Modelling the spatial distribution of species ('Species Distribution Modelling', SDM) is a very popular activity these days. The ususal approach is to relate observational data to independent variables (mostly climate, soil and terrain, human activity and land cover) via some machine learning technique.

Dangerous

SDM can technically extrapolate beyond the observation domain in the geographical space or the functional space of the independent variables, and thereby easily generate beautiful maps. This seduces may investigators to do so and treat SDM-output as a quantitative prediction. As with any data-driven method here is no justification for this: SDM should be seen as an explorative research activity (to generate ideas about how a system might work) or interpolation, and not as an inferential activity (to test hypotheses or over-interpret the significance of model-coefficients) or for extrapolation. It is moreover very hard to calculate meaningful uncertainty estimates of SDM-output, and there is confusion about the proper way to deal with various artefacts in the observation data and the effects of these on SDM-output. Given these ingredients, it does not come as a surprise that the accuracy and precision of SDM-output is often quite disappointing (if reported at all) for those who plan to use it in nature conservation.

But promising

Considering the above, you probably think of me as a trouble-seeker if I confess that I'm quite interested in these methodological problems with SDM ... but that's not my (main) motivation: I am in fact interested in using SDM to create a measurement equation for conceptual population-dynamics models. In this way I hope to create animal abundance reanalysis data sets for specific populations or geographical regions. But before we can start to work on this, first some of the aforementioned methodological problems have to be solved.

Collaborators

Workflows for SDM

Some of the EcoGRID activities visualized One of the main projects which makes it possible for me and most of the people above to work on SDM is called EcoGRID and financed by the National Dutch Authority for Data concerning Nature (GAN). In EcoGRID we collaborate with experts from the organisations that collect species observations in the Netherlands (joined in the foundation VOFF) to enhance the process of species distibution modelling and establish the uncertainty of SDM-output. We concentrate currently on the intelligent placement of pseudo-absences and cross-validation procedures to test the predictive accuracy of SDM.

Linking SDM to ecological dynamics

In the Balgzand area (part of the Dutch Wadden sea). Intensive observations on some tidal food-rich areas are collected 24/7 via a high-resolution video system. In combination with these local observations, a tracking bird radar collects the movement of waders in the neigbourhood. Through these observations we are trying to learn more about the behavioural and foraging ecology of waders, and link statistical methods to describe bird distribution with dynamic models. This research is part of a larger NWO-project. It is mainly Adriaan Dokter who works on this.

Evening sun over Balgzand

Just a pretty picture from Balgzand.

some resources for species distribution modelling