机器学习算法辅助的GPS信噪比观测值土壤湿度反演

Retrieving Soil Moisture Using Signal-to-noise Ratio of GPS Signal by Assisted Machine Learning Algorithm

利用BP神经网络和支持向量回归机两种机器学习算法,构建基于机器学习算法的GNSS卫星反射信号土壤湿度反演模型,并与线性回归统计模型和实测数据进行对比分析。结果表明:基于机器学习算法的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,反演模型的决定系数分别为0.928 3和0.913 1,均方根误差为0.026 6和0.032 6,线性回归统计模型的决定系数分别为0.553 2和0.859 8,均方差根误差分别为0.093 9和0.041 6。说明利用回归算法定量估测土壤湿度明显优于线性回归统计模型,且基于支持向量回归机的土壤湿度反演模型定量估测土壤湿度优于基于BP神经网络算法的土壤湿度反演模型,证明了该方法的可靠性,为土壤湿度的实时反演研究提供了一种新方法。

We construct the soil moisture retrieval model of GNSS satellite reflection signal based on machine learning algorithm with the use of BP neural network and support vector machine,and compared with the linear regression statistical model and the measured data.The results show that the error between soil moisture prediction of retrieval model based on the regression algorithm and reference value of soil moisture is small,the determination coefficient are 0.928 3 and 0.913 1,respectively,the root mean square error are 0.026 6 and 0.032 0,and the determination coefficient of the linear regression statistic model are 0.553 8 and 0.859 2,respectively,the root mean square error are 0.093 9 and 0.041 6.Those demonstrates that the quantitative estimates of soil moisture by using regression algorithm is better than the linear regression model,and the soil moisture retrieval model based on support vector regression is superior to the soil moisture retrieval model based on BP neural network algorithm,which proves the reliability of the method and provides a new method for real-time retrieval of soil moisture.