要旨
Tree-ring chronologies have been widely used in studies of tree mortality where variables of recent growth act as an indicator of tree physiological vigour. Comparing recent radial growth of live and dead trees thus allows estimating probabilities of tree mortality. Sampling of mature dead trees usually provides death-year distributions that may span over years or decades. Recent growth of dead trees (prior to death) is then computed during a number of periods, whereas recent growth (prior to sampling) for live trees is computed for identical periods. Because recent growth of live and dead trees is then computed for different periods, external factors such as disturbance or climate may influence growth rates and, thus, mortality probability estimations. To counteract this problem, we propose the truncating of live-growth series to obtain similar frequency distributions of the “last year of growth” for the populations of live and dead trees. In this paper, we use different growth scenarios from several tree species, from several geographic sources, and from trees with different growth patterns to evaluate the impact of truncating on predictor variables and their selection in logistic regression analysis. Also, we assess the ability of the resulting models to accurately predict the status of trees through internal and external validation. Our results suggest that the truncating of live-growth series helps decrease the influence of external factors on growth comparisons. By doing so, it reinforces the growth–vigour link of the mortality model and enhances the model’s accuracy as well as its general applicability. Hence, if model parameters are to be integrated in simulation models of greater geographical extent, truncating may be used to increase model robustness.
