Climatic tolerance or geographic breadth: what are we measuring?

Abstract

In their recent letter, Olalla-Tárraga et al. (2015; hereafter ‘OT&al') applied phylogenetic path analysis to investigate the determinants of range size in terrestrial mammals. They concurred with Di Marco & Santini (2015; hereafter ‘DM&S') in identifying the predictive importance of human pressure, but disagreed that this role prevails over biological traits, criticizing some conceptual and methodological aspects of DM&S. OT&al found that climatic niche is the primary predictor of range size, while human pressure and biological traits were of secondary importance. Here, we discuss that the two studies are not directly comparable, and we address the criticisms to DM&S. First, OT&al used a smaller set of predictors than DM&S (see Table 1 in DM&S vs. Table 1 in OT&al). In particular, they only used human footprint to represent human pressure, which was the worst-performing among four human predictors tested in DM&S. Selecting a single underperforming predictor may have caused OT&al's model to underestimate the predictive role of human pressure, making direct comparison with DM&S difficult. Second, the use of phylogenetic comparative analysis, as implemented in OT&al, is necessary to correct for the nonindependence in regression-based models, because species share evolutive history (Purvis, 2008). However, these analyses rely on the evolutionary hypotheses behind the phylogenetic tree and on the evolutionary process assumed (Freckleton et al., 2002). Conversely, machine learning (ML) models, as implemented in DM&S, have much less assumptions, for example they do not assume independence between species. ML models do not require the definition of a priori hypotheses and allow to fit complex interactions, which cannot be easily handled by regression-based models. ML are indeed part of the analytical toolkit for ecological (Cutler et al., 2007) and conservation (Murray et al., 2014) research, and have been found to have similar predictive performance when compared to phylogenetic approaches (Bielby et al., 2010) Third, OT&al erroneously reported that DM&S did not consider diet breadth. In fact, DM&S found that the effect of diet breadth in predicting range size for separate taxonomic orders was of lower importance with respect to human pressure (Fig. S6 in DM&S). Additionally, while we agree with OT&al that climatic niche breadth is an intrinsic trait, we disagree that this can be measured by looking at current species ranges. Climatic tolerance is a physiological characteristic, typically measured through experiments (Sunday et al., 2011). Conversely, current species distribution largely depends on biogeographical, ecological and human constrains, rather than climatic tolerance alone. At one extreme, the Javan rhino (Rhinoceros sondaicus) nowadays experience a very narrow climatic range (within Ujung Kulon National Park), yet the species could arguably survive in different climates given its historical distribution in South-East Asia (Groves & Leslie, 2011). We maintain that climatic variables, as measured in DM&S (i.e. single values), represent extrinsic environmental conditions. Conversely, the realized climatic niche, as measured by OT&al (i.e. ranges of values), represents an unknown fraction of the species climatic tolerance (i.e. a fraction of the fundamental niche; Soberón & Nakamura, 2009). Additionally, larger geographic ranges are more likely to experience larger ranges in climate variables by chance, thus introducing circularity in the use of climatic niche as a range size predictor (Fig. 1). In conclusion, despite not being directly comparable, the analyses of OT&al substantially concur with those of DM&S and do not contrast with their original interpretation. We thank Luigi Maiorano and Oscar Venter for their suggestions.