Methods
We collected data on synanthropic free-tailed bat building use in the Taita-Taveta county of south-eastern Kenya (East Africa), between February and April 2022 (Figure 1). Topography of the county spans low-lying savannah plains to mountain ranges, with elevations between 700 m (in the plains), up to 2200 m (highest peak of the Taita Hills)40. Human land-use in this county spans a gradient of development, from rural areas with predominantly traditional building practices, to more urbanised areas in the townships of Mwatate, Maktau, Voi, and Wundanyi, with predominantly modern building practices. Bat-borne viruses of zoonotic interest have been detected from free-tailed bats in this area, including coronaviruses and Bombali virus (genus: ebolavirus ) 41–43.
We assessed buildings for synanthropic free-tailed bat occupancy within ten, 1x1 km sites (Figure 1). Sites were chosen to represent the gradient of human landscapes utilised by free-tailed bats (traditional-style housing to modern-style housing, Figure 2), and each was centred on a single building roost identified to contain synanthropic free-tailed bats. Building footprint maps, derived from satellite imagery collected in 2020 and 2021, were used to identify all buildings within sites 24. As public attitude towards bats is negative in this region – being associated with witchcraft and witch doctors (Mwasi & Mwakachola pers. comm.) – we endeavoured to become familiarised and trusted by the community prior to surveying sites and conducted all surveys with a local field assistant.
We individually evaluated each building within our 1 km2 sites for occupation by synanthropic free-tailed bats. To do this, we (1) asked building owners whether they had recently seen or heard bats inside the building during the day, (2) assessed the building for signs of bat use (e.g., bat faeces on the ground, stained ceiling, staining around external roost entrance points, and smell, Appendix S1), and where possible (3) assessed the building for physical presence of bats, for sighting and/or auditory confirmation. We noted features of each building that could impact roosting suitability for synanthropic free-tailed bats (building type, roof design, and, where possible, presence of a ceiling) (Figure 2). We classified modern buildings as those with walls and flooring built from finished materials (e.g., cement and tiles), and traditional as those built from unfinished materials (e.g., compacted earth). Buildings were considered occupied if: the building owners confirmed bat occupation, the building had signs of bat occupation, or bats were seen or heard inside the building. While multiple bat species can occupy buildings in this region, synanthropic free-tailed bats are the most common and are distinctive in their building use (Appendix S1). Buildings that were occupied by species other than free-tailed bats were noted, but not included in analyses of bat building use.
To identify building-level attributes of bat building use, we modelled the response in building occupation relative to 1) building type (modern or other), and 2) roof structure (triangular or other). Models were generalized linear models (GLMs) with maximum likelihood (ML) estimation and a binomial distribution with a logit-link, fit using the mgcv package in R. We performed checks of standardised residuals to evaluate model fit, as per Wood (2017) 44. Note that these indicators of occupation may reflect past or current occupation by synanthropic free-tailed bats, but nevertheless provide an indication of building suitability.
To identify landscape-level attributes of bat-human exposure risk, we modelled response in the number of occupied buildings per site, relative to 1) the total number of buildings available, 2) the proportion of those buildings that were a modern-build style, and 3) the proportion of those buildings with a triangular roof style. To better reflect the landscape of active roosts (and therefore, the landscape of human exposure risk), we ran models on an additional dataset where building occupation was rated based on the collective weight of evidence indicating current free-tailed bat occupation; 0=very unlikely; 1-2=possible; 3=likely; 4=very likely; 5=certain (Appendix S1). Highly weighted evidence for occupation included sighting and/or auditory confirmation of free-tailed bats by the authors (inclusion into category 5, certain of occupation). Moderately weighted evidence for occupation included owner confirmation of occupation, and signs of occupation. Low-weighted evidence included building suitability, as per findings relating to building-level attributes, described above. Buildings were considered presently occupied if they were categorised as four or greater (occupation very likely). All levels of evidence were evaluated in addition to knowledge on where and how many bats were roosting, to indicate synanthropic free-tailed bats, as detailed in Appendix S1. Models were fitted as above, but with a Poisson distribution and log link.
To provide an empirical estimate on landscape-scale building-roost density, we calculated density as: 1) the total number of occupied buildings (occupation category >=4) divided by the total site area, and 2) the average of fixed-bandwidth kernel estimates, estimated using the spatstat package in R 45. Kernel estimates have the advantage of explicitly incorporating the distribution of buildings into the density estimate, and can therefore account for spatial heterogeneity in building aggregation46. Kernel values were estimated using roost building location with Gaussian kernel smoothing and a smoothing bandwidth of 0.347. Bandwidth was selected by comparing projected kernel density values to expected density values based on building distances and survey area. Kernel averages were calculated per site (pixel size = 0.008969 x 0.00896 meters).
Given the specific interest in the transition from traditional- to modern-style housing, all analyses were repeated on a subset of data that included traditional- and modern-style houses only, and houses with triangle and flat roofing only (Appendix S3).