METHODS

Study Site. — Data were collected in a mixed-pine and hardwood stand (hereafter, “pine stand”) at the Lake Louise Field Station (LLFS) in Lowndes County, Georgia. LLFS is a 76.9-ha plot located in a rural area 15 km south of Valdosta, Georgia. The pine stand is a 10.4-ha section of forest consisting predominantly of slash (Pinus elliottii) and loblolly pine (P. taeda). Hardwood forest surrounds the pine stand on all sides, from which it is partitioned by dirt roads. Hardwood encroachment (by water oak [Quercus nigra] and, to a lesser degree, by southern live oak Q. virginiana]) has occurred along the margins of the pine stand, with some pockets of hardwoods throughout the interior. Prescribed burning of the pine stand has been conducted at intervals of 1 to 5 yrs since 1998 (Riggs et al. 2010).

Burrow Survey. — Following a prescribed burn in March 2019, the pine stand was exhaustively surveyed for tortoise and armadillo burrows. A team of 4 surveyors walked the length of the pine stand on a north-to-south bearing in overlapping belt transects, repeating this process from the eastern to the western edges until no unmarked burrow could be located.

Burrows of each species were differentiated by the presence or absence of a large sandy mound, known as an apron, and the shape of the burrow entrance. Tortoise burrows have a large apron and characteristic half-moon shaped entrance (Hansen 1963; Kinlaw and Grasmueck 2012), whereas armadillo burrow entrances are typically circular to oval and have little to no apron (McDonough et al. 2000; Sawyer et al. 2012).

The activity status of each burrow was determined using criteria similar to that described in Cox et al. (1987). Burrows with freshly disturbed soil at the burrow entrance or apron were considered to be active and occupied by the respective excavator species, while inactive (but not abandoned) burrows did not display disturbed soil, were not blocked by debris or spider webs, retained their shape and, for tortoise burrows, did not exhibit an overabundance of vegetation on the apron. Abandoned burrows were those with severely degraded, misshapen, or partially collapsed entrances; blocked by debris or spider webs; and with dense vegetation typically present on the apron of tortoise burrows. As a result of changes in activity status (i.e., active burrows becoming inactive and vice versa) for some burrows during preliminary data collection, active and inactive burrows were grouped into a single category deemed “possibly occupied”.

Burrow locations were georeferenced in Universal Transverse Mercator (UTM) coordinates with 200 fixes per burrow using a centimeter grade Trimble Geoexplorer 6000 Series (GeoXT) handheld GNSS unit. Burrows were physically marked with a ground flag and metal tag displaying a unique identification number.

Camera Trap Configuration. — All camera traps consisted of a Bushnell NatureView HD Cam (Model# 119740) fastened to a bipod camera mount approximately 1 m above the burrow (see Butler 2020 and Lamb 2021). An external viewer, a feature unique to the NatureView HD, was used to ensure that the burrow entrance was the focal point of each camera’s field of view, with some surrounding ground visible on all sides.

We monitored entrances of 10 tortoise and 10 armadillo burrows (n = 20) from October 2019 to December 2020. Camera traps were evenly distributed across the pine stand using a random burrow selection function applied in conjunction with a quadrat system, using custom scripts in R (R Core Team 2021). We first overlaid a 3 × 4-ha quadrat grid onto a shapefile of the pine stand using the “quadrats” function within the Program R package ‘spatstat’ (Baddeley et al. 2015). The random selection function was then employed to select one armadillo burrow and one tortoise burrow for monitoring within each quadrat. During the random selection process, the two most southeastern quadrats were combined with the quadrats on their northern border because of the limited number of available cameras, limited number of existing burrows within those quadrats, and small quadrat area (Fig. 1).

View Figure

• Download as Powerpoint
• Download Figure

To assess the effects of burrow activity status on associate event counts, we selected one burrow of each status (i.e., possibly occupied and abandoned) so that burrows of opposite type and status were monitored within each quadrat (Fig. 1). The number of burrows monitored was limited by camera trap availability. In total there were 4 combinations of burrow type and activity status with 5 replicates of each. All camera traps remained at their assigned burrows for the duration of the sampling period.

