What do we gain from simplicity versus complexity in species distribution models

By Cory Merow, Matthew J. Smith, THOMAS C. EDWARDS JR., Antoine Guisan, S. McMahon, S. Normand, Wilfried Thuiller, Rafael O. Wüest, N. E. Zimmermann, Jane Elith. 2014

593 citations recorded by the source corpus.

Species distribution models (SDMs) are widely used to explain and predict species ranges and environmental niches. They are most commonly constructed by inferring species' occurrence–environment relationships using statistical and machine-learning methods. The variety of methods that can be used to construct SDMs (e.g. generalized linear/additive models, tree-based models, maximum entropy, etc.), and the variety of ways that such models can be implemented, permits substantial flexibility in SDM complexity. Building models with an appropriate amount of complexity for the study objectives is critical for robust inference. We characterize complexity as the shape of the inferred occurrence–environment relationships and the number of parameters used to describe them, and search for insights into whether additional complexity is informative or superfluous. By building ‘under fit’ models, having insufficient flexibility to describe observed occurrence–environment relationships, we risk misunderstanding the factors shaping species distributions. By building ‘over fit’ models, with excessive flexibility, we risk inadvertently ascribing pattern to noise or building opaque models. However, model selection can be challenging, especially when comparing models constructed under different modeling approaches. Here we argue for a more pragmatic approach: researchers should constrain the complexity of their models based on study objective, attributes of the data, and an understanding of how these interact with the underlying biological processes. We discuss guidelines for balancing under fitting with over fitting and consequently how complexity affects decisions made during model building. Although some generalities are possible, our discussion reflects differences in opinions that favor simpler versus more complex models. We conclude that combining insights from both simple and complex SDM building approaches best advances our knowledge of current and future species ranges.

Subfields: Ecological Modeling, Ecology, Nature and Landscape Conservation

View Paper: https://www.semanticscholar.org/paper/35f35103d3310d019475d9e3e479472c4a0b8de7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Ecological Modeling

In Ecology

In Nature and Landscape Conservation

Citations in the corpus

Spatial patterns and climate relationships of major plant traits in the New World differ between woody and herbaceous species (2018, background): “Species distribution models have improved significantly in recent years (Merow et al., 2014; Thuiller, Lavorel, Sykes, & Ara ujo, 2006; Thuiller, Munkemuller, Moller, Fiedler, & Berthold, 2010) and range maps are increasingly available for many plant species worldwide.”
Testing species assemblage predictions from stacked and joint species distribution models (2019, background): “However, very complex models are not always desirable as this could limit the transferability when extrapolating beyond the sampled environmen‐ tal conditions (Merow et al., 2014; Wüest, Münkemüller, Lavergne, Pollock, & Thuiller, 2018; Yates et al., 2018).”
Outstanding Challenges in the Transferability of Ecological Models. (2018, background): “Additionally, they could be useful in exploring nonlinear and dynamic associations of species with indirect predictors across landscapes, seasons, or years [40], to help better accommodate nonstationarity.”
A comprehensive evaluation of predictive performance of 33 species distribution models at species and community levels (2019, background): “They range from purely data-driven SDMs allowing for very flexible predictor functions (e.g., random forest and generalized additive models) to more rigid ones (e.g., generalized linear models; Guisan et al. 2002, James et al. 2013, Merow et al. 2014; Table 2, Feature A).”
Predicting range shifts of Asian elephants under global change (2019, background): “However, when a large set of predictor variables are considered for analysis, mainly due to the multidimensional abundance of environmental compo‐ nents, the model complexity increases and may lead to the devel‐ opment of heavily parameterized and overfitted models (Merow et al., 2014).”
Biological and geophysical feedbacks with fire in the Earth system (2018, background): “However, deterministic relationships between climate and vegetation were assumed in the first global vegetation models, which arose in the 1980s (Prentice et al 2007, Woodward 1987), and are still employed in species distribution models for a wide variety of purposes (Merow et al 2014).”
Two alternative evaluation metrics to replace the true skill statistic in the assessment of species distribution models (2019, background): “Instead, good model performance may result from model overfitting as species presences simply coincide with specific conditions, combinations or transformations of environmental variables in overly complex models (Merow et al. 2014; Fourcade et al. 2018).”
ENMeval 2.0: Redesigned for customizable and reproducible modeling of species’ niches and distributions (2021, background): “This commonly leads to overfit models that reflect noise in the data (Elith & Graham, 2009; Merow et al., 2014) and have poor transferability when applied to novel conditions (Guevara et al., 2018).”
Comparing sample bias correction methods for species distribution modeling using virtual species (2021, background): “…is not often discussed, but has been shown to be an v www.esajournals.org 18 March 2021 v Volume 12(3) v Article e03422 important factor affecting predictions across geographic regions and time periods (Warren and Seifert 2011, Syfert et al. 2013, Merow et al. 2014, Bell and Schlaepfer 2016).”