Using remote camera traps to assess mammal and bird assemblages across a complex environmental landscape

Animals must navigate a complex mosaic of habitat types, both natural and artificial. As artificial habitats (e.g., agricultural fields) become increasingly abundant in many landscapes, species will be affected differently, depending on their habitat preferences. We investigated the diversity, richness, abundance, and biomass of mammals and birds with remote camera traps that optimized the capture of both large and small animals. Camera traps allowed us to capture natural rates of mammals and birds, which is difficult to obtain using human observers who can affect the behavior of animals and are limited in their spatio-temporal scope and ability to assess nocturnal communities. Our camera trap arrays were established along two transects in a local conservation reserve; one transect ran from an agricultural field to an upland forest and another from a wetland to an upland forest. Over the 6-week study our cameras recorded 2,245 images, within which we observed 483 individuals comprising 16 species of mammals and birds. Our data showed that species composition and abundances were only marginally different between the two transects, with species common to both transects not exhibiting any statistical difference in abundances. Coyotes and armadillos were unique to the riparian transect, and many more bird species were present along the riparian transect than the agricultural transect. Diversity, richness, and total community biomass did not differ significantly between the two transects nor along each transect but there were non-significant trends in predicted directions. Our results revealed that fewer species use the forest immediately adjacent to the agricultural field, but more species use the wetland and the forest immediately adjacent to the wetland. Our results corroborate other studies revealing that certain species are more common in forested areas but also that some species thought to prefer forested areas may actually be more habitat generalists than previously thought.


0 8
Our study was conducted at Lindenwood University's Daniel Boone Field Station, generate a subset of models [33,37]. We considered models with the lowest AIC c values 2 2 1 to be the models of best fit, and we ranked models based on weight of each of the best models was less than 0.9, indicating other models were 2 2 7 supported by the data, so we performed model averaging which provides more robust 2 2 8 model variances and increases the reliability of parameter estimates [34]. We included a 2 2 9 final subset of models that had cumulative model weights of ≥ 0.95 [38]. To determine 2 3 0 the relative importance of each term in the models, we calculated the normalized Akaike 2 3 1 weight for each parameter (w ip ), which is the sum of the w im in which that parameter is 2 3 2 present (w ip = 1 indicates a parameter present in all models). We calculated the 2 3 3 confidence intervals of the slopes between each parameter and diversity, species richness, 2 3 4 and biomass to determine when those parameters may have significant effects.

3 5
Additionally, we performed ANOVAs to determine if diversity, richness, and biomass 2 3 6 differed along each transect (see above section "environmental variables" for binning 2 3 7 procedure), and we used two-sided, Welch's t-tests to determine differences between 2 3 8 transects. We recorded at least one mammal or bird from 92 % of the plots we sampled; 38 of 42 2 7 2 plots along the agricultural transect and 39 of 42 plots along the riparian transect. In total, 2 7 3 we captured 2,245 images that included at least one mammal or bird, which we estimated   The model selection process varied between diversity, species richness, and biomass 3 1 5 (Table 1A, B, C, respectively). In the model selection for diversity, three models had 3 1 6 cumulative weights ≥ 0.95 (Table 1A), with the best-fitting model containing distance, 3 1 7 transect, maximum temperature, average temperature, litter depth, transect*maximum 3 1 8 temperature, transect*litter depth, and distance*litter depth. The model weight was less 3 1 9 than 0.9 for this model, but the next best-fitting model, which included canopy cover, had 3 2 0 much lower weight and a Δ AIC c above two (Table 1A). In the model selection analysis 3 2 1 for species richness, four models had cumulative weights of ≥ 0.95 (Table 1B) second best-fitting model was less than two. The model selection for biomass produced 3 2 6 four models with cumulative model weights ≥ 0.95 (Table 1C), with the best-fitting 3 2 7 model containing distance, transect, maximum temperature, average temperature, 3 2 8 transect*distance, transect*maximum temperature (Table 1C). The second best-fitting 3 2 9 model, which included litter depth, had a comparably high model weight and a Δ AIC c 3 3 0 value less than two. Distance is the only parameter common to the best fitting models for 3 3 1 diversity, richness, and biomass. Common parameters between diversity and richness 3 3 2 include distance, litter depth, and the distance-litter depth interaction.

