Estimating village-level potato yields using satellite remote sensing and ground data in Uttar Pradesh: An ensemble approach

Yield estimation is essential for effective crop management, market regulation, and agricultural planning. Uttar Pradesh, a leading contributor to India’s potato production, requires reliable yield estimates to ensure food security and stabilize market dynamics. This study proposes a robust, village-level yield estimation framework that leverages satellite remote sensing, meteorological data, ground observations, and advanced machine learning techniques. The approach addresses key limitations of traditional yield estimation methods, such as high data acquisition costs, low spatial resolution, and variability in data quality, which hinder accurate yield forecasting at fine spatial scales. To overcome these challenges, the framework employs a multi-source ensemble approach that is scalable, cost-effective, and reliable. It integrates: Sentinel-1 Synthetic Aperture Radar (SAR): Provides critical information on surface roughness and soil moisture, especially valuable during the planting and tuber development stages. Sentinel-2 Multispectral Imagery: Offers high-resolution optical data for calculating vegetation indices such as NDVI, LSWI, and LAI, which are indicative of crop health and biomass accumulation. Sentinel-3 FAPAR Products: Serve as proxies for photosynthetic activity, reflecting the physiological status of the crop. Meteorological Data from IMD (Indian Meteorological Department): Includes rainfall and temperature variables to account for climatic influences on crop growth. Ground-based Crop yield data Provide accurate, location-specific yield data that enhance model calibration and validation. Given the absence of suitable historical yield data, the model uses current-season ground yield data for validation. These ground yield data were strategically collected based on categorized satellite and ground inputs to represent diverse ground realities, thereby improving the model’s training and overall estimation accuracy. This research demonstrates the power of integrating remote sensing and machine learning to address yield estimation challenges. The proposed framework is not only effective for potato yield estimation in Uttar Pradesh but is also scalable and adaptable to other crops and regions. By enabling data-driven decision-making, it supports more efficient and sustainable agricultural practices, enhances food security, and contributes to greater resilience in the agricultural sector.