Modeling and allocating forestry survival
Abstract
The forest industry has a major economic impact on the Southeastern United States and loblolly pine (Pinus taeda L.) is the primary commercial species. Consequently, many management tools have been developed to aid in the management of loblolly pine. These tools include growth and yield models that predict stand growth and corresponding wood yield. Hence, growth and yield systems, which generally include survival, basal area, height, and volume models, have garnered considerable research attention in recent years. Prediction of surviving stems per unit area, which is critical in forecasting wood yield, is an important component of these growth and yield systems. The importance of survival prediction can be demonstrated using whole stand and individual tree or dbh class survival predictions for projecting stand tables, which are used for generating stock tables. Our study, which uses permanent plot loblolly pine data, builds upon the existing foundation of forestry survival models: both whole stand and individual tree, and assesses the impact of mortality allocation in stand table projection algorithms. We develop a generalized methodology for deriving flexible whole stand survival models for the continuum of a stand’s development by merging traditional survival analysis and existing whole stand survival methods. In addition, we demonstrate a methodology for modeling interval-censored individual tree survival data and show that the model derivation naturally leads to the complementary log-log survival function. Our individual tree survival model accounts for heterogeneity that occurs within and among plots by using multilevel modeling techniques. Moreover, since logistic regression is the most common technique used for modeling individual tree survival, we document the utility of using a multilevel individual tree logit model. Lastly, the multilevel logit individual tree survival model is used in projecting stand tables and assessed with a commonly used stand table projection algorithm.