the model.

DOUGLAS T. BOLGER ET At

Great-horned Owls (Bubo virginianus), and Red-tailedHawks (Buteo jamaicensis) are potential rodent pred-ators that appear to be as abundant or more abundantin fragments than in unfragmented habitat. In partic-ular, subsidized predators such as housecats have beenshown to be potent predators on native small mammals(Hubbs 1951, Fitzgerald and Karl 1979, Liberg 1984),with the ability to reduce population sizes of rodents(Davis 1957, Erlinge et al. 1983). Predation rates mayincrease due to reduced shrub cover. We do not knowif interactions with human commensal rodents, Rattusrattus or Mus musculus, are important (King 1957, Lid-icker 1966, Blaustein 1980), but the significant nega-tive correlation of Rattus with native rodent diversityis suggestive (Table 3).

It is worth considering in more detail the potentialrole of fire in this system. Fire is an important ecolog-ical and evolutionary influence on mediterranean cli-mate floras such as the chaparral and coastal sage scrubplant assemblages. Many chaparral and coastal sage

~ shrubs appear to have life history characteristics that

are adaptations to periodic fire. Frequent or intense firesare thought to change plant community compositionfrom chaparral and coastal sage scrub communities tograssland (Zedler et al. 1983). As we have shown, theshrub cover in our sites declines through time, whilegrass and herbaceous cover increases. This may be part-ly due to the influence of small human-caused fires.Fire records do not exist for our sites, however, so wecannot evaluate the role of fire in these vegetativechanges. It would seem reasonable to expect fires tobe more frequent in fragments due to the proximity to

ignition sources. However. it seems likely that fireswould be detected and extinguished rapidly in thesedeveloped areas, thus limiting the size of the fires. Im-mediate effects of fire on rodent density are not severeand recovery is rapid (Cook 1959, Price and Waser1984). Nevertheless, changes in vegetation due to firewould be expected to influence the density of indi vidualrodent species depending on the habitat requirementsof the species and the particular fire-induced change invegetation. In terms of vegetation structure, fire mightbe expected to favor species that prefer open habitat,such as Perognarhus and Reirhrodonromys (Meserve1976a, Price and Waser 1984), and to disfavor speciesthat prefer higher shrub cover such as Peromyscus cal-ifornicus and Neoroma fuscipes (Meserve 1976a). Ingeneral the effect of fire on vertebrate communities isnot well described or understood in these habitats. Thiswill present a problem for managing the effects of firein this system.

The apparent infrequency of recolonization of frag-ments indicates that rodent populations in fragmentsare isolates. This isolation creates the conditions underwhich a local extinction can occur. The decline in den-sity that leads to extinction may be a combination ofstochastic variation and increased mortality and de-

InC~

~~'0

.8E~

z

0 20 40 60

Number of captures

FIG. 5. Plot of occurrence vs. relative abundance in un-fragmented habitat. The vertical axis is the number of frag-ments, out of the total of 25. in which each species wascaptured. The horizontal axis is the number of captures ofeach species recorded in the three mainland control plots (724total trap nights). NL = Neotoma lepida. NF = Neotomafuscipes. PE = Peromyscus eremicus. PC = Peromyscus cal-ifomicus. RM = Reithrodontom.vs megalotis, and PF = Pe-

rognathus fallax.

BO

0 20 40 60 80 100

AGE (yr)

FIG. 4. Percentage shrub cover (%SHRUB) as a functionof fragment age.

~

g~"0"u.='0

5"of0

~Q..

FIG. 6. Incidence functions of (a) the three most abundantnative rodent species. (b) the three less abundant species. and(c) the two introduced rodent species. Incidence is the pro-ponion of fragments within a panicular area class in whichthe species occurred. Fragment area class intervals were cho-sen to equalize the number of fragments in each of the fourintervals (n = 6. 7. 6. and 6. respectively).

