论文部分内容阅读
以石台县为研究地,结合Rapideye高分遥感影像和不同森林类型样地林木地上生物量调查数据,采用Pearson双变量相关分析方法筛选模型变量,分别用多元线性回归和随机森林算法建立不同森林类型的遥感地上生物量估测模型,并进行模型估测精度对比分析。结果表明,叶绿素红边模型(CRM)与叶绿素绿波模型(CGM)2个指数与针叶林、阔叶林生物量在0.01水平上的相关性极显著,且在其多元线性回归模型和随机森林模型中两者均被挑选为建模变量。另外,与生物量相关性较强的纹理特征主要集中的红光波段和红边波段,且仅MEAN、VAR、SM3个滤波对生物量估测贡献较大,可作为建模变量。阔叶林、针叶林和针阔混交林3种森林类型的地上生物量模型估测精度均表现为随机森林模型优于多元线性回归模型。随机森林模型生物估测绝对均方误差在12.8760~36.5363之间,相对均方误差在20.20%~45.95%之间;多元线性回归生物量估测绝对均方误差在22.0425~46.4494之间,相对均方误差在34.58%~58.42%之间。
Taking Shitai County as study area and Rapideye high-resolution remote sensing image and forest floor biomass survey data of different forest types, Pearson’s bivariate correlation analysis method was used to screen the model variables. Multivariate linear regression and stochastic forest were used to establish different forest Type of remote sensing aboveground biomass estimation model, and compared the accuracy of model estimation. The results showed that there were significant correlations between the two indices of chlorophyll red edge model (CRM) and chlorophyll green wave model (CGM) and the biomass of coniferous forest and broad-leaved forest at the level of 0.01. And in the multiple linear regression model and random Both of the forest models were chosen as modeling variables. In addition, the texture features that are highly correlated with biomass are mainly concentrated in the red band and the red band, and only MEAN, VAR, and SM3 filters contribute more to the biomass estimation and can be used as modeling variables. The estimation accuracy of aboveground biomass models of three forest types, ie, broadleaved forest, coniferous forest and coniferous and broadleaf mixed forest, showed that the random forest model was superior to the multiple linear regression model. The absolute mean square error (RMSE) was 12.8760 ~ 36.5363, and the relative mean square error was 20.20% ~ 45.95%. The absolute mean square error of multiple linear regression was estimated to be between 22.0425 and 46.4494 Square error between 34.58% ~ 58.42%.