Village level identification of sugarcane in Sangali, Maharashtra using open source data

Abstract

Agriculture and crop monitoring are very important for an agrarian country like India. This study is done in June Khed village in the Sangli district of Maharashtra, India to assessing the efficiency of an open source cloud-based remote sensing platform Google Earth Engine (GEE), in the village-scale identification of sugarcane. The ground-truth data was collected and the efficiency of Landsat-8 and Sentinel-2 satellite data was assessed in driving GEE’s inbuilt Machine Learning classifiers: Classification and Regression Tree (CART), Support Vector Machine (SVM) and Random Forest (RF). Results were validated with the ground truth data and the official data. Of the methods used, SVM outperformed RF and CART with the lowest relative deviation (+9.2%), highest F1- score (0.8) and overall accuracy (78%), using the Sentinel-2 data. Results also indicated the in-situ use of observation data with high spatio-temporal resolution data. The validated model was then up-scaled for the Walwa Taluka level, to map sugarcane area that can be used for further agriculture tasks such as crop monitoring and yield prediction, leading to better management of crop and better formulating of sugarcane farmer policy.

Keywords

• Sugarcane
• Remote Sensing
• Google Earth Engine
• Machine Learning
• Crop classification