occupancy with persistence by assuming that speciesthat occur in fewer fragments do so because their pop-ulations have gone extinct, and not because they wereinitially absent from the fragment. In other words, wehave assumed that every species was in every fragmentat the time the fragments were created. Because of thesampling effect of fragment area, this assumption willnot be completely met (Bolger et al. 1991); the smallerthe area included in a fragment, the more likely it isthat species will, by chance, be initially absent fromthe fragment. This is a source of possible bias becausethe rarer species are the species most likely to be absentfrom a fragment at the time it is created, and so theirextinction vulnerability may be exaggerated. However,several facts indicate the assumption is approximatelymet. The number of species recorded from the mainlandplots reveals that very small plots contain most of thespecies pool. Consequently, we would expect all butthe smallest fragments to have contained populationsof most or all native rodents at the time they werecreated. Also. for the most part this analysis deals onlywith the larger fragments, because only those fragmentswith> 1 ha of shrub habitat support any of these species(with the exception of Spruce canyon). The averagearea of shrub habitat in these fragments is 19 ha, sub-stantially larger than our mainland plots and conse-quently more likely to have initially contained the en-tire species group. Furthermore, when the fragmentswere isolated, the amount of shrub habitat in them wasoften much greater than it is presently (Fig. 4), and soit is more likely they initially contained the entire spe-cies pool than the present area of habitat in the fragmentwould indicate.

Despite the limitations of our analyses. it appearsthat the native rodent species that are initially moreabundant, and which thus have larger initial populationsizes in fragments, are more resistant to local extinctionthan less-abundant species. This agrees with the pre-dictions of stochastic demography (Richter-Dyn andGoe11972, Leigh 1981. Goodman 1987), and with pre-vious empirical analyses of extinction vulnerability inbirds (Terborgh and Winter 1980, Diamond 1 984b,Soule et al. 1988, Bolger et al. 1991).

There is clearly an asymptote in the relationship be-tween extinction vulnerability and relative abundanceof native rodents (Fig. 5). N. fuscipes and R. megalotis.which have roughly the same relative abundance, occurin 9 and 10 fragments, respectively. However. Pero-myscus californicus, which was much more abundantin mainland sites, also occurs in 10 fragments (Fig. 5).This asymptote may arise because only 12 of the 25fragments are capable of supporting any native rodentspecies. We hypothesize that the remaining 13 frag-ments are so small, and/or old, disturbed. and isolatedthat they will not support populations of native rodentsregardless of their density in undisturbed habitat. Con-sequently, despite its higher density, P. californicus

creased fecundity due to edge effects. Although we donot have data bearing on the factors causing extinc-tions. we hypothesize that isolation combined with thereductions in density attributable to edge effects leadsto local extinction in a positive-feedback fashion (Gil-pin and Soule 1986).

Relative extinction vulnerability of native rodents

Our analysis of relative extinction vulnerability hasseveral important caveats. In the analysis we equated

DOUGLAS T. BOLGER ET AL &:ologio:aJ APflIK:ation~Vol. 7, No. ~

562

cannot achieve a higher occupancy than the other spe-cies.

relationship holds for other taxa and habitats. and atother spatial scales. then density provides an easilymeasured indicator of extinction vulnerability. Such anindicator would be useful in situations where detaileddemographic data are unavailable.

ACKNOWLEDGMENTS

We would like to thank Lee McClenaughan for advice. loanof traps. and assistance in rodent identification. Ted Case.Mark McPeek. Trevor Price. and Adam Richman providedvaluable criticism of the manuscript. We thank Victoria Pen-nick. David Zippin. Lela Carney. Gerard Zegers. Paul Heady.and Kathy Heady for field assistance; and the San DiegoCounty Advisory Commission for Fish and Wildlife for fund-ing.

Implications for urban conservation

We have demonstrated that the native rodent faunais vulnerable to collapse in representative habitat frag-ments in San Diego County. Apparently many popu-lations in these fragments are not viable on the spatialand temporal scales considered in this study. Theseresults make several significant points in regard to thepreservation of native rodent populations in urban/sub-urban landscapes. The first is the relatively large min-imum area requirement for these small-bodied rodentpopulations. Canyon fragments under 2S ba that havebeen isolated for at least 30 yr support very few pop-ulations of native rodents. In general. larger and youn-ger fragments support more species. but our data cannotprovide definitive guidelines that would insure againstthe loss of native rodents. Complicating the predictionof persistence is the problem of disentangling the ef-fects of area. isolation. age. and disturbance on speciesdiversity. It is safe to say that fragments in the 25-80ha range sustain native rodent populations better thansmaller fragments. at least over the time period con-sidered in this study. To support populations for inter-vals >80 yr. it is likely that larger areas would berequired.

