Modeling Tree Mortality in Low- to Medium-Density Uneven-Aged Hardwood Stands Under a Selection System Using Generalized Estimating Equations

Abstract

Abstract Tree mortality was modeled through time in uneven-aged northern hardwood stands managed under a selection system using long-term remeasurement data. Two models were fitted and compared: a traditional logistic regression model that predicted the probability of individual tree mortality over discrete time periods and a logistic regression model estimated using generalized estimating equations (GEEs) to account for autocorrelation in the longitudinal data. Model evaluation was based on the mean prediction error, mean absolute prediction error, variance of prediction errors, and mean square error calculated using an independent validation data set. The GEE model produced smaller evaluation statistics, especially for the smaller diameter classes. The predicted probability of mortality from the two models was compared with the observed mortality across 5-cm diameter classes for two time periods. Our results indicate that the GEE model was better able to capture the change in the probability of mortality over time, especially for the smaller diameter classes, than the more traditional logistic model.