Methods for estimating aboveground biomass and its components for Douglas-fir and lodgepole pine trees

Abstract

Estimating aboveground biomass and its components requires sound statistical formulation and evaluation. Using data collected from 55 destructively sampled trees in different parts of Oregon, we evaluated the performance of three groups of methods to estimate total aboveground biomass and (or) its components based on the bias and root mean squared error (RMSE) that they produced. The first group of methods used an analytical approach to estimate total and component biomass using existing equations and produced biased estimates for our dataset. The second group of methods used a system of equations fitted with seemingly unrelated regression (SUR) and were superior to the first group of methods in terms of bias and RMSE. The third group of methods predicted the proportion of biomass in each component using beta regression, Dirichlet regression, and multinomial log-linear regression. The predicted proportions were then applied to the total aboveground biomass to obtain the amount of biomass in each component. The multinomial log-linear regression approach consistently produced smaller RMSEs compared with both SUR methods. The beta and Dirichlet regressions were superior to both SUR methods except for Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco) branch biomass, for which the simple SUR method produced smaller RMSE compared with the beta and Dirichlet regressions.

Aboveground Biomass Equations for Small Trees of Brutian Pine in Turkey to Facilitate Harvesting and Management

“…However, including height in addition to DBH improves the accuracy of such models [8,23]. A variety of model forms such as power function (e.g., Picard et al [23]), simple logarithmic model resulting from logarithmic transformation of power function (e.g., Poudel and Temesgen [9]), and exponential model (e.g., Ritchie et al [24]; Poudel and Temesgen [8]) have been used to relate such dendrometric variables to AGB. A simple linear model in the form of Equation (1) was first tested.…”

“…To ensure the additivity of such equations, the constrained seemingly unrelated regression has been popular. Recently, component biomass has been estimated as the product of predicted proportion obtained from different generalized linear models [5,8,9,25] and the predicted total biomass obtained from the method described in the previous section. In this study, the proportion of biomass in different tree components was estimated using three generalized linear models: beta regression, Dirichlet regression, and multinomial logistic regression.…”

“…The beta distribution is a continuous distribution defined on the unit interval. It has been used in forestry to model percent canopy cover [27], riparian percent shrub cover [28], component biomass proportions [5,8,9], and to model basal area mortality due to fire [29].…”

“…The third group of methods, fairly new in biomass estimation, predicts the proportions of biomass in each component using generalized linear models such as beta, Dirichlet, and multinomial logistic regression. Predicted proportions are then applied to the observed total AGB [8] or the total AGB obtained from fitting a separate equation [9].…”

“…Common independent variables of biomass models are DBH and tree height, although some studies have used crown dimensions as well. Poudel and Temesgen [8] indicated that factors that affect growth in tree diameter and height (e.g., genetics, site quality, environmental factors, stand density, tree and stand age) also affect AGB and component biomass. Therefore, there is a need to develop tree-level biomass models using data from areas within the natural distribution of this species in the Mediterranean Region.…”