Selecting from correlated climate variables: a major source of uncertainty for predicting species distributions under climate change

Abstract

Correlative species distribution models are frequently used to predict species’ range shifts under climate change. However, climate variables often show high collinearity and most statistical approaches require the selection of one among strongly correlated variables. When causal relationships between species presence and climate parameters are unknown, variable selection is often arbitrary, or based on predictive performance under current conditions. While this should only marginally affect current range predictions, future distributions may vary considerably when climate parameters do not change in concert. We investigated this source of uncertainty using four highly correlated climate variables together with a constant set of landscape variables in order to predict current (2010) and future (2050) distributions of four mountain bird species in central Europe. Simulating different parameterization decisions, we generated a) four models including each of the climate variables singly, b) a model taking advantage of all variables simultaneously and c) an un‐weighted average of the predictions of a). We compared model accuracy under current conditions, predicted distributions under four scenarios of climate change, and – for one species – evaluated back‐projections using historical occurrence data. Although current and future variable‐correlations remained constant, and the models’ accuracy under contemporary conditions did not differ, future range predictions varied considerably in all climate change scenarios. Averaged models and models containing all climate variables simultaneously produced intermediate predictions; the latter, however, performed best in back‐projections. This pattern, consistent across different modelling methods, indicates a benefit from including multiple climate predictors in ambiguous situations. Variable selection proved to be an important source of uncertainty for future range predictions, difficult to control using contemporary information. Small, but diverging changes of climate variables, masked by constant overall correlation patterns, can cause substantial differences between future range predictions which need to be accounted for, particularly when outcomes are intended for conservation decisions.

Using species distributions models for designing conservation strategies of Tropical Andean biodiversity under climate change

“…For each species we 220 drew 10,000 pseudo-absences from the countries where the species was reported (according to 221 our database). This was done to avoid over-fitting of the models whilst maintaining a good 222 discrimination between presence and absence of the species (Isaac et exception (Braunisch et al, 2013;Dormann, 2007;Phillips, 2008). Excess predictors in a Maxent 227 model can cause over-fitting and hence bias the responses under future scenarios by over-228 weighting certain drivers over others (Warren and Seifert, 2010).…”

Forest conversion can help to mitigate impacts of climate change on common forest birds

“…Due to the use of a detailed spatial land use and climate data model, the performance indicated by AUC was relatively high as were the model fits indicated by D 2 . Although the estimated error in abundance revealed a certain level of uncertainty and the fact that predictions of species distributions and abundances under climate change are subject to different sources of uncertainty (e.g., Barbet-Massin et al 2010;Synes and Osborne 2011;Braunisch et al 2013), we strongly believe that our model results at least allow relative comparisons between the current situation and the scenarios. Our predictions of population sizes were used to comparatively evaluate different scenarios with the current reference condition.…”

Paleodistribution modeling in archaeology and paleoanthropology

“…Rigorous modeling methods already exist for analyzing multidimensional data, addressing multicollinearity, testing for variable importance, estimating response curves, associating response variables with multiple predictors, and mitigating or at least understanding effects of sample size and sample bias (Braunisch et al, 2013;Elith and Franklin, 2013;Elith and Leathwick, 2009;Franklin, 2010a). Methods are also being developed to help identify non--analog environments -combinations of different climate and other variables that did not occur in a different time period -in order to identify where and why model projections may be highly uncertain (Elith et al, 2010;Fitzpatrick and Hargrove, 2009;Mesgaran et al, 2014).…”