Oil palm land cover classification using WorldView satellite and machine learning algorithms from QGIS ENMAP-box plugin

Abstract

Mapping oil palm within a plantation is a crucial task for efficient oil palm management. Using satellite data and machine learning techniques to accurately map land cover classes help in identifying and differentiating between oil palm and other types of land cover. Pertaining to this research, WorldView satellite imagery was employed to assess the land cover classification using machine learning algorithms from the QGIS EnMAP-Box plugin. Machine learning classifiers used for the analysis of land cover classification are random forest (RF), XGBoost, and LGBM. In order to assess how well each classifier performed, accuracy assessments were conducted, namely kappa accuracy, overall accuracy, in addition to user accuracy, as well as producer accuracy. The classification map generated by the best-performing classifier was then utilized. The result shows RF achieved the highest accuracy of 83.1% overall as well as kappa accuracy of 79.5%, compared to XGBoost and LGBM classifiers. RF also shows acceptable user accuracy and producer accuracy in most classes, such as mature oil palm and water. The best classification map obtained from RF serves as an inventory for effective plantation management, facilitating tasks such as oil palm tree identification, replanting programs, and other operations. This comprehensive approach plays a crucial role in monitoring and overseeing the development of oil palm for sustainable production.