Mapping Cacao Plantations Using Random Forest Classification and Sentinel-2A Imagery in Batulappa District, Pinrang Regency, Indonesia

Putri Ayu Andirah (Faculty of Agricultural Technology, Hasanuddin University)
Haerani (Faculty of Agricultural Technology, Hasanuddin University, Makassar)
Suhardi (Faculty of Agricultural Technology, Hasanuddin University, Makassar)
Husnul Mubarak (Faculty of Agricultural Technology, Hasanuddin University, Makassar)

Salaga Journal
Volume 04, No. 1, June 2026

Remote sensing and Geographic Information System (GIS) technologies provide effective tools for mapping plantation crops and supporting sustainable land management. Cacao is an important plantation commodity in Indonesia, particularly in Pinrang Regency, South Sulawesi. This study aimed to map cacao and non-cacao land cover in Batulappa District using Sentinel-2A imagery and the Random Forest algorithm. Three input approaches were evaluated: an RGB band composite, the Normalized Difference Vegetation Index (NDVI), and Gray-Level Co-occurrence Matrix (GLCM) texture features. Ground-truth data were divided into training and validation datasets, and classification accuracy was assessed using a confusion matrix, including overall accuracy, user accuracy, and producer accuracy. The RGB band composite produced the highest overall accuracy of 85.38%, followed by GLCM with 75.47% and NDVI with 74.06%. For the cacao class, the RGB approach achieved a user accuracy of 80.00% and a producer accuracy of 86.96%, with an estimated cacao area of 4,516.80 ha, or 46.90% of the study area. These results indicate that the Sentinel-2A RGB band composite combined with Random Forest classification outperformed NDVI and GLCM for mapping cacao plantations in Batulappa District.