Application of Random Forest Algorithm on Sentinel-2A Imagery for Garlic Land Classification Based on Growing Phase in Sembalun

Abstract

Garlic production in Indonesia is not sufficient to meet the consumption demands, which caused the government to implement a garlic import policy. Garlic productivity needs to be increased to reduce imports and achieve garlic self-sufficiency in 2030. Sembalun is one of the centers of garlic production in Indonesia. This study aims to classify garlic fields in Sembalun based on the garlic plant growing phase. The data used in this study are Sentinel-2A Level-1C images in July 2021 with four bands of 10 m resolution and NDVI value, as well as drone image data as ground truth. The algorithm used to perform image classification is Random Forest. This study used two dataset scenarios with the best model accuracy in predicting new data is 65.90% in the second scenario using the NDVI feature. The classification model without using the NDVI feature gives an accuracy value of 58.40%. Based on the accuracy value, the model with the NDVI feature can provide better predictions.