OUP user menu

Habitat patch size modulates terrestrial mammal activity patterns in Amazonian forest fragments

Darren Norris, Fernanda Michalski, Carlos A. Peres
DOI: http://dx.doi.org/10.1644/09-MAMM-A-199.1 551-560 First published online: 16 June 2010


Understanding how environmental change influences the behavior of organisms is central for both ecological understanding and species conservation. We used camera traps to monitor the diurnal variation in activity of 3 ubiquitous terrestrial mammals in neotropical forests—nine-banded armadillos (Dasypus novemcinctus), common opossums (Didelphis marsupialis), and red-rumped agoutis (Dasyprocta leporina)—across a fragmented forest landscape of the southern Brazilian Amazon. Results from a total of 3,086 camera-trap days distributed across 21 forest patches (ranging in size from 2 to 14,480 ha) and 2 undisturbed continuous forest areas were used to test the effects of a series of abiotic and forest disturbance variables on species activity. An information theoretic analysis revealed significant predictors of the temporal distribution of activity that varied among species. Habitat fragmentation affected the activity of both nocturnal species, but effects of habitat patch area depended on interactions with disturbance variables for the common opossum. Of the 3 species investigated, D. novemcinctus exhibited the greatest variation in activity in relation to forest patch size. Armadillos were strictly nocturnal in forest areas > 1,000 ha, whereas their foraging activity switched to a cathemeral pattern, with up to 60% of all photos recorded during the day in smaller forest patches (< 1,000 ha). In contrast, the time since forest patches had been isolated was the only significant predictor of activity patterns for agoutis, a diurnal species exhibiting a bimodal activity pattern. Our results support the hypothesis that behavioral plasticity is an important determinant of species persistence in small forest remnants dominated by edge effects.

Key words
  • Amazon forest
  • anthropogenic change
  • behavior
  • Dasyprocta leporina
  • Dasypus novemcinctus
  • Didelphis marsupialis
  • diel activity
  • habitat loss
  • terrestrial mammal

Patterns of diel behavior can directly influence individual fitness, and behavioral responses to time of day are often predictable in vertebrates so the temporal distribution of animal activity is an important niche dimension (Kronfeld-Schor and Dayan 2003). Understanding the extrinsic determinants of circadian rhythms of activity therefore is relevant to understanding how species adapt to and persist in their environments following human perturbation (Buchholz 2007; Lima and Zollner 1996).

Despite sensorial constraints imposed by phylogenetic inertia, the temporal allocation of activity budgets of diurnal, nocturnal, and cathemeral mammal species is known to vary within and between species so that proximate determinants are expected to fluctuate in response to environmental changes (Kronfeld-Schor and Dayan 2003; Roll et al. 2006). Because of such behavioral plasticity, mammals are a useful study group for investigations into impacts of environmental change on temporal activity. It is therefore surprising that the influence of habitat loss on mammalian activity patterns is an area of research that has remained relatively unexplored.

Habitat loss has a myriad of effects on species and their interactions such that identifying factors that lead to either positive or negative effects is often challenging (Fahrig 2003). In addition, habitat loss does not operate in isolation and often is associated with a range of additive or synergistic effects of anthropogenic perturbations, including selective logging, wildfires, and hunting (Laurance and Cochrane 2001; Peres 2001; Peres and Michalski 2006). In neotropical forests direct influences of habitat loss on species richness, abundance, and biomass of terrestrial mammals depend primarily on forest patch size (Galetti et al. 2009; Michalski and Peres 2007). However, few studies document less obvious impacts on differential allocation of diel activity (hereafter referred to as diel activity, activity patterns, temporal activity, and activity budgets interchangeably to discuss temporal activity patterns over a 24-h period) under the differing environmental conditions resulting from habitat loss, fragmentation, and the concomitant additive or synergistic effects of anthropogenic perturbations.

