Antoine Guisan, University of Lausanne, Switzerland, discusses his article: Spatially-nested species distribution models (N-SDM): An effective tool to overcome niche truncation for better inference and projections
Setting the scene
Species distribution models (SDMs) relate species observations to mapped environmental conditions to estimate the ecological niche (i.e., the ensemble of suitable conditions) and predict the spatial distribution of species. As such, they are key tools for projecting the impact of global change on species and have been used in many biodiversity assessments. However, when the data used to fit the model does not encompass the whole species range (i.e., subrange), the estimated ecological niche can be truncated, which can lead to wrong or inaccurate spatial predictions, especially in other areas or time periods.
When fitted extensively, to cover the whole species range, users often have to rely on climatic variables at coarse resolution (often ≥1 km). But such data cannot capture the fine environmental requirements of species that are often only revealed at local scales. A solution is to combine SDMs at multiple (usually two) spatially-nested scales to produce spatially-nested SDMs (N-SDMs). In our paper, we review the development and use of N-SDMs and discuss when they are (or are not) needed.
Main findings: What are N-SDMs and what are they used for?
N-SDMs thus combine a coarse resolution SDM fitted over the entire species range with a fine resolution SDM fitted to the extent desired for more detailed inferences. Hence, models are fitted to different extents and resolutions, the finer SDM being spatially nested within the coarser SDM. Species with large ranges but complex climatic dependencies, which require the consideration of both macro- and micro-climatic conditions and other fine-grained environmental predictors (e.g., alpine species), can thus be expected to be better predicted using N-SDMs.
We identified and reviewed six main approaches to fit N-SDMs (two of them are shown above), from simple combinations of initial data or final predictions to more advanced statistical models. Yet, there are still relatively few studies comparing different N-SDM approaches and those existing show that the approach to use often depends on the study objectives. If projection is the aim, a sequential or integrated approach may be more well-suited, but when management decision is the aim, keeping predictions separate may be more informative. We identified several successful applications of N-SDMs, and also discussed cases where they are not needed, such as when a coarse resolution is sufficient to model the whole distribution of the species with large coverage, or when a species has a limited range that can be entirely modelled using fine-resolution data at the regional level.
Why do N-SDMs matter?
Our review clearly shows that using geographic coverage that is too small to model the distribution of species can result in biased SDMs and lead to incorrect predictions and projections, especially in other areas or time periods. N-SDMs can overcome this problem and have already been proven useful in several applications, such as anticipating biological invasions, downscaling large scale SDMs, assessing the potential effects of global change on species distributions, quantifying the effect of environmental predictors on species distributions at different spatial scales, and supporting conservation decisions. We expect N-SDMs to play a growing role in supporting assessments to reach global biodiversity conservation targets. To achieve these goals, anticipation of future biodiversity patterns will be more needed than ever, and thus N-SDMs will be key in obtaining robust future projections.
