OUP user menu

Effects of habitat disturbance on a Peromyscus leucopus (Rodentia: Cricetidae) population in western Pennsylvania

Alicia V. Linzey, Aaron W. Reed, Norman A. Slade, Michael H. Kesner
DOI: http://dx.doi.org/10.1644/11-MAMM-A-130.1 211-219 First published online: 16 February 2012

Abstract

Many species of wild mammals occur in habitats that have been disturbed by fragmentation or degraded in quality. Previous researchers have hypothesized that demographic characteristics of populations may shift with changes in environmental conditions, with self-regulatory ability increasing with environmental suitability. We studied responses of white-footed mice (Peromyscus leucopus) to habitat disturbance. Given that optimal habitat for this species is deciduous woodland, we predicted that populations in habitats disturbed by cutting woody vegetation would be lower and more variable in density than in undisturbed habitat, density and stability of populations in disturbed habitat would increase over time, survival would be higher in undisturbed than in disturbed habitat, and populations in undisturbed habitat would show a greater degree of self-regulation. This 6-year study in western Pennsylvania involved 3 replicated study sites (each 3.8 ha), with woody vegetation removed on half of each site prior to beginning the study. Density in disturbed treatment averaged 65% of density in undisturbed habitat. There were no differences between treatments in survival or in population growth rates over time. Population trends over time were similar between treatments, populations in disturbed habitat did not become more stable with time, and density did not converge with that of undisturbed habitat. Although populations in the undisturbed habitat were clearly self-regulating, those in disturbed habitats were not. Despite expectations that demographic performance will align with environmental suitability, it may be difficult to ascribe a particular demography to a habitat generalist such as P. leucopus.

Key words
  • demography
  • density
  • disturbance
  • Peromyscus leucopus
  • stability
  • suboptimal habitat
  • white-footed mouse

A landscape approach to mammalian population dynamics recognizes that spatial and temporal variation in biotic and abiotic factors can play a dominant role in determining demographic patterns (Lidicker 1995). Habitat disturbance results in a spatiotemporal mosaic of patches at different successional stages. Although small mammals have been the subject of many demographic studies, information on long-term dynamics of forest-dwelling species inhabiting early successional habitats is limited. Although some ecologists have found high immigration and low recruitment in such suboptimal habitats (i.e., source-sink dynamics—Adler et al. 1984; Pulliam 1988; Pulliam and Danielson 1991), others have found evidence of regulation (i.e., inverse relationship between density and reproductive effort) that could only develop if individuals are resident and interactive (Adler and Tamarin 1984; Linzey and Kesner 1991). This range of observations is not surprising, given that a landscape composed of a mosaic of habitats that differ in quality would invite similarly variable demographic patterns (Lidicker 1995). Adler and Wilson (1987) hypothesized that self-regulatory ability may increase with environmental suitability and that demographic structure may shift with changes in environmental conditions. For a species that prefers late-successional habitat, environmental suitability would be expected to improve with time in a habitat recovering from disturbance.

Optimal habitat for a given species can be defined as the habitat in which individuals can potentially achieve highest fitness (Morris and Davidson 2000). However, as density increases individuals may achieve higher fitness by dispersing to a suboptimal habitat that has lower density and therefore greater access to existing resources per individual (Fretwell and Lucas 1970; Holt 1985; Runge et al. 2004). Although the source-sink concept has been used to describe colonization of successively poorer habitats by dispersing individuals, a suboptimal habitat is not necessarily a sink habitat (defined as a population maintained solely by dispersal—Pulliam 1988). In fact, suboptimal habitats can support resident populations that are self-regulatory (Linzey and Kesner 1991).

Many species of wild mammals occur in habitats that have been disturbed by fragmentation or degraded in quality. Theoretical models that describe responses of populations to disturbance have included effects of singular events (i.e., pulsed perturbations), with patterns after disturbance including changes in density, and a period of population instability followed by a return to preperturbation patterns (Royama 1992). Although empirical studies that support these models have focused on population responses to periodic increases in resources (Elias et al. 2004; Krohne et al. 1988; Marcello et al. 2008; Ostfeld et al. 1999), a sudden decline in resources might have a similar impact. If the decline is due to habitat destruction, population resilience would be a function of the rate at which the habitat returns to suboptimal and then to optimal conditions. Insights from studies that follow populations in incrementally improving habitats, such as would occur during succession for species preferring forested habitat, are needed to understand population interactions across landscapes characterized by dynamically changing spatial and temporal patterns. Such studies also contribute empirical data that can be used to further evaluate theoretical models that describe long-term dynamics of populations subsequent to pulsed disturbances.

