Detecting isolation-by-distance versus barrier
To identify IBD patterns among populations, we analyzed the correlation between pairwise genetic distances and geographic locations of all populations using BARRIER 2.0 (Manni et al. 2004). BARRIER applies a geometric approach and produces the length and width of barriers. We also applied Mantel tests to determine concordance and statistical robustness between geographic and genetic distances. Mantel tests compared geographic distance matrices (produced according to the multiple dispersal scenarios described above) and the N-distance matrix. Mantel tests were performed on the 5 species with 4 scenarios of dispersal, respectively, while the BARRIER software accepts coordinate as input. We also calculated AIC for linear correlation models of genetic distance and geographic distance under different scenarios with the AIC function in R core functions (R Core Team 2020). Values were formatted into a ΔAIC matrix to infer the best scenario for IBD pattern.
As noted by Wright (1943), an IBD pattern would be disrupted by non-random dispersal in reaction to specific environments. This observation led ultimately to the development of the idea of isolation-by-environment (IBE; Nosil et al. 2005). To test for IBE, we selected 20 environmental variables—including 19 bioclimatic variables from the public database CHELSA (Karger et al. 2017, Karger et al. 2018; Supporting information) and the GMTED2020 (Danielson and Gesch 2011) elevational dataset mentioned above—that are expected to be correlated with factors consequential to bird habitat. A negative correlation or no correlation between the climatic variables and genetic distance, therefore, would dispute IBE. To avoid overfitting in the environmental analyses, we first remove highly correlated variables. We randomly sampled 10000 grids to characterize each candidate variable and performed PCA analysis of their correlation with one another (Supporting information). Seven of the variables—Bio04 (temperature seasonality), Bio07 (annual range of air temperature), Bio12 (annual precipitation), Bio13 (precipitation of the wettest month), Bio16 (mean monthly precipitation of the wettest quarter), Bio18 (mean monthly precipitation of the warmest quarter), and elevation, were the least related to one another. Using these 7 variables, we tested for IBE by applying Mantel tests and linear regression on variable differences versus N-distances among genetic sample points using the mantel function in thevegan package (Oksanen et al. 2019) and lm function in R core functions (R Core Team 2020). A significant correlation would support IBE among populations.
To examine the effects of recent climate changes on the distribution of the 5 species, we performed environmental niche modeling (ENM), using the same 7 environmental variables and MAXENT 3.4.4 (Phillips et al. 2006) with default parameters. Variable values from the last glacial maximum (LGM, ~21,000 years before present), the final stages of the LGM (BAW, ~14,600 years ago at the maximum of the Bølling-Allerød warming; Karger et al. 2018), and present day (CHELSA; Karger et al. 2017) were used. The elevation above sea level and land contours for modeling and mapping were from Lambeck et al. (2014) (96 meters lower than present for the BAW), and Siddall et al. (2003) (125 meters lower than present for the LGM). Ten replicates were run, and cross-validation was used to test the model. To build the models, record data from the China Bird Report Center (CBRC, http://www.birdreport.cn/, retrieved on 2 February 2020) and eBird database (Sullivan et al. 2009; retrieved in May 2021) for the 5 species were used. Point occurrence records were reduced to one record per 15 km radius. Records with longer observation and higher count were selected during the thinning process. The final collection consisted of 3931 records in total (1480 forI. mcclellandii, 312 for A. hueti, 1079 for L. lutea , 663 for P. pectoralis and 397 for S. torqueola ). For graphic illustration of the distribution of the species over time, we collected the logistic output raster. We also prepared a raster map of summed logistic output of all 5 species for an overview.