Because many factors can influence activity patterns, how can we best understand influences across broad spatial scales? For example, behavioral questions relating to effects of habitat loss are best addressed across a gradient of habitat patches ranging from a few to thousands of hectares so it is appropriate to examine species that persist within this range of patch sizes. Michalski and Peres (2007) found a variety of mammal species that persist and even thrive across such a range, including the red-rumped agouti (Dasyprocta leporina, Linnaeus, 1758—hereafter agouti), the common opossum (Didelphis marsupialis, Linnaeus, 1758—hereafter opossum), and the nine-banded armadillo (Dasypus novemcinctus, Linnaeus, 1758—hereafter armadillo).

These species, or closely related congeners, are ubiquitous across neotropical forests (Emmons and Feer 1997) and exhibit pronounced variation in patterns of activity (McDo-nough and Loughry 1997; Sunquist et al. 1987), which makes them optimal study species for investigations into environmental influences on activity patterns. Previous studies have reported that abiotic factors such as day length and temperature commonly influence mammalian activity patterns (Kronfeld-Schor and Dayan 2003; Roll et al. 2006). A number of abiotic factors, such as temperature (Clark 1951; McDo-nough and Loughry 1997; McManus 1971), luminosity (Aliaga-Rossel 2004), and rainfall (Smythe 1978), also have been shown to influence the activity of at least 1 of our 3 study species. However, no previous studies present details of how anthropogenic disturbance variables, such as habitat fragmentation, hunting, logging, and burning, influence the activity of terrestrial neotropical mammals.

Using data from a 12-month camera-trapping campaign across a fragmented forest landscape of southern Amazonia, we address the questions of whether diel activity patterns of 3 ubiquitous mammal species in neotropical forests are primarily influenced by either abiotic factors (day length, temperature, lunar luminosity, and rainfall) or factors associated with habitat perturbation (habitat patch area, time since isolation, logging, fire, and hunting).

Materials and Methods

Study area.—Field surveys were carried out over 12 months (July-December 2003 and July-December 2004) within 23 forest sites, across a 34,200-km2 landscape surrounding the town of Alta Floresta, Mato Grosso, Brazil (09 °53′S, 56 °02′W; Fig. 1). We maximized the spatial independence between preselected sites by establishing a minimum edge-to-edge distance >1 km (X̄= 24.7 km ± 13.3 SD, range = 1.2-67.3 km, n = 253 pairwise comparisons). Deforestation in this landscape resulted from an agricultural resettlement scheme dating from the early 1980s. The Alta Floresta landscape currently consists of forest remnants of various sizes, shapes, and levels of structural and nonstructural forest disturbance, surrounded by an open-habitat matrix dominated by cattle pastures (Michalski et al. 2008; Peres and Michalski 2006).

Fig. 1

Location of the study region in Alta Floresta, northern Mato Grosso, Brazil, showing the 21 surveyed forest patches (solid areas) and 2 continuous, undisturbed forest sites used as controls (solid circles). Forest patches <30 ha are circled. Gray and white areas on either bank of the Teles Pires River represent forest and nonforest cover, respectively.

The mean annual rainfall is 2,350 mm, and the evapotranspiration is <1,000 mm/year, providing a 1,350- tol,400-mm/ year surplus, except for the dry season (May-September), which typically results in a hydrological deficit of 250-300 mm (RADAMBRASIL 1983). We distributed our sampling evenly across the same time of year (July-December) in both 2003 (11 sites) and 2004 (12 sites) to minimize any potentially confounding effect of rainfall seasonality.

Day length does not vary >1.2 h across the year at the latitude of our study area. The shortest days occur in June and longest days in December, with a minimum and maximum day length of 11.5 and 12.7 h (United States Naval Observatory Calendar, http://aa.usno.navy.mil/). Further details on the study landscape are presented in Michalski et al. (2008).

Camera trapping.—Camera traps baited with scent lure (Hawbaker's Wild Cat Lure 2; Minnesota Trapline Products, Pennock, Minnesota) were used to record continuously the activity of the target species for a total of 30 consecutive days within each of the 23 forest sites surveyed, including 21 forest patches ranging widely in size (2-14,480 ha) and 2 undisturbed continuous forest areas. The 30-day period ensured that sampling included a full moon phase in each area surveyed. We distributed camera traps within each fragment on a hexagonal grid to minimize variation in density of camera-trap stations that were placed ∼500 m apart, thereby ensuring a survey area of approximately 130 ha in forest sites > 1,000 ha. Therefore fewer traps were placed in small fragments (Table 1), but even the smallest site was sampled by 2 cameras (X̄= 4.2 camera traps/site ±3.1 SD).

