Geospatial Time Series Analysis for Coastal Systems: AI-Powered NARX Neural Networks Integrating Remote Sensing for Advanced Shoreline Change Prediction
Nada Mansour, Tharwat Sarhan, Mahmoud El-Gamal, Karim Nassar, Mahmoud E. Abd-Elmaboud
Abstract
Advanced NARX neural network with Bayesian optimization achieved exceptional shoreline prediction accuracy (RMSE: 0.07-0.52m). Integration of multi-temporal Landsat data enabled precise shoreline delineation across 158 transects with NRMSE of 0.116595. Model 5's multi-parameter integration (wave height, tidal range, displacement) demonstrated superior performance with Performance Index values ranging from 0.000149 to 0.000857. Spatial analysis quantified critical erosion zones (-7.0 m/year) and accretion areas (+24.48 m/year) across 21 distinct coastal sectors. The decision matrix enabled targeted protection strategies based on vulnerability thresholds (20 m/year High, 5-20 m/year Moderate, 0-5 m/year Low). The implementation framework aligns with SDGs through data-driven coastal protection and climate adaptation strategies.