LAUSR

Abstract Tree mortality is caused by complex interactions between multiple biotic and abiotic factors. Processes of tree mortality that are not induced by natural disturbances are often reflected in distinct radial growth patterns of trees, which typically serve as reliable indicators of impending tree mortality. However, it remains unclear whether empirical mortality models that are based on tree size and growth result in more realistic projections of forest succession in dynamic vegetation models (DVMs). We used a combination of tree-ring and inventory data from unmanaged Swiss natural forest reserves to derive species-specific survival models for six Central European tree species (Abies alba, Fagus sylvatica, Larix decidua, Picea abies, Pinus cembra and Quercus spp.). We jointly used 528 tree-ring samples and inventory data from eight forest reserves. We implemented the estimated parameters of the survival models into the DVM ForClim and performed simulations of forest succession that were validated using the inventory data of the forest reserves. Size- and growth-dependent variables (i.e., diameter at breast height and mean ring width) over the last few years prior to tree death were reliable predictors to distinguish between dying and living trees. Very low mean ring widths over several preceding years as well as small and large trees, respectively, reflected low survival probabilities. However, the small sample sizes of small and large trees resulted in considerable uncertainty of the survival probabilities. The implementation of these survival models in ForClim yielded plausible projections in short-term simulations and for some sites improved the predictions compared to the current ForClim version. Stand basal area, however, tended to be overestimated. Long-term simulations of ForClim based on the empirical survival models resulted in realistic predictions only if the uncertainty of the predicted survival probabilities was considered. We conclude that the combination of different data sources in combination with the consideration of intra-specific trait variability yields robust predictions of tree survival probabilities, thus paving the way towards better tree mortality models and more reliable projections of future forest dynamics.