Evaluation of site frequency spectrum-based demographic inference methods for use in conservation contexts

Genomic methods for inferring historical effective population size (Nₑ) trajectories offer valuable tools for conservation, yet their reliability under conditions typical of conservation datasets—small sample sizes, reduced-representation SNP data, and recent demographic change—remains poorly characterised. We evaluated the performance of two widely used site frequency spectrum (SFS)–based methods, Stairway Plot 2 and Epos, for reconstructing recent demographic histories relevant to conservation management. Using forward-time simulations in SLiM, we generated 609 unique population trajectories across four demographic scenarios (decline, expansion, stability, and bottleneck) with 20 replicates each, varying sample sizes (20–200 individuals) and number of loci (1,000–50,000 SNPs). Simulated genetic data were provided to both inference methods, and outputs were assessed for computational performance, trajectory reconstruction accuracy, temporal reliability, and Nₑ estimation error. Both methods reliably detected sustained declines and expansions, with correct trajectory reconstruction exceeding 80% overall. Epos was computationally faster and better at identifying stable trajectories, while Stairway Plot 2 was more accurate for Nₑ estimation and bottleneck detection. Both methods produced inflated Nₑ estimates in the most recent ~15 generations, and bottleneck scenarios were consistently difficult to reconstruct. Increasing sample size improved inference more than increasing SNP density. SFS-based demographic inference can effectively identify directional population trends under realistic conservation conditions but should not be relied upon for contemporary Nₑ estimation or short-lived bottleneck detection. We recommend prioritising individual sampling over marker density, trimming recent estimates, and integrating SFS-based results with complementary methods such as linkage disequilibrium–based estimators for robust conservation decision-making.