Collinearity in ecological niche modeling: Confusions and challenges

Abstract

Ecological niche models are widely used in ecology and biogeography. Maxent is one of the most frequently used niche modeling tools, and many studies have aimed to optimize its performance. However, scholars have conflicting views on the treatment of predictor collinearity in Maxent modeling. Despite this lack of consensus, quantitative examinations of the effects of collinearity on Maxent modeling, especially in model transfer scenarios, are lacking. To address this knowledge gap, here we quantify the effects of collinearity under different scenarios of Maxent model training and projection. We separately examine the effects of predictor collinearity, collinearity shifts between training and testing data, and environmental novelty on model performance. We demonstrate that excluding highly correlated predictor variables does not significantly influence model performance. However, we find that collinearity shift and environmental novelty have significant negative effects on the performance of model transfer. We thus conclude that (a) Maxent is robust to predictor collinearity in model training; (b) the strategy of excluding highly correlated variables has little impact because Maxent accounts for redundant variables; and (c) collinearity shift and environmental novelty can negatively affect Maxent model transferability. We therefore recommend to quantify and report collinearity shift and environmental novelty to better infer model accuracy when models are spatially and/or temporally transferred.

Global occurrence data improve potential distribution models for Aedes japonicus japonicus in non‐native regions

“…Collinearity between annual precipitation (bio12) and precipitation in the warmest and coldest quarters (bio18, bio19) also occurred. Because multicollinearity does not affect Maxent model performance, we refrained from excluding single highly correlated variables. Moreover, De Marco and Nóbrega even suggest that the Maxent fitting process takes advantage of existing collinearity in finding a best set of parameters.…”

A checklist for maximizing reproducibility of ecological niche models

“…Transferring a model across space and/or time may lead to extrapolation if the projected environments are novel relative to training environments. Several studies have found that environmental novelty 48,94,95 (D6) and collinearity shift (D7; changes of collinearity structure of covariates 77,96 ) reduce predictive performance, and recommended quantifying the novelty of the projected environments and the collinearity shift between the calibrated and to include bug fixes or revised default settings. For instance, the default transformation method of Maxent raw output was changed from 'logistic' to 'cloglog' between versions 3.3 and 3.4 80 .…”

“…The information regarding environmental data necessary to reproduce studies is similar across biological research, such as in studies of relationships between species richness and environmental gradients 108 . The modelling algorithm details in the checklist are applicable to other studies that use statistical models, such projected environments 96,97 . Further, different algorithms use different strategies to extrapolate (clamping, truncation, extrapolation 94,98 ); for example, the default clamping function in Maxent uses the marginal values in the calibration area as the prediction for more extreme conditions in transfer areas 22 .…”

“…Inclusion of elements of the checklist (32 in total) varied widely, ranging from fully reported (100%; C5 algorithm name) to not reported at all (0%; D7 collinearity shift), though documentation of the importance of this latter element is still limited in the literature 77,96 . Completeness of information across the checklist also varied among papers, ranging from 24% to 89%, averaging 54% (s.d.…”

Predicting the dynamic distribution of Sphagnum bogs in China under climate change since the last interglacial period

“…The collinearity between variables in the SDMs may lead to over-fitting phenomenon. However, Xiao Feng et al reports that the strategy of excluding highly correlated variables has little impact because Maxent accounts for redundant variables [30]. Besides, the nineteen environmental variables are regular bioclimatic variables, and the growth and distribution of Sphagnum plants are susceptible to precipitation and temperature.…”