Crown defoliation improves tree mortality models

The objective of this study was to evaluate the predictive power of crown d efoliation, assessed in 5% classes, in predicting year-to-year tree mortali ty. A visual analysis of Swiss Forest Health Inventory (SFHI) data suggeste d an exponential increase in the mortality rate with increasing defoliation . We verified this trend using a logistic regression model with defoliation , social position and their interaction as explanatory variables. We fitted our model to SFHI data for the years 1990-1997 (annual mortality rate=0.32 %), and validated the model with data from long-term forest ecosystem monit oring sites for the years 1995-1998 (annual mortality rate=0.48%). Several indicators of prediction accuracy showed that regression models with total defoliation achieved 40-50% higher accuracies than models with unexplained defoliation, i.e. the portion of defoliation that held crews are unable to attribute to known causes. The logistic regression model with total defolia tion correctly predicted 33% of the dead trees in the calibration data set, and 57% in the validation data set. This prediction accuracy was calculate d with a deterministic method, using a predicted threshold probability abov e which trees were assumed to die. Our study suggests that including defoli ation has the potential of considerably improving the prediction accuracy o f models that predict tree mortality based on competition indicators and tr ee size alone. (C) 2001 Elsevier Science B.V. All rights reserved.