思茅松天然林单木生物量地理加权回归模型构建

Modeling Individual Biomass of Pinus kesiya var. langbianensis Natural Forests by Geographically Weighted Regression

通过调查云南省思茅区思茅松天然林63株思茅松单木的地上部分干、枝、叶生物量数据,并测定其中30株的根系生物量数据。基于普通最小二乘模型选型,采用地理加权回归的方法构建思茅松单木树干生物量、树枝生物量、树叶生物量和地上部分生物量,以及根系生物量和整株生物量模型。结果表明:(1)地理加权回归模型(GWR)的决定系数(R2)大于普通最小二乘(OLS)模型,且GWR模型拟合的R2值除树叶生物量模型外,其余生物量维量模型均大于0.950;Akaike信息指数(AIC)值小于普通最小二乘(OLS)模型,平均相对误差(EE)和平均相对误差绝对值(RMA)的绝对值除树枝生物量外均小于OLS模型,说明GWR模型拟合效果优于OLS模型;(2)地理加权回归模型拟合在一定程度上克服了OLS在拟合生物量模型中存在的异方差问题。

The stem,branch,leaf biomass of 63 sampling trees,and root biomass of 30 trees at Simao pine （Pinus kesiya var.langbianensis） natural forest were investigated in Simao district of Yunnan Province.Based on the model selected by ordinary least square （OLS）,the models of the tree stem,branch,leaf biomass,aboveground biomass,root biomass and whole tree biomass were built by geographically weighted regression （GWR）.The results showed that：（1） the values of the coefficient of determination （R2） of GWR were greater than that of OLS models,and the R2 of the GWR models were greater than 0.950 except the leaf biomass model.Akaike＇s information criterions （AIC） of GWR were less than that of OLS models,the absolute value of the mean relative error （EE） and the mean absolute relative error （RMA） were less than that of OLS model except the branch biomass model.So the fitting effect of GWR outperforms OLS models.（2） For individual tree biomass models,GWR overcame the heteroscedasticity of the OLS models at a certain extent.