Spatiotemporal Predicting of Susceptible Coastal Areas Using SAR-Optical Data Integration and Geospatial Artificial Intelligence

Coastal landslides create significant risks for human populations and infrastructure in susceptible coastal areas; therefore, strong and scalable monitoring tools are required for efficient risk reduction. This article offers a comprehensive framework for predicting the vital signs of coastal landslide danger throughout the study area using multifeature data fusion and deep-learning methods. We derived monthly composites of normalized difference vegetation index (NDVI), normalized difference water index (NDWI), slope, and elevation across 2023 using Shuttle Radar Topography Mission digital elevation model data and Sentinel-2 surface reflectance photos. Data preprocessing and feature aggregation were performed within the Google Earth Engine (GEE) system to ensure high temporal consistency and computational efficiency. The long short-term memory neural networks were used to forecast these environmental variables temporally. While dynamic indices showed safe mean absolute percentage errors of 49.92% for NDVI and 48.11% for NDWI, the model showed minimal prediction error for static variables (slope and elevation). Combining the expected characteristics, a rule-based risk categorization helped to produce a spatially explicit coastal landslide susceptibility map, hence stressing priority areas for hazard control. The findings highlight the potential for proactive landslide risk assessment by integrating deep learning with multisource satellite data, which can aid in the development of data-driven methods for sustainable coastal management. © 2025 Elsevier B.V., All rights reserved.