We studied responses of white-footed mice (Peromyscus leucopus) to habitat disturbance, specifically in regard to the effects of time since disturbance on population dynamics and on components of individual fitness. Although these mice are found in a wide variety of situations, their optimal habitat is mature woodlands with a shrubby understory (Baker 1968; Krohne 1989; Wolf and Batzli 2002). Marginally habitable patches were created by cutting and removing resulting woody debris; population parameters were evaluated as these disturbed areas underwent succession. Adjoining undisturbed patches served as controls. Because undisturbed patches were in habitat considered suboptimal for this species (Krohne 1989), an additional comparative study area was established in optimal woodland habitat. We predicted that populations in cut habitat would be smaller and more variable in density than in uncut habitat; density and stability of populations in cut habitat would increase over time; survival would be higher in uncut than in cut habitat; and populations in uncut habitat would show a greater degree of self-regulation (adjustment of reproductive effort with density). Although not part of the formal experimental design, we expected that the woodland habitat would have highest density, greatest stability, and highest survival. We also compared components of individual fitness, asking whether reproductive effort and body mass differ in disturbed habitat. Finally, we examined whether demographic parameters suggest that the disturbed habitat is acting as a habitat sink.

Materials and Methods

Study sites.—The study was conducted within Yellow Creek State Park, approximately 20 km southeast of Indiana, Indiana County, Pennsylvania (40°34′N, 79°02′W), from spring 1998 to autumn 2003. The experiment employed a blocked design involving three 3.8-ha sites (Lake, Myer, and Hunter) that, prior to experimental manipulation, had an abundant shrub growth (primarily Cornus amomum), scattered patches of small trees (Cornus florida, Malus domestica, Rhus typhina, and Acer rubrum), and single rows of large trees (Acer rubrum and Quercus sp.) that represent remnants of fencerows. These sites were last cleared in the mid-1970s. In September 1997, half of each site was disturbed by cutting woody vegetation to 15-cm height but leaving trees larger than 10.2-cm diameter at breast height. Between early October and late December 1997, all woody debris was gathered and burned on each site. A 4th site (Woods; 1.8 ha) was established in mature 2nd-growth forest dominated by Quercus, Carya ovata, and Acer rubrum.

Distances between experimental sites (in km) were approximately 0.6 (Hunter to Myer), 2.6 (Hunter to Lake), and 2.8 (Myer to Lake). The Woods site was 0.9 km from Lake, 2.5 from Myer, and 2.6 from Hunter. Movements of individual white-footed mice between study grids were infrequent (3 of 2,331 mice), indicating that sampled populations were independent. Trap stations within each site were located in a grid pattern at 15-m intervals. Experimental sites had 200 trap stations arranged as 10 rows of 20 traps (Lake and Myer) or 15 rows of 13 or 14 traps (Hunter). In both designs, 100 trap stations were located in the cut half and 100 in the uncut half. The Woods site had 100 trap stations, with 10 rows of 10 traps.

Estimates of shrub cover (percent cover in squares formed by 4 trap stations), as well as diameter at breast height of trees, were completed in 1999 and 2003. Shrub cover in the cut treatment increased from 7% (1999) to 50% (2003). Cover also increased in the uncut treatment, going from 37% (1999) to 60% (2003). Total diameter at breast height of trees increased in both treatments (7% in the cut treatment and 13% in the uncut treatment). Total diameter at breast height in the cut treatment was mostly due to large trees that were confined to specific areas on each grid; total diameter at breast height in the uncut treatment included trees with a greater range of sizes and distributions.

Population studies.—Population monitoring began in spring 1998 and continued through fall 2003. We conducted 36 trapping sessions, with 3 at each site in both spring (May–June) and fall (October–November). Each session involved 5 nights of trapping in 6 days (usually 3 successive nights, a 1-night break, and 2 final nights). The time between the beginnings of successive 5-night trap sessions averaged 18 days. During spring, traps were opened late in the afternoon, left open overnight, and checked beginning shortly after sunrise. In order to avoid trap mortality during fall, traps were opened late in the afternoon, and checked and closed beginning at 2300–2400 h.

