Hypothesis-driven species distribution models for tree species in the Bavarian Alps



Question: What are the main drivers for tree species distribution in the Bavarian Alps? What are the species-specific habitat requirements? Are predictions in accordance with expert knowledge?

Location: Bavarian Alps (Southern Germany).

Methods: To describe tree species–environment relationships, we established species distribution models for the 14 most common tree species of the region. We combined tree species occurrence data from forest inventories and a vegetation database with environmental data from a digital elevation model, climate maps and soil maps. For modelling, we used generalized additive models (GAM) combined with techniques to account for spatial autocorrelation and uneven coverage of environmental gradients. We developed parsimonious models to judge whether statistical models correspond to models based on expert knowledge.

Results: Conceptual models were generally in accordance with expectations. Variables based on average temperatures were the most important predictors in most models. Proxies for soil properties such as water and nutrient availability were statistically significant and generally plausible, but appeared largely redundant for model performance. Altitudinal limits of tree species were generally well represented by models. Most species responded differently to summer and January temperatures, indicating that temperature variables are proxies not only for energy balance, but also for frost damage and drought. Although model building benefits considerably from collation with expert knowledge, there are limitations.

Conclusions: Meaningful species distribution models can be obtained from noisy data sets covering only a small fraction of species ranges. Models calibrated with such data sets benefit from hypothesis-driven model building rather than strict data-driven model building. Hence, misleading explanations and predictions can be avoided and uncertainties identified. Nevertheless, projections based on climate scenarios can be substantially improved only with models calibrated on a wider data set. Ideally, environmental gradients should cover the whole niche space of a species, or at least include regions with analogous climate.