多元线性回归与神经网络模型在森林地上生物量遥感估测中的应用

Remote Sensing Estimation of Forest Aboveground Biomass Based on Multiple Linear Regression and Neural Network Model

利用遥感影像构建森林生物量估测模型,能够快速、实时估算区域森林生物量。采用吉水县TM影像以及森林资源调查固定样地数据,构建估算森林地上生物量的多元线性回归模型及BP神经网络模型,并对两种模型进行了比较。结果表明:两种模型对样地生物量的预测值大部分比实测值小,多元线性回归模型预测值与实测值的偏差幅度比BP神经网络模型更大,偏差幅度分别为-110.24~38.09 t·hm-2、-35.12~26.17 t·hm-2;多元线性回归模型与BP神经网络模型的决定系数(R2)分别为0.470和0.869,均方根误差(RMSE)分别为30.52和12.69 t·hm-2,预测精度分别为50.07%和71.65%。因此,运用BP神经网络模型估测森林地上生物量优于多元线性回归模型。

With the TM image and the permanent plot data of forest management inventory in Jishui County,the multiple linear regression and BP neural network model were established to estimate forest aboveground biomass. The most biomass predicted values of the two models were smaller than the measured values. However,compared with the BP neural network model,the deviation of regression model between the predicted value and the measured value was greater,and the deviation values were-110.24-38.09 t·hm^-2,and-35.12-26.17 t·hm^-2,respectively. The R2,root mean square error（ RMSE）and prediction accuracy of multiple linear regression model were 0. 470,30. 52 t·hm^-2,and 50.07%,and those of BP neural network model were 0. 869,12. 69 t · hm^-2,and 71.65%,respectrvely. Therefore,using the BP neural network model to estimate the forest aboveground biomass was better than the multiple linear regression model.