结合机器学习的GNSS-IR多卫星双频组合土壤湿度反演

GNSS-IR multi-satellite dual-frequency combination soil moisture inversion combined with machine learning

为提高土壤湿度反演精度,并克服基于单一卫星、单一频点开展土壤湿度反演存在的不足,本文提出了一种结合机器学习的GNSS-IR多卫星双频组合土壤湿度反演方法,将GNSS卫星L1、L2频点上SNR作为数据源进行土壤湿度反演研究,采用BP和RBF神经网络算法构建土壤湿度预测模型,并与一元线性回归预测模型进行对比分析。试验结果表明:(1)相对于单卫星而言,多卫星组合的土壤湿度反演增加了有效卫星利用率,并提高了土壤湿度反演的精度;(2)多星组合的L1、L2双频均值融合延迟相位观测值与土壤湿度的相关系数为0.956,均优于L1、L2频点反演结果;(3)BP、RBF神经网络模型预测结果精度均优于ULR模型预测结果。

In order to improve the accuracy of soil moisture inversion and overcome the shortcomings of soil moisture inversion based on single satellite and single frequency point,this paper proposes a soil moisture inversion method based on GNSS-IR multi-satellite dual-frequency combination combined with machine learning.By using the SNR of GNSS satellite L1 and L2 frequency points as the data source,the soil moisture inversion is studied.The BP and RBF neural network algorithms are used to construct the soil moisture prediction model,and compared with the linear regression prediction model.The experimental results show that:(1)Compared with the single satellite,the soil moisture inversion of multi-satellite combination increases the effective satellite utilization rate and improves the accuracy of soil moisture inversion.(2)The correlation coefficient between L1 and L2 dual-frequency mean fusion delay phase observations of multi-satellite combination and soil moisture is 0.956,which is better than the inversion results of L1 and L2 frequency points.(3)The prediction accuracy of BP and RBF neural network models is better than that of ULR model.