Larger burrow entrances are correlated with longer burrow lengths (Hansen 1963), and ultimately a greater volume of space; therefore, only adult tortoise burrows (entrance width > 23 cm; Alford 1980) were considered eligible for monitoring. Similarly, for armadillos, the distinction between feeding burrows (length < 25 cm; McDonough et al. 2000) and shelter or nest burrows (length > 25 cm) was made during our initial survey and feeding burrows were excluded from monitoring efforts.

Camera parameters were standardized across all cameras (see Butler 2020 and Lamb 2021 for details), which were set to operate 24 hrs/d, recording a sequence of two images and a 10-sec video, consecutively and instantaneously, upon trigger, with a 2-sec interval (i.e., wait time before next trigger can occur) between triggers. All images and videos were time stamped.

Memory cards were exchanged every 2–4 wks, photos and videos were reviewed by eye, and vertebrate animals were identified to species, when possible, using field guides and confirmation from experts. There was limited visibility of surrounding ground around burrow entrances in each camera’s field of view (∼ 15 cm), so we considered any vertebrate present in an image or video to be visiting the burrow for some purpose. Many photos could not be identified to species as a result of poor image quality or because identifying characteristics of the animal were out of frame. In these instances, vertebrates were identified to the lowest possible taxonomic level. For statistical analyses, we grouped species into 3 taxonomic classes: birds, mammals, and herpetofauna. We also assigned photos to feeding guilds: carnivore, insectivore, omnivore, herbivore, and granivore (as determined from animaldiversity.org [University of Michigan 2020]) to investigate the potential of burrows as foraging locations.

A database of event counts was generated using the ‘recordTable’ function in “camtrapR” (Niedballa et al. 2016), with consecutive observations of the same species condensed to a single event for triggers occurring within 10 min, because of our inability to distinguish individuals of both excavator and associate species. Burrows were compared using 4 measures: 1) richness of vertebrate taxa, 2) event frequency (counts), 3) Shannon–Wiener diversity index (H), and 4) relative abundance index (RAI; number of visits by a species per 100 trap-nights; O’Brien et al. 2003). No animals were handled during this study.

Environmental Data. — Microhabitat data were collected at each monitored burrow for use in multivariate analyses to identify environmental factors that influence associate event counts. The following variables were chosen based on microhabitat characteristics investigated by Bhandari (2019):

1. Light intensity was measured at ground level using a lux meter. Five measurements were taken for each burrow (one directly above the burrow entrance and at 2.5 m from the entrance to the north, south, east, and west) on clear sunny days between 1200 and 1400 hrs and measurements were averaged for each burrow;

2. Similar to the procedure for light intensity, proportion of canopy cover was evaluated using a canopy densitometer to visually estimate 5 canopy cover measurements (see criteria for light intensity data collection) and averaged for inclusion in analyses;

3. Visibility was estimated using a 43 × 31-cm black-and-white checkerboard placed at the burrow entrance at ground level and calculated as the percent of visible black squares. Four visibility estimates were taken at a distance of 5 m from the north, south, east, and west of each burrow and averaged before converting to proportions;

4. Distance from the nearest habitat edge was calculated using the ‘nncross’ function in “spatstat”. This was included as a covariate because the pine stand’s edges represented ecotonal habitat between pine and hardwood forest where we expected greater associate event counts because of a greater abundance of resources, greater habitat variation, and increased vertebrate diversity in these areas (Harris 1988). We also expected a negative relationship between associate event counts and increasing distance from edges; and

5. Vegetative density measurements were derived from LiDAR data, downloaded as a LAS file from the National Oceanographic and Atmospheric Administration Office for Coastal Management. LAS data were processed in ArcGIS 10.4.1, with the aid of the LAS Point Statistics as Raster utility and Spatial Analyst Toolbox. Vegetative density was measured as the proportion of above ground returns in 10 × 10-m pixels. The vegetative density raster was saved as a TIFF file, imported into R via the “rgdal” package (Bivand et al. 2021), and then into “spatstat” where vegetative density data were extracted at each camera position.

Beyond these 5 variables, we included the following variables that we suspected might influence associate event counts: burrow type (tortoise or armadillo), burrow activity status (possibly occupied or abandoned), burrowing density (number of burrows per unit area), richness of tree species, proportion of hardwood trees, and RAI of excavator species (i.e., combined number of visits by tortoises and armadillos per 100 trap-nights at individual burrows as an indicator of the degree to which burrows were maintained). UTM easting and northing coordinates were also included as covariates to test whether there were linear trends in the response variables along the X or Y axis.

