![Masters Theses](../../assets/md5images/ebc0648ded2cd429f55e0a2fe4407546.gif)
Masters Theses
Date of Award
12-2023
Degree Type
Thesis
Degree Name
Master of Science
Major
Forestry
Major Professor
Sheng-I Yang
Committee Members
Consuelo Brandeis, Donald G. Hodges
Abstract
Accurately predicting the occurrence of harvest activities is important to monitor the dynamics of natural resources in a region. Harvesting in forestry is influenced by various site factors, which are interconnected and may contribute to the variation observed in harvesting. Survival analysis composed of a group of analytical approaches provides a possible solution to achieve the goal. Although survival analysis has been used to predict harvest occurrence in the past, such application is more commonly studied for predicting tree mortality in forest modeling. Recently, random survival forest (RSF) as an extension of the original random forest algorithm has proven useful to capture the complex relationships among variables. However, to our knowledge, the predictability of the RSF models for harvest occurrence has not been explicitly studied in forestry. Thus, the main objective of this study was to employ the RSF algorithm to examine the temporal evolution of tree harvest times, accounting for stand and environmental variables. Specifically, the predictability of the RSF model was compared with the Cox proportional hazard (Cox) model, a popular model in survival analysis. Important predictor variables were identified, which were then used to construct the reduced models (i.e., models with a reduced number of predictors). Data collected from permanent plots in the southern Appalachian region maintained by the USDA Forest Service Forest Inventory and Analysis (FIA) program were utilized in analysis.
Results showed that the RSF model consistently outperformed the Cox model based on prediction accuracy. Among 11 variables examined, ownership, elevation, and slope are top three predictor variables. It was found that reduced models built with (1) ownership, elevation, and slope or (2) elevation, slope, state, and forest type can produce satisfactory prediction accuracy compared to the full models (i.e., the models with all variables included). The findings of this work provide insights for forest managers and policy makers to utilize survival analysis methods in understanding harvest activities at the regional scale.
Recommended Citation
Senevirathne, Dinuka Madhushan, "Predicting Harvest Activities in Mixed-species Forests using a Random Survival Forest Algorithm. " Master's Thesis, University of Tennessee, 2023.
https://trace.tennessee.edu/utk_gradthes/10133