Mangrove Vegetation Mapping using Google Earth Engine, Open-Access Satellite Data, and Machine Learning

Monitoring the distribution of mangrove vegetation through remote sensing data can cover a large area and provide effective and efficient data. However, in mapping mangroves using remote sensing, they often experience misclassification due to the complexity of the mangrove ecosystem, such as abundant species in the mangrove class and mixing with other vegetation in the mangrove ecosystem. Accurate mapping of mangrove vegetation is very important for monitoring coastal area resources. Mangrove vegetation maps can be generated from Sentinel-1 and Sentinel-2 data using the Random Forest classifier. In addition, it is possible to download vast data in open source by utilizing the GEE platform for data downloading and preprocessing on cloud computing facilities. The final classification results are evaluated comprehensively through different analyses. The classification results obtained a high average overall accuracy, Kappa coefficient, and Intersect of Union, respectively 96.94, 0.95 and 0.67. Overall, based on the qualitative (i.e., visual interpretation) and quantitative (i.e., statistical accuracy assessment) evaluation criteria, the proposed method confirms its applicability in producing accurate mangrove vegetation maps. Furthermore, the comparison results prove the contribution of multi-sensor remote sensing data (i.e., SAR + optical), the effectiveness of seasonal downscaling, and the role of the vegetation indexes. © 2025 American Institute of Physics Inc.. All rights reserved.