基于地理加权回归模型的亚热带地区乔木林生物量估算

基于浙江省碳汇样地调查数据,以乔木林生物量(含地上和地下生物量)为因变量,将筛选的与因变量相关性较高的因子作为解释变量,采用地理加权回归和协同克里格方法对乔木林生物量进行估算,对比分析两种估测方法的精度。结果表明:基于地理加权回归方法构建的乔木林生物量估算模型(R2adj=0.820 4,RMSE=23.021 5 t/hm2)精度优于协同克里格方法(R2adj=0.726 3,RMSE=28.054 9 t/hm2),同时使用地理加权回归方法的乔木林生物量预测值的变异系数(Cv=0.618 9)高于协同克里格法(Cv=0.585 4),由此可知地理加权回归方法因考虑了待估变量的局部变异,比协同克里格方法具有更好的拟合结果,预测精度较高。

Biomass Estimation of Arbor Forest in Subtropical Region Based on Geographically Weighted Regression Model

Accurate estimation of arbor forest biomass is of great significance for the study of forest ecological function and carbon storage. Because of the spatial heterogeneity of the survey factors, the geographically weighted regression method can estimate the local regression of variables and show a good application advantage. Based on the survey data of carbon sinks in Zhejiang Province, taking the biomass of arbor forest (including aboveground and belowground biomass) as dependent variable and factors with high correlation with dependent variable as the explanatory variables, the biomass of arbor forest was estimated by using the geographically weighted regression and co-Kriging methods and compared the accuracy of the two estimation methods. The results showed that the accuracy of arbor forest biomass estimation model (R2adj was 0.8204, RMSE was 23.0215t/hm2) constructed by geographically weighted regression method was better than that of co-Kriging method (R2adj was 0.7263, RMSE was 28.0549t/hm2).The coefficient of variation (Cv was 0.6189) of the prediction value of biomass of arbor forest using geographically weighted regression method was higher than that of the co-Kriging method (Cv was 0.5854). Because of considering the local variation of the estimated variables, the geographically weighted regression method had better fitting results than co-Kriging method, and the prediction accuracy was high. This study can provide a reference for estimating the forest biomass and other forest parameters in a wide range of tree stands by using the geographically weighting regression method.