Resolution in species distribution models shapes spatial patterns of plant multifaceted diversity

Abstract

Species distribution models (SDMs) are statistical tools that relate species observations to environmental conditions to retrieve ecological niches and predict species' potential geographic distributions. The quality and robustness of SDMs clearly depend on good modelling practices including ascertaining the ecological relevance of predictors for the studied species and choosing an appropriate spatial resolution (or ‘grain size'). While past studies showed improved model performance with increasing resolution for sessile organisms, there is still no consensus regarding how inappropriate resolution of predictors can impede understanding and mapping of multiple facets of diversity. Here, we modelled the distribution of 1180 plant species across the European Alps for two sets of predictors (climate and soil) at resolutions ranging from 100‐m to 40‐km. We assessed predictors' importance for each resolution, calculated taxonomic (TD), relative phylogenetic (rPD) and functional diversity (rFD) accordingly, and compared the resulting diversities across space. In accordance with previous studies, we found the predictive performance to generally decrease with decreasing predictor resolution. Overall, multifaceted diversity was found to be strongly affected by resolution, particularly rPD, as exhibited by weak to average linear relationships between 100‐m and 1‐km resolutions (0.13 ≤ R2 ≤ 0.57). Our results demonstrate the necessity of using highly resolved predictors to explain and predict sessile species distributions, especially in mountain environments. Using coarser resolution predictors might cause multifaceted diversity to be strongly mispredicted, with important consequences for biodiversity management and conservation.

MaxEnt brings comparable results when the input data are being completed; Model parameterization of four species distribution models

“…These methods, however, are challenged by a number of methodological problems that make their comparison controversial. Among these issues are providing a balance between goodness‐of‐fit and model complexity (Araújo et al, 2019 ; Warren & Seifert, 2011 ), the spatial bias of the input data, and manipulating them for evaluating model performance (Chauvier et al, 2022 ; Hijmans, 2012 ; Phillips et al, 2009 ). Generally, for most SDMs, particularly for complex machine learning ones, using a set of default parameters has been recommended based on a comprehensive model tuning [for example see Phillips & Dudík, 2008 for the MaxEnt and Elith et al, 2008 for boosted regression trees].…”

“…For example, in the first step, improving the sampling design can reduce bias and inaccuracy in the geographical distribution of the collected data (Araújo & Guisan, 2006 ; Chauvier et al, 2021 ). At this level, ensuring that the collected data correctly represent the actual distribution of the species (Guillera‐Arroita et al, 2015 ; Tessarolo et al, 2014 ) and that the scale of modeling and independent variables are consistent with sampling precision (Chauvier et al, 2022 ; Guisan et al, 2007 ; Wiens et al, 2009 ), and reducing unbiased recognition of the taxonomy of the species (Lorestani et al, 2022 ; Rocchini et al, 2011 ) improve results of an SDM analysis.…”

Applying landscape metrics to species distribution model predictions to characterize internal range structure and associated changes

“…Applying landscape metrics to SDM outputs adds another layer of complexity, since the accuracy of SDM predictions also varies depending on the spatial resolution and scale considered (e.g. Chauvier et al, 2022). Here, we defined a patch as a minimum of one isolated pixel because of the broad-scale nature of the study.…”

Transnational conservation to anticipate future plant shifts in Europe

“…All generated EIV soil property layers showed excellent evaluations with Spearman r > 0.82 (see Table S5). The generated EIV soil property layers are proxies of soil nitrogen and substrate composition, and have been shown to be excellent predictors of plant species distribution in SDMs 54,57 . The obtained EIV soil property layers were proxies of soil nitrogen (NITROGEN) and substrate composition (CALCAREOUS%).…”

“…Finally, all final layers (i.e. 13 per species) were each aggregated by mean to a 1 km resolution, as stacked SDMs provide more meaningful predictions of species diversity when species distributions are aggreagated from high to lower resolution 19,57,91 .…”