Aiming at the problems of uneven quality,poor reliability and unstable model inversion results of single characteristic parameter data in GNSS reflected signal remote sensing,this paper proposes a GNSS-IR soil moisture inversion method that combines isolated forest and deep learning.The experimental results show that the frequency characteristic parameters of GNSS SNR are not suitable for the inversion of soil moisture,while the amplitude and phase characteristic parameters are highly correlated with soil moisture,which can be used for the inversion of soil moisture.The inversion results of the fusion amplitude and phase characteristic parameters of the three deep learning models of CNN,DBN and GRU are in good agreement with the measured soil moisture.Compared with the single feature parameter inversion method using only amplitude or phase,the inversion accuracy of the proposed method is improved by 21.4％～55.8％,and the correlation coefficient is improved by 4％～9.1％.
万方数据