Comparative analysis of machine learning algorithms and statistical models for predicting crown width of Larix olgensis

Crown width is one of the most important crown dimensions that influence tree growth and survival. Accurate crown width prediction is vital for forest management. However, measuring crown width is time-consuming and labor-intensive, making it necessary to construct a convenient and accurate crown width prediction model. Nowadays, machine learning technologies have already been increasingly used to accurately predict tree growth, but there is still a lack of systematic and comprehensive comparison. This paper provided a comparative analysis of various machine learning methods (i.e., the Linear Regression, the Least Absolute Shrinkage and Selection Operator, the k-NearestNeighbors, the Random Forest, the Gradient Boosting Decision Tree, the Support Vector Regression, the Voting Regressor and the Multi-Layer Perceptron), simple crown width-diameter of breast height model, generalized crown width-diameter of breast height model and nonlinear mixed-effect crown width model for estimating the crown width of Larix olgensis in terms of the coefficient of determination, root mean squared error, mean absolute error, and mean absolute percentage error using hold-out validation and 10-fold cross-validation. The study showed that machine learning algorithms performed better than common nonlinear regression and nonlinear mixed-effect models. Specifically, the voting regressor and random forest algorithm yielded the highest prediction quality of crown width among the different models. In addition, from a practical point of view, the advantage of machine learning is that its implementation does not require crown width measurements. On the contrary, the calibration of the mixed-effect model requires prior information, which limits its use.