The second point is that isolation exacerbates vul-nerability. Dispersal. colonization. and rescue of pop-ulations is apparently too infrequent to maintain rodentpopulations in more than balf of the fragments sur-veyed. despite the relatively short distances betweenfragments. Apparently the urban matrix is relativelyimpervious to native rodents. To link populations inseparate fragments and achieve the oft-hypothesizedbenefits of gene ftow. supplementation of depleted pop-ulations. and recolonization following local extinc-tions. habitat corridors will be necessary. It seems like-ly that relatively small patches could still support pop-ulations of rodents if they were connected by move-ment corridors. However. even in the presence ofcorridors. populations in small fragments will be vul-nerable to edge effects.

Coastal sage scrub and chaparral vegetation are verysusceptible to degradation through trampling. clearing.and burning. A consequence of this is that shrub coverdeclines sharply with fragment age (Fig. 4). More ro-bust vegetation types. such as forest. appear to be moreresistant to physical disturbance. Management of smallparcels of coastal sage scrub and chaparral for the pur-pose of conservation must include protection againstthis type of degradation through prevention. monitor-ing. and restoration. Corridors. too. would have to beprotected from disturbance.

Finally. our results for both rodents and birds areconsistent with the view that density or relative abun-dance is a reliable metric for ranking relative extinctionvulnerability within a group of similar species. H this

LrTERA TURE CITED

Alberts. A. C.. A. D. Richman. D. Tran. R. Sauvajot. C.McCalvin. and D. T. Bolger. 1993. Effects of habitat frag-mentation on populations of native and exotic plants insouthern California coastal scrub. Pages 103-110 in J. E.Keely. editor. Interface between ecology and land devel-opment in California. Southern California Academy of Sci-ences. Natural History Museum of Los Angeles County.Los Angeles. California. USA.

Beauchamp. R. M. 1986. A ftora of San Diego County. Cal-ifornia. Sweetwater River Press. National City. California.USA.

Blaustein. A. R. 1980. Behavioral aspects of competition ina three-species rodent guild of coastal southern California.Behavioral Ecology and Sociobiology 6:247-2SS.

Bolger. D. T.. A. C. Alberts. and M. E. Souli. 1991. Oc-currence patterns of bird species in habitat fragments: sam-pling. extinction. and nested species subsets. AmericanNaturalist 137:ISS-I66.

Brown. J. H. 1971. Mammals on mountaintops: non-equi-librium insular biogeography. American Naturalist 105:467-478.

Brown. J. H.. and A. Kodric-Brown. 1977. Turnover ratesin insular biogeography: effect of immigration on extinc-tion. Ecology 58:44S-449.

Burgess. R. L.. and D. M. Sharpe. editors. 1981. Forest islanddynamics in man-dominated landscapes. Springer-Verlag.New York. New York. USA.

Case. T. J. 1975. Species numbers. density compensation andcolonizing ability of lizards on islands in the Gulf of Cal-ifornia. Ecology 56:3-18.

Cook. S. F.. Jr.. 19S9. The effects of lire on a population ofsmall rodents. Ecology 40:102-108.

Davis. D. E. 1957. The use of food as a buffer in a predator-prey system. Journal of Mammalogy 38:466-472.

Diamond. J. M. 1972. Biogeographic kinetics: estimation ofrelaxation times for avifaunas of southwest Pacific islands.Proceedings of the National Academy of Sciences (USA)":199-203.

-. 1975. Assembly of species communities. Pages 342-444 in M. L. Cody and J. M. Diamond. editors. Ecologyand evolution of communities. Harvard University Pre!\s.Cambridge. MA. USA.

-. I 984a. "Normal" extinctions of isolated popula-tions. Pages 191-246 in M. H. Nitecki. editor. Extinctions.University of Chicago Press. Chicago. Illinois. USA.

