基于随机森林回归方法的水稻产量遥感估算

为寻求高效的水稻产量估算方法,以2017年长春市九台和德惠地区的采样点为样本,遥感数据和气象数据为特征变量,通过对产量与特征变量间的相关性分析与特征变量之间的主成分分析和袋外数据(out-of-data,OOB)变量的重要性分析对特征变量进行选择,以选择后的特征变量为输入变量建立水稻产量估算的随机森林回归(RFR)模型。结果表明:特征变量优选后的RFR模型对水稻产量估算的精度更高,决定系数R~2和平均相对误差MRE分别为0.950和0.060;并将该模型应用到农安地区,以多元逐步回归模型作为比较模型,表明RFR模型的水稻产量估算精度明显优于多元逐步回归模型,RFR模型的R~2和MRE分别为0.730和0.090,多元逐步回归模型的R~2和MRE分别为0.530和0.120。

Remote sensing estimation of rice yield based on random forest regression method

In order to find an efficient method to estimate rice yield, the sampling points in Jiutai and Dehui areas of Changchun City in 2017 were taken as study objects. Remote sensing data and meteorological data were taken as characteristic variables. The correlation analysis between yield and characteristic variables, principal component analysis between characteristic variables and OOB importance analysis were employed to select the characteristic variables. The optimal characteristic variables were taken as the input variables to establish the random forest regression(RFR)model. The results showed that the RFR model with optimized characteristic variables had higher estimation accuracy in rice yield estimation. The coefficient of determination R2 and the mean relative error(MRE)were 0. 950 and 0. 060 respectively. The model was applied to the Nongan area and the multiple stepwise regression model was used as a comparative model. The results showed that the RFR model was significantly better than the multiple stepwise regression model. The R2 and MRE of RFR model were respectively 0. 730 and 0. 090, and the R2 and MRE of multiple stepwise regression model were respectively 0. 530 and 0. 120.