View this table:
Table 1

Summary of the camera-trap survey effort used to monitor 3 species of terrestrial mammals in each of 4 size categories of forest patches surrounding the town of Alta Floresta, Mato Grosso, Brazil.

Count of independent photos (day, night)
Site classPatch size (ha)nCamerasaTotal camera-trap daysDasypus novemcinctusDasyprocta leporinaDidelphis marsupialis
12.4–7.362 (2–2)372.1348 (9, 39)16 (15, 1)40 (4, 36)
214.9–98.272(2–3)488.0163 (4, 59)31 (31, 0)24 (2, 22)
3106.2–899.855 (3–7)739.4150 (5, 45)26 (26, 0)78 (4, 74)
41,763.3–144,80559(9–10)1,486.7152 (0, 52)58 (56, 2)49 (1, 48)
Total233,086.26213 (18, 195)131 (128, 3)191 (11, 180)
  • a Mean (and range) of camera traps used per patch in each class.

Because previous studies have shown differences in activity between juveniles and adults (McDonough and Loughry 1997), we limited our analysis for all species to adults and subadults, which were assigned to age class on the basis of body size. Because we were unable to identify individuals unambiguously, consecutive photos of the same species were defined as independent occurrences if the interval between photos was >30 min. Although it is possible that we recorded the same individual on the same day, the majority of photos were recorded with an interval >24 h between consecutive images from the same camera (70%, 63%, and 63% for agouti, armadillo, and opossum, respectively). Of those recorded within 24 h, the mean (range) interval was 4.5 h (0.5-11.7 h), 8.0 h (0.5-22.4 h), and 8.7 h (0.5-22.6 h) for agouti, armadillo, and opossum, respectively. As such, a minimum 30-min interval reduced the temporal dependence between photos and any systematic bias in our analysis. Additional details on camera-trapping methodology are presented in Michalski and Peres (2007).

Abiotic variables.—Because we did not measure abiotic variables at each camera location, qualitative indexes were derived to quantify the relative influence of nocturnal illumination, maximum daytime temperature, and total precipitation on the timing of photos. A standardized (to the average over our 12-month study period) nocturnal illumination index was calculated following Schwitzer et al. (2007), where for each night, the duration of moonshine between the set and rise of civil twilight was multiplied with a value corresponding to the illuminated fraction of the moon, and multiplied by a factor C between 0 (heavily overcast) and 1 (clear) to account for cloud cover and weather condition. Data for moon phase, illuminated fraction, civil twilight, and moonrise and moonset were obtained from the United States Naval Observatory Calendar (http://aa.usno.navy.mil/), using the Alta Floresta geographical coordinates (09 °52′S, 56 °06′W, GMT -4 h).

Values for maximum temperature (°C) and total precipitation during the 24-h period prior to each photo were collected via a Stevenson weather station located at the Alta Floresta Airport (location = 9 °52′S, 56 °06′W, SYNOP id = 82965) and supplied by the Plymouth State Weather Centre (2009). Following the technique used to generate the nocturnal illumination index we also derived qualitative indexes (standardized to the average over our 12-month sampling period) to evaluate the relative effects of temperature and precipitation.

Disturbance variables.—We quantified 4 variables—forest patch isolate age (years), history of fire disturbance, logging intensity, and hunting pressure—within patch disturbance. We obtained isolation age (X̄= 16.0 years ± 8.3 SD, range = 0-27 years) from 11 biannual Landsat (http://glovis.usgs.gov/) images (1984-2004). When forest patches already had been isolated in the earliest image, we used interviews to determine their isolation age, defined as the number of years since a forest patch had been disconnected from the continuous forest or from a neighboring patch > 10,000 ha. Isolation age was scored as 0 for control sites or patches > 10,000 ha. Principal component analysis was used to derive a single measure for each of logging intensity, burn severity, and hunting pressure. Data on the intensity and extent of disturbance within each site were based on site inspections and information obtained from local interviews to determine the composite history of logging intensity (4 correlated variables), burn severity (4 variables), and hunting pressure (5 variables). We then conducted 3 principal component analyses to characterize these 3 disturbance variables and used the scores of the 1st axis in subsequent analyses (Michalski and Peres 2007).

