Comparison of community structure
We found remarkable similarities in the community structures of the 1982 and 2018 bird censuses (Fig. 1; Table S1). The overall community structure as suggested by the species abundance distributions (SAD) did not change between 1982 and 2018 (Fig. 1). Parameter estimates for the log-normal SAD model for 1982 were\(\mu=1.5,\ \mathrm{\text{CI}}=\left(1.33-1.61\right)\) and\(\sigma=1.1,\mathrm{\text{CI}}=\left(0.99,1.21\right)\), and\(\mu=1.64,\ \mathrm{\text{CI}}=\left(1.51-1.8\right)\) and\(\sigma=1.1,\ \mathrm{\text{CI}}=\left(1.00,1.17\right)\) for 2018. Community’s alpha diversity was highly correlated between the time periods (Rho=0.98; Fig. 1)
Although our results show strong similarity in the abundance of almost all species in 1982 and 2018, we did find eight outlier species out of the 275 total (Venables & Ripley 2002; decreases for Odontophorus stellatus, Myrmotherula brachyura, Brotogeris cyanoptera, Cacicus cela,and increases for Thamnophilus schistaceus, Isleria hauxwelli, Myrmoborus myotherinus, Monasa nigrifrons ; Table S2). The non-parametric bootstrap test showed that the number of observed outliers (8) is not significantly different than expected by chance (p = 0.19). The regression analysis showed no change in community-wide abundances between the two time periods (Fig. 2A: the Likelihood Ratio Test (LRT) failed to reject Ho: slope = 1 versus Ha slope \(\neq\)1, p-value = 0.92. Slope=1.00 (95% CI= 0.92-1.08), R-squared = 0.69). The linear model without phylogenetic dependency had better support than all regressions accounting for phylogenetic covariances using 1000 trees (average delta BIC = BIC(with phylogeny) -BIC(no phylogeny) = 114.84; 2.5% quantile = 88.3, 97.5% quantile = 310.5).
The non-phylogenetic regression model for overall biomass was better than the phylogenetic model in 320 out of 1000 trees (average delta BIC = BIC(with phylogeny) -BIC(no phylogeny) = -43.8; 2.5% quantile = -135.3, 97.5% quantile = 378.6). The phylogenetic regressions showed a decline in biomass (Fig. S1, mean slope = 0.67, mean 95% CI = (0.59,0.76)). For all the 687 regressions where the phylogenetic dependency model was better, a Likelihood Ratio Test (LRT) rejected Ho: slope = 1 versus Ha slope \(\neq\) 1, p-value <0.00001. A decrease in small species is mainly responsible for the observed pattern, after correcting for phylogenetic non-independence (Fig. 2B). Our guild analyses showed high similarity in the abundances of insectivores, frugivores, omnivores, and raptors (Fig. S2, Table 1), and declines in granivores (Fig. S2, Table 1). We also found little change in ground, mid-canopy and canopy species (Fig. S3, Table 1). We found little or no change in the abundance of social understory species (Fig. S4A). However, we found notable declines in social canopy species (Fig. S4B). We observed declines of all bamboo specialists concomitant with bamboo patches die-offs (Fig. S5A, Table 1). River edge specialists did not change in spite of changes to river edge habitat generated by shifts of the Manu River (Fig. S5B, Table 1).
In addition to the high similarity in the abundance of most species, we also found higher-than-expected-by-chance similarity in the distribution of the territories of most species, even after controlling for observer bias. The high similarities prevail even after partitioning the species according to different ecological axes (Fig. 3). Mean territory distribution overlap amongst all species was ~.40 and the median was 0.48 (Fig. S6, P < 0.0001 for both the mean and the median). Not only have species been constant in abundance through time, but the distribution of territories was mostly the same 36 years later (Fig. 1; Supplementary pdf map files 1 and 2). Neither phylogeny nor functional traits explained the temporal variation in species’ KDEs correlation coefficients. A null model with an intercept and no phylogenetic correlation structure was >4 BIC points better than any of the other models tested, even after propagating the phylogeny estimation uncertainty.