【摘要】 为了准确预测兴安落叶松地上生物量，以小兴安岭201株兴安落叶松地上生物量作为研究对象，以胸径（D）和树高（H）为变量，构建随机森林（RF）、人工神经网络（ANN）、支持向量回归（SVR）和梯度提升回归树（GBRT）等4种机器学习模型，并将机器学习算法的预测结果与传统二元生物量模型的预测结果进行对比分析。结果表明：对比传统生物量模型，4种机器学习算法的拟合效果与检验精度均有了大幅度提高。模型拟合精度由高到低的顺序为随机森林、梯度提升回归树、人工神经网络、支持向量回归、传统生物量模型；RF模型在各模型中的拟合精度最高，相对于传统生物量模型，RF模型的确定系数（R²）提升了3.72%,均方根误差(R_MSE)降低了44.47%,平均绝对误差(M_AE)降低了42.81%,相对误差绝对值(M_PB)降低了42.80%,赤池信息准则值降低了18.17%。模型检验精度由高到低的顺序为随机森林、人工神经网络、梯度提升回归树、支持向量回归、传统生物量模型；RF模型在各模型中的预测精度最高，与传统生物量模型相比，RF模型的确定系数（R²）提升了1.08%,均方根误差(R_MSE)降低了10.95%,平均绝对误差(M_AE)降低了10.34%,相对误差绝对值(M_PB)降低了10.34%,赤池信息准则值降低了5.20%。因此，相对于传统生物量模型，4种机器学习算法模型均可以提高兴安落叶松地上生物量的预测精度，RF模型的预测精度最高。更多还原

【Abstract】 In order to accurately predict the aboveground biomass of Larix gmelinii in the Xing’an, 201 samples of L. gmelinii aboveground biomass in the Xiaoxing’an Mountains were taken as the research object. Four machine learning models, including random forest（RF）, artificial neural network（ANN）, support vector regression（SVR）, and gradient boosting regression tree（GBRT）, were constructed using diameter at breast height（D） and tree height（H） as variables. The prediction results of the machine learning algorithms were compared and analyzed with the prediction results of the traditional binary biomass model. The results showed that compared with the traditional biomass model, the fitting effect and test accuracy of the four machine learning algorithms were significantly improved. The order of model fitting accuracy from high to low was random forest, gradient boosting regression tree, artificial neural network, support vector regression, and traditional biomass model. Among all models, the RF model had the highest fitting accuracy. Compared with the traditional biomass model, the RF model increased the coefficient of determination（R²） by 3.72%, reduced the root mean square error(R_MSE) by 44.47%, reduced the mean absolute error(M_AE) by 42.81%, reduced the absolute value of relative error(M_PB) by 42.80%, and reduced the Akaike information criterion(A_IC) value by 18.17%. The order of model test accuracy from high to low was random forest, artificial neural network, gradient boosting regression tree, support vector regression, and traditional biomass model. Among all models, the RF model had the highest prediction accuracy. Compared with the traditional biomass model, the RF model increased the coefficient of determination（R²） by 1.08%, reduced the root mean square error(R_MSE) by 10.95%, reduced the mean absolute error(M_AE) by 10.34%, reduced the absolute value of relative error(M_PB) by 10.34%, and reduced the Akaike information criterion(A_IC) value by 5.20%. Therefore, compared with the traditional biomass model, all four machine learning algorithms can improve the prediction accuracy of L. gmelinii aboveground biomass, and the RF model has the highest prediction accuracy.更多还原