Sentinel-1A Data Analysis for Rice Classification Utilizing Random Forests and Support Vector Machine

Abstract

Listen

Translate article

Rice is known to be one of the most important crops in India and many other nations, particularly in Asia, therefore accurate rice area estimation has an important role in many activities, ranging from human nutrition to environmental concerns. As a result, the determination of cultivation area remains a hot topic among researchers from numerous disciplines, planners, and decision makers. Using Sentinel-1A SAR (Synthetic Aperture Radar) satellite data, this study attempts to evaluate the effectiveness of random forest (RF) and support vector machines (SVM) algorithms for rice crop classification. According to the findings, rice fields can easily be distinguished from other crops in the research area by using the temporal characteristics of the rice crop as reflected in the VH backscatter patterns. The total precision and Kappa coefficient produced by RF showed 85.7% and 0.74, respectively, when the classification outcomes were compared to the ground reference data. These values were somewhat higher than those obtained by SVM (81.2% overall accuracy and 0.68 Kappa coefficient). The government’s rice area statistics were used to compare the analysis results; the relative difference in rice area for RF and SVM, respectively, came out to be +1.40% and -4.63%. In summary, the RF algorithm is highly recommended for the accurate differentiation of rice fields from neighbouring classes in conditions of identical climate, soil, and topography with similar methods of cultivation. On the other hand, Sentinel-1A SAR provides a valuable data set at cost-free for similar studies.