BDS Multi-frequency Soil Moisture Retrieval Considering the Amplitude Stability of Reflected Signal

With the successful networking of BeiDou Navigation Satellite System (BDS), BDS B1I/B2I/B3I signals enriched GNSS-IR data sources. GNSS-IR technology used to monitor soil moisture content (SMC) is constantly developing, but there are still the following problems in the current research. Firstly, the elevation angle range used for inversion is mostly determined based on experience, ignoring the problem that the amplitude of reflected signal varies greatly within a certain elevation angle range due to the influence of antenna gain, soil roughness, which affects the inversion Performance; Secondly, using single-star data, the inversion accuracy of linear regression model established by least squares algorithm is poor. Therefore, this research proposed a BDS multi-frequency SMC inversion method that takes into account the amplitude stability of reflected signal. Based on the sliding window method, a stationary signal window is obtained intelligently as the data source for subsequent inversion. On this basis, joint inversion of multi-frequency signals is carried out. In addition, RANSAC algorithm is used to estimate model parameters to reduce noise interference. The experiment shows that (1) Using the window with stable amplitude change as the elevation angle range of the experiment is useful to improve the retrieval accuracy of SMC. Compared with the retrieval results of the empirical elevation angle range of 5–30°, the correlation coefficient increases in the range of 5.2–34.9%, and the root mean square error decreases in the range of 19.2–52.9%; (2) B1I/B2I/B3I triple-frequency signal fusion can reflect the soil moisture information near the measurement station more comprehensively; (3) Compared with the least square linear regression algorithm, the SMC estimate effect of RANSAC algorithm is better. Finally, the SMC was inverted using the delayed phase fusion of the BDS triple-frequency signal. The correlation coefficient can reach 0.9850, the root mean square error is 0.0066 cm3/cm3, and the average absolute error is 0.0053 cm3/cm3. Compared with the traditional method, this method has significantly improved.