Files
Abstract
Projecting forest inventory plays an essential role in forest management. In this study I focused on techniques for projecting forest inventories, such as projection models, stand table projection techniques, and southern annual forest inventory system (SAFIS) sample plot updates. I developed a new stand table projection model whose model form was derived based on the same assumptions as the Pienaar & Harrison equation (1988). The new stand table model is a two random effects model. It significantly outperformed the Pienaar & Harrison stand table projection model using data for Consortium for Accelerate Pine Plantation Studies (CAPPS). The new stand table modeling technique is an integration of a new expectation function, maximum likelihood estimation, and Empirical Best Linear Unbiased Predictor (EBLUP). I proposed the quantile regression estimator for parameters of percentile growth models. According to extensive simulation analyses, the new estimator favorably compared with ordinary least squares in terms of the first order and second order statistics, especially when error terms are heteroscedastic. Simulation results indicated that the gain from quantile regression was approximately proportional to heteroscedasticity. In forest biometrics, it is often the situation that only one observation is available for predictions due to investment and biological limitations. Accordingly, mixed models are not necessarily superior to projection models. However, mixed models are appropriate for updating SAFIS sample plots since multiple observations will become available as the inventory cycle repeats. EBLUP can provide the best prediction in comparison with any other methodologies available through a weighted scheme that uses information on individual observations and the population mean.