A simulation-based method for selecting calibration areas for ecological niche models and species distribution models

Marlon E. Cobos,A. Townsend Peterson,Fernando Machado-Stredel

doi:10.21425/f5fbg48814

Abstract

Ecological niche models and species distribution models (ENM and SDM, respectively) are tools that have seen massive use and considerable improvement during the last twenty years. The choice of calibration areas for such models has strong effects on model outcomes and model interpretation, as well as on model transfer to distinct environmental settings. However, approaches to selecting these areas remain simple and/or unlinked to biological concepts. Such models should be calibrated within areas that the species of interest has explored throughout its recent history, the accessible area (M). In this paper, we provide a simulation approach for estimating a species’ M considering processes of dispersal, colonization, and extinction in constant current climate or glacial-interglacial climate change frameworks, implemented within a new R package we developed called grinnell. Using the avian genus Aphelocoma, we explored different parameterizations of our simulation, and compared them to current approaches for M selection, in terms of model performance and risk of extrapolation using the algorithm Maxent and mobility-oriented parity analyses. Model calibration exercises from all approaches resulted in at least one model meeting optimal performance criteria for each species; however, we noted high variability among taxa and M selection methods. More importantly, M hypotheses derived directly from simulations of key biological processes, rather than being based on simple proxies of those processes, and as such are better suited to erecting biologically appropriate contrasts in model calibration, and to characterizing the potential for model extrapolation more rigorously. Major factors in our simulations were environmental layer resolution, dispersal kernel characteristics, and the inclusion of a changing framework of climatic conditions. This contribution represents the first simulation-based method for selecting calibration areas for ENM and SDM, offering a quantitative approach to estimate the accessible area of a species while considering its dispersal ability, along with patterns of change in environmental suitability across space and time.

Highlights

Ecological niche models (ENM) and species distribution models (SDM) inevitably involve some level of model transfer from areas and conditions that were sampled to broader areas and sets of conditions (Yates et al 2018), even when models are interpreted across the model calibration area only
The calibration areas resulting from our simulations were consistent with the distribution of the studied Aphelocoma species, including potential unsuitable regions where these taxa have not been recorded
The 110 parameterizations for Aphelocoma species took a total of 1851 hours of computational time, with A. californica simulations taking most of that time (60.5%; Supplementary Table S4)

Summary

Introduction

Ecological niche models (ENM) and species distribution models (SDM) have become a popular technique in the emerging area of distributional ecology (Araújo et al 2019), with publications based on these tools in evolutionary biology (e.g. Warren et al 2008, Varela et al 2011, Saupe et al 2018), ecology (e.g. Anderson 2017, Zurell 2017, Osorio‐Olvera et al 2020), conservation biology (e.g. Guisan et al 2013, Freeman et al 2018), and public health (e.g. Peterson 2014, Quiner and Nakazawa 2017). Barve et al (2011) demonstrated that calibration area choice matters, and linked a biological concept (i.e., the set of sites that has been accessible to the species over relevant time periods) and the area over which models should be calibrated. Simulations for selecting ENM and SDM calibration areas across the region, termed η(M), is the range of environments that the species has explored and tested.

Objectives

Methods

Results

Conclusion