Dispersal and extrapolation on the accuracy of temporal predictions from distribution models for the Darwin's frog

• Published in: Ecological applications 2017-07, Vol.27 (5), p.1633-1645
• Main Authors: Uribe-Rivera, David E., Soto-Azat, Claudio, Valenzuela-Sánchez, Andrés, Bizama, Gustavo, Simonetti, Javier A., Pliscoff, Patricio
• Format: Article
• Language: English
• Subjects: Animal Distribution; Animals; Anura - physiology; Argentina; Biodiversity conservation; Chile; Climate Change; Climate models; Climatic zones; Climatology; Conservation of Natural Resources; Darwinism; Ecological modeling; ecological niche modeling; Extinct species; extinction risk; Mathematical extrapolation; model transferability; Models, Biological; no‐analogue climates; Predictive modeling; range dynamics; Time Factors
• ISSN: 1051-0761; 1939-5582
• DOI: 10.1002/eap.1556

Climate change is a major threat to biodiversity; the development of models that reliably predict its effects on species distributions is a priority for conservation biogeography. Two of the main issues for accurate temporal predictions from Species Distribution Models (SDM) are model extrapolation and unrealistic dispersal scenarios. We assessed the consequences of these issues on the accuracy of climate-driven SDM predictions for the dispersal-limited Darwin's frog Rhinoderma darwinii in South America. We calibrated models using historical data (1950–1975) and projected them across 40 yr to predict distribution under current climatic conditions, assessing predictive accuracy through the area under the ROC curve (AUC) and True Skill Statistics (TSS), contrasting binary model predictions against temporal-independent validation data set (i.e., current presences/absences). To assess the effects of incorporating dispersal processes we compared the predictive accuracy of dispersal constrained models with no dispersal limited SDMs; and to assess the effects of model extrapolation on the predictive accuracy of SDMs, we compared this between extrapolated and no extrapolated areas. The incorporation of dispersal processes enhanced predictive accuracy, mainly due to a decrease in the false presence rate of model predictions, which is consistent with discrimination of suitable but inaccessible habitat. This also had consequences on range size changes over time, which is the most used proxy for extinction risk from climate change. The area of current climatic conditions that was absent in the baseline conditions (i.e., extrapolated areas) represents 39% of the study area, leading to a significant decrease in predictive accuracy of model predictions for those areas. Our results highlight (1) incorporating dispersal processes can improve predictive accuracy of temporal transference of SDMs and reduce uncertainties of extinction risk assessments from global change; (2) as geographical areas subjected to novel climates are expected to arise, they must be reported as they show less accurate predictions under future climate scenarios. Consequently, environmental extrapolation and dispersal processes should be explicitly incorporated to report and reduce uncertainties in temporal predictions of SDMs, respectively. Doing so, we expect to improve the reliability of the information we provide for conservation decision makers under future climate change scenarios.