Abstract: China's BeiDou-3 (BDS-3) has abundant frequency band data, which makes the system promising for application in the field of Global Navigation Satellite System Reflectometry (GNSS-R). In this paper, we systematically carry out the study of BDS-R water level inversion, firstly, we compare the performance of two algorithms of spectral analysis and nonlinear fitting in water level inversion, and the results show that the overall accuracy of nonlinear fitting is better than that of spectral analysis; then, we introduce Savitzky-Golay filtering algorithm to denoise the inversion results for various errors, and the numerical values show that the difference between the results before and after denoising and the measured water level The average RMSE of the difference between the results before and after denoising and the measured water level is reduced by 7.4 cm, and the inversion accuracy of the full frequency band of BeiDou is comparable to that of GPS L5.