Data collection methods employed mark-and-recapture livetrapping, with each grid station having 1 small Sherman live trap (5.0 × 6.5 × 16.5 cm; H. B. Sherman Traps, Inc., Tallahassee, Florida). Each captured white-footed mouse was permanently marked for individual recognition by toe clipping, sexed, aged by pelage (juveniles gray, subadults undergoing molt, and adults brown), examined for reproductive condition, and released at the point of capture. Marking techniques conformed to guidelines developed by the American Society of Mammalogists (Sikes et al. 2011). Project methods were approved by the Institutional Animal Care and Use Committee at Indiana University of Pennsylvania.

Statistical methods.—We estimated population size (reported as number of individuals/1.9 ha) using the numbers of individuals captured and capture probability estimates from a multistate model (Brownie et al. 1993) for the Lake, Hunter, and Myer sites (states = cut and uncut treatments) and a Cormack-Jolly-Seber (Pollock et al. 1990) model for the Woods site. We tested for normality of the residuals from final models with the Anderson-Darling test in Minitab 14 Statistical Software (Minitab, Inc. 2005). However, we were not concerned with departures from normality because visual inspection of dot plots indicated that residuals were symmetrically distributed and our large sample sizes ensured that sample means were normally distributed. We ran every possible combination of sex, treatment (cut or uncut), and time (month or season) for survival, capture probability, and movement between treatments in program MARK (White and Burnham 1999). We tested 3 data structures for time: where parameters were constant across time periods, varied among each time period, or varied seasonally (i.e., capture probability was the same for the 3 sampling periods in a season but differed among years and season). We then evaluated the effect of the treatment on survival and recapture probability by setting the parameters equal in a design matrix, and comparing the results to models with treatment-specific estimates. We tested goodness-of-fit for the full model with program UCARE (Pradel et al. 2003) and adjusted the overdispersion parameter (ĉ), if necessary, before model selection. If the goodness-of-fit test indicated overdispersion, we used an adjusted value of ĉ (χ2/d.f.) in model selection.

We used Akaike's information criterion (corrected for small sample size [AICC]) to select the most-parsimonious model. We considered any model with an AICC value > 2 more than the best competing model to have insufficient support. We used model averaging if multiple models received support (Buckland et al. 1997).

We estimated population size by dividing the number of individuals captured (ni) by the capture probability for the same period (p̂i). These estimates were separated by treatment and sex if the most-parsimonious model included these variables. We estimated variance using

Embedded Image

and the 95% confidence interval (95% CI) for each estimate (Williams et al. 2001).

Population growth rate (λ) was used as an indicator of population stability and was measured using Pradel's temporal symmetry method (Pradel 1996) to estimate the Woods populations and the multistate method (Nichols et al. 2000) for the remaining sites. We ran all combinations of treatments for the Pradel model in MARK. We ran multistate models with capture histories reversed for the remaining sites. We estimated λ by dividing the survival probability (Φt) from the “forward” multistate model by the seniority estimates (the probability that an individual was in the population at the previous time step) from the following trap period in the “reverse” model (γ+1). Model selection was completed as above. To test whether the differences between cut and uncut areas became smaller through time at each of the 3 sites with paired grids, we calculated differences in estimated numbers for each of the 36 individual trapping sessions (3 sessions per season). We then regressed these differences against time.

The estimated numbers of males and females from our MARK procedures were analyzed for differences between treatments (cut and uncut), among sites (Lake, Hunter, and Myer only), between sexes, and between seasons using Minitab 14 Statistical Software (Minitab, Inc. 2005). We initially tested for all main effects and for the interactions among treatments, sites, and sexes. In effect, this used season as a blocking factor. We then removed the highest-order interaction term that was not significant at the 0.10 level and repeated. The final analyses included any factor that was involved in a significant interaction or main effect. We also repeated these analyses using the inverses of variances of estimated numbers as weighting factors. Weighted analyses differed little from unweighted and were complicated by the need to omit some estimates with errors of zero or to make arbitrary choices for weights. Hence we present the results from the unweighted analyses.