Burrowing density was calculated by creating a point pattern of all burrow locations then using the ‘density’ function in “spatstat” (with the default method of bandwidth selection based on the study area geometry) to return a kernel-smoothed surface representing burrowing density (i.e., number of burrows per meter) across the pine stand (Fig. 1). Burrowing density values at monitored burrows were then extracted for use in our analyses.

Tree species richness and proportion of hardwoods at monitored burrows were determined by universal Kriging interpolation (Oliver and Webster 1990) using lattice data collected at 186 points within the pine stand. Interpolated surfaces of tree richness and hardwood proportion were generated using the ‘krige’ function within the R package “gstat” (Pebesma 2004). Surface maps were converted to raster layers and values were extracted at monitored burrow locations using the ‘extract’ function within the package “raster” (Hijmans 2020).

Statistical Analysis: Differences in Associates Between Burrow Types. — Richness between burrow types was compared using a Mann-Whitney U-test because the vertebrate richness was nonnormally distributed across armadillo burrows. A Student’s t-test was used to compare mean Shannon-Wiener diversity between armadillo and tortoise burrows because these values were normally distributed and homoscedastic. We used Levene’s tests to check for equal variance of species richness and Shannon-Wiener diversity between burrow types.

To examine differences in associate event counts between burrow types, we tested for fit to a proportional model using Chi-square tests where expected counts were determined using the proportion of total trap-nights at each burrow type within the R function ‘chisq.test’. For these analyses, all observations of tortoises at tortoise burrows and armadillos at armadillo burrows were removed from the data set so that only observations of associate taxa remained.

Statistical Analysis: Differences in Community Composition Between Burrow Types. — To assess whether armadillo and tortoise burrows supported different vertebrate communities, we used nonmetric multidimensional scaling (NMDS) on a distance-based measure of dissimilarity. A data matrix of Bray-Curtis dissimilarity values (Bray and Curtis 1957), displaying pairwise dissimilarity between all possible pairs of monitored burrows, was generated from individual burrow RAI values for each observed taxon using ‘metaMDS’ from the R package “vegan” (Oksanen et al. 2020), and plotted using “ggplot2” (Wickham 2016). Homoscedasticity was checked using the ‘betadisper’ function in “vegan”, a multivariate analog to Levene’s test for equal variance.

Significance testing was performed using a permutational analysis of variance (PERMANOVA) specific to distance matrices. We used the ‘adonis’ function from “vegan” with 999 permutations to test for a statistical difference in mean dissimilarity between community composition of armadillo and tortoise burrows.

Statistical Analysis: Burrow and Microhabitat Effects on Associate Event Counts. — To examine relationships between burrow and microhabitat predictors and measures of vertebrate event counts, respectively, we performed correlational principal components analyses (PCA) using the ‘princomp’ function within the R package “MASS” (Venables and Ripley 2002).

To quantify the effects of changes in predictor variables on associate event counts, we used linear models (LMs) for continuous variables and Gaussian, or Poisson, generalized linear models (GLMs) for discrete variables, depending on the distribution of each model’s residuals. Response variables tested against predictor variables in our regression models included vertebrate richness, Shannon-Wiener diversity index, and 3 categories of event counts: total associate events, events by each taxonomic class, and events by each feeding guild. For models that used taxonomic class and feeding-guild event counts as response variables, we removed armadillo observations from mammal and omnivore counts occurring at armadillo burrows and tortoise observations from herpetofauna and herbivore counts occurring at tortoise burrows. Log trap-nights were included as an offset term in all models that used count data as a response variable to account for variation in sampling effort (due to camera malfunctions, battery depletion, and user error). Stepwise forward and backward model selection was conducted by employing Akaike Information Criterion (AIC; Akaike 1974) using the ‘stepAIC’ function within the R package “MASS” to reduce the number of variables included in each model and thereby only retain predictor variables that were informative. Residual plots were used to determine that the assumptions of LMs and GLMs were met. The significance level was set to α = 0.05 for all statistical tests.

