Anthropogenic range contraction, temporal sampling bias, and the
performance of species distribution models
Abstract
Anthropogenic range contraction (ARC) dynamically alters species ranges,
which can be problematic for correlational species distribution
modelling (SDM). We used a virtual ecologist approach to simulate ARC
and temporally-biased sampling, demonstrate its link to two principal
forms of data misrepresentation, and investigate the outcomes to SDM
performance. Occurrences sampled before ARC were mismatched against
contemporary anthropogenic pressures (‘occurrence-habitat mismatching’),
which impeded predictions of species responses to those pressures and
produced models that underestimated the impact of human activity on
species distribution. Occurrences sampled after ARC were truncated
representations of historically realised niches (‘niche truncation’),
which were carried over onto predictions of species-environment
relationships, but were overcome by incorporating an appropriate
anthropogenic predictor. Given the pervasive global human footprint and
associated ARC, these data misrepresentations are likely inherent in
most occurrence datasets and should be interrogated during model
development. Accounting for these effects differs depending on the type
of data misrepresentation.