To estimate the numbers by season for the 3 experimental grids, we ran analyses of variance (ANOVAs) on the estimates by session and used the least-squares mean numbers for 3 sessions per season. To estimate the variance of that mean, we summed the variances for each of the 3 estimates per season and divided by 9. We reasoned that the mean for a season was the sum of 3 estimates, divided by 3. Thus, assuming that errors in the 3 estimates of numbers were independent, the variance of that mean equals the sum of the variances multiplied by (1/3)2. We did this for each sex and then added both the mean estimates of numbers and the estimates of variance to get comparable figures for total numbers and their variances. Density differences between the Woods and experimental grids were evaluated using the Mann-Whitney Latest. Ancillary data (body mass, reproductive effort, and age and sex structure) were tested for normality and transformed if appropriate. Significance tests used ANOVAs and z-tests if raw or transformed data were normally distributed and Mann-Whitney U-tests if data were still not normally distributed after transformation.

We used phase-space plots to estimate whether populations were self-regulating. These plots portray the degree to which reproductive effort is adjusted in response to density (Schaffer and Tamarin 1973). The number of different animals (adults and subadults) in reproductive condition in spring and fall seasons was summed to derive a yearly value, which was expressed as the proportion of the total number of adult and subadult animals present. Females were considered in reproductive condition if they had enlarged nipples, a perforate vagina, or were pregnant. Males were considered in breeding condition if testes were enlarged or scrotal, or both.

Results

We captured a total of 2,330 different white-footed mice 8,440 times during the study (Hunter n = 517, Lake n = 474, Myer n = 767, and Woods n = 572). Of these different animals, 3 males moved between grids: 1 juvenile and 1 subadult moved from Hunter to Myer (0.6 km), and 1 subadult moved from Woods to Lake (0.9 km). Other small mammal species captured (in descending order of capture frequency) were Blarina brevicauda, Microtus pennsylvanicus, Zapus hudsonius, Tamias striatus, Sorex, Synaptomys cooperi, Glaucomys volans (Woods only), Mus musculus, Microtus pinetorum (Woods only), and Condylura cristata (Woods only).

A comparison of population density estimates (numbers/ 1.9-ha treatment) reveals, with few exceptions, a consistent pattern of higher density in the uncut than in the cut treatment (Table 1). The results of ANOVA confirm this observation, with a significantly higher density in the uncut than in the cut treatment (F1,128 = 32.64, P = 0.001; Fig. 1). There were no significant interactions among variables. Other significant differences include more males than females (54:46; F1,128 = 4.57, P = 0.034) and significantly higher overall density in fall than in spring (F11,128 = 31.10, P = 0.001). However, neither sex ratio nor seasonal density exhibited a significant interaction with treatment. Grids also differed significantly (F2,128= 8.71, P = 0.001), with the difference due to higher mean density on Myer (38.9 versus 28.2 on Hunter and 28.8 on Lake). Density in optimal habitat (Woods) was significantly higher than on the uncut sides of experimental sites (U24 = 121, P = 0.004; Fig. 1). Population trends were similar between treatments, and experimental grids were similar to woodland habitat. As is typical of white-footed mouse populations in this region, there was an annual cycle of density, with peaks in fall and troughs in spring. The only exception to this pattern was fall 1999 and spring 2000. Highest densities were attained in fall 1998 and fall 2000 and lowest in spring 1999 and spring 2003.

Fig. 1

Population density (N̂1.9-ha treatment ± SD) on cut, uncut, and Woods by season/year. Density in the uncut treatment was significantly higher than density in the cut treatment. Density in Woods was significantly higher than in the uncut treatment.

View this table:
Table 1

Estimates of population density (N̂/treatment ± SD) by grid, treatment, and season. Uncut treatment has significantly higher density than cut treatment. Woods has significantly higher density than experimental grids.

