Assessing how bats respond to habitat attributes requires an integrative approach to reliably predict direct community-level effects. We focused on hipposiderid and pteropodid bats because of their diverse resource use patterns, body size ranges, and dispersal abilities. We combined an array of bat species-level characteristics with key rainforest stand characteristics that may covary with habitat use. Twelve stations were sampled in the Lomami and Yangambi landscapes in the Democratic Republic of the Congo. We investigated whether the species-level flight ability of bats and rainforest stand characteristics can affect bat commuting flights and community-level estimates of both species detection and habitat occupancy. We captured bats for 108 trap-nights. Three sampling events (early evening, middle of the night, and early morning) were replicated for each survey night. Hipposiderids showed an early evening flight peak, while flight activity of pteropodids was constant throughout the night, but increased around the middle of the night. Species capture probability decreased with higher wing loading in hipposiderids and was negatively correlated with higher wing aspect ratio in pteropodids. Forest occupancy of hipposiderids increased along the gradient towards waterways, while pteropodid occurrence was not directly linked to measured forest stand variables. This suggests a consequence of habitat patterns at larger spatial scales, which would need clarifying through additional data collection. We discuss these findings in terms of resource-use strategies of clutter-tolerant and clutter-intolerant species. We argue that the occurrence of specific bat species and their habitat use patterns can serve as surrogate measures of ecosystem health.

The study design tested the status of bats as mobile link organisms that can facilitate within-community variation in species-specific traits such as mobility, home range, and rainforest habitat requirements. This study focused on two bat families, Hipposideridae and Pteropodidae. We used a stratified sampling design to account for bat community composition, including replication of sites across the Lomami and Yangambi landscapes in the Democratic Republic of the Congo. We selected two sites per landscape. To capture local variation, rainforest stand descriptions and bat surveys were conducted using a nested plot design established along transect segments following slopes. This design was based on a set of six (20 × 5 m) sampling points, hereafter “station”, comparable to a forest stand as a sampling unit. At each sampling point, we recorded forest clutter indicators. We recorded stem density, including lianas with diameter at breast height (dbh) ≥ 5 cm, and basal area of woody stems with dbh ≥ 10 cm. We determined the Euclidean distance between each sample point and the nearest waterways.

Bats were captured in Lomami (October 2018-January 2019) and in Yangambi (October 2019-January 2020). We used one set of six traps per station simultaneously. To reduce the effects of night time, observer, and trap shyness, we used a rotating survey scheme between stations within the site. To fully standardize the trapping effort, surveys were repeated at the same station seven days apart during different time periods. We used temporal sampling occasions, henceforth “event”, as part of the night replication. We fixed the sampling time for the fly-out period so that the sampling duration (a 4-h period corresponds to one event) was the same for all events per site:  early evening (1800-2200 h), middle of the night (2200-0200 h), and early morning (0200-0600 h) in Central Africa Time (UTC+0200 h). The total bat survey effort was 108 trap-nights (i.e., 3 trap events × 3 replicate nights × 12 stations). Captured bats were morphometrically measured and released unharmed.

Recorded data on bats and rainforest stand characteristics were pooled at the station level. Bat records from three events were combined into a single-season dataset for statistical robustness. We built two matrices based on 108 trap-nights for each bat family. We evaluated relationships between within-community variation in species-specific traits (aspect ratio, wing loading, and potential commuting time slots) and capture, separately for hipposiderid and pteropodid records (i.e., all 1s for captures and all 0s for no captures). Because aspect ratio and wing loading were continuous variables, time slots were treated as rank variables with values of 1 (early evening), 2 (middle of the night), and 3 (early morning). We used wing loading as a key determinant of hipposiderid flight behavior and aspect ratio as a parameter of pteropodid flight performance to infer commuting flight patterns. We simultaneously examined predictors with Bayesian multivariate binary logistic regression. We used the wald.test function in the aod R package to test for the overall effect of time slots on bat captures. We modeled bat capture probability distributions over three events with response variables of 1 and 0.

We analyzed repeated presence/absence data from a collection of spatial units (stations) over time (events). The model was based on the assumption that non-detection (or non-capture) can be distinguished from absence through repeated sampling, and that using pooled data for all species observed during sampling can improve species-specific occupancy estimates. The models built included two sub-models, namely an occupancy sub-model and a detection sub-model to estimate detection probability to account for imperfect species detection and sparse recording. Since this modeling approach requires a fixed list of possible species, we generated a species occurrence matrix containing the “true” observation state at each station. We included previous unpublished data from both Lomami and Yangambi in the analyses to control for potential false-negative detections. Species observed at least once were assigned an encounter history of 1, otherwise 0 for unobserved but previously recorded, to control for false-positive errors in the estimates. We created separate matrices for bat species characteristics and habitat descriptors. Species-specific covariates were rescaled using z-score normalization before running the models to facilitate model convergence. Values were back-transformed into their original units when discussing model predictions. We fit single-season occupancy models with multi-species data (hipposiderids and pteropodids separately) using the RPresence package. We included species covariates for detection probability and habitat covariates for occupancy probability. We estimated the probability of occupancy at a given station and the probability of detection (or capture) for each species. We selected the most parsimonious model with a change in the corrected Akaike information criterion score (i.e., ΔAIC_c < 2) from a set of candidate models with the lowest AIC_c. Species richness estimates were derived from occupancy probabilities.

Please see the README document and the accompanying published article: Mande C, Van Cakenberghe V, Kirkpatrick L, Laudisoit A, De Bruyn L, Gembu G-C, Verheyen E. Drivers affecting habitat use in Afrotropical hipposiderid and pteropodid bats. Biotropica, Accepted. doi:10.1111/btp.13242