Evaluating how species niche modelling is affected by partial distributions with an empirical case

Abstract

Ecological niche models (ENMs) will successfully identify a species' ecological niche, provided that important assumptions are fulfilled, namely environment equilibrium and niche equality across the distribution. Violations may seriously affect ENM reliability, leading to erroneous biogeographic conclusions and inappropriate conservation prioritisation. We evaluate the robustness of ENMs against incomplete knowledge of distribution with a real example, the threatened Iberian lizard Podarcis carbonelli, whose distribution was gradually discovered over a long time period. We used several ENM methods for presence-only data (Maxent, ENFA, Bioclim, and Domain) to infer the realised ecological niche at two spatial resolutions (1 km and 200 m). The distribution data were split into four partial datasets corresponding to separate subranges: Central System (CS); Viseu-Aveiro (VA); Atlantic coast (AC); and Doñana (DO). We then accumulated the datasets following the species discovery sequence: CS + VA, CS + VA + AC, and CS + VA + AC + DO. Niche equivalence and similarity between partial models were compared using Ecospat. ENMs were strongly affected by the violation of niche equilibrium; only the VA subrange forecasts the complete species range. ENMs were also sensitive to the violation of niche equality: only VA models were similar to the Iberian model, altitude being the most important variable followed by annual precipitation, maximum temperature in July, and annual radiation. When the ENMs were applied only to the first subrange discovered (CS), only the VA area was predicted, while the other subranges might have remained unknown, thus compromising conservation strategies. As assumptions of niche equilibrium and equality were violated, likely owing to the species' ecological multimodality, the models generated were biased and of limited predictive value. ENMs are useful tools in biogeography and conservation, but only if their basal assumptions are achieved. Partial models may be useful if they are considered as representing different suitable habitats.