RESULTS

Burrow Survey. — A total of 234 burrows were recorded within the pine-stand (Fig. 1). Armadillo burrows (n = 145) were more numerous than tortoise burrows (n = 82) and 7 tortoise burrows were identified as co-opted (i.e., consistently occupied and maintained by an armadillo). Of the tortoise burrows, 57 were possibly occupied, 17 were abandoned, and activity status was uncertain for 8 burrows. For armadillos, possibly occupied burrows (n = 108) were also the most abundant category. Twenty-five armadillo burrows were identified as abandoned and activity status was uncertain for the remaining 12 burrows.

Camera Trap Data. — Camera traps were operational for 7064 trap-nights over the 14-mo sampling period. Sampling effort per burrow ranged from 174 to 405 trap-nights (mean ± SD = 353.2 ± 14.37 trap-nights). Trap nights varied as a result of camera malfunction, battery depletion, and user error. Camera traps were triggered 18,302 times, representing 7281 burrow events from 40 vertebrate taxa (Table 1). In total, 14 mammal, 14 bird, 9 reptile, and 3 amphibian species were observed visiting burrows, comprising 64.5% (n = 4694), 14.4% (n = 1049), 21% (n = 1533), and 0.05% (n = 4) of identifiable events, respectively. Twenty-eight taxa were observed utilizing both burrow types, while 3 were only observed at armadillo burrows and 9 were only observed at tortoise burrows (Table 1). Additionally, we observed 11 taxa that had not been previously documented utilizing armadillo burrows and 4 that had not been documented utilizing tortoise burrows (Table 1). Grouping taxa by feeding guilds, 1.4% of observations were of carnivores (n = 101), 0.4% granivores (n = 28), 17.3% herbivores (n = 1210), 17.9% insectivores (n = 1258), and 62.8% omnivores (n = 4394).

Table 1. Table of taxa observed visiting tortoise and/or armadillo burrows. For each taxon, its feeding guild and total number of observations (n. obs.) is provided, along with its relative abundance index at armadillo burrows (RAI ARM), and relative abundance at tortoise burrows (RAI GT).

View Fullscreen

The most frequently observed associate taxa across all burrows were mice (Muroidea spp.), hispid cotton rats (Sigmodon hispidus), and Carolina wrens (Thryothorus ludovicianu; Table 1). At armadillo burrows the most frequently observed associate taxa were also mice, hispid cotton rats, and Carolina wrens, while at tortoise burrows mice, armadillos, and hispid cotton rats were the most observed associates (Table 1).

Event frequency was greater across tortoise burrows (n = 4638, 63.70%) for both tortoises (n = 957, 13.14%) and associate taxa (n = 3678, 50.52%) compared with armadillo burrows that were visited less frequently (n = 2643, 36.30%) by armadillos (n = 182, 2.50%) and associate taxa (n = 2459, 33.77%). Events per burrow (including visits from respective excavator species) ranged from 48 to 585 for armadillo burrows (264.10 ± 47.24 visits), and 59 to 1372 for tortoise burrows (463.50 ± 113.07 visits), while events by associate taxa ranged from 46 to 554 for armadillo burrows (245.90 ± 46.05 visits) and 57 to 655 for tortoise burrows (367.80 ± 55.77 visits).

Burrow Comparisons. — Armadillo burrows were visited by 31 taxa and richness ranged from 12 to 17 species (14.50 ± 0.67), while 37 taxa were observed visiting tortoise burrows, with richness ranging from 10 to 25 (17.30 ± 1.41). Shannon-Wiener diversity ranged from H = 1.49 to H = 2.09 across armadillo burrows (1.77 ± 0.06), and H = 1.20 to H = 2.24 across tortoise burrows (1.79 ± 0.11). However, these differences were not statistically significant for median species richness (U = 29.5, p = 0.1) or mean Shannon-Wiener diversity (t = −0.13065, df = 18, p = 0.8). In contrast, there was strong evidence against a proportional model for total associate events between burrow types (χ² = 453.46, df = 1, p < 0.001), which was greater than expected at tortoise burrows (r_p = 15.59) and less than expected at armadillo burrows (r_p = −14.5; Table 2). Comparisons of associate event counts for taxonomic classes and feeding guilds were also significantly greater than expected at tortoise burrows and less than expected at armadillo burrows (Table 2), but only after removing an outlier data point attributed to what was likely a single juvenile tortoise, identified by its relative size and short durations between events, that produced 108 observations at an armadillo burrow between 12 and 29 June 2020, disproportionately increasing herpetofauna and herbivore counts.

