Comparison of Google Earth Engine (GEE)-based machine learning classifiers for mangrove mapping

Providing fast, up-to-date and accurate mangrove extent map is essential to support mangrove monitoring and management actions. Conventional scene by scene remote sensing image classification approach is inefficient and time consuming. Google Earth Engine (GEE) provides a cloud platform to access and seamlessly process large amount of freely available satellite images. The GEE also provides a set of the state-of-the-art classifiers for pixel-based classification that can be used for mangrove mapping. This study aims to compare the performance of three machine learning classifiers available in GEE, namely support vector machine (SVM), random forest (RF), and classification and regression trees (CART), for discriminating mangrove and non-mangrove objects. We used Landsat 8 OLI scene over Agats Papua area, Indonesia, as the main image for this study (path 102 row 64) acquired at October 19, 2014. A total of 838 points samples were collected representing mangroves (244), non-mangroves (161), water bodies (311), and cloud (122) class. These training areas were used by all classifiers in GEE to classify the image into targeted land cover classes. The classification results show that mangrove objects could be detected by all classifiers. To assess the accuracy of the map produced, the map results were compared to the visually interpreted mangrove map of the study area and previous mangrove map by Giri et al. (2011) as the references. From this comparison we found that SVM outperformed the classification results of CART and RF significantly. Most of the miss-classification were occurred at the shadow objects beneath the clouds on the image, where shadows were miss-classified as mangroves. This study shows the potentials of GEE for mangrove extent mapping. Future works need to focus on the sample refinement and testing the applicability of the classifiers at other mangrove sites. © 2021 Elsevier B.V., All rights reserved.