Abstract: Objectives: The Global Navigation Satellite System-Interferometric Reflectometry (GNSS-IR) technique is an effective "remote sensing water level gauge", capable of retrieving water level information for oceans, rivers, lakes,and reservoirs against the global reference frame. However, when the observation environment of the station is complicated and the capacity of receiving satellite signals is limited, the number and length of the signal-to-noise ratio sequences of the water surface reflections are insufficient, which reduces the inversion accuracy of GNSS-IR water level monitoring at a single station. Methods: Therefore, the paper proposes a GNSS-IR multi-station fusion water level monitoring method based on robust estimation. It reduces the influence of outlier rough values of all stations by optimizing the inversion points of satellite arcs of all stations. In order to verify the effectiveness of the proposed method, we carried out experiments using data from three monitoring stations receiving reflective signals from different water surface in the upper reservoir of the Xilongchi pumped storage power station. Results: The results show that the correlation coefficient between the fused water level inversion results and water level gauge measurements is as high as 0.983 for the sudden rise and fall process of the pumped storage power station, which has a daily water level change of 18.7 m. Moreover, the accuracy of the fusion water level results from multiple stations is significantly better than that of single-station water level monitoring. The relative accuracy of the fusion inversion results is 0.06, which is 30% - 70% higher than that of the single station monitoring results. Conclusions: It indicates that the method significantly enhances the performance of GNSS-IR water level monitoring in scenarios with insufficient single-station data, and improves the robustness and stability of the water level monitoring model.