One of the most fruitful sub-fields in ecology is using climate variables to predict species’ geographic distributions. For the uninitiated, species distribution modelling assumes that species are limited in their distributions to suitable climate zones. By studying the environmental conditions where species are known to occur, you can infer the total geographic distribution by calculating the suitability of unsampled regions based on the environmental. Furthermore, using the same principle, species distribution modelling can forecast the effect of future climate change of the distribution of life on earth.
Unfortunately, studies have shown that these fancy climate-based techniques cannot consistently outperform much simpler ones based on spatial phenomena. For instance, spatial interpolation between point occurrences outperforms sophisticated climate-based predictions. Similarly, elaborate climate-based predictions perform no better than expected from random chance.
The trouble lies in the spatially-structured world we live in. Species distributions, especially at large spatial scales, are spatially-autocorrelated due to constrained dispersal. Similarly, climate variables are also spatially structured because the meteorological processes at proximal regions are generally more similar than those at distant sites.
When trying to link species distributions to climate conditions, the challenge lies is separating spatial and environmental correlations in species distributions. Specifically, we should identify three patterns in the geographical species distributions.
- We must first identify ‘true’ correlations with the environment, which are independent of spatial patterns (E|S).
- Next, we must identify the environmental-associations that also have a strong spatial structure (E∩S). This is known as exogenous spatial autocorrelation because it is due to autocorrelation is the underlying variables.
- Finally, we need to identify spatial patterns that are completely independent of environmental conditions (S|E). This is called endogenous spatial autocorrelation because it supposedly stems from spatial processes, such as dispersal.
In our latest study just published online at Ecography, we set out to quantify the degree of environmental correlation, exogenous and endogenous spatial autocorrelation in the distributions of 4 423 species of amphibians, reptiles, birds and mammals in Africa. Continue reading