Data analysis.—All analysis was performed in R (R Development Core Team 2008) with associated analytic packages. Circular summaries (Lund and Agostinelli 2007) were used to determine the mean overall timing of species activity as recorded by camera traps. A resampling approach was used to derive unbiased estimates of the 95% confidence intervals (95% CIs), whereby the circular mean from a random sample of 100 data points from each species was calculated 10,000 times with replacement to generate reliable estimates of 0.025 and 0.975 quantiles.

We grouped patches into classes on a log scale (1: <10 ha; 2: 10 ha >100 ha; 3: 100 ha > 1,000 ha; and 4: > 1,000 ha) to enable us to derive metrics for comparison with previous studies and investigate the univariate influence of patch size class on the activity patterns of the 3 species. We used circular analysis of variance (function “circular aov ”—Lund and Agostinelli 2007) to investigate the variation in the response of timing of activity (photo time) with patch size class. As a further test of the influence of patch area on species activity we used generalized linear models with quasi-binomial errors to test if patch area (log-transformed) predicted the proportion of daytime photos (total daytime/total photos). Because our area estimates for control sites were arbitrary values, they were excluded from the generalized linear model analysis.

To investigate the influence of abiotic and disturbance variables on species activity patterns we used linear mixed-effect models (Bates and Maechler 2009). As our response variable for this analysis we used a linear scale to quantify activity patterns, defined as the time (min) from sunset for armadillos and opossums and time from sunrise for agoutis. The influence of abiotic (maximum temperature, rainfall, day length, and nocturnal luminosity) and disturbance variables (isolate age, logging intensity, burning severity, and hunting pressure) were investigated within an information theoretic modeling framework. This approach enables multimodel inference where models are ranked and scaled by some information criterion to enable an understanding of model uncertainty over the set of candidate models (Burnham and Anderson 2002). Our uncorrelated (Spearman's rs < 0.56) abiotic and disturbance variables were modeled separately, but unclassed forest patch area (log-transformed) was included in both model sets to enable us to investigate any interactions with forest patch size. Because our area estimates for control sites were arbitrary values, they were excluded from this analysis. We used linear mixed-effect models with our abiotic and disturbance variables as fixed effects and forest patch identity as a random effect, enabling us to account for unmeasured variation within patches.

We evaluated models based on their information content, as measured by the corrected Akaike information content for small sample sizes (AICCHurvich and Tsai 1989), to determine the probability of the best approximating model (Burnham and Anderson 2002). We included in our analyses all possible candidate models; therefore, all variables were on equal footing to calculate relative variable weights (Burnham and Anderson 2002). A reduced subset of models for a 95% confidence set, based on the sum of Akaike weights from largest to smallest that resulted in the sum of >0.95, was used to calculate slopes and unbiased SEs (Burnham and Anderson 2002).


Our camera-trapping effort revealed similar abundances of the 3 study species across the forest patches surveyed. From a total effort of 3,086 camera-trap days we recorded 213 (0.07 photos/trap day), 191 (0.06 photos/trap day), and 131 (0.04 photos/trap day) independent photos of armadillo, opossum, and agouti, respectively. We found significant interspecific differences in mean timing of activity across our study region (F2,532 = 230.73, P < 0.001). Agoutis were predominantly diurnal compared with the mainly nocturnal armadillo and opossum (Table 1; Fig. 2). The average time recorded for agouti photos was 0954 h (95% CI = 0852-1112 h; range = 0440-2214 h) with only 3 photos recorded during the night (2 of which occurred within 40 min of civil twilight). Of the 3 species surveyed, armadillos had the broadest range of activity with photos recorded in 19 of 24 h with a mean time of 2220 h (95% CI = 2139-2302 h; range = 1740-1556 h) compared with a mean of 2252 h (95% CI = 2205-2340 h; range = 1657-0604 h) for opossums.

