Comparison of Different Supervised Classification Algorithms for Mapping Paddy Rice Areas Using Landsat 9 Imageries

Abstract

Rice is known to be one of the most essential crops in Turkey, as well as many other countries especially in Asia, whereas paddy rice cropping systems have a key role in many processes ranging from human nutrition to environment-related perspectives. Therefore, determination of cultivation area is still a hot topic among researchers from various disciplines, planners, and decision makers. In present study, it was aimed to evaluate performances of three classifications algorithms among most widely used ones, namely, maximum likelihood (ML), random forest (RF), and k-nearest neighborhood (KNN), for paddy rice mapping in a mixed cultivation area located in Biga District of Çanakkale Province, Turkey. Visual, near-infrared and shortwave infrared bands of Landsat 9 acquired in dry season of 2022 year was utilized. The classification scheme included six classes as dense vegetation (D), sparse vegetation (S), agricultural field (A), water surface (W), residential area – base soil (RB), and paddy rice (PR). The performances were tested using the same training samples and accuracy control points. The reliability of each classification was evaluated through accuracy assessments considering 150 equalized randomized control points. Accordingly, RF algorithym could identify PR areas with over 96.0% accuracy, and it was followed by KNN with 92.0%.