黔中马尾松木荷混交林树高-胸径模型OA北大核心CSTPCD

[Objective]To establish a tree height-diameter model for mixed forests of Pinus massoniana and Schima superba,introduce tree species as dummy variables into the model,and consider the spatial autocorrelation and heterogeneity of residuals of the model,in order to provide theoretical basis for the construction of the tree height-diameter model of mixed forests and the scientific management of mixed forests.[Method]Based on the survey data of 727 groups of tree height-diameter in mixed forests of P.massoniana and S.superba in Kaiyang County,Guizhou Province,we constructed ordinary least squares(OLS),generalized additive model(GAM),linear mixed model(LMM),geographically weighted regression model(GWR),and geographically weighted regression kriging(GWRK)models for tree height-diameter-whole-forest model,on the basis of which tree species was introduced as a dummy variable,and five model residuals spatial autocorrelation and spatial heterogeneity were selected for global Moran'I,local Moran'I,and intra-block variance analyses with the coefficients of determination(R2),mean squared error(MSE),and the Akaike information criterion(AIC)to evaluate the models.[Result](1)The fitting accuracies of the whole-forest base models of P.massoniana-S.superba mixed forest were OLS＜GAM＜LMM＜GWR＜GWRK in descending order.(2)The fitting accuracies of the models were higher than those of the whole-forest base models after introducing tree species as a dummy variable into the models.(3)The global Moran'I of the residuals of the OLS and GAM models was significant at the α=0.05 level(Z＞1.96),and there were more hot spots in the local Moran'I distribution maps,which showed strong spatial autocorrelation.In contrast,the global Moran'I of the residuals of the LMM,GWR and GWRK models is insignificant at the α=0.05 level(-1.96≤Z≤1.96)and there are more cold spots in the local Moran'I distribution plot,indicating that spatial autocorrelation of the model residuals has been eliminated.(4)The intra-block variance of the residuals of the five models show a tendency to increase with the lag distance,but the GWR and GWRK models have smaller intra-block variance,which can better reduce the heterogeneity of the model residual space.[Conclusion]The OLS and GAM models do not have high fitting accuracy and cannot eliminate spatial autocorrelation and heterogeneity of model residuals,so they are not the best choices for modeling tree height-diameter.The LMM,GWR,and GWRK models perform well in improving the model fitting accuracy and decreasing the spatial autocorrelation,but the GWR and GWRK models are more significant in decreasing the spatial heterogeneity,and they are the most appropriate of the tree height-breast diameter models.[Ch,2 fig.3 tab.38 ref.]