An observation-domain fault detection and exclusion method considering measurement correlation for GNSS RTK in landslide monitoring

With the global navigation satellite system (GNSS), especially the real-time kinematic (RTK) technology being widely used in landslide monitoring, the quality of GNSS satellite observations remains unstable in complex environments. The conventional fault detection and exclusion (FDE) method ignores the correlation among observations, and the RTK double differences (DD) model results in a high degree of correlation between test statistics. This may lead to contamination effects that adversely impact the accuracy of fault identification. This study proposes an adaptive FDE method that combines correlation coefficients weighted based on observation-domain correction (CWOD-FDE) for landslide monitoring. Unlike conventional satellite-by-satellite exclusion, our method classifies faults by observation type to preserve redundancy, introducing a maximum fault mode constraint and correlation-based weighting with observation coefficients to suppress contamination and improve detection reliability. Experimental results based on real landslide monitoring data indicate that the proposed CWOD-FDE method reduces the fault anomaly rate from approximately 50 % to 10 %, effectively minimizing false alarms and the erroneous exclusion of valid observations. The number of usable carrier-phase observations increased from 18 to 20, enhancing the redundancy of observation. In terms of positioning accuracy, the root mean square error (RMSE) in the East, North, and Up directions are 0.037 m, 0.031 m, and 0.058 m, respectively, representing improvements of 89.5 %, 91.3 %, and 92.4 % compared to extended Kalman Filter (EKF) positioning, and 74.5 %, 91.9 %, and 88.3 % compared to traditional FDE method. This study provides an effective FDE method within the observation domain for GNSS landslide monitoring.