Monitoring of coastal changes using satellite imagery: A case study of Unawatuna

Project Work 2

Abstract

Coastal environments are dynamic landscapes shaped by natural forces and human activities, necessitating effective monitoring to support sustainable management. This project focuses on developing a methodology to monitor sand fluctuations along the coastline of Unawatuna, Sri Lanka, from 2019 to 2023, using satellite imagery and machine learning. Sentinel-2 data, with its multi-spectral features and 20m spatial resolution, served as the foundation for classifying land cover types such as Sand, Water, Forest, and Buildings. The dataset was manually annotated, enhanced by high-resolution references, and used to train machine learning models. While the project highlights the utility of open-source tools like QGIS and R for integrating geospatial analysis with machine learning, challenges such as data resolution limitations and the need for improved classification accuracy remain.

Three models — Classification and Regression Tree (CART), Random Forest (RF), and Support Vector Machine (SVM) — were implemented and evaluated for their performance in detecting sand areas. The Random Forest model emerged as the most effective, achieving high accuracy in Sand classification and successfully identifying temporal changes in sand coverage. Results indicate a notable reduction in sand between 2020 and 2021, followed by partial recovery in subsequent years.