Statistical models to identify stand development stages by means of stand characteristics

Abstract

Stand development stages differ mainly in terms of stand structure, stand density, and mortality patterns. As the fulfilment of socio-economic forest functions often depends on stand structure and density, knowledge of the frequency and distribution of stand development stages is needed for optimal forest management. Development stages have been previously identified only qualitatively by experts in forest ecology, but this study developed and compared statistical models to identify development stages by means of stand characteristics. Data from the Austrian National Forest Inventory with 4761 observations of stand development stages were used as the training data set for quadratic discriminant analysis and multinomial logistic regression. The models differ only marginally in terms of the hit ratio and the overall kappa statistic (both determined by means of an independent test data set). The quadratic discriminant analysis has the advantage that the user can reduce or even avoid the influence of the group size on the group-specific model performance by using equal prior probabilities. Furthermore, the discriminant analysis showed the best model behaviour in terms of the explanatory variables and performed best in identifying the stages that were infrequent in the training data set.