Decision support for identifying spruce forest stand edges with high probability of wind damage

Abstract

In multiple-purpose forestry, windstorms affect both the ecological and economic values that forest stands represent. Silvicultural treatments and forest planning can reduce the probability of wind damage. A tool for the identification of spruce stands with a high probability of wind damage, aimed at helping forestry practitioners target such measures, is presented. Initial assessments of the annual probability of wind damage of exposed stand edges were made for a landscape of about 1200 ha in southern Sweden, using the WINDA system of models. In the calculations, each edge was classified as having either a high or a low annual probability of wind damage. Decision tree methodology using easily accessible variables was employed for identifying the edges classified as having a high probability of wind damage. Since in a multiple-use situation, the risk preferences of decision makers differ, one decision tree was constructed for each of three threshold probabilities of wind damage: 5, 10, and 20%. This corresponds to disturbance intervals of 20, 10, and 5 years or less in each case. The decision trees were found to correctly classify 64–71% of the high-probability stand edges, the misclassification rate for the low-probability stand edges being 12–26%. Alternative cost-matrices were used to take account of user-preferences regarding misclassification rates in the model output. Among the most important predictor variables used in the decision trees were stem taper, gap size in front of the stand edge, and the direction of wind exposure. In an evaluation landscape located 250 km from the parameterization landscape, the decision trees were found to correctly classify 44–50% and 0–83% of the high-probability stand edges with and without use of cost-matrices, respectively. For the evaluation landscape, a statistically significant difference between classifications produced by the decision tree approach and a set of randomly classified stand edges was obtained just for four of the decision trees. This result was explained in terms of the high degree of complexity of the underlying processes, limitations in the parameterization data set, and differences between the landscapes involved. Decision trees of the type described can thus provide help in practical forestry within and nearby the landscape used for constructing the decision trees. In general, the presented methodology appears to be suitable for developing management decision support tools.