Oil palm mapping based on machine learning and non-machine learning approach using Sentinel-2 imagery

Abstract

Oil palm is a plantation commodity that has high economic value and investment opportunities. Mapping oil palm areas is important to determine the extent and location of oil palm distribution to improve the region's economy. This study aims to map oil palm land cover using the machine learning approach (Decision Tree (DT) and Support Vector Machine (SVM)) and non-machine learning approach (Maximum Likelihood Classifier (MLC)) and to extract other land covers, such as built-up areas, fields, water bodies, and other Vegetation. The Sentinel 2A satellite imagery data is used with a spatial resolution of 10 meters to monitor objects above the earth's surface on a large scale. The results show that the three methods can map the oil palm area with an overall accuracy above 90% and kappa value of 0.66 for Decision Tree, 0.94 for Support Vector Machine methods, and 0.92 for Maximum Likelihood Classification. The conclusion is that the total area of mapped oil palm is 1073.88 Ha (Decision Tree), 936.64 Ha (MLC), and 1204.56 Ha (SVM). This study shows that the accuracy of the machine learning approach using the SVM method is higher for oil palm mapping.