Figure legends
Figure 1. Overall community structure in 1982 vs 2018. The left panel shows the comparison of rank abundance plots for all species detected in 1982 (red) vs 2018 (blue). The solid points show the observed species abundances. The solid lines show the fit of the best rank abundance model (lognormal in both cases) and the colored polygons show a 95% CI of the fits computed using parametric bootstrap. The right panels show the hotspots of alpha diversity within the plot in 1982 and 2018 based on a 25 x 25 m grid. Each map presents the sum of the standardized kernel density estimates for all the species’ spot maps for both the 1982 census and the 2018 census. This plot is analogous to Terborgh et al.’s12 map of superimposed distributions for all species (their Figure 3).
Figure 2. Regression of abundance (panel A) and change in abundance in relation to body mass (panel B). Panel A shows the results of linear models where the results from 2018 were regressed against the results of 1982. This panel shows a simple linear regression fit and its CI (white dotted line and grey polygon). The solid black line indicates the 1:1 relationship which represents no change between the two time periods. The x axis in panel B represents the 0.1, 0.25, 0.5, 0.75 and 0.9 quantiles of the biomass distribution of all bird species present in Cocha Cashu, while the y axis shows the difference in biomass between 1982 and 2018 time periods for each species. The parenthesis “(“ indicates an open interval end while square bracket indicates closed interval end.
Figure 3. Temporal comparison (from 1982 to 2018) of the occupancy estimates via non-parametric bootstrap using the following ecological partitioning axes: ecological guild, foraging strata, sociality and habitat specialization: Bamboo/No bamboo and River-edge/No river-edge. Is the correlation in spatial distribution between time periods, per ecological partitioning axis, greater than what would be expected by chance? To answer this question, for each species we first estimated its KDE from 1982 using the digitized old spatial observations and computed its spatial correlation with the KDE derived with the 2018 spatial data. Then, we generated via non-parametric bootstrap 10,000 samples of KDE correlations under a Null model (see methods). The observed and the Null bootstrap distributions of the correlations per ecological partitioning axis are shown side by side in orange and blue, respectively. Note that in all comparisons, the observed median correlation is larger than the median correlation expected at random (see results of the non-parametric bootstrap test and Fig. S6). The rightmost panels with maps compare the distribution of Understory and Canopy Flock territories from 1982 (blue polygons) with those from 2018 (colored points). The 2018 colored points correspond to 6965 direct observations of geo-referenced flock activity.