Statistical analyses
To evaluate how roost features at different scales influence tree-level abundance and 3-D density of bats, the measures most likely to be relevant for transmission of virus between tree-roostingPteropus , we fitted generalized additive models (GAMs) with restricted (residual) maximum likelihood (REML) estimation, Poisson distribution with a log link, and random effects of roost site, subplot and survey session with the mgcv package in R (Wood 2017). Roost site and subplot were fitted with random effects smoothers to account for high variability within and between roost sites, and survey session with a cyclic cubic regression spline to allow for seasonal variation. We accounted for nesting of subplots within roosts by including an autoregressive model for errors in the model (Yang et al. 2012; Laurinec 2017). We constructed a candidate set of GAMs comprising of a null model of random site, subplot and survey session effects only, alongside models with roost features hypothesised to impact tree-level measures: number of trees tagged within subplots, total roost abundance, total roost area, total subplot abundance, tree preference (whether the tree was regularly occupied: occupied in at least 80% of surveys; or irregularly occupied: occupied in less than 80% of surveys, as referred in Lunn et al. (2021)), and proportion of trees occupied per subplot, fit with the same set of random model effects relevant to that scale. We compared GAMs with the Akaike information criterion (AIC) and considered models within 2 ΔAIC units to be competitive (Burnham & Anderson 2002). We performed checks of standardised residuals to evaluate model fit, as per Wood (2017).

Results

The dataset includes tree-level abundance and tree-level 2-D density estimates from 2,522 spatially referenced trees, and tree-level 3-D density estimates from 480 of these trees. Measures were repeated monthly for 13 months; however, 52 tagged trees were cut or had fallen during the survey period, giving a final dataset of 32,206 tree-level abundance and 2-D density estimates, and 6,240 tree-level 3-D density estimates. Mean tree-level abundance was 3.75 bats per tree across all trees (interquartile range: 0.00-3.00 bats) and 13.35 bats per tree across all occupied trees (3.00-15.50). A full set of summary data are available at: < https://github.com/TamikaLunn/FF-roost-structure >.