Sampling design may obscure species–area relationships in landscape‐scale field studies

We investigated 1) the role of area per se in explaining anuran species richness on reservoir forest islands, after controlling for several confounding factors. We also assessed 2) how sampling design affects the inferential power of island species–area relationships (ISARs) aiming to 3) provide guidelines to yield reliable estimates of area‐induced species losses in patchy systems. We surveyed anurans with autonomous recording units at 151 plots located on 74 islands and four continuous forest sites at the Balbina Hydroelectric Reservoir landscape, central Brazilian Amazonia. We applied semi‐log ISAR models to assess the effect of sampling design on the fit and slope of species–area curves. To do so, we subsampled our surveyed islands following both a 1) stratified and 2) non‐stratified random selection of 5, 10, 15, 20 and 25 islands covering 1) the full range in island size (0.45–1699 ha) and 2) only islands smaller than 100 ha, respectively. We also compiled 25 datasets from the literature to assess the generality of our findings. Island size explained ca half of the variation in species richness. The fit and slope of species–area curves were affected mainly by the range in island size considered, and to a very small extent by the number of islands surveyed. In our literature review, all datasets covering a range of patch sizes larger than 300 ha yielded a positive ISAR, whereas the number of patches alone did not affect the detection of ISARs. We conclude that 1) area per se plays a major role in explaining anuran species richness on forest islands within an Amazonian anthropogenic archipelago; 2) the inferential power of island species–area relationships is severely degraded by sub‐optimal sampling designs; 3) at least 10 habitat patches spanning three orders of magnitude in size should be surveyed to yield reliable species–area estimates in patchy systems.