Model uncertainty in statistical modelling and regionalisation - assessing the effect of climate change on species distribution

Presenting person: Dr. Boris Schröder, Institute of Geoecology, University of Potsdam
Th. 2008-05-29 (16:15), H 6

Statistical species distribution models are a useful tool to determine the habitat factors that govern the spatial and temporal distribution of species and to assess the effect of changing environmental conditions on species distributions. This talk presents species distribution models for rock ptarmigan, an alpine bird species. A multi-scale approach considering territory scale and larger scales with 1 km2 as well as 100 km2 grain size shows the scale dependency of environmental predictors. To analyse the effect of climate change on the spatial distribution of this species, scenarios are used to predict the future habitat in 2030, 2050 and 2070. Predictions are calculated with several methods (i.e. standard methods such as GLM, GAM, CART as well as sophisticated machine learning and ensemble forecasting methods such as Boosted Regression Trees and Random Forest) to analyse the high amount of model uncertainty related to the statistical approach. All models are checked for residual spatial autocorrelation and internally validated by bootstrapping to receive unbiased estimates of model performance.
The second part of the talk will focus on the applicability of the methods applied in this study in related disciplines such as soil landscape modelling and predictive eco-geomorphological modelling.

Export as iCal: