Sunflower mapping using machine learning algorithm in Google Earth Engine platform

Abstract

The sunflower crop is one of the most pro sources of vegetable oil globally. It is cultivated all around the world including Haryana, in India. However, its mapping is limited due to the requirement of huge computation power, large data storage capacity, small farm holdings, and information gap on appropriate algorithms and spectral band combinations. Thus, the current work has been done to identify an appropriate machine learning (ML) algorithm (after comparing random forest (RF) and support vector machine (SVM) reported as the best classifiers for land use and land cover) and best band combinations (among the six combinations (including Sentinel-Optical, Sentinel-SAR, and combined-Optical-SAR in single data and time series manner) for Sunflower crop mapping in Ambala and Kurukshetra districts of Haryana using Google Earth Engine (GEE) cloud platform. GEE cloud-computing system combined with RF and SVM provided Sunflower map with an accuracy ranging from 0.0% to 90% in various bands and classifiers combinations but was the highest for the RF with single date optical data. The SVM classifier tuned with parameters like kernel type, degree, gamma, and cost provided better overall accuracy for the classification of land use and land cover along with Sunflower ranging from 98.09% to 98.44% and Kappa coefficient ranging from 0.96 to 0.97 for optical data and combination of SAR and optical time series. The platform is efficient and applicable for a larger part of the country to map Sunflower and other crops with currently identified combinations of satellite data and methodology due to the availability of satellite images, advanced ML algorithms, and analytical modules on a single platform.