Fig. 2

Activity patterns across size classes of forest patches. The proportion of independent photos recorded over a 24-h cycle presented in 2-h bins (e.g., 23,24 includes all photos between 2300.00 and 0059.59 h) from 3 species of terrestrial mammals—A) Dasyprocta leporina, B) Dasypus novemcinctus, and C) Didelphis marsupialis—monitored with camera traps in 21 forest patches and 2 areas of continuous forest surrounding the town of Alta Floresta, Mato Grosso, Brazil.

Although the distribution of activity (expressed as the proportion of photos within 2-h bins) varied between forest patch size classes across all species (Fig. 2), the mean timing of activity varied significantly only between forest patch size classes for armadillos (F3209 = 3.23, P = 0.023) in which larger patches were associated with nighttime photos recorded later (mean values by patch size: 1 = 2157 h, 2 = 2224 h, 3 = 2215 h, and 4 = 2316 h).

Overall, agouti exhibited peaks of activity in the mornings (0500-0559 h and 0800-0859 h) and in the evenings (1700-1759 h). In areas >10 ha early morning activity was more pronounced, whereas in areas <10 ha activity appeared to be distributed more evenly throughout the day (Fig. 2). We found no significant difference in mean time of photos between forest patch size classes for agouti (X̄= 1306 h, 0955 h, 0901 h, and 0902 h; F3,127 = 2.38, P = 0.073). Similarly, the mean time of photos did not vary significantly between size classes for opossum (X̄= 2233 h, 2212 h, 2233 h, and 2353 h; F3,87 = 1.88, P = 0.13).

Of the 3 mammal species, armadillos were the only one switching their timing of activity (Table 1). Further analysis revealed that forest patch area had a significant influence on the proportion of daytime photos per patch only for armadillos (generalized linear model P = 0.015, deviance explained = 24.84%) with the proportion of daytime photos declining nonlinearly with patch area (log-transformed; Fig. 3). We did not record any daytime photos of armadillos in areas > 1,000 ha, whereas armadillos were active during both day and night in areas < 1,000 ha, with most of daytime photos (72%) recorded in areas <100 ha. We also found considerable variability in activity patterns in areas < 1,000 ha, with the proportion of daytime photos ranging from 0 to 0.67 (Fig. 3).

Fig. 3

Proportion of daytime photos for nine-banded armadillos (Dasypus novemcinctus) as a function of forest patch size. A generalized linear model (predictions = dashed line) demonstrates a significant decline in the proportion of daytime photos as forest patch area (n = 21) increases.

Proximate influences on activity.—When considered together with disturbance variables, forest patch area influenced activity of armadillos and opossums (Table 2). However, the sum of weights for forest patch area exceeded 0.5 only for armadillos when modeled together with abiotic variables. The number of candidate models retained for armadillos (24 of 31 and 22 of 31), opossums (25 of 31 and 13 of 31), and agoutis (22 of 31 and 25 of 31) suggests that forest patch area effects are modulated by interactions with both habitat disturbance and abiotic variables.

View this table:
Table 2

Model weights and parameter (slope) estimates from information-theoretic analysis of species activity. Influences on species activity patterns were studied in 21 forest patches surveyed with scented camera traps. Variables were: biotic: area, isolate age, LI = logging intensity, BS = burn severity, HP = hunting pressure; and abiotic: area, luminosity = nocturnal luminosity; Temp = maximum daily temperature, day = day length, rain = total daily rainfall. Activity was defined as time difference (min) from sunset (civil twilight) for Dasypus novemcinctus(n =197) and Didelphis marsupialis (n= 173) and time difference from sunrise for Dasyprocta leporina(n = 103).

