Classification Model of Sugarcane Growth Phase from Multi-temporal Sentinel 1 Imagery Using Random Forest Algorithm

Abstract

The Special Region of Yogyakarta, a designated sugarcane center, demands special attention for effective extensification efforts, necessitating spatial insights into sugarcane farming. Monitoring of sugarcane fields served to obtain information on the growth phases of sugarcane and its distribution for agricultural extensification strategies. For this reason, it is necessary to carry out image classification using the Random Forest reliable algorithm to classify sugarcane growth phases in multi-temporal Sentinel 1 images. The sugarcane planting calendar Map is conducted from the image classification outcomes and then tested for its accuracy for evaluation. The classification process involves analyzing each image captured monthly throughout 2020, with a dataset comprising 9690 sample pixels across six classification classes: buildings, vegetation, water bodies, rice fields, sugarcane phase class 1, and sugarcane phase class 2. The results show that the Sentinel 1 image consisting of 13 images has an average classification model accuracy of 65.38%. Notably, the image classification achieved its pinnacle performance in October, boasting the highest overall accuracy level at 73.33%, accompanied by an RMSE value of 2.05.