LakeMyerHunterWoods
CutUncutCutUncutCutUncut
Spring 19987.1 ± 0.93.6 ± 1.118.4 ± 1.522.2 ± 1.719.1 ± 2.312.6 ± 1.842.2 ± 3.7
Fall 199815.3 ± 2.635.9 ± 3.048.1 ± 3.2101.7 ± 5.249.4 ± 3.660.6 ± 4.173.3 ± 2.4
Spring 19994.7 ± 1.75.8 ± 1.37.0 ± 0.611.4 ± 0.81.8 ± 0.56.8 ± 1.318.6 ± 1.9
Fall 199914.6 ± 2.125.7 ± 2.122.0 ± 2.231.5 ± 2.711.2 ± 1.122.0 ± 1.744.2 ± 3.6
Spring 200019.8 ± 3.527.6 ± 2.331.5 ± 3.034.1 ± 3.29.9 ± 2.026.3 ± 3.667.9 ± 6.6
Fall 200047.0 ± 4.263.4 ± 3.741.2 ± 1.967.9 ± 2.731.5 ± 1.648.1 ± 2.268.5 ± 3.1
Spring 200117.2 ± 4.111.4 ± 0.89.3 ± 0.017.7 ± 0.03.3 ± 0.06.3 ± 0.032.6 ± 4.9
Fall 200117.2 ± 2.327.5 ± 2.212.1 ± 0.623.8 ± 1.011.6 ± 1.018.1 ± 1.354.8 ± 3.3
Spring 20028.0 ± 1.024.6 ± 3.18.3 ± 0.911.5 ± 1.217.6 ± 3.220.2 ± 3.148.5 ± 4.9
Fall 200214.5 ± 3.624.8 ± 2.412.8 ± 1.522.6 ± 2.28.4 ± 1.214.4 ± 1.732.8 ± 2.8
Spring 20031.5 ± 1.24.4 ± 1.44.7 ± 0.05.3 ± 0.02.0 ± 0.05.7 ± 0.017.8 ± 5.5
Fall 200318.5 ± 3.720.3 ± 1.121.3 ± 1.636.7 ± 2.518.2 ± 1.526.7 ± 1.842.2 ± 4.2

The most-parsimonious models from the multistate analysis indicated no difference in survival or recapture probability between cut and uncut treatments (Table 2). Goodness-of-fit tests from program UCARE indicated no overdispersion (all P > 0.24). Reverse capture histories also showed no difference between treatments for survival, therefore no difference in population growth rates (our estimate of population stability) over time between populations in cut and uncut treatments (Table 3).

View this table:
Table 2

Results of modeling survival and movements between treatments on experimental sites where w is model weight and K is number of parameters. Survival varied over time, but not between treatments. Movements between cut and uncut treatments did not differ. AICC = Akaike's information criterion (corrected for small sample size); ΔAICC = difference in AICC value between a model and the model with the most support.

ModelaAICcΔAICcwKDeviance
Hunter
Φ c=u (t); Pc=u(season); Ψc=u(.)1,628.900.6648687.0
Φ c=u (t); Pc=u(season); Ψc=u(.) Ψu=c(.)1,630.41.50.3149686.3
Φ c=u (t); Pc(.) Pu(.); Ψc=u(.)1,635.66.70.0260666.5
Lake
Φ c=u (t); Pc=u(season); Ψc=u(.)1,639.100.3760780.9
Φ c=u (t); Pc(.) Pu(.); Ψc=u(.)1,639.30.10.3538830.9
Φ c=u (t); Pc(.) Pu(.); Ψc=u(.) Ψu=c(.)1,640.91.70.1639830.3
Φ c=u (t); Pc=u(season); Ψc=u(.) Ψu=c(.)1,641.52.30.1261780.9
Myer
Φ c=u (t); Pc=u(season); Ψc=u(.)2,295.100.4448844.5
Φ c=u (t); Pc=u(season); Ψc=u(.); Ψu=c(.)2,295.70.630.3250840.8
Φ c=u (t); Pc(.) Pu(.); Ψc=u(.); Ψu=c(.)2,309.014.0062827.9
Woods
Φ(t); p(t)1,777.200.7869513.1
Φ(t); p(.)1,781.03.80.1236589.9
  • a Φ = survival; p = probability of capture; Ψ = movement between treatments; t = time (each trap period); season = seasonal group (1 estimate for fall or spring); = constant; c = cut; u = uncut; c=u indicates a single estimate for both the cut and uncut treatments; c-u indicates movement from cut to uncut treatments.

View this table:
Table 3

Model selection for Pradel models where w is model weight and K is number of parameters. There was no difference in seniority between treatments and therefore no difference in population growth rates over time between populations in cut and uncut treatments. AICC = Akaike's information criterion (corrected for small sample size); ΔAICC = difference in AICC value between a model and the model with the most support.

