Remote Sensing-Based Crop Identification and Acreage Estimation of Rabi Wheat in Anand, Gujarat

Abstract

The Study evaluates the performance of supervised and unsupervised classification techniques for crop identification using Sentinel-2 imagery. Four supervised classifiers—Random Forest (RF), Minimum Distance (MD), Support Vector Machine (SVM), and Smile Cart (sCART)—were assessed, with RF achieving the highest overall average accuracy (91%) and kappa value (87%) across two cropping seasons. The unsupervised classification method, utilizing the Isoclustering algorithm, recorded an average accuracy and kappa value of 84% in the first season and 80% in the second season. Acreage estimation revealed RF to be the most reliable, estimating 69,000 hectares (2019-20) and 64,000 hectares (2020-21), closely aligning with district statistical yield data. In contrast, sCART and SVM classifiers demonstrated lower accuracies of 46% and 36%, respectively. The study underscores RF's superiority in crop identification and acreage estimation, offering valuable insights for agricultural planning and management.