Development and comparative diagnosis of conventional (linear/nonlinear) and artificial intelligence techniques-based predictive models for estimating timber volume of Tectona grandis

Abstract

This study aimed to develop volume estimation models which will be robust and useful for predicting merchantable volume of teak trees in different teak growing regions of the world. The data was statistically simulated based on various published models for different teak growing conditions in different parts of the world. A total of thirteen models comprising nine conventional (linear and nonlinear) and four Artificial Intelligence (AI) techniques-based models, thus two Support Vector Machine (SVM) techniques and two Artificial Neural Network (ANN) techniques, were fitted to the data. Several statistical model selection criteria including Efron’s pseudo R-squared, Root Mean Square Error, Mean Absolute Bias, Nash-Sutcliffe Efficiency, Index of Agreement and Akaike Information Criterion were used to evaluate and rank the models’ performances from best to worst. All AI techniques-based models were superior over conventional models in performance, and the overall best model was SVM technique followed by an ANN technique. Among conventional models, allometric models generally fitted the data better than linear regression type models, with model 𝑀5 being the best while 𝑀2 was the worst. Combination of tree diameter at breast height (dbh) and height as predictors of tree volume was shown to improve model prediction accuracy for teak trees irrespective of the model involved. On the basis of the varied nature of the data used for model fitting, the developed models would be useful in making reliable predictions of teak timber volume for different teak growing regions across the world. The models have wide application potential and may be recommended for use in managing teak plantation inventory in different parts of the world.