Detection of a recent bottleneck
We used a model-selection approach to detect whether the black-faced spoonbill had experienced a recent population bottleneck. Two demographic models were constructed for the recent demography of the black-faced spoonbill. The null model assumed that the effective population size (N e) of the black-faced spoonbill had been constant (constant size model). The alternative model assumed that the black-faced spoonbill had experienced a recent bottleneck event (recent bottleneck model) and therefore contained the following five parameters: N e before the bottleneck,N e for the bottleneck, N eafter bottleneck, the date of the initiation of the bottleneck and the date of the end of the bottleneck. We used a method implemented inFastsimcoal2 (Excoffier et al., 2013) to conduct model selection and parameter inference for the selected model. First, we generated the observed folded (minor allele) site frequency spectrum from the 215,722 unlinked autosomal SNPs of the black-faced spoonbill with the scripteasySFS (https://github.com/isaacovercast/easySFS#easysfs). For the constant population size model, the prior for theN e was set to 100-1,000 haploid individuals. Assuming a generation time of ten years and that the bottleneck event occurred between the 1950 and the 1980s(La Touche, 1931),(Austin, 1948), the priors for the dates of bottleneck initiation (T bot) and termination (T endbot) were set to 2-5 and 2-10 generations ago, respectively. Bracketing the long-term N e(1/2 to 2×) since the last glacial maximum estimated from theSMC ++, we set the prior of N e for the pre-bottleneck (N anc) to 7,500-30,000 haploid individuals. Considering black-faced spoonbill’s population size of 288- 4000 since 1988, the priors for bottleneck (N bot) and post-bottleneck N e(N cur) were set to 2-100 and 100-1,000 haploid individuals respectively. Using uniform random samples from the priors of the two demographic models, we generated 100,000 folded SFSs for each model from the coalescent simulations and ran 40 optimization cycles to estimate each parameter and its composite likelihood in each simulation. The set of parameters with the highest likelihood was used for model selection. The simulation procedure described above was repeated 1,000 times, and a total of 5,000 sets of parameters were generated for each model. The parameters of the model from the set of simulations with the highest estimated likelihood was chosen as the best estimate of parameters for a given model. The Akaike information criterion (AIC) was used to compare the best simulation of the two demographic models to the observed folded site frequency spectrum (SFS). We used Δ AIC and AIC weight (w ) to evaluate which model better fitted the observed folded SFS. Then we ran the parameter estimation procedure under the best model to obtain 100 bootstrapped folded SFSs. Based on these bootstrapped folded SFS, we ran the parameter estimate procedure for each bootstrapped folded SFS again to compute the confidence interval of each parameter of the selected model.