-. I 984b. Historic extinctions: a Rosetta Stone for un-derstanding prehistoric extinctions. Pages 824-862 in P. S.Martin and R. G. Kline. editors. Quaternary extinctions. aprehistoric revolution. University of Arizona Press. Tucson.Arizona. USA.

Dickman. C. R.. and C. P. Doncaster. 1987. The ecology ofsmall mammals in urban habitats. I. Populations in a patchyenvironment. Journal of AnimaJ Ecology 56:629-640.

May l\)\)~ EFFECTS OF HABITAT FRAGMENTATiON ON ROOl:.NlS

Middleton. J., and G. Merriam. 1981. Woodland mice in afarmland mosaic. Journal of Applied Ecology 18:703-710.

Munz, P. A., and D. D. Keck. 1959. A California flora. Uni-versity of California Press. Berkeley. California. USA.

Price, M. V., and N. M. Waser. 1984. On the relative abun-dance of species: postfire changes in a coastal sage scrubrodent community. Ecology 65: 1161-1169.

Richter-Dyn, N., and N. S. Goel. 1972. On the extinction ofa colonizing species. Theoretical Population Biology 3:406-433.

Robinson. G. R., R. D. Holt, M. S. Gaines. S. P. Hamburg,M. L. Johnson, H. S. Fitch. and E. A. Martinko. 1992.Diverse and contrasting effects of habitat fragmentation.Science 257:524-526.

Soule, M. E.. and A. J. Sloan. 1966. Biogeography and dis-tribution of the reptiles and amphibians on islands in theGulf of California. Mexico. Transactions of the San DiegoSociety of Natural History 14: 137-156.

Soule, M. E.. B. A. Wilcox. and C. Holtby. 1979. Benignneglect: a model of faunal collapse in the game reservesof East Africa. Biological Conservation 15:206-272.

Soule. M. E.. D. T. Bolger. A. C. Alberts. R. Sauvajot. J.Wright, M. Sorice, and S. Hill. 1988. Reconstructed dy-namics of rapid extinction of chaparral requiring birds inurban habitat islands. Conservation Biology 2:75-92.

Terborgh. J. 1975. Faunal equilibria and the design of wild-life preserves. Pages 369-380 in F. Golley and E. Medina.editors. Tropical ecological systems: trends in terrestrialand aquatic research. Springer-Verlag, New York. NewYork. USA.

Terborgh, J., and B. G. Winter. 1980. Some causes of ex-tinction. Pages 119-131 in M. E. Soule and B. A. Wilcox.editors. Conservation Biology: an evolutionary-ecologicalperspective. Sinauer Associates, Sunderland, Massachu-setts. USA.

van Apeldoorn. R. C., W. T. Oostenbrink. A. van Winden.and F. F. van der Zee. 1992. Effects of habitat fragmen-tation on the bank vole. Cleithrionomys glareolus, in anagricultural landscape. Oikos 65:265-274.

Whitcomb, R. F., C. S. Robbins. J. F. Lynch. B. L. Whitcomb,M. K. Klimkiewicz, and D. Bystrak. 1981. Effects of forestfragmentation on avifauna of eastern deciduous forests.Pages 125-205 in R. L. Burges and D. M. Sharpe. editors.Forest island dynamics in man-dominated landscapes.Springer- Verlag: New York. New York, USA.

Wilcove. D. S.. C. M. McLellan, and A. P. Dobson. 1986.Habitat fragmentation in the temperate zone. Pages 237-256 in M. E. Soule. editor. Conservation biology: the sci-ence of scarcity and diversity. Sinauer Associates. Sun-derland, Massachusetts. USA.

Wilcox. B. A. 1980. Insular ecology and conservation. Pages95-117 in M. E. Soule and B. A. Wilcox. editors. Con-servation biology: an evolutionary-ecological perspective.Sinauer Associates, Sunderland. Massachusetts. USA.

Willis. E. O. 1974. Populations and local extinctions of birdson Barro Colorado island, Panama. Ecological Monographs44:153-169.

