DATA-DRIVEN CROP FORECASTING FRAMEWORK USING SATELLITE IMAGERY AND MACHINE LEARNING FOR FOOD SECURITY PLANNING

Abstract

Ensuring reliable crop yield forecasting is a central challenge for modern food security frameworks as global agricultural systems face rising uncertainty from climate variability, soil degradation, resource scarcity, and unpredictable extreme weather events. Recent advancements in satellite-based remote sensing and machine learning offer new pathways for constructing robust, data-driven crop forecasting models capable of operating at regional and national scales. This study develops an integrative forecasting framework that fuses multispectral satellite imagery, vegetation indices, meteorological variables, and high-resolution soil datasets with supervised machine learning techniques to predict crop yields with greater accuracy and spatial precision. By incorporating temporal profiles of the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Land Surface Temperature (LST), rainfall anomalies, evapotranspiration, and soil moisture, the framework captures both spatial and temporal crop-growth dynamics. Multiple models, including Random Forests, Gradient Boosting, and Temporal Convolutional Networks, were evaluated to determine the optimal predictive architecture for multi-crop environments. The results demonstrate that combining spectral profiles with meteorological time series significantly improves forecasting accuracy over conventional statistical or remote-sensing-only approaches. The study underscores that integrated satellite–ML forecasting systems are critical tools for food security stakeholders, enabling proactive planning, risk mitigation, and data-driven policy formulation. This framework also provides a scalable foundation for national food security agencies to manage crop monitoring, optimize resource allocation, and reduce vulnerability to agricultural instability.