Objective  This paper aims to study the application of multiple machine learning algorithms in the prediction of bark thickness, to compare and analyze the influence of different individual tree factors on the prediction of bark thickness, and to provide new methods for the prediction of bark thickness.

  Method  Four machine learning algorithms (neural network, support vector regression, decision tree, random forest) were constructed based on the bark thickness data of Dahurian larch (Larix gmelinii) in Daxing’anling Mountains of northeastern China. Their performance in predicting bark thickness was compared with six traditional bark thickness models. The determination coefficient (R²), root mean square error (RMSE), mean absolute error (MAE) and Akaike information criterion (AIC) were used to evaluate different models and algorithms.

  Result  (1) Among the six basic models, Model 5 showed better prediction results. In the comparison between the basic model and the machine learning models, all the machine learning models fitted better accuracy than the traditional model Model 5 except for the CART4 model. (2) Among the machine learning models, the fitting and prediction accuracy of ANN4 and SVR3 were similar, and RF4 was the best. (3) The input variables of RF4 were diameter at breast height (DBH), tree height (H), and relative tree height (H_r). Based on the training samples, the R² of random forest increased from 0.675 2 to 0.723 4, RMSE decreased from 0.575 5 to 0.531 0 compared with Model5. Based on the testing samples, the R² of random forest increased from 0.666 9 to 0.710 5, RMSE decreased from 0.616 9 to 0.544 6 compared with Model 5.

  Conclusion  Compared with the basic bark thickness model, random forest, support vector machine regression and artificial neural network in machine learning algorithm can improve the prediction accuracy of bark thickness. The prediction effect of random forest is slightly better and suitable for the prediction of bark thickness of Dahurian larch in this region.