BP神经网络辅助的GNSS反射信号土壤湿度反演

BP neural network assisted soil moisture retrieval using GNSS satellite reflected signal

针对如何快速准确地估算区域尺度上的土壤湿度问题,该文首先从高质量GPS接收机接收的信噪比观测值中,提取L2C反射信号的振幅和相位作为输入,并采用Noah陆面模型计算土壤湿度值作为期望值,构建基于BP神经网络算法的GNSS卫星反射信号土壤湿度反演模型。实验结果表明:基于BP神经网络算法的GNSS卫星反射信号土壤湿度反演方法获取的土壤湿度结果与土壤湿度参考值误差较小,线性回归的决定系数R2为0.909 1,均方根误差为0.028 7;进一步与线性回归统计模型比较发现,利用BP神经网络模型定量估测土壤湿度明显优于线性回归统计模型,证明了该方法的可靠性。

Aiming at the problem of how to estimate the soil moisture on the regional scale quickly and accurately,this article first extracted L2 Creflected signal amplitude and phase as input,which came from the observation value of signal-to-noise coming from high quality GPS receiver and used Noah land surface model to calculate soil moisture as expected value to build soil moisture inversion model of GNSS satellite reflected signal based on BP neural network algorithm;the test result showed that the difference between soil moisture reference and predictive value of soil moisture inversion model which based on BP neural network algorithm was small,the regression coefficient of decision was 0.909 1,the root mean square error was 0.028 7;and by comparing with linear regression model,it was found that quantitative estimates of soil moisture by using BP neural network model was better than the linear regression model,which proved the reliability of this method.