在多树种、结构复杂的阔叶混交林中,如何基于优势木胸径构建高精度地位指数模型是混交林立地质量评价中亟待解决的科学问题。以河北省木兰围场山杨与白桦阔叶混交林为研究对象,基于70块标准地调查数据(每块标准地面积为0.06 hm²),利用非线性最小二乘法、BP神经网络、非线性混合效应3种参数估计方法分别构建山杨与白桦阔叶混交林地位指数模型,采用均方误差(MSE)、均方根误差(RMSE)、平均偏差百分比(MPB)、决定系数(R²)、调整后的决定系数($R_{\mathrm{adj}}^{2}$)、Akaike信息准则(AIC)、贝叶斯信息准则(BIC)和负2倍对数似然值(-2LL),比较不同参数估计方法对模型预测精度的影响。结果表明:1)在5个候选方程中,以优势木胸径为自变量的Logistic方程是山杨与白桦阔叶混交林的最优地位指数基础模型,其模型的MSE、RMSE、MPB、R²、$R_{\mathrm{adj}}^{2}$分别为9.071 7、2.269 6、11.972 9、0.583 8、0.569 9;2)与非线性最小二乘和BP神经网络相比,基于非线性混合效应构建的山杨与白桦阔叶混交林地位指数模型具有更高的预测精度,其模型的MSE、RMSE、MPB、R²、$R_{\mathrm{adj}}^{2}$分别为5.477 4、1.779 4、9.161 4、0.782 8、0.761 7。以优势木胸径为自变量构建的地位指数模型可用于评价混交林立地质量。

In broad-leaved mingled forest with multiple tree species and complex structures,constructing a high-precision site index model based on the diameter at breast height of dominant trees is a key scientific challenge in evaluating the site quality of these forests.The Populus davidiana and Betula platyphylla broad-leaved mingled forest in Mulan Weichang,Hebei Province is selected as the research object.Based on the survey data from 70 standard plots(each standard land area is 0.06 hm²),using the nonlinear least squares method,BP neural network,and nonlinear mixed-effects model,three parameter estimation methods are used to construct the site index models of P.davidiana and B.platyphylla broad-leaved mingled forest.Using Mean Square Error(MSE),Root Mean Square Error(RMSE),Mean Percentage Bias(MPB),Coefficient of Determination(R²),Adjusted Coefficient of Determination($R_{\mathrm{adj}}^{2}$),Akaike Information Criterion(AIC),Bayesian Information Criterion(BIC),and Negative Twice the Log-Likelihood(-2LL),the impact of different parameter estimation methods on model prediction accuracy is compared.The results show that:1)Among the five candidate equations,the logistic equation with the diameter at breast height of dominant trees as the independent variable is the optimal site index base model for the mingled P.davidiana and B.platyphylla broad-leaved forest,and the MSE,RMSE,MPB,R²,$R_{\mathrm{adj}}^{2}$ of its model are 9.071 7,2.269 6,11.972 9,0.583 8,0.569 9,respectively.2)Compared with the nonlinear least squares method and BP neural network,the site index model of P.davidiana and B.platyphylla broad-leaved mingled forest based on the nonlinear mixed-effects has higher prediction accuracy.The MSE,RMSE,MPB,R²,$R_{\mathrm{adj}}^{2}$ of the model are,5.477 4,1.779 4,9.161 4,0.782 8,0.761 7,respectively.The site index model constructed with the diameter at breast height of dominant trees as the independent variable can be used to evaluate the site quality of mingled forests.