Genetic Clustering
We estimated genetic clusters using discriminant analysis of principal components (DAPC; Jombart et al. 2010) and sparse non-negative matrix factorization (sNMF; Frichot et al. 2014) on the SNP matrix. For DAPC in the R package ‘adegenet’ (Jombart 2008), we chose the optimal number of clusters by minimum Bayesian Information Criterion (BIC) and optimal assignment using DAPC cross-validation (Fig. S2–4). For sNMF in ‘LEA’ (Frichot and François 2015), we chose the regularization factor α and the number of clusters K using minimum median cross-entropy from 100 replicates (Fig. S5–6). For a preliminary test of IBD, we performed a spatially-aware analysis of population clustering and admixture using the ‘conStruct’ package (Bradburd et al. 2018), which accounts for geographic genetic variation using spatial layers to represent clusters that account for allelic covariance attributed to continuous (e.g., IBD) versus discrete (e.g., barriers) sources (Bradburd and Ralph 2019). We tested 1–4 layers with 4 chains of 250k generations. We took a layer to be significant when it accounted for more than 20% of the allelic covariance matrix (Fig. S7).