Prediction of InSAR time-series deformation using deep convolutional neural networks

Predicting deformation is crucial to issue early warnings of abnormal conditions and implement timely remedial actions. Herein, we propose a data-driven method based on deep convolutional neural networks (DCNN) to predict interferometric synthetic aperture radar (InSAR) time-series deformation. We conducted experiments at the Hong Kong International Airport built on reclaimed lands. The results showed that the DCNN was able to predict the linear settlement of the reclaimed lands and nonlinear thermal expansion of the buildings. The mean internal error (0.3 mm) was negligible compared with the millimetre-level accuracy of the monitored deformation, indicating that the DCNN approximates the monitored deformation values very well. The root mean square error of the predicted deformation in the subsequent year was 3 mm after validation using ground data, which was comparable to the accuracy of the monitored deformation. The results demonstrated the effectiveness of the DCNN for short-term prediction of InSAR time-series deformation, which can be potentially used in early warning systems.