Opportunistic datasets perform poorly in Ecological Niche Modelling: a case study from a polymorphic lizard

In: Wildlife research, Band 49, Heft 8, S. 749-759

Abstract

Context Among processes involved in colour polymorphism, geographic variation in morph composition and frequency has been attracting interest since it reflects morph local adaptation. A recent study in the Pyrenees associated the pattern of geographic variation in morph frequency of the common wall lizard with the divergence in climatic niches, supporting the hypothesis that morphs represent alternative local climatic adaptations. However, the Pyrenees represent only a small portion of the species range. Aims We modelled the ecological niches of Italian morphs using the same procedure adopted for the Pyrenees to check whether the effects detected at local scales (i.e. the Pyrenees) were repeatable at regional scales (i.e. Italy). This generalisation is needed to investigate how natural selection maintains locally adapted polymorphisms. Methods We classified each locality (120 populations) according to the presence/absence of morphs, and independent Ecological Niche Models (ENMs) against the same background were fitted. Receiver Operating Curves accounting for sampling biases, equivalency and similarity tests were used to check and compare models accounting for spatial distribution of data. Key results Morph-specific ENMs did not reproduce any of the patterns detected in the Pyrenees. Any difference among morphs disappeared after controlling for morph spatial distribution. Since occurrence points of the rarest morphs were a subsample of the occurrence points of the most common morph, it is not possible to separate the effects of true ecological differences among morphs from the effects of the spatial distribution patterns of morph occurrence. Conclusions Using presence data not specifically collected for ENM comparisons does not allow reliable assessments of morph niche segregation. Our analysis points out the need to be very cautious in ecological interpretations of ENMs built on presence/background or presence-only data when occurrences are spatially nested. Implications When dealing with data not specifically collected according to a targeted design, it is not legitimate to compare ENMs with completely nested occurrence points, because this approach can not exclude the possibility that ENM differences were the result of a spatial subsampling. This type of bias is probably largely underestimated, and it may lead to serious misinterpretations as shown in this study.