Prediction of airport runway subsidence using SBAS-InSAR and LSTM networks optimized by EnKF

The prediction of ground settlement is a critical component of disaster prevention, mitigation, and airport runway maintenance. However, prevailing methodologies are constrained by challenges such as the inefficiency of manual data acquisition and the latency associated with the backpropagation process in time-series neural networks. This study introduces a precise and efficient method for runway subsidence forecasting that integrates the Ensemble Kalman Filter (EnKF) with a Long Short-Term Memory (LSTM) network, implemented within an encoder–decoder (Seq2Seq) framework. This hybrid approach is designed to enhance model adaptability and mitigate prediction latency. The model was validated using subsidence data acquired from Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR) analysis of Sentinel-1A imagery covering the Chengdu Shuangliu Airport runway. Comparative experiments demonstrated that although both the proposed LSTM-EnKF and a standard LSTM model can effectively predict runway settlement, the LSTM-EnKF model exhibited a performance improvement of approximately 12 % on the selected evaluation metrics. Moreover, the LSTM-EnKF model demonstrated superior adaptability to environmental fluctuations, while the encoder–decoder architecture proved crucial for enhancing long-term prediction accuracy. These advancements hold significant potential for automating infrastructure monitoring and optimizing maintenance planning within the aviation industry.