Sunflower mapping using machine learning algorithm in Google Earth Engine platform.

Area of Science:

• Agricultural Remote Sensing
• Machine Learning Applications
• Geospatial Analysis

Background:

• Sunflower is a globally significant source of vegetable oil, with cultivation widespread, including in Haryana, India.
• Accurate crop mapping is hindered by computational demands, data storage needs, small farm sizes, and a lack of knowledge on optimal algorithms and spectral bands.
• Existing land use and land cover classification methods often rely on computationally intensive processes and may not be optimized for specific crop mapping.

Purpose of the Study:

• To identify the most effective machine learning algorithm (Random Forest vs. Support Vector Machine) for sunflower crop mapping.
• To determine the optimal spectral band combinations and data types (Optical, SAR, combined, time series) for accurate sunflower mapping.
• To evaluate the efficiency and applicability of the Google Earth Engine (GEE) cloud platform for large-scale crop mapping.

Main Methods:

• Comparison of Random Forest (RF) and Support Vector Machine (SVM) algorithms for land use/land cover classification.
• Evaluation of six different spectral band combinations, including Sentinel-Optical, Sentinel-SAR, and combined Optical-SAR data, in single and time-series formats.
• Utilized the Google Earth Engine (GEE) cloud platform for data processing and analysis of sunflower crop mapping in Haryana, India.

Main Results:

• The Random Forest classifier with single-date optical data yielded the highest accuracy among initial combinations (0.0% to 90%).
• The Support Vector Machine (SVM) classifier, optimized with specific parameters, achieved superior overall accuracy (98.09%–98.44%) and Kappa coefficients (0.96–0.97).
• SVM demonstrated high accuracy for classifying land use, land cover, and sunflower using optical data and combined SAR-optical time series.

Conclusions:

• The Google Earth Engine (GEE) platform, coupled with optimized machine learning algorithms like SVM and appropriate satellite data combinations, provides an efficient solution for accurate sunflower crop mapping.
• The developed methodology is scalable and applicable to larger regions in India for mapping sunflower and potentially other crops.
• This study addresses the limitations of computational power and data handling for precise agricultural monitoring using advanced remote sensing techniques.