Urbanization is one of the most pervasive landscape transformational processes responsible for novel selection agents promoting functional community homogenization. Bats may persist in those human environments, but the mechanisms responsible for their adaptability and the spatial scales in which landscape imposes environmental filtering remain poorly studied in the Neotropics. We tested the hypothesis that landscape composition interacts with the spatial scale to affect the functional diversity of phyllostomids in an urban-rural gradient. Based on functional traits, we calculated indices of functional richness, divergence, evenness, community-weighted means of morphological traits, and classified species into functional groups. We evaluated the changes of those variables in response to forest, grassland, and urbanized areas at 0.5, 1.25 and, 2 km scales. The number of functional groups, functional richness, and functional evenness tended to be higher in areas far from cities and with higher forest cover, whereas functional divergence increased in more urbanized areas. Our results show that the mean value of wing loading in the assemblage was negatively associated to landscape transformation at several spatial scales. However, environmental filtering driven by grass cover was particularly robust at the 500 m scale, affecting big-sized species with long pointed wings. Retaining natural forest in cattle ranging systems at ~12 km² appears to favor bat abundance evenness among functional types in the urban-rural landscape. Recognizing the scale of the effect on phyllostomid functional responses appears to be a fundamental issue for elucidating the spatial extent to which phyllostomid conservation planning in urban-rural landscapes should be addressed.

We selected five sampling sites with different landscape compositions and structures, hereafter named landscape units (LU). An LU refers to the spatial extent from which we measured landscape metrics, whereas a site refers to the centroid of each LU in which phyllostomid bats were surveyed. Sites were selected based on the following criteria: (a) enough tree cover to place mist-nets with canopy cover above 80% and (b) a separation between any two sites of at least 4.3 km to reduce spatial autocorrelation bias. At each site, we set two to three mist-nets of 12×3 m, separated at least 50 m from each other. Each mist-net remained open from 1730 h until 0100 h for every field survey. We visited each site five times at similar times during the extent of the study, five replicates of observations per site, selecting nights with moon phase levels lower than 60% . Sampling effort at each site ranged from 288 to 300 m²×nights, and the total sampling effort was 1,484 m²×nights.

To avoid bias in the quantification of species abundance and to assess the independence of bat captures among mist-nets at each site, we marked all captured individuals with a unique numerical code using a rabbit tattoo. This mark was placed in the right wing at the lower part of the plagiopatagium. We identified all individuals to species level using taxonomic references based on external characters. Those individuals that we could not identify in the field were collected for accurate taxonomic identification using dental and cranial characters, and by comparison with specimens in the Museo de Historia Natural – Universidad de Los Llanos, MHN-U.

Bat functional traits — We used one categorical (trophic guild - Tg) and four morphological traits related to body size: (body mass – Bm) and wing morphology (wingspan - Ws; aspect ratio - Ar; wing loading - Wl). We used information from the literature to assign trophic categories to each bat species. This categorical assignment was confirmed by collecting samples of feces from captured individuals.

We placed the bats in a paperboard tube and measured Bm using a digital balance (0.1g precision). Wing morphological traits were measured as follow: (a) we took a photo of the left wing fully extended in a ventral position against a paper with millimeter scale. Then, (b) those images were processed using ImageJ 1.6 (Abramoff et al., 2004) to measure Ws and wing area as described by Norberg & Rayner (1987). Finally, (c) Ar was calculated as the quotient of the Ws squared and wing area, and Wl was estimated by dividing Bm into the product of gravitational acceleration, 9.81 m/s², and wing area.

Land use classification and predictor variables — We processed DigitalGlobe satellite images from Google Earth taken between 2014 and 2015, with a resolution of 2 m² per pixel, to characterize landscape composition and structure. We used a supervised classification method of ArcGIS 10.1 to generate maps with three land uses: built-up areas, grasslands, and forest. We did not include in the analysis other land uses and landscape elements such as crops, water bodies, or bare ground because they covered less than 5% of each area at a 2 km radius. We defined concentric buffers around the sampling sites to delimit landscape units (LU) at three spatial scales, 0.5, 1.25 and 2 km of radius. At each spatial scale we measured the percentage of built area, grassland, and forest cover, number of forest patches, mean forest patch size, forest patch size standard deviation, and forest patch densities. 

We used LINKTREE analysis to estimate the number of functional groups for the whole assemblage. We calculated indices of functional richness, evenness, divergence and CWM using phyllostomid numerical traits. Then, using GLMs, we modeled the changes in those variables in response to landscape composition and structure at each spatial scale.