1. Species–area (SA) models have often been used to predict biodiversity loss resulting from habitat loss. This application of SA models hinges on two fundamental assumptions: the resultant landscape matrix is inhospitable to the taxa of interest; and edge effects do not factor into extinction risks. Despite growing consensus that these assumptions are unrealistic, the SA approach continues to be used in assessments of biodiversity decline and conservation planning.
2. We propose an overhaul of the SA approach by accounting for taxon-specific responses to landscape-specific matrix quality and deleterious effects of habitat edges. We pitted nine variants of an improved SA model (calibrated for edge and/or matrix) against two variants of the conventional model (calibrated with island or continental z values) to predict species extinction and endangerment in 15 tropical biodiversity hotspots.
3. The matrix-calibrated SA model received the highest Akaike’s Information Criterion weight (birds: 66·8%; mammals: 63·3%), which reflects the weight of evidence in support of it being the most parsimonious model given the set of candidate models and data considered. Additionally, the matrix-calibrated (MC) model produced species extinction predictions that were the most accurate and least biased.
4. The second best model (for both birds and mammals) was one that simultaneously corrected for matrix and edge effects.
5. The conventional SA model (particularly when calibrated with an island z value) performed worse than the matrix-calibrated and/or edge-corrected models.
6. Synthesis and applications. Our results suggest that accounting for the landscape matrix per se is a sufficient and significant improvement to the SA approach in terms of assessing species extinction risks from land-use change. More importantly, given that the MC model was also the most parsimonious model (in that it requires only one additional model parameter than the conventional SA model), it could prove to be a cost-effective heuristic tool for conservation scientists and decision makers to accurately evaluate extinction risks resulting from land-use decisions. We argue that, henceforth, the MC model, which takes account of both the extent of deforestation and quality of the resultant matrix, should replace the conventional SA model for predicting biodiversity loss.