Utilizing the Power of Global Navigation Satellite System (GNSS) Data, and Transformer Model for Efficient Prediction of Land Subsidence in the Cities of Bandung and Yogyakarta

Land subsidence poses a significant threat to infrastructure, environmental stability, and public safety, particularly in urban regions such as Bandung and Yogyakarta. This study leverages Global Navigation Satellite System (GNSS) data and advanced deep learning models, specifically the Temporal Fusion Transformer (TFT). The proposed methodology integrates GNSS data preprocessing, feature engineering, and multi-horizon forecasting to develop a scalable and interpretable forecasting framework. Experimental results demonstrate the superiority of the TFT model over traditional transformer architectures, achieving a Mean Squared Error (MSE) of 0.000109, Root Mean Squared Error (RMSE) of 0.010420, Mean Absolute Error (MAE) of 0.008929, and an R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) value of 0.997831 for Yogyakarta’s JOGS station. In contrast, the traditional transformer model exhibited significantly higher error values, with an MSE of 0.321957 and an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of -0.010883, indicating poor predictive capability. The findings highlight the advantages of TFT in capturing complex temporal dependencies and long-term subsidence patterns, making it a valuable tool for geospatial analysis.