• Citer

Possibilities and limitations of hierarchical modeling of species communities to uncover the climate niches of African tree species

par Ridder, Grace
Promoteur Hardy, Olivier J.
Co-Promoteur Dauby, Gilles
Publication Non publié, 2021-08-31

Mémoire

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Methods for studying the biodiversity of tropical ecosystems are continually evolving forboth species’ distribution modeling and the types of data available while the limits ofeach have yet to be explored. We investigated the potential of the recently developedhierarchical modeling of species communities (HMSC) framework to infer speciesenvironmental associations when applied on herbarium data. We applied HMSC to adataset of 2,963 records comprising of 50 tree species from western Central Africa(6°S-6°N, 10°E-15°E), subdivided in 167 (0.5°×0.5°) grid cells. The herbarium data wasextracted from the RAINBIO database and six environmental variables from theWorldClim database were used as predictors. An additional fixed effect was added toaccount for variability in sampling effort and a spatially explicit random effect was usedto account for spatial structure. The full model with real data had reasonableexplanatory power (AUC = 0.84) and suggested that minimum annual temperature waslargely responsible for the variation in tree species distribution. However, furtherinvestigation revealed that model fit between models run with shifted environmentaldata and real data do not differ greatly suggesting possible over-fitting due to spatialautocorrelation. Furthermore, simulations with artificial species distribution data withknown environmental associations confirm that the method generates an excess offalse positives, which is only partially compensated when accounting for sampling effortinformation and spatial effects. Therefore, we conclude that using HMSC on RAINBIOtypedata may produce inaccurate results which is a problem that should beinvestigated in other species distribution modeling methods.