Machine Learning-Augmented Tidal Signature Extraction from GNSS Interferometric Reflectometry (GNSS-IR) Observations

Abstract

GNSS Interferometric Reflectometry (GNSS-IR) is a promising method for sea level monitoring that utilizes reflected GNSS signals to estimate water surface height. It serves as a low-cost and passive alternative to traditional tide gauges, especially beneficial in remote or under-monitored areas. However, conventional spectral techniques used to extract tidal signals from GNSS-IR data often struggle with multipath interference, environmental noise, and rigid model assumptions. To address these limitations, this study proposes a machine learning-based framework that applies three models Random Forest Regression (RFR), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM) to enhance the accuracy of tidal height estimation using GNSS signal-to-noise ratio (SNR) data. The approach was tested using data from a GNSS station in Port Klang, Malaysia, with tide gauge measurements serving as reference values. Among the models tested, RFR achieved the best performance with a root mean square error (RMSE) of 0.6451 meters and a coefficient of determination (R²) of 0.7949. This indicates a strong correlation between the predicted and actual tide data. The findings demonstrate that machine learning techniques, particularly RFR, offer significant improvements over traditional methods and can play a crucial role in enhancing GNSS-IR-based sea level monitoring in regions with limited observational infrastructure.