Abstract: Rapid land subsidence is a kind of geological disaster, which is related to the sustainable development of society and even threatens the safety of human life and property. InSAR technology can obtain long-term and large-scale surface deformation data, analyze potential land subsidence problems, and then provide reliable means for preventing geological disasters. How to predict land subsidence based on InSAR data has always been the key direction and problem that researchers focus on. Thus, on the basis of previous research on land subsidence prediction, a land subsidence prediction method was proposed, which combines the Auto Regressive Integrated Moving Average (ARIMA) model with the Long Short-Term Memory (LSTM) model in deep learning. The difference between the deformation data and the prediction result of the ARIMA model was obtained and then LSTM was used to train and predict the difference. Taking the InSAR monitoring data of Hangzhou Bay from 2017 to 2019 as an example, the method was verified. The result shows that compared with the traditional single prediction algorithm, the root mean square error of the method is reduced by at least 2.23 mm, and the mean absolute error is reduced by at least 0.98 mm, and the average prediction accuracy is improved by at least 15.19%, which verifies the feasibility of this method and provides some ideas and methods for early warning of ground subsidence.

Key words: land subsidence, InSAR, LSTM, ARIMA, coastal zone