Model selection methods for species distribution modelling

Abstract:

Species distribution modelling, where the spatial distribution is modelled as a function of environment, is widely used in ecology - for predicting rare species and predicting climate change impacts. This is currently a rapidly changing field in which many methods from different disciplines have been proposed, but there is currently little understanding how these methods relate to each other. I will discuss two main areas of work aiming to shed light on something or other.

Two methods for species distribution modelling - MAXENT, the most widely used method at the moment, and Poisson point process models, a method only recently proposed - will be shown to be equivalent. This has many practical implications for the way MAXENT is currently used, for example, we can now provide the first definitive method for identifying the spatial resolution to use in analysis. A second area of research is related to using the LASSO in model selection - how should we estimate the shrinkage parameter? I will discuss some preliminary results from species distribution modelling of a Eucalypt distribution in the Blue Mountains.

About the speaker: Ian Renner is a PhD student within the School of Mathematics and Statistics at UNSW. His supervisor is Dr David Warton.