|Wunder, J; Bigler, C; Reineking, B; Fahse, L; Bugmann, H: Optimisation of tree mortality models based on growth patterns, Ecological Modelling, 197, 196-206 (2006)|
Forest succession is often modelled using "gap models" that simulate the establishment, growth and mortality of individual trees. However, many mortality submodels that are currently used in gap models are based on theoretical assumptions and have not been tested with empirical data sets. Except for disturbance-induced mortality; these models predict the time of individual tree death using stress thresholds (ST). They often include a simple stress memory that keeps track of low diameter growth over the recent growth history of each tree, which may lead to increased mortality. In the present study, we optimised the parameter values for a range of commonly used classical ST models. We used the geometric mean of the averages of the correctly classified living and dead trees as our optimisation and model performance criterion. Furthermore, we compared the performance of the ST models with that of recently derived logistic regression models based on growth patterns as predictor variables. Tree-ring data from dead and living Norway spruce (Picea abies) trees of subalpine forests at three study sites in Switzerland were used to calibrate and validate the ST models. The optimisation increased the performance-of the classical ST models by 61-153%. Surprisingly, the model without any stress memory showed the highest performance and thus exceeded the performance of more "realistic" models, i.e., those considering a stress memory. Despite these tremendous improvements, the optimised ST models did not attain the performance of the logistic regression models. Therefore, we conclude that even optimised classical ST models are inferior to regression models with regard to predicting the time of tree death. A considerable change in the simulated forest succession is to be expected if classical ST models that are still used in many gap models are replaced by logistic regression models based on field data. (c) 2006 Elsevier B.V. All rights reserved.