Machine Learning XGBoost Method for Detecting Mangrove Cover Using Unmanned Aerial Vehicle Imagery

Abstract

The mangrove ecosystem can be understood as a unique and different type of ecosystem that can benefit the surrounding ecosystem from the socio-economic and ecological perspective. The purpose of this study is to classify mangrove cover in Tanjung Lapin Beach, about 18.3 hectares, North Rupat District Bengkalis Regency, Riau Province, by applying machine learning XGBoost methods of UAV images by producing interpretations of mangrove cover in the research area. The use of machine learning with a high level of accuracy resulting from the XGBoost method is expected to help the availability of spatial data in identifying better mangrove forest cover. The data obtained from the orthomosaic results from the 3,500 tiles image is used as a reference for making sample points for the analysis process using the XGBoost method, with 224 sample points of mangrove objects visually recognized as training data. Regarding training data, the XGBoost method's iteration result obtained 99% overall accuracy and Kappa accuracy of about 0.98. It means the analysis process continues to the mangrove object cover detection stage. Based on the detection results, it was obtained about 11.9 hectares of mangrove forest cover (64% of the total study area). It has 68 sample points as test data used as an accuracy test tool from the detection results of mangrove objects, where an overall accuracy of 87% and kappa accuracy of 0.82 were obtained. This shows the successful use of the XGBoost method in identifying the mangrove's cover.