ModelaAICCΔAICcwKDeviance
Hunter
γc=u(t); Pc(.) Pu(.); Ψc-u(.) Ψu-c(.)1,685.40139725.3
γc=u(t); Pc(season) Pu(season); Ψc-u(.) Ψu-c(.)1,709.123.7061699.8
Lake
γc=u(t); Pc(.)Pu(g);Ψc-u(.)Ψu-c(.)1,752.60140817.6
γc=u(t); Pc(.)Pu(g);Ψc-u(.)Ψu-c(.)1,774.722.1074761.2
Myer
γc=u(t); Pc(.)Pu(.);Ψc-u(.)Ψu-c(.)2,397.70139877.8
γc(t); γu(.)Pu(g);Ψc-u(.)Ψu-c(.)2,428.831.1074832.3
Woods
γ(g*t); p(g*t); λ(g)5,466.00169574.4
γ(g*t); p(t); λ(g)5,504.038.0036612.5
  • a γ = seniority; p = probability of capture; Ψ = movement between treatments; t = time (each trap period); season = seasonal group (1 estimate for fall or spring); g = sex; = constant; c = cut; u = uncut; c=u indicates a single estimate for both the cut and uncut treatments; c−u indicates movement from cut to uncut treatments.

Modeling indicated that the probability of moving between cut and uncut treatments was similar; few of the most-parsimonious models included a treatment effect on movement (Table 2). A total of 195 animals (8%) moved between cut and uncut sides of the 3 experimental grids (Hunter n = 60, Lake n = 54, and Myer n =81). Of these 195 individuals, 79 were 1st captured on the cut treatment and 116 on the uncut treatment. Forty-six individuals (2%) moved from one treatment to the other and did not return (20 from cut to uncut [14 males : 6 females] and 26 from uncut to cut [13 males: 13 females]). Thirty-five animals (2%) moved back and forth with significant portions of their ranges in both sides. The remaining animals were caught only twice, made a single foray and returned to their range in either the cut or uncut, or occupied the area near the transition and were occasionally captured at an edge station (7.5 m from the cut-uncut boundary).

Neither male body mass (adults and subadults) nor female body mass (nonpregnant adults and subadults) differed between cut and uncut treatments (X̄ ± SE; males: 16.9 ± 0.14 g versus 16.7 ± 0.11 g, respectively, F1,846 = 1.37, P = 0.243; and females: 17.5 ± 0.22 g versus 16.8 ± 0.19 g, respectively, F1,437 = 2.63, P = 0.105). Although both sexes showed some significant differences and interactions among seasons, grids, and years, there were no consistent trends.

Proportional contributions of males and females (both young and adults) to the overall population in the cut treatment were not significantly different from those in the uncut treatment (females: n = 36 for each age group, z = −0.98 and 1.59, P = 0.325 and 0.111 for young and adults, respectively; males: n = 36, z = −0.20 and −1.22, P = 0.842 and 0.223). A comparison between the 1st and 2nd halves of the study (1998–2000 versus 2001–2003) indicated that female age structure was stable between those periods in both treatments (cut: n = 18 for each treatment; z = 0.07 and 1.34, P = 0.944 and 0.176 for young and adults, respectively; uncut: n = 18, z = 0.79 and 1.35, P = 0.448 and 0.176). However, although the adult male component of the population did not change between the 1st and 2nd halves of the study (cut: n = 18, z = 1.71, P = 0.087; uncut: n = 18, z = 0.74, P = 0.460), young males in both treatments constituted a greater proportion of the population in 2001–2003 than in 1998–2000 (cut: n = 18, z = −2.26, P = 0.024; uncut: n = 18, z = −1.98, P = 0.048). A comparison of sex-age structure between experimental sites and optimal Woods habitat indicated that Woods had a higher proportion of adult females (n = 36 for experimental sites and 12 for Woods, U = 299.0, P = 0.049), but young females did not differ (n = 36 and 12, U = 233.0, P = 0.698), nor were there differences between 1st and 2nd halves of the study in either habitat (experimental sites: n = 18 for young and adults, U = 180.0 and 218.5, P = 0.584 and 0.074; Woods: n = 12, U = 22.0 and 23.0, P = 0.589 and 0.485). In the case of males, there was no difference between experimental sites and Woods for adult males (n = 36 and 12, U = 286.5, P = 0.094), but young males contributed a significantly higher proportion of the population on experimental sites (n = 36 and 12, U = 309.0, P = 0.026). Finally, there were no temporal shifts in young or adult male proportional abundance between halves of the study (experimental sites: n = 18 for each time period, U = 218.0 and 199.5, P = 0.079 and 0.254 for young and adults, respectively; Woods: n = 6, U = 20.0 and 19.0, P = 0.818 and 0.937).

