Open AccessPredictive ability of a process‐based versus a correlative species distribution model
Author(s) -
Higgins Steven I.,
Larcombe Matthew J.,
Beeton Nicholas J.,
Conradi Timo,
Nottebrock Henning
Publication year - 2020
Publication title -
ecology and evolution
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.17
H-Index - 63
ISSN - 2045-7758
DOI - 10.1002/ece3.6712
Subject(s) - correlative , species distribution , transferability , environmental niche modelling , range (aeronautics) , computer science , process (computing) , distribution (mathematics) , ecology , domain (mathematical analysis) , artificial intelligence , machine learning , biology , habitat , mathematics , engineering , operating system , philosophy , linguistics , mathematical analysis , logit , ecological niche , aerospace engineering
Abstract Species distribution modeling is a widely used tool in many branches of ecology and evolution. Evaluations of the transferability of species distribution models—their ability to predict the distribution of species in independent data domains—are, however, rare. In this study, we contrast the transferability of a process‐based and a correlative species distribution model. Our case study uses 664 Australian eucalypt and acacia species. We estimate models for these species using data from their native Australia and then assess whether these models can predict the adventive range of these species. We find that the correlative model—MaxEnt—has a superior ability to describe the data in the training data domain (Australia) and that the process‐based model—TTR‐SDM—has a superior ability to predict the distribution of the study species outside of Australia. The implication of this analysis, that process‐based models may be more appropriate than correlative models when making projections outside of the domain of the training data, needs to be tested in other case studies.