基于地理加权回归模型的思茅松生物量遥感估测

Remote Sensing Estimation of Biomass of Pinus kesiya var.langbianensis by Geographically Weighted Regression Models

通过调查云南省景谷县思茅松林120株单木数据,构建思茅松单木生物量模型。结合2005年景谷县TM影像数据及2006年森林资源二类调查小班数据,采用普通最小二乘模型(OLS)和地理加权回归模型(GWR)的方法构建思茅松生物量遥感估测模型。结果表明:地理加权回归模型比普通最小二乘模型具有更好的拟合效果,其决定系数(R2)显著高于OLS模型,Akaike信息指数(AIC)相比降低7.832;两种模型通过独立样本检验可以看出,模型预估精度从OLS模型的72.70%提高至GWR模型的75.06%;通过GWR模型反演计算,研究区内思茅松林单位面积生物量为49.02t/hm^2,比实测数据低1.229%,与实测数据基本吻合,且估算误差优于OLS模型;基于GWR模型估算的景谷县思茅松林总生物量为2.101×107t。可见基于地理加权回归方法估测森林地上生物量的方法是有效的,能提高森林生物量遥感估测模型的拟合和预估精度,可以用于思茅松林的生物量的遥感估算。

The Biomass model of Simao pine （ Pinus kesiya var. langbianensis） was built based on the data collected from 120 Simao pine sampling trees, Landsat TM images in 2005 and the data of forest resource in- ventory in 2006 in Jinggu County, Yunnan Province. Then the remote sensing biomass estimation Model of Simao Pine were built by the ordinary least square （ OLS ） and geographically weighed regression （ GWR）.The results showed that ： GWR model had a better fitting effect than OLS, in which coefficient of determina-tion （/？） was significandy bigger than the OLS model ,Akaike information index （A/C） reduced by 7. 832; It was obviously depicted from the sample test of independence that model prediction accuracy was improved from 72. 70% （ OLS） to 75. 06 % （ GWR）. The unit-area biomass was 49. 02t / hm2 by inversion, and basical-ly consistent with the measured data; it was lower than the measured data 1. 229% ,and less than the esti-mation value of OLS. The total biomass of Simao pine in Jinggu County was 2. 101 X 107 t based on GWR model. The study indicated that forest aboveground biomass estimation based on geographically weighed re-gression （GWR） model could improve effectively the fitting accuracy of forest biomass estimation model,and could be used to estimate the biomass of Simao pine forest by remote sensing.