Leveraging GNSS tropospheric products for machine learning-based land subsidence prediction

Abstract

Land subsidence is a hazardous phenomenon that requires accurate prediction to mitigate losses and prevent casualties. This study explores the utilization of the Long Short-Term Memory (LSTM) method for time series prediction of land subsidence, considering various contributing factors such as groundwater levels, soil type and slope, aquifer characteristics, vegetation coverage, land use, depth to the water table, proximity to exploiting wells, distance from rivers, distance from faults, temperature, and wet tropospheric products. Due to the high spatial variability of wet tropospheric parameters, utilizing numerical weather models for extraction is impractical, especially in regions with a sparse network of synoptic stations. This hinders obtaining accurate prediction results because wet tropospheric products play a significant role in subsidence prediction and cannot be ignored in the subsidence prediction process. In this study, Global Navigation Satellite Systems (GNSS) tropospheric products, including Integrated Water Vapor (IWV) and EvapoTranspiration (ET), are employed as alternatives. Two scenarios were considered: one incorporating GNSS products alongside other parameters, and the other relying solely on the remaining parameters in the absence of GNSS tropospheric products. Ground truth data from Interferometric Synthetic Aperture Radar (InSAR) displacement measurements were used for evaluation and testing. The results demonstrated that the inclusion of GNSS tropospheric products significantly enhanced prediction accuracy, with a Root Mean Square Error (RMSE) value of 3.07 cm/year in the first scenario. In the second scenario, the absence of wet tropospheric information led to subpar predictions, highlighting the crucial role of wet tropospheric data in spatial distribution. However, by utilizing tropospheric products obtained from GNSS observations, reasonably accurate predictions of displacement changes were achieved. This study underscores the importance of tropospheric indices and showcases the potential of the LSTM method in conjunction with GNSS observations for effective land subsidence prediction, enabling improved preventive measures and mitigation strategies in regions lacking synoptic data coverage.

Deep Learning CNN-GRU Method for GNSS Deformation Monitoring Prediction

“…However, it is difficult to handle state equations and system errors in complex environments. The grey model is suitable for analyzing and modeling short GNSS time series with limited datasets and incomplete information [21][22][23]. Nevertheless, it is only suitable for short-term and exponential growth predictions.…”