Statistical analyses
The main statistical comparisons tested with our empirical data were: 1) whether frequency of occupation is greater for subplots in ‘core’ areas of the roost compared with subplots in irregularly occupied ‘peripheral’ areas (defined by occupation greater than or less than 80% of surveys respectively (Appendix S1); 2) whether bat occupation decreases with distance from the roost centre (per species); 3) whether bat species segregate in vertical space; and 4) whether dominant individuals occupy the centre of roosts, and subdominant individuals the outer area (per species). We also provide qualitative comparisons of 5) seasonal patterns of abundance and occupancy per species; and 6) whether bat species segregate in horizontal space.
We utilised generalized additive models for all statistical comparisons to allow for nonlinearity, with random effects modelled with smooth functions. Roost site and subplot were modelled using a standard random effects smoothing function. Session was modelled using a cyclic cubic regression spline in cases where seasonality in the time series was evident (all comparisons except those involving the proportion of male black, male grey-headed and combined male bats per tree), otherwise session was modelled with a standard random effects smoothing function. We accounted for non-independence (nesting) of random effects by including an autoregressive model for errors in the model (Yang et al. 2012; Laurinec 2017). For the comparisons involving evaluation of species, models were run separately for each species owing to differences in seasonality of occupation (and so, differences in the fit of cyclic cubic regression splines). Error distribution for comparisons were specified according to data type and extent of zero-inflation (as per Crawley 2013). We fit the models and performed checks of standardised residuals in R (Version 4.0.2), using the ‘mgcv’ package (functions ‘gamm’ and ‘gam.check’) (as per Wood 2017). See Appendix S1 in the Supporting Information for more detailed information on modelling decisions and a summary table of comparisons. Summarised data and annotated R code are available on GitHub at: < https://github.com/TamikaLunn/FF-roost-ecology >.

Results

From our review of management, recovery and restoration documents published by state government, we highlighted 31 commonly held understandings relevant to flying-fox roosting structure (Table 2). From our systematic search for empirical literature we generated a total of 79 search results. Of these, 52 were removed through screening (10 being outside the Australian mainland, 4 on non-Pteropus species, and 38 focused on topics other than roost structure). An additional 18 published studies and 4 honours/PhD theses were included from citations and the author’s reference collections, giving 49 included studies in total (Appendix S1). Lastly, we generated an empirical dataset consisting of 13 monthly repeat measures from 2,522 trees across eight roost sites. Roost sites contained 118-474 measured and tagged trees each, with an average of 2 (sparsely structured) to 75 (densely structured) trees per 20x20 meter subplot. Tree roosting height and count was recorded for 9,056 trees out of 32,206 repeat measures. (Note that our total repeat measures were less than 32,786 owing to cases of tree removal through the duration of the survey.) We report model outputs of main interest in the main text, but see Appendix S2 in the Supporting Information for full model output.
Below, and in Table 2, we synthesise how commonly held understandings compare with existing literature and new data from our study.