Table 2. Chi-square Goodness of Fit (to a proportional model) table listing expected vs. observed event frequency, Pearson standardized residuals (r_p), χ² statistics, and p-values for associate species classes and guilds between armadillo (ARM) and tortoise (GT) burrows.

View Fullscreen

Measures of associate events were generally greater for tortoise burrows, and showed greater range as opposed to armadillo burrows, which had more narrowly distributed ranges. Despite associate events at tortoise burrows accounting for a greater proportion of the total event count, ratios of event counts by taxonomic classes and feeding guilds were relatively similar for each burrow type, respectively. For class-level event counts, the deviation between expected and observed frequencies was greatest for mammals and birds, both of which showed greater than expected event counts at tortoise burrows (Table 2). Deviation between expected and observed event counts at the guild level was greatest for omnivores and insectivores, which were also present at tortoise burrows at greater than expected frequencies (Table 2).

There was little evidence of dissimilarity between armadillo and tortoise burrow community composition. Nonmetric multidimensional scaling (NMDS) square-root-transformed Bray-Curtis dissimilarity values, producing a slightly greater than desired stress value of 0.22 after 20 runs. The results of our PERMANOVA showed community dissimilarity between burrow types was only marginally significant (p = 0.08) and substantial overlap between point clouds representing armadillo and tortoise burrow communities was observed in a spider plot (Fig. 2), indicating little difference in community composition.

View Figure

• Download as Powerpoint
• Download Figure

Environmental Effects on Burrow Event Counts. — Relationships between variables were visualized using PCA biplots (Fig. 3). For predictor variables (Fig. 3A), the first two principal components (PCs) only accounted for 41% of the variation in microhabitat data, and no strong correlations (r > 0.8) were found between predictor variables. For response variables (Fig. 3B), PC1 and PC2 accounted for 68% of the variation, and strong correlations existed between insectivore event counts and bird event counts (r = 0.97, df = 18, p < 0.001), omnivore and mammal event counts (r = 0.99, df = 18, p < 0.001), omnivore and total associate event counts (r = 0.97, df = 18, p < 0.001), and mammal and total associate event counts (r = 0.97, df = 18, p < 0.001). These relationships were likely due to significant overlap between contributions from each category of associates (i.e., insectivore observations were mostly of birds, omnivore observations were mostly mammals, and the total count of associate events was heavily influenced by mammals).

View Figure

• Download as Powerpoint
• Download Figure

Effects of burrow location were detected in regression analyses. Specifically, easting positively affected vertebrate richness (β = 0.025, t = 3.42, p = 0.003) and herbivore event counts (β = 0.025, t = 2.91, p = 0.01), while northing showed varied effects; it positively affected mammal (β = 0.69, t = 2.56, p = 0.03), insectivore (β = 0.31, t = 2.68, p = 0.02), omnivore (β = 0.65, t = 2.45, p = 0.04), and total associate event counts (β = 0.97, t = 2.54, p = 0.03), but negatively affected diversity (β = −1.44 × 10⁻³, t = −3.18, p = 0.008) and herpetofauna (β = −0.15, t = 3.74, p = 0.003) event counts.

Burrows that were assigned a possibly occupied activity status during our initial survey showed consistently positive effects on several response variables, including vertebrate richness (β = 2.74, t = 2.22, p = 0.04), total associate event counts (β = 106.6, t = 2.57, p = 0.03), mammal (β = 122.8, t = 4.39, p = 0.003), and omnivore event counts (β = 112.6, t = 4.11, p = 0.004). However, herpetofauna event counts were negatively affected for possibly occupied burrows (β = −10.26, t = −2.32, p = 0.04). Consistent positive effects were detected for tortoise burrows for vertebrate richness (β = 4.22, t = 3.17, p = 0.006), diversity (β = 0.19, t = 2.37, p = 0.03), total associate event counts (β = 202.9, t = 5.56, p = 0.004), mammal (β = 179.8, t = 6.23, p ≤ 0.001), carnivore (β = 0.53, t = 2.29, p = 0.02), herbivore (β = 4.39, t = 3.01, p = 0.01), and omnivore event counts (β = 169, t = 5.97, p ≤ 0.001). Richness of tree species positively affected bird (β = 68.29, t = 2.64, p = 0.02), herpetofauna (β = 34.28, t = 3.36, p = 0.007), insectivore (β = 96.03, t = 3.27, p = 0.01), and total associate event counts (β = 266.8, t = 2.8, p = 0.02).

