Spatio-temporal prediction of time-series InSAR Land subsidence based on ConvLSTM neural network

Objective:  Existing spatio-temporal prediction methods of land subsidence have some problems, such as poor ability to capture time-series features and ignoring spatial neighborhood features, which lead to poor reliability of spatio-temporal prediction of land subsidence.   Method:  In this study, a spatio-temporal prediction of land subsidence based on convolutional long short-term memory (ConvLSTM) neural network is proposed, which can capture time-series features and spatial neighborhood features. Beijing Capital International Airport (BCIA) was selected as the experimental area. Firstly, based on Small Baseline Subset Interferometric Synthetic Aperture Radar (SBAS-InSAR), Sentinel-1A images were used to obtain the spatio-temporal InSAR data of land subsidence at BCIA, and then a ConvLSTM model was constructed to predict the land subsidence in the next year. Permanent scatters InSAR (PS-InSAR), SBAS-InSAR and benchmark results were used to cross-verify the reliability of time-series InSAR results. Time-series InSAR land subsidence data were segmented by sliding windows to form a many-to-one data set model. Combined with wavelet transform and evaluation indexes to determine the optimal time step of the prediction model, the ConvLSTM spatiotemporal prediction model of time-series InSAR land subsidence was established.   Results:  The R² of the predicted and real results reached 0.997. Meanwhile, the performance of the model was further evaluated based on the image evaluation index structural similarity (SSIM) and multi-scale structural similarity (MS-SSIM), and the SSIM and MS-SSIM reached 0.914 and 0.975, respectively.   Conclusion:  In addition, support vector regression (SVR), multilayer perceptron (MLP), convolutional neural network, CNN) and the Long-Term Memory neural network (LSTM) model were compared and analyzed, and each index revealed that the model proposed in this paper was optimal. The ConvLSTM spatial-temporal prediction model was used to predict that the maximum cumulative subsidence at BCIA will reach 157 mm by November 2022. This study can provide key technical support for the early prevention of urban land subsidence.