Oil Palm Plantation Land Cover and Age Mapping Using Sentinel-2 Satellite Imagery and Machine Learning Algorithms

Abstract

Nowadays, there are various techniques and methods used in land cover classification using remote sensing data especially in oil palm monitoring. This study discussed the oil palm mapping using satellite imagery (Sentinel-2) and classification of land cover features using machine learning algorithms such as linear support vector classifier (LSVC), random forests (RF) and deep neural network (DNN). A total 13218 sampling points (80% of the total sampling points used as training samples and 20% applied as testing samples) were randomly selected in the study area which were then classified into six land cover features; water, bare soil, forest, immature oil palm (the age of 2-8 year), mature oil palm (age >8 year) and built-up area. These data were validated by using spectral reflectance, Google Earth Pro and ground checking. The accuracy assessment was conducted by a confusion matrix method. The results showed that classification of land features using DNN with batch size 32 and epoch 100 has the highest accuracy which is 99.35% for overall accuracy and 98.49% kappa accuracy. This study demonstrated various machine learning algorithms that may be used to detect and classify the maturity of oil palm trees, which is vital to record in tree inventories for effective plantation management.