Canopy cover consistently showed negative effects on associate event counts. Increased canopy cover negatively affected bird (β = −1.15, t = −2.95, p = 0.01), herpetofauna (β = −42.9, t = −2.49, p = 0.03), carnivore (β = −2.19, t = −1.96, p = 0.05) and insectivore (β = −1.48, t = −3.18, p = 0.01) event counts. A greater proportion of hardwood trees negatively affected bird (β = −260.2, t = −2.39, p = 0.04), herpetofauna (β = −64.11, t = −2.29, p = 0.04), and insectivore (β = −334.7, t = −2.77, p = 0.02) event counts, but positively affected carnivore event counts (β = 2.52, t = 2.69, p = 0.007).

DISCUSSION

Our study is the first to compare vertebrate associates between armadillo and tortoise burrows in a syntopic population. In the southeastern United States, the armadillo has been largely regarded as a nuisance species. However, our study, along with others (Taber 1945; Clark 1951; Platt et al. 2004; Butler 2020; DeGregorio et al. 2022) supports the concept that armadillo burrows are utilized by associate species. Although armadillos may be beneficial in areas where tortoises are absent or have been extirpated, their removal may still be warranted in areas where tortoises occur because of their negative impacts on tortoise populations (Smith et al. 2012). Additionally, the results of the present study suggest tortoise burrows have a greater potential impact on local biodiversity, and restoring tortoise populations may provide a greater ecological benefit than allowing armadillos to colonize historical tortoise habitat because the effects on local biodiversity from armadillo burrows alone remains unclear.

Our results differed from other camera-trapping studies that investigated tortoise burrow associates because mammals and birds accounted equally for the majority of species richness whereas another recent study found bird richness to be greater than mammal richness (Dziadzio and Smith 2016), and another reported herpetofauna richness to be greater than bird richness, with mammal species being least numerous (Murphy et al. 2021). Conversely, a recent study of desert tortoise (Gopherus agassizii) burrow associates reported similar results to ours, showing species richness was equally contributed by mammals and birds but less so by herpetofauna (Agah et al. 2017). For armadillo burrows in our study site, birds and mammals again contributed nearly equally toward overall species richness. This differs slightly from the results of two other camera-trapping studies of armadillo burrow associates, one of which found birds contributed the most toward species richness, with mammals and herpetofauna being represented equally but less than birds (Butler 2020), and another that reported birds were the richest class of associates, followed by mammals and then herpetofauna (DeGregorio et al. 2022). These differences may be due to variations in geographic location and habitats where studies were conducted, although any causal links remain unclear.

Although richness, diversity, and community composition overlapped between burrow types, tortoise burrows were visited at greater-than-expected frequency by all vertebrate classes and feeding guilds, suggesting many species had stronger preferences for tortoise burrows. Although a lack of statistical significance was found for bivariate tests between burrow types, after controlling for other microhabitat variables, positive effects were detected for tortoise burrows in regression models for increased richness and diversity. The results of bivariate tests may have been confounded by prescribed burning of the pine stand, which may have caused armadillo burrows to become more accessible to a greater number of species because their burrow locations are typically associated with dense vegetation (Bhandari 2019). This implies armadillos may not be perfect ecological surrogates of tortoises in open-canopy pine-forest ecosystems. However, burrow-specific proportions of associate events were consistent between burrow types, and associate vertebrates may have more opportunities to benefit from armadillo burrows in our study site because they were more numerous (Fig. 1).

Associate event counts were affected by several microhabitat and burrow characteristics. Although not consistent across all responses, burrow variables that repeatedly showed positive effects on associate event counts were burrow type (= tortoise burrows) and burrow status (= possibly occupied). Most likely, these results reflect the necessity of high-quality refugia in this environment. Although not measured in our study, tortoise burrows have greater overall volume than armadillo burrows, allowing occupancy by more individuals, and are created from a greater degree of bioturbation, which may have increased micronutrient availability and led potentially to greater increases in local biodiversity (Lohrer et al. 2004; Meysman et al. 2006; Erwin 2008). Additionally, greater thermal gradients associated with the greater depths of tortoise burrows may increase associate preference for them during the summer months.