Forest area (log10 ha)Isolate age (years)LIBSHPForest area (log1o ha)LuminosityTempDayRain
MaΣwbSlope (±1 SE)cΣWibSlope (±1 SE)cΣWibΣWibΣWibMaΣWibSlope (±1 SE)cΣWibSlope (±1 SE)cΣWibΣWibΣWib
Dasypus novemcinctus240.6622.00 (21.30) (18.60)
Didelphis marsupialis250.6016.80 (18.50)0.370.260.250.44130.471.0019.10 (20.10)0.410.380.21
Dasyprocta leporina220.300.869.78 (5.06)0.270.300.34250.270.430.380.210.28
  • a Confidence set of models with the sum of the Akaike weights (wi-s) from largest to smallest ≤0.950.

  • b The sum of Akaike weights (wi-s) for all models with a given variable.

  • c Slope and unconditional standard error (SE) for variables with sum of Akaike weights >0.5.

Comparing the sum of model weights revealed that for agoutis the time since forest patches were isolated (years) had the strongest influence on activity timing (sum of weights = 0.86); we did not detect any other strong disturbance or abiotic influences on the activity of this species. Of the abiotic variables, the nocturnal luminosity index was the only variable to influence activity strongly, and only for opossums (Table 2).


Our results present the 1st demonstration of the influence of habitat structure on the activity patterns of terrestrial forest mammals across a hyper-fragmented Amazonian landscape, thereby enabling us to quantify empirically the relationship between observed behavior and environmental predictors. By contrasting species responses we provide insights into potential adaptive benefits of the variation in diel patterns observed for these species in environments dominated by human perturbation.

Dasypus novemeinetus.—The overall mean timing of activity we found for nine-banded armadillos differs from studies in other regions. Peak activity in North American populations of armadillo was around sunset with a mean activity (±1 SD) for adult armadillos of 1949 ± 2 h (McDonough and Loughry 1997). In Alta Floresta the mean activity time from all nine-banded armadillo photos was 2220 h or 4 h 10 min after sunset, and even in the smallest forest patch size class (where activity occurred earliest) the mean activity time was 2157 h. These results agree with those of previous studies (Loughry and McDonough 1998) that also described later activity in a Brazilian population of armadillos when compared with a population from North America. The activity of armadillos varies with temperature, with some authors suggesting that temperature is an important limiting factor to the geographic expansion of armadillos across North America (Layne and Glover 1978, 1985; Taulman and Robbins 1996). Therefore, it seems likely that differences in temperature contribute to differences in activity observed between sites.

Previous studies suggest a gradient of activity, with armadillos becoming active later (i.e., closer to midnight) nearer the equator (0.71 h before sunset in Tall Timbers, Florida, 30 °N [Loughry and McDonough 1998]; 1.79 h after sunset in Poço das Antas Reserve, 22 °S [Loughry and McDonough 1998]; and 4.17 h after sunset at 9 °S [present study]). Although maximum temperatures are on average similar between study areas (X̄ = 31.2 °C ± 2.2 °C SD during our study compared with 32.56 °C ± 2.3 °C from Tall Timbers [McDonough and Loughry 1997]), the more subtropical climate in Florida results in cooler nights, which could explain why armadillos are active closer to sunset in their nonpeninsular Florida study site. However, standardized surveys across a variety of latitudes are required to determine if other factors such as natural predation pressure, differences in habitat, hunting pressure, or a combination of these are not contributing to temporal differences in activity observed among sites (Loughry and McDonough 1998).

In contrast to the comparison across sites, abiotic factors did not significantly influence the activity of armadillos within our Amazonian study area. Of the 3 species studied, forest patch area had the strongest effect on the activity of armadillos. Not only did the mean timing of activity vary significantly between forest fragment size classes, with armadillos being active earlier in smaller areas, we also found more daytime activity in smaller fragments. We also found that the influence of habitat fragmentation on the proportion of daytime photos was not linear, with half of all daytime photos recorded in fragments <10 ha. Previous studies in North America have shown that armadillos can be active throughout the day, with activity patterns varying in relation to a variety of abiotic and biotic factors such as temperature and prey availability (Clark 1951; McDonough and Loughry 1997). However, given the spatial scale of our study, we were unable to detect any significant effects from other disturbance and abiotic variables investigated.

