Robust County-Level Corn Yield Estimation Using Ensemble Machine Learning and Multisource Remote Sensing

Abstract

Advancement of sustainable agriculture techniques and ensuring food security depend on accurate and reliable crop yield forecasting. Accurate and reliable crop yield forecasting is essential for sustainable agricultural planning and global food security. However, data quality issues—such as missing values and temporal misalignments in remote sensing datasets—often challenge the robustness of machine learning models. This study proposes a robust yield prediction framework that integrates multi-source data, including MODIS-based Gross Primary Production (GPP), vegetation indices (NDVI, EVI), climate variables, and soil properties, to estimate maize yield at the county level across the U.S. Corn Belt. Two ensemble models, Random Forest and Extreme Gradient Boosting (XGBoost), are trained and evaluated under both clean and simulated degraded data conditions. XGBoost achieved the highest accuracy (RMSE = 14.58, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.84), while Random Forest demonstrated strong robustness (RMSE = 15.10, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.82), even when early-season NDVI was missing or GPP time series were temporally shifted. Feature importance analysis identified late-season GPP and soil organic matter as the most influential predictors. These findings demonstrate the potential of ensemble learning models to deliver reliable and interpretable yield forecasts, even under imperfect data conditions, making them practical tools for real-world precision agriculture applications. This work provides a practical framework for implementing AI-driven solutions in large-scale, real-world remote sensing-based agricultural monitoring and yield forecasting systems.

Topics & Concepts