Optimizing Mangrove Classification with Data Fusion: Machine Learning Approaches for Enggano Island, Indonesia

Mangrove ecosystems are crucial in coastal protection, carbon sequestration, and biodiversity. Accurate mapping is vital for the conservation and sustainable management of these species, especially in vulnerable areas like Enggano Island, Indonesia. This study evaluates the performance of machine learning algorithms in GEE to model mangrove distribution on Enggano Island, Indonesia, using multi-source data, including optical data (Sentinel-2), radar data (Sentinel-1), and elevation data filtering (FABDEM). Three input configurations were developed to explore the best combination of data: (1) visual and infrared bands from Sentinel-2, (2) Sentinel-2 band ratios and spectral indices, and (3) a fusion of Sentinel-2 optical data with Sentinel-1 SAR data. Several machine-learning algorithms, including Random Forest (RF), Classification and Regression Trees (CART), Minimum Distance (MD), Gradient Tree Boost (GTB), K-Nearest Neighbor (KNN), and Support Vector Machines (SVM), were assessed using accuracy, precision, recall, and F1 score. Results showed that the third configuration, which combined Sentinel-2 optical bands, band ratios, and Sentinel-1 radar polarimetric, provided the best performance with the highest overall accuracy (OA 95.19%) using the Random Forest algorithm. This approach demonstrated superiority in overcoming mangrove classification challenges, such as cloud cover, seasonal variability, and spectral similarity with non-mangrove vegetation. These results support the importance of mangrove monitoring in small islands and tropical regions, contributing to ecosystem conservation and coastal disaster mitigation.