Didelphis marsupialis.—The activity of the other nocturnal mammal species—the common opossum—varied with forest patch size and nocturnal luminosity, with opossums active later with increasing patch size and luminosity levels. Because patch area did not have any influence on activity timing when considered together with abiotic variables, the influence of area effects on opossum activity seems to depend on interactions with disturbance variables.

Opossums were active later with higher levels of nocturnal luminosity. Compared with armadillos, opossums rely more on visual cues to facilitate movement -and forage, and their nocturnal foraging activity is facilitated by increased retinal sensitivity from an intraocular reflecting structure, the tapetum lucidum (Ollivier et al. 2004; Volchan et al. 2004). We propose 2 possible explanations for the difference in activity with lunar luminosity: with increased visibility associated with higher nocturnal luminosity opossums extend bouts of foraging activity; or a lunarphobic response, which is more common in nocturnal mammals (Bearder et al. 2006) in that on brighter nights opossums refrain from activity until food requirements overcome the fear or discomfort of foraging. The 2nd explanation appears most likely, but additional studies are required to confirm this.

Dasyprocta leporina.—Our results agree with previous studies of similar species in the same genus that show that agoutis have a bimodal pattern of activity with peaks shortly after sunrise and before sunset (Lambert et al. 2009; Smythe 1978). Also, foraging activity of agoutis varies in response to available resources and temperature (Jorge and Peres 2005; Smythe 1978). The Central American agouti (Dasyprocta punctata) at Barro Colorado Island, Panama, is known to extend activity during periods of food scarcity or under a full moon (Aliaga-Rossel 2004). Radiotracking studies in Barro Colorado Island also have shown occasional nocturnal activity, which was positively correlated with both fruit availability and higher daytime temperatures (Lambert et al. 2009).

In our study the diurnal distribution of agouti activity was strongly influenced only by isolate age (sum of weights = 0.86), with timing of activity increasing in older patches. Because we found no significant relationship between timing of activity and indices of maximum daily temperature or nocturnal luminosity, we hypothesize that limited resource availability in older patches is causing agoutis to extend foraging activity. This hypothesis is supported by habitat surveys in the same patches showing that older sites have a higher density of small-seeded softwood trees (Michalski et al. 2007), which provide a poorer quality and quantity of the seeds and fruits that the agoutis depend on (Jorge and Peres 2005; Smythe 1978).

Contrasting influences of proximate predictors.—A variety of proximate factors can mask or modulate endogenous circadian rhythms in mammals, including season (Donati and Borgognini-Tarli 2006), nocturnal luminosity (Fernandez-Duque and Erkert 2006; Schwitzer et al. 2007), temperature (Chappell 1980), competition (Wasserberg et al. 2006), predation (Colquhoun 2006; Griffin et al. 2005), prey availability (Theuerkauf et al. 2003), resource availability (Lambert et al. 2009; Thies et al. 2006), habitat type (Presley et al. 2009; Schwitzer et al. 2007), and anthropogenic disturbance (Di Bitetti et al. 2008; Kitchen et al. 2000). Across a fragmented Amazonian forest landscape we found that activity patterns were influenced primarily by habitat perturbations. Of the variables we investigated within an information theoretic analysis, forest patch area (for armadillos and opossums) and time since patch isolation (for agoutis) significantly influenced the distribution activity, whereas nocturnal luminosity was the only abiotic variable with a significant influence on opossum activity.

Although it was possible to investigate only a subset of all possible proximate factors, our results provide an important step in understanding not only how abiotic and disturbance variables influence the activity of species in a deforestation frontier but also how species persist in human-dominated landscapes. A number of variables were not considered, but our focus on within-patch disturbances is justified because these factors will have an overarching influence on other biotic factors, such as the presence of predators and population densities, that also may modulate activity patterns (Colquhoun 2006; Griffin et al. 2005; Wasserberg et al. 2006). Factors such as occurrence of predators and intra- and interspecific densities vary with the size of forest patch and disturbance variables in our study area (Michalski and Peres 2005, 2007). Although we cannot exclude the possibility of other biotic influences including other landscape factors, such as distance to the nearest forest patch, our analysis shows that forest patch size and isolation age are likely to be important drivers of changes in the activity of our study species in fragmented forest landscapes. But why did the activity of all 3 generalist species vary with the influence of habitat perturbations?

