Boosting for estimating spatially structured additive models

Abstract

Spatially structured additivemodels offer the flexibility to estimate regression relationships for spatially and temporally correlated data. Here, we focus on the estimation of conditional deer browsing probabilities in the National Park Bayerischer Wald. The models are fitted using a componentwise boosting algorithm. Smooth and non-smooth base learners for the spatial component of the models are compared. A benchmark comparison indicates that browsing intensities may be best described by non-smooth base learners allowing for abrupt changes in the regression relationship. © 2010 Springer-Verlag Berlin Heidelberg.