AI-BASED FILTERING OF VESSEL-INDUCED NOISE IN SIPHON-TYPE RIVER WATER LEVEL MEASUREMENTS AT PHU AN HYDROLOGICAL STATION

Abstract

Accurate river water level observations are fundamental for hydrological analysis, tidal assessment, and flood forecasting. At the Phu An Hydrological Station, water level measurements recorded by a siphon-type gauge system are frequently contaminated by high-frequency oscillations caused by vessel traffic on the river. These vessel-induced disturbances introduce significant noise into the observed time series, leading to errors in identifying tidal extrema and reducing the reliability of water level data, particularly under increasing hydrodynamic variability associated with climate change an artificial intelligence (AI)-based filtering framework to mitigate vessel-induced noise in river water level observations. The proposed approach integrates signal preprocessing and deep learning techniques to model the intrinsic temporal dynamics of natural river water levels while suppressing non-hydrological fluctuations. Specifically, a deep learning model is trained to learn the smooth and continuous behavior of water level variations driven by tides and river flow, enabling the reconstruction of a standardized and noise-free water level hydrograph from raw observations affected by vessel-induced disturbances. The experimental results demonstrate that the proposed AI-based method effectively reduces high-frequency noise and significantly improves data stability without distorting the underlying hydrological signal. The corrected water level series shows improved accuracy in capturing tidal peaks and troughs, thereby improving the reliability of hydrological observations at Phu An Station. The proposed methodology provides a practical and scalable solution for improving water level measurements at river stations influenced by intensive vessel traffic and offers strong support for hydrological forecasting, climate change impact assessment, and sustainable water resource management.