Determining the ultimate (adaptive) functions for behaviors is challenging, even if the proximate factors stimulating that behavior are understood. Although we were able to show that habitat perturbations have a causal influence on species activity, the adaptive significance remains ambiguous. For example, the nocturnal activity of fossorial mammals such as armadillos is at least partly to avoid heat stress of diurnal temperatures (Johansen 1961). Therefore increased diurnal activity in small fragments could be considered maladaptive (i.e., lead to declines in the genetic or demographic viability of populations), or indicate populations under stress, or both. Alternatively it is possible that the increase in daytime activity is related to food acquisition. Small forest areas are dominated by edge effects that drive changes in floral and faunal communities (Laurance et al. 2002; Murcia 1995). Area and edge effects will change the availability of arthropods in smaller forest areas compared with large forest areas > 1,000 ha and also affect insect availability in the surrounding matrix. Daytime activity in armadillos therefore could be a response to changes in food availability in the surrounding matrix, which ultimately enables them to persist at high densities in small forest areas (Michalski and Peres 2007).

Our findings that the activity of all 3 generalist species varies in response to habitat perturbation (habitat size and isolation age) suggest that changes in behavior are not maladaptive. Although habitat loss and fragmentation across the region is relatively recent (<30 years—Michalski et al. 2008), the generation times of all 3 species are relatively short, and if the behaviors were maladaptive we would expect a negative impact on the relative abundance of these species. However, habitat loss and accompanying anthropogenic perturbations did not have negative effects on the abundances of these species across our study area (Michalski and Peres 2007). These species are known to tolerate influences of edge effects, anthropogenic disturbance, and habitat conversion to various degrees (McDonough et al. 2007; Michalski and Peres 2007; Norris et al. 2008; Vaughan and Foster Hawkins 1999). Therefore we conclude that behavioral plasticity facilitates the persistence of these species in largely deforested regions in which small, edge-dominated forest fragments become prevalent.

Conclusions.—The variation in the diel activity periods of the 3 mammal species addressed in this study may have been expected based on results elsewhere (Gómez et al. 2005; Inbar and Mayer 1999; Lambert et al. 2009; McDonough and Loughry 1997; Sunquist et al. 1987). However, what is surprising is that habitat fragmentation affected the timing of activity for both nine-banded armadillos and common opossums, particularly considering the contrasting phylogeny and differences in life-history strategies of these 2 species.

Based on body size we would expect that the 3 mammal species studied would be resilient to the effects of forest patch erosion (Peres 2001). An increase in forest patch size had either no effect (for agoutis and opossums) or a negative effect (armadillos) on species abundances across the areas that we surveyed (Michalski and Peres 2007), and 2 of the species (armadillo and opossum) are well documented as tolerating human-dominated landscapes, including cities and towns (Guerra et al. 2007; Taulman and Robbins 1996). Yet allometry alone is not sufficient to explain the observed differences in abundance-area relationships. Why do the larger bodied armadillos persist and even appear to thrive in small forest patches? Our findings that the activity of contrasting species varies with the size of forest patch, and that armadillos are the most flexible of the 3 species in diel activity patterns—to the extent of switching from nocturnal to cathemeral activity in small forest patches—strongly suggests that behavioral plasticity in activity patterns contributes to the ability of these species to adapt to changes induced by, or associated with, habitat loss. In light of this, we hypothesize that plasticity in activity patterns facilitates persistence in a fragmented forest landscape dominated by human perturbation.


This study was funded by the Natural Environment Research Council and the World Wide Fund for Nature-Brazil (United States Agency for International Development grant NT 746/2003) and Conservation International. FM was funded by a Brazilian Ministry of Education (CAPES) Ph.D. studentship. We are deeply indebted to all landowners in Alta Floresta, particularly R. Silva and V. Riva, for providing access to their properties. G. Araújo and A. Araújo provided invaluable assistance during fieldwork.


  • Associate Editor was Rodrigo A. Medellín.

Literature Cited

View Abstract