Microhabitat variables influenced vertebrate assemblages visiting burrows because increased canopy cover and proportion of hardwood trees negatively affected bird and herpetofauna event counts. This may be due to these variables obscuring aerial visibility of burrows for birds, which rely on vision for foraging (Fernandez-Juricic et al. 2004), as well as decreased thermal stress from shading effects making burrows less of a necessity for herpetofauna (Taylor et al. 2020). As in studies of associates of other Gopherus species, mammals were observed at burrows frequently (Lovich et al. 2014; Agha et al. 2015, 2017). Mammalian event counts were higher than expected at tortoise burrows, and positively associated with possibly occupied burrows, suggesting that mammals observed in our study showed greater interest in well-maintained tortoise burrows. Potentially, burrows that were identified as possibly occupied are higher quality sources of refugia for mammals, are more easily detectable, or may be a preferred foraging resource. Herpetofauna event counts were also higher than expected at tortoise burrows, although the deviation from the expected event frequency for herpetofauna was much lower when compared with that of birds and mammals. The positive association between herpetofauna event counts and abandoned burrows may imply the occurrence of negative interactions between herpetofauna and burrow engineers. Additionally, this may suggest that herpetofauna have a preference for an increased level of debris (e.g., leaf litter) within burrows. Lastly, this may indicate that herpetofauna are less reliant on burrows as a food resource (because abandoned burrows will not produce the cascading effects provided by tortoise dung) or as refugia (because other sources of refugia that are similar in quality to an abandoned burrow are likely available). This result is consistent with results from Murphy et al. (2021), which found that most reptiles in their study were positively associated with pine stump holes rather than maintained tortoise burrows, suggesting that reptiles are reliant on other refuges.

Microhabitat variables also affected burrow event counts by feeding guilds. Associate event counts of carnivores and omnivores were positively associated with tortoise burrows, particularly those that were possibly occupied, which is consistent with the concept of burrows serving as a food resource for these guilds facilitated by cascading effects of tortoise dung increasing the abundance of coprophagus arthropods and, in turn, animals within higher trophic levels (Young and Goff 1939; Milstrey 1986). Burrow event counts for herbivores were positively associated with burrows near habitat edges and tortoise burrows, but were not associated with activity status. This result may be due to the increased diversity of plant species associated with the ecotonal edges of the pine stand, and the lack of association with activity status suggests herbivores are not reliant on burrows for foraging. However, Kaczor and Harnett (1990) reported increased recruitment from seed and greater cover of forbs on tortoise aprons compared with nonexcavated plots, which may have influenced herbivore event counts in our study. Finally, burrows with greater canopy cover showed a negative association with carnivores and insectivores, possibly due to decreased thermal stress from shading leading to decreased burrow use by prey species.

It is clear that both armadillo and tortoise burrows provide important resources supporting local biodiversity and maintaining ecosystem processes (Hacker and Gaines 1997; Hooper et al. 2005). Such effects represent a benefit to the increasing presence of armadillos in the United States because many tortoise burrow associates in our study were observed visiting armadillo burrows. With tortoise abundance continuing to decline (Hermann et al. 2002; McCoy et al. 2006; Smith et al. 2006), there is potential for ecological redundancy provided by armadillos (Lawton and Brown 1994). However, we recommend priority be given toward restoring tortoise populations and habitat because tortoise conservation is more in line with the preservation of native ecosystems and the tortoise would serve as a more effective umbrella species for several other vertebrates of conservation concern (Diemer and Speake 1983; Jackson and Milstry 1989; Layne and Jackson 1994; Roznik and Johnson 2009; Johnson et al. 2017).

Future studies of burrow associates using camera traps can be improved by monitoring greater numbers of burrows, including burrow dimensions (depth, entrance height and width, and length), time since excavation, soil nutrient availability, and temperature as explanatory variables in regression analyses, and should include monitoring of random sites that may be needed to accurately assess the effects of burrows on local biodiversity.

Our study provides a framework for investigating the community-level effects of an introduced excavator and enhances our understanding of interactions between medium-sized burrowers in novel ecosystem networks and the effects of syntopic burrowing species on local biodiversity. Additional research is needed to make informed decisions about armadillo control efforts and optimize ecological functioning of open-canopy pine forests.