Developing a new model for predicting the diameter distribution of oak forests using an artificial neural network

Abstract

The parameters of the probability density function (PDF) may be estimated using the parameter prediction method (PPM) and the parameter recovery method (PRM). However, these methods can suffer from accuracy issues. We developed and evaluated the prediction accuracy of two PPMs (stepwise regression model and dummy variable model) and an artificial neural network (ANN) to predict diameter distribution using data collected from 188 oak forest plots. The results demonstrated that the Weibull distribution performed well in fitting the diameter distribution. Compared with the stepwise regression model, the PPM model with stand type as a dummy variable reduced the predictional errors in estimating the parameters b and c of the Weibull distribution, but the prediction accuracy of the diameter distribution showed no significant improvement. Compared with the two PPM models, the ANN model with diameter class (C), average diameter (D) and stand type (T) as input variables decreased the RRMSE by 2.9% and 4.33% in estimating diameter distribution, respectively. The satisfactory prediction accuracy and simple model structure indicated that an ANN worked well for the prediction of the diameter distribution with few requirements and high practicality.