Dual-window fusion filtering with adaptive anomaly correction for real-time GNSS landslide monitoring

Abstract Real-time landslide early warning relies on high-fidelity global navigation satellite system (GNSS) displacement monitoring, but sensor data is often corrupted by trend-like distortions and long-duration anomalies from environmental and hardware interference. These issues hinder accurate motion tracking, increase false alarms, and overload central fusion systems. To address this, we propose dual-phase anomaly suppression algorithm (DASA), a lightweight edge-deployable filtering approach for GNSS-based landslide monitoring. DASA extends the forward–backward fusion paradigm with three key innovations: (1) segment-wise pseudotrend denoising using robust signal baseline extraction; (2) fluctuation phase detection integrating bidirectional mutation evidence; and (3) drift-aware fusion combined with volatility-guided interpolation to suppress abrupt corrections. Instead of fixed weights or abrupt resets, DASA modulates state updates based on directional trust and signal volatility, enabling smooth realignment after disturbances. Field validation using a two-year dataset from Niushou Mountain shows that DASA significantly improves trajectory stability, accelerates convergence to true motion, and reduces overreaction to non-physical anomalies. This method provides a scalable, computation-efficient solution for embedded GNSS monitoring nodes in landslide-prone environments.