-. 1979. The composition of avian communities in re-manescent woodlots in southern Brazil. Papeis A vulsos Zo-ology 33:1-25.

Zedler. P. H.. C. R. Gautier. and G. S. McMaster. 1983. Veg-etation changes in response to extreme events: the effectof a short interval between fires in California chaparral andcoastal scrub. Ecology 64:809-818.

Dickman. C. R.. and C. P. Doncaster. 1989. The ecology ofsmall mammals in urban habitats. II. Demography and dis-persal. Journal of Animal Ecology 58:119-127.

Erlinge, S. 1983. Predation as a regulating factor on smallrodent populations in southern Sweden. Oikos 40:36-52.

Fitzgerald. B. M.. and B. J. Karl. 1979. Foods of feral housecats (Felis cattus L.) in forest of the Orongorongo Valley.Wellington. New Zealand Journal of Zoology 6:107-126.

Gilpin. M. E., and M. E. Soul~. 1986. Minimum viable pop-ulations: processes of species extinction. Pages 19-34 inM. E. Soul~. editor. Conservation biology: the science ofscarcity and diversity. Sinauer Associates, Sunderland.Massachusetts. USA.

Goodman. D. 1987. Consideration of stochastic demographyin the design and management of biological reserves. Nat-ural Resource Modeling 1:205-235.

Henderson. M. T.. G. Merriam. and J. Wegner. 1985. Patchyenvironments and species survival: chipmunks in an ag-ricultural mosaic. Biological Conservation 31:95-105.

Hubbs. E. L. 1951. Food habits of feral house cats in theSacramento Valley. California Fish and Game 37: 177-189.

Janzen. D. H. 1986. The eternal external threat. Pages 286-303 in M. E. Soul~. editor. Conservation biology: the sci-ence of scarcity and diversity. Sinauer Associates. Sun-derland. Massachusetts. USA.

King, J. A. 1957. Intra- and interspecific conflict of Mus andPeromyscus. Ecology 38:355-357.

Laurance. W. F. 1994. Rainforest fragmentation and thestructure of small mammal communities in tropical Queens-land. Biological Conservation 69:23-32.

Leigh. E. G.. Jr. 1981. The average lifetime of a populationin a varying environment. Journal of Theoretical Biology90:213-239.

Liberg. O. 1984. Food habits and prey impact by feral andhouse-based domestic cats in a rural area in southern Swe-den. Journal of Mammalogy 65:424-432.

Lidicker. W. Z. 1966. Ecological observations on a feralhouse mouse population declining to extinction. EcologicalMonographs 36:27-50.

Lovejoy. T. E., R. O. Bierregaard. A. B. Rylands, J. R. Mal-colm. C. E. Quintela. L. H. Harper, K. S. Brown. A. H.Powell. H. O. R. Schubart. and M. B. Hays. 1986. Edgeand other effects of fragmentation on Amazon forest frag-ments. Pages 257-285 in M. E. Soul~. editor. Conservationbiology: the science of scarcity and diversity. Sinauer As-sociates. Sunderland. Massachusetts. USA.

Lynch. J. F.. and D. F. Whigham. 1984. Effects of forestfragmentation on breeding bird communities in Maryland.USA. Biological Conservation 28:287-324.

MacAnhur. R. H., and E. O. Wilson. 1967. The theory ofisland biogeography. Princeton University Press, Princeton,New Jersey, USA.

M'Closkey. R. T. 1972. Temporal changes in populationsand species diversity in a California rodent community.Journal of Mammalogy 52:657-676.

Merriam. G., and A. Lanoue. 1990. Corridor use by smallmammals: field measurements for three experimental typesof Perom.vscus leucopus. Landscape Ecology 4: 123-131.

Meserve, P. L. 1972. Resource and habitat utilization byrodents of the coastal sage scrub community. Dissenation,University of California. Irvine. California. USA.

-. 19760. Habitat and resource utilization by rodentsof California coastal sage scrub community. Journal of An-imal Ecology 45:647-666.

-. I 976b. Food relationships of a rodent fauna in aCalifornia coastal sage scrub community. Journal of Mam-malogy 57:300-319.