Predicting species distribution at range margins: testing the effects of study area extent, resolution and threshold selection in the Sahara–Sahel transition zone
Article first published online: 23 JUL 2013
© 2013 John Wiley & Sons Ltd
Diversity and Distributions
Volume 20, Issue 1, pages 20–33, January 2014
How to Cite
Vale, C. G., Tarroso, P., Brito, J. C. (2014), Predicting species distribution at range margins: testing the effects of study area extent, resolution and threshold selection in the Sahara–Sahel transition zone. Diversity and Distributions, 20: 20–33. doi: 10.1111/ddi.12115
- Issue published online: 14 DEC 2013
- Article first published online: 23 JUL 2013
- National Geographic Society. Grant Numbers: 7629-04, 8412-08
- Fundação para a Ciência e Tecnologia. Grant Number: PTDC/BIA-BEC/099934/2008
- FCT. Grant Numbers: SFRH/BD/72522/2010, SFRH/BD/42480/2007
- ecological niche factor analysis;
- global models;
- maximum entropy;
- regional models;
- species distribution models
Compare the performance of continental and regional models in predicting species distributions at range margins. Selection of study area extent, resolution and threshold affects ecological model predictions. At range margins of species distribution, local populations may be restricted to suboptimal environments distinct from the species' global range, which may be missed by continental models.
Africa and West Africa.
We analysed differences in predicted distributions at range margins of three widespread African species that in West Africa occur in peripheral populations restricted to particular habitats. We made comparisons between models built with data from the complete and restricted range of species' distributions (Africa and West Africa, respectively), with coarse and fine resolutions (10 × 10 km and 1 × 1 km, respectively), and classified with three thresholds of species presence (minimum training presence, 10th percentile training presence and maximum training sensitivity plus specificity thresholds). We predicted the species' distributions and quantified environmental variable importance and profile using maximum entropy and estimated niche breadth parameters with ecological niche factor analysis.
We found differences between model types in niche breadth estimates and also in response curves of the most important variables, suggesting that fine resolution models are more accurate at selecting marginal habitats in West Africa than in Africa. The predictions of species distributions differed with model extent, resolution and threshold analysed. Models built with the complete species environmental range and with coarse resolution tended to overestimate species distributions at the edge, but accuracy increased when more restrictive thresholds were used. In West Africa, independently of the resolution, the threshold value was less important for maximizing agreement between predicted probabilities and observed distribution.
At range margins of species distributions, regional models with precise data and conservative thresholds should be preferred over continental models with coarser resolution to identify suitable areas for peripheral populations.