3
The model selection analysis for diversity included eight explanatory parameters in 3 3 4 the model of best fit (Table 1A). Diversity decreased at plots with higher average  (Table 1A). Likewise, diversity did not differ significantly between   The model selection analysis for species richness included three parameters for the 3 5 5 model of best fit (Table 1B). Species richness increased slightly with increased litter 3 5 6 depth along the riparian transect but not the agricultural transect ( Figure 5G). The 3 5 7 difference in these slopes was not significant though (Table 1B). Species richness did not   The model selection analysis for biomass included six parameters in the model of 3 6 9 best fit (Table 1C). Biomass decreased at plots with maximum temperatures along the 3 7 0 agricultural transect but increased with maximum temperature on the riparian transect 3 7 1 ( Fig 6C). The difference in these slopes was not significant however (Table 1C)   multiple cameras with overlapping fields of view also reduced the probability of missing 3 8 5 animals due to failure of a single camera to trigger, which occurs for a variety of reasons 3 8 6 [27,41]. Thus, our multiple-camera setup (Fig 2) allowed us to estimate abundance, 3 8 7 diversity, and biomass across a wider range of mammal and bird species than would be 3 8 8 possible with just a single camera at each plot.

8 9
There were no differences in the abundances of species that were found on both 3 9 0 transects. The marginally significant difference in species composition and abundance 3 9 1 found by the MRPP between the transects was likely driven by the presence of armadillos 3 9 2 and coyotes on the riparian transect but not on the agricultural transect as well as by the 3 9 3 increased number of birds on the riparian transect. Several of our predictions regarding 3 9 4 species-specific trends in abundances were supported, despite no significant differences 3 9 5 in the abundances of most mammal species between the two transects (Fig 4A, B).

9 6
Armadillos were only present on the riparian transect, corroborating studies that found 3 9 7 armadillos avoid open habitats and grasslands [11,14]. The two species of squirrels at 3 9 8 our site are considered habitat generalists and are quite adaptable to novel environments 3 9 9 [15, 42-45] so it was unsurprising they occurred in high abundance on both transects 4 0 0 (Table 1)  agricultural transect, contrary to our prediction that they would be found about equally 4 1 8 between the transects (Fig 4A, B).

1 9
Contrary to our prediction and despite higher species richness on the riparian 4 2 0 transect, diversity was nearly equal between the two transects ( Fig 6A). Two main 4 2 1 reasons explain this pattern of diversity. First, the distribution of species is more even 4 2 2 along the agricultural transect (Fig 4A, B). Squirrels dominate in abundance on both 4 2 3 transects, but are relatively more dominant on the riparian transect than on the 4 2 4 agricultural transect (Fig 4A, B). This unevenness in species abundances likely overrides 4 2 5 the presence of two more species on the riparian transect. Second, two species of 4 2 6 squirrels and many species of birds were respectively lumped into squirrel or bird groups 4 2 7 in the diversity analysis. Lumping birds into a single group results in a greater 4 2 8 underestimation in diversity on the riparian transect than on the agricultural transect. On 4 2 9 the agricultural transect, the only birds captured were the tufted titmouse, wild turkeys, 4 3 0 and the northern cardinal. In addition to those species, several coopers hawks, blue jays, 4 3 1 wood thrushes, and barred owls were found on the riparian transect.

3 2
The parameters in the best-fitting model for diversity included distance, transect, 4 3 3 maximum and average temperature, leaf-litter depth, transect*maximum temperature, 4 3 4 transect*leaf-litter depth, and distance*leaf-litter depth. Of those parameters, average 4 3 5 temperature and the distance*leaf-litter depth had the strongest effects, in which the 95% 4 3 6 CI of the slopes for each parameter did not cross zero (Table 1A). Diversity decreased 4 3 7 with increased average and maximum temperatures (Fig 5A, B), a trend that is often

