Displacement time series forecasting and anomaly detection based on EGMS-PSInSAR data towards effective bridge monitoring

Abstract

Monitoring bridges is essential due to their critical role in infrastructure networks. With its high temporal resolution, Differential Synthetic Aperture Radar Interferometry (DInSAR) emerges as promising remote sensing technique for this purpose. This study demonstrates the application of historical displacements time series provided by the European Ground Motion Service (EGMS) with daytime average temperature data for displacement time series forecasting and anomaly detection. This application was applied to 15 bridges in Wroclaw city, Poland covering 1441 points across two Sentinel-1 orbit geometries. Three forecasting models— Seasonal Autoregressive Integrated Moving Average (SARIMA), Long Short Term Memory (LSTM), and Prophet—were evaluated for their predictive performance. Since thermal expansion is common in bridges, an exogenous temperature variable was incorporated into each model, resulting in six predictive models. Root Mean Squared Error (RMSE) was used to assess prediction accuracy, with results showing that a displacement prediction accuracy on the level of 2 mmcan be achieved. LSTM and Prophet models performed the best, achieving RMSE values between 1.5 mm and 1.6 mm, outperforming SARIMA. Moreover, an approach for detecting anomalous displacement was proposed based on confidence intervals, using Student's t-distribution and standard deviation to establish a 90% confidence margin. This study highlights the benefits of combining DInSAR time series data with machine learning models for accurate displacement time series prediction and anomaly detection, contributing to more effective bridge monitoring and infrastructure management.