The proportion of all males that were subadults at the time they 1st appeared on the study sites was compared between treatments, and between experimental sites and Woods. There was no difference between cut and uncut treatments (n = 36 for each treatment, z = −0.64, P = 0.261) or between experimental sites and Woods (n = 72 for experimental sites and 12 for Woods, z = 1.30, P = 0.097). However, there was a greater number of male subadults on experimental grids in the fall season (n = 36 for each treatment, z = 1.88, P = 0.030).

Reproductive effort by adults and subadults (n = 1,116 females, n = 1,297 males) did not differ between cut and uncut treatments (females: n = 67 for cut and n = 66 for uncut treatment, U = 2,312.5, P = 0.649; males: n = 64 and 71, U = 2,562.0, P = 0.203), nor between experimental sites and Woods (females: n = 133 and 24, U = 1,706.5, P = 0.595; males: n = 135 and 24, U = 1,784.5, P = 0.431). To determine whether reproductive effort was being adjusted in response to density, the proportion of females and males in reproductive condition in each year was plotted against yearly mean population density (Fig. 2). Lines joining these points in time sequence follow a clockwise trajectory for the uncut treatment, but not for the cut treatment.

Fig. 2

Phase-space plots of reproductive effort versus mean density (N̂/1.9 ha). A = females in the cut treatment; B = females in the uncut treatment; C = males in the cut treatment; D = males in the uncut treatment.

Discussion

Results of this study show that this population did not conform to theoretical predictions that perturbations will result in population instability and that populations recovering from disturbance will experience increases in density and stability over time. There was no indication that white-footed mouse populations in disturbed habitat were less stable than those in the uncut treatment or woodland. Population trends over time were similar between treatments, and experimental grids were similar to woodland habitat. White-footed mouse populations in the cut treatment did not become more stable with time, and density did not converge with that of the uncut treatment. There was no difference in population growth rates over time between populations in cut and uncut treatments. Moreover, survival in cut and uncut treatments was not significantly different.

Theories of density-dependent habitat selection provide a framework for understanding distribution of individuals between habitats of differing quality (Fretwell and Lucas 1970; Morris et al. 2004; Rosenzweig and Abramsky 1985). When populations are increasing, dispersal is predicted to take place into sequentially poorer habitats, where the ratio between density and available resources is equivalent to that in optimal habitat (thus, individuals have the potential to achieve higher fitness than if they had remained in better habitat). Our results align with these predictions in that components of individual fitness (body mass and reproductive effort) did not differ between cut and uncut treatments.

Although there were many demographic similarities between treatments in our study, density of white-footed mice was significantly higher in the uncut treatment than in the cut treatment throughout the study, a finding consistent with previous observations that white-footed mouse densities are positively correlated with habitat complexity (especially understory trees and logs—Seamon and Adler 1996). Although some components of the habitat had recovered by the end of the study (i.e., shrub cover), densities remained lower in the cut treatment. This was not due to differences in canopy-sized trees, which were present in restricted areas of both treatments on all experimental grids. The habitat component that was removed from the cut treatments and that did not have sufficient time to recover was trees smaller than 10 cm in diameter. This midstory canopy would have ameliorated the abiotic environment and provided some protection from aerial predators. Seed resources would have been diminished during shrub regrowth and insect foods may have been affected by exposure to harsher abiotic conditions.

The degree of intrinsic self-regulation by populations may increase with environmental suitability (Adler and Wilson 1987). Previous studies indicate that populations living in suboptimal habitats can be self-regulating (Linzey and Kesner 1991), but that there appears to be a habitat-quality threshold below which population regulation fails (Adler and Tamarin 1984). Adler and Tamarin (1984), in a study of white-footed mice in mainland and island habitats (both suboptimal), found that the population on the mainland was self-regulating but the population on the island was not. Based on phase-space plots of reproductive effort relative to density, the population on our cut treatment was not self-regulating. Therefore, although both habitats were suboptimal, the cut treatment fell below a habitat-quality threshold for population self-regulation.

