The comparison of ARIMA and LSTM in forecasting of long-term surface movements derived from PSINSAR

Abstract

In recent years, airports, serving as vital transportation hubs, have faced the challenge of limited available land in megacities. As a result, airport construction on reclaimed areas has become a common solution. However, over time, these areas are exposed to soil behaviors like settlement and uplift, leading to surface movements. Detecting and monitoring these movements consistently is crucial to prevent potential disasters. Interferometric Synthetic Aperture Radar (InSAR) has emerged as a powerful tool for monitoring surface movements with high temporal and spatial resolution based on satellite properties, unlike traditional point-based methods. In particular, time series InSAR methods, such as Persistent Scatterer Interferometry (PSI), have been developed to monitor surface movements over a period of time. However, in addition to observing past surface movements, forecasting future movements is also of great importance. In this context, various forecasting methods have been explored, among which Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) have gained significant popularity due to their successful performance. In a recent study, these two methods were applied to forecast surface movements at Istanbul Airport, utilizing time series data obtained from the freely available Sentinel-1 SAR images. The performance of the ARIMA and LSTM models was evaluated using well-established metrics including root mean square error (RMSE) and mean absolute error (MAE). Both ARIMA and LSTM are suitable for forecasting surface movements, but LSTM exhibited a marginally better fit to the data compared to the ARIMA model.