Coastal Change and Shoreline Prediction using Deep Learning

Coastlines undergo continuous transformation due to natural forces and anthropogenic influences, resulting in considerable variability in shoreline positions across time. Tracking and forecasting these shifts is critical for sustainable coastal governance, hazard reduction, and safeguarding ecosystems. This research introduces a hybrid methodology for shoreline forecasting that leverages multi-date satellite data from Landsat 5, 7, and 8, acquired via Google Earth Engine, to examine extended shoreline dynamics across Indian coastal zones. Land–water delineation is accomplished through the Modified Normalized Difference Water Index (MNDWI) alongside K- Means clustering, subsequently refined using morphological processing to eliminate noise and improve accuracy. Shoreline contours are then derived from the processed imagery and subjected to temporal examination. A deep learning architecture incorporating ConvLSTM with an integrated attention mechanism is utilized to simultaneously capture spatial and temporal characteristics, enabling precise projection of future shoreline locations. Furthermore, linear regression is employed to assess overarching trends and validate the prediction reliability. The framework additionally produces change maps identifying zones of coastal erosion and sediment deposition. Experimental outcomes confirm that the developed system accurately predicts shoreline variability, delivers stable results, and serves as a dependable tool for coastal surveillance and informed decision-making in at-risk coastal communities.