Loss of only the smallest patches will reduce species diversity in most discrete habitat networks.

Under many global-change scenarios, small habitat patches are the most vulnerable to destruction. For example, smaller ponds are at greater risk in a drying climate and their loss would remove any obligate aquatic individuals present. We asked what proportional loss of species diversity from metacommunities comprised of discrete habitat patches should be expected from attrition (complete loss) of only the smallest patches under such a premise. We analyzed 175 published datasets for different taxonomic groups (vertebrates, invertebrates, and plants) and habitat types (islands, habitat islands, and fragments). We simulated the destruction of only the smallest patches to an approximate 20% of total area (range: 15.2%-24.2%) and analyzed species loss. Mean [± 95% CI] species loss was 12.7% [10.8, 14.6], although 18.3% of datasets lost no species. Four broad patterns of species loss were evident, reflecting underlying differences in minimum area requirements and the degree of species turnover among patches. Regression modeling showed species loss increased with greater species turnover among patches (βSIM ) and decreased with greater area scaling of diversity (i.e., larger power-law island species-area relationship exponents). Losses also increased with greater numbers of single-patch endemics and with increasing proportions of patches destroyed. After accounting for these predictors, neither taxonomic group nor habitat type increased explained variation in species loss. Attrition of the smallest patches removed species in >80% of metacommunities, despite all larger patches and >75% of total area remaining intact. At both 10% and 20% area reduction, median species loss across all datasets was around 50% higher than predicted from methods based on the species-area relationship. We conclude that any mechanism of global change that selectively destroys small habitat patches will lead to imminent extinctions in most discrete metacommunities.