Robust learning-based prediction for timber-volume of living trees

Abstract

Many existing timber volume calculating applications are applied by solving a single or multi-variables formulation, while most of those methods are based on the felled trees, which lead to a number of living trees is cut down yearly in different areas. In this paper, a novel learning-based nonlinear timber volume predicted model is proposed, which based on the least squares support vector machine (LSSVM) algorithm, and a modified particle swarm optimization (MPSO) algorithm is used to optimize the parameters involved in the LSSVM. Specifically, the initial weight coefficient in classical particle swarm optimization (PSO) is modified, such that the global optimal solution can be obtained more fleetly and accurately, meanwhile, the timber volume predicted model is established based on the modified algorithm. The experiments are carried out on our collected data, which are obtained from Xiashu plantation of Jurong in Jiangsu Province of China. Three kinds of trees, named Populus, Liriodendron and Soapberry, are selected as the experiment samples. The historical timber volume data of the same kinds, used as the training set in the proposed MPSO-LSSVM model, are obtained from the management of Xiashu plantation. The two properties from the manually measured data, including tree height and diameter at breast height (DBH), are used as the input parameters in the testing set of MPSO-LSSVM. Furthermore, the virtual trees, generated by computers, provide a novel approach to estimate the predicted accuracy of the learning-based model in forest inventory. The experiment results in comparisons with the solutions from volume equation, taper function, felled trees and the virtual trees demonstrate the availability and efficiency of the proposed model in prediction of timber volume.