Suboptimal habitats have frequently been referred to as habitat “sinks.” Although not meeting the definition of a sink (a population maintained solely by dispersal), our disturbed sites did show some characteristics that align with those of a sink population. However, there was no indication that the cut treatment was more of a sink than the uncut treatment. It appeared that individuals perceived differences in treatments and either chose one side or the other (i.e., were habitat selectors) or were confined there by intraspecific interactions (only 1.5% of animals had ranges that straddled the boundary between treatments). Movement of animals between treatments was approximately equal. Young males did not constitute a greater proportion of the population in cut than in the uncut treatment. An examination of colonization patterns by age and sex indicated that there was no difference between treatments in the proportion of males that were subadults at the time they 1st appeared on the study sites. This pattern of demographic similarity, but limited exchange of individuals between habitats, is reminiscent of earlier studies of white-footed mouse populations at sharp transitions between disturbed and undisturbed habitats (Adler et al. 1984; Linzey 1989; Wegner and Merriam 1979).

We did find, however, demographic differences between white-footed mouse populations in the experimental sites and the woodland habitat. Density was significantly lower in the experimental sites, suggesting that habitats with abundant shrub cover have lower carrying capacity than those with fewer shrubs and with a high tree-canopy layer. The population age structure differed, with the experimental sites having a significantly lower proportion of adult females and a significantly higher proportion of young males. Colonization of experimental sites in fall by subadult males was significantly higher than in optimal habitat. Because the optimal-habitat study site was not replicated, we cannot be sure that these observations will apply to other woodland habitats, but they do align with previous studies and with predictions for comparative demographic performance (Adler and Wilson 1987; Martell 1983; Sullivan 1979).

We can think of several possible reasons why our population did not conform to theoretical predictions of pulsed-perturbation models. Because such models can consider a limited number of variables, factors other than the experimental perturbation could obscure underlying patterns. Density of our populations peaked in fall 1998 and in spring and fall 2000 at all sites and in both treatments, suggesting that populations were synchronized and responding to influences other than our treatments. Although weather has been shown to influence populations of white-footed mice in Ohio (Lewellen and Vessey 1998), a previous western Pennsylvania study indicated that weather contributed only 3.4% to variation in numbers (Kesner and Linzey 1997). Abundance of acorn mast has a strong influence on white-footed mouse population patterns in eastern North America (e.g., Clotfelter et al. 2007; Elias et al. 2004; Ostfeld et al. 1996). In northern West Virginia, oak mast was scarce in 1997, abundant in 1998–2000, and scarce again in 2001–2002 (West Virginia Division of Natural Resources 1970–2010). Because there is spatial synchrony in mast production patterns over a large area (Haynes et al. 2009; Lusk et al. 2007), these data may have relevance for our populations. Another possibility is that our disturbance was not a “pulse” in the true sense of the word. Narrowly, a pulse is an event that occurs over a relatively short period of time, with conditions quickly returning to “normal.” At our sites, although shrub cover was equal to that in undisturbed habitat by the end of the study, one habitat component (understory trees) was still absent. In fact, the entire area encompassing the experimental grids had most recently been cleared more than 20 years prior to the beginning of this study, but the white-footed mouse population density in the uncut treatment was still significantly lower than in woodland habitat. Future tests of pulse-perturbations models should use very long-term data sets, such as those developed at Long-Term Ecological Research Program sites (e.g., Konza Prairie in Kansas and Sevilleta National Wildlife Refuge in New Mexico).

Acknowledgments

This study was supported by grants from the School of Graduate Studies and Research to AVL and MHK, and by the Department of Biology, Indiana University of Pennsylvania. AVL and MHK gratefully acknowledge the cooperation of the Pennsylvania Department of Natural Resources, which provided the manpower and equipment to cut the study sites, and Yellow Creek State Park for hosting our studies over the years. Without the assistance of numerous students, this study would not have been possible. We especially thank graduate students B. Jones and E. Trexler, who assisted with clearing woody debris from the study sites. This manuscript benefitted from comments by G. Schnell and an anonymous reviewer.

Footnotes

  • Associate Editor was Bradley J. Swanson.

Literature Cited

View Abstract