Integrated Machine Learning Framework for Short-Term and Long-Term Shoreline Forecasting and Coastal Risk Assessment in Sri Lanka

Abstract

Coastal erosion and shoreline instability pose significant environmental and socio-economic challenges, particularly for island nations such as Sri Lanka. Accurate prediction of shoreline dynamics is essential for sustainable coastal management and disaster mitigation. This study proposes a hybrid spatiotemporal framework for shoreline forecasting and coastal erosion risk mapping by integrating shoreline geometry with environmental forcing variables. Official shoreline datasets obtained from the Coast Conservation and Coastal Resource Management Department of Sri Lanka are combined with atmospheric and oceanographic data from ERA5 and Copernicus Marine datasets. Shoreline positions are represented using standardized transect-based vectors, enabling consistent temporal analysis. A multi-output Gradient Boosting model is used for short-term forecasting, while a multi-model ensemble approach is employed for long-term prediction. Model validation using hindcast testing demonstrates reliable performance in capturing shoreline movement patterns. The framework further identifies erosion hotspots and classifies coastal risk zones. The proposed system provides a scalable and automated solution for shoreline monitoring, forecasting, and decision support.