Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
A framework for species distribution modelling with improved pseudo-absence generation
Ist Teil von
Ecological modelling, 2015-09, Vol.312, p.166-174
Ort / Verlag
Elsevier B.V
Erscheinungsjahr
2015
Quelle
Alma/SFX Local Collection
Beschreibungen/Notizen
•The method for pseudo-absence generation strongly affected output SDM.•Environmental profiling of the background provided reliable models and improved AUCs.•TS and RSEP resulted in the most adequate methods for pseudo-absence data generation.•We propose the AUC-driven method to obtain a suitable background distance threshold.•We provide a modelling framework written in the open source R. language
Species distribution models (SDMs) are an important tool in biogeography and phylogeography studies, that most often require explicit absence information to adequately model the environmental space on which species can potentially inhabit. In the so-called background pseudo-absences approach, absence locations are simulated in order to obtain a complete sample of the environment. Whilst the commonest approach is random sampling of the entire study region, in its multiple variants, its performance may not be optimal, and the method of generation of pseudo-absences is known to have a significant influence on the results obtained. Here, we compare a suite of classic (random sampling) and novel methods for pseudo-absence data generation and propose a generalizable three-step method combining environmental profiling with a new technique for background extent restriction. To this aim, we consider 11 phylogenetic groups of Oak (Quercus sp.) described in Europe. We evaluate the influence of different pseudo-absence types on model performance (area under the ROC curve), calibration (reliability diagrams) and the resulting suitability maps, using a cross-validation approach. Regardless of the modelling algorithm used, random-sampling models were outperformed by the methods that incorporate environmental profiling of the background, stressing the importance of the pseudo-absence generation techniques for the development of accurate and reliable SDMs. We also provide an integrated modelling framework implementing the methods tested in a software package for the open source R environment.