Spatiotemporal analysis and prediction of shoreline dynamics: A comparative study of machine learning and traditional forecasting approach

Understanding shoreline dynamics is critical for effective coastal management. This study analyzes the 1990–2024 evolution of Egypt’s Damietta coastline using satellite imagery and the Digital Shoreline Analysis System (DSAS). Results show a dominant accretion trend, with a mean accretion rate of 3.0 m/year and mean erosion of −0.95 m/year, resulting in a net shoreline advance of 28.9 km. To evaluate predictive capability, a standard Kalman filter model was tested but showed limited accuracy for long-term forecasting, yielding a high error (RMSE = 102.7). A significant improvement was achieved using a Long Short-Term Memory (LSTM) deep learning model, which combined historical shoreline data with hydrodynamic forcings (wave height, tidal range). The LSTM model demonstrated superior performance, reducing the RMSE by 92.78% and achieving high accuracy (RMSE = 9.89, MAPE = 0.012) for the 2020–2024 validation period. Its projection to 2044 identified 15 distinct coastal sectors with divergent future erosion and accretion patterns. This research concludes that while geostatistical methods are useful for historical analysis, reliable long-term shoreline forecasting requires integrating physical processes into models. The LSTM’s enhanced accuracy advocates a shift in coastal management from reactive planning to proactive, data-driven strategies based on dynamic, process-aware modeling for sustainable resilience.