Abstract:
This dissertation examines how remote sensing and advanced statistical and machine learning methods can improve crop yield estimation at the farm scale. It addresses the lack of reliable yield data in developing and low-income countries, where timely and accurate estimation is essential for food security, farm income, and policy decisions. The study combines high-resolution Sentinel-2 imagery, UAV-based vegetation indices, crop height, solar radiation, and soil properties to build yield models for cotton and wheat. The results show that integrating multiple indicators improves estimation accuracy compared with single-variable approaches. Hyperparameter-tuned machine learning models further enhance predictive performance and reduce dependence on any single metric. The dissertation demonstrates that publicly available satellite data and low-cost UAV sensors can provide practical, scalable, and accurate tools for agricultural yield estimation and decision-making for wider real-world use.