5
The parameters in the best-fitting model for biomass included distance, transect, 4 4 6 maximum and average temperature, transect*distance, and transect*maximum 4 4 7 temperature. The slopes of all these parameters had 95% CI that included zero, indicating 4 4 8 that none of the parameters significantly affected biomass. However, biomass increased 4 4 9 gradually from the hay field to the upland forest (Fig 6E), a pattern we would expect if 4 5 0 fewer species are using the hay field [8]. The diversity of mammals and birds found near 4 5 1 the hay field was relatively low (Fig 6B). The concentration of biomass farther from the 4 5 2 hay field and the lower species diversity supports the idea that the field may not be used 4 5 3 by many species. On the other hand, biomass on the riparian transect was higher near the 4 5 4 wetland and at intermediate distances (Fig 6F), and the diversity on the riparian transect 4 5 5 decreased with distance from the wetland (Fig 6C), supporting the idea that many species 4 5 6 are using the wetlands or areas adjacent to the wetlands. Wetlands are likely more useful 4 5 7 for many organisms than the hay field. The wetland provides water, potentially a greater 4 5 8 diversity of plants for herbivores to consume, and a greater diversity and abundance of 4 5 9 prey for predator species. In fact, only on the riparian transect did we observe carnivores.

6 0
The relationship between biomass and temperature was opposite between the 4 6 1 transects (Fig 5E, F). Biomass decreased with temperature on the agricultural transects 4 6 2 and increased with temperature on the riparian transect. This pattern was more 4 6 3 pronounced for maximum temperature (Table 3; Fig 5E, F), which varied more between 4 6 4 the transects than average temperature (Fig 3D, E). Biomass was likely lower at warmer 4 6 5 temperatures along the agricultural transects because plots with higher maximum 4 6 6 temperatures were found near the hay field, where biomass was low.

6 7
An important caveat in the use of camera traps for wildlife studies is that it can be 4 6 8 difficult to accurately determine density and abundances of animals that cannot be 4 6 9 individually identified. Species with unique individual markings (e.g., tigers and jaguars) 4 7 0 can be used in mark-recapture methods [22,49]. When researchers are unable to identify 4 7 1 individuals from markings, it becomes impossible to determine if each photograph 4 7 2 represents separate individuals. However, researchers can take measures to reduce 4 7 3 counting an individual more than once. We set our camera traps to have a five second 4 7 4 delay after a photo was taken and when more than one image of the same species (or 4 7 5 genus for Sciurus and Peromyscus) were taken within 10 min of each other at the same 4 7 6 plot, they were considered the same individual, unless they were was obviously different 4 7 7 individuals (e.g., different body size, antlered versus non-antlered deer). Many applied to animals that can be individual identified, but still provide comparably reliable 4 9 0 estimates on "unmarked" individuals. Indeed, their ability to capture rare and cryptic 4 9 1 species far out-stripes traditional methods. Finally, while camera traps have been less 4 9 2 utilized to non-invasively determine phenotypic traits, their ability to do so can greatly 4 9 3 increase their capabilities as an ecological tool [29]. In our study, we used camera traps to 4 9 4 estimate body size and, hence, the amount of biomass moving through each of the plots.

9 5
As far as we know, this is the first study to utilize camera traps to obtain body sizes, and 4 9 6 we conclude that future studies using camera traps can obtain body size and incorporate 4 9 7 that information into their analyses.

9 8
In conclusion, we found that camera traps worked to estimate diversity, richness, 4 9 9 abundance, and biomass of large and small mammals and birds. We found slight, non-5 0 0 significant differences in diversity, abundances, and biomass between a transect that 5 0 1 traverses from a wetland to an upland forest and a transect that traverses from an 5 0 2 agricultural field to an upland forest in eastern Missouri. Fewer species were found near 5 0 3 the hay field while more species were found near the wetland. Armadillos and coyotes 5 0 4 were unique to the riparian transect, while birds were more common and diverse on the 5 0 riparian transect. Virginia opossums were more common along the agricultural transects, 5 0 6 but no species were unique to the agricultural transect.     Δ AICc, a model weights (w im ) are found on the lower part of the table. We used model-averaging to obtain parameter estimates, which are displayed on the right section of the table.