A dynamic approach for corn yield prediction to ensure agricultural resilience in the U.S. Midwest

Authors: MITRA, ALAKANANDA - University Of Nebraska; KARKI, SAGUN - University Of Nebraska; Fleisher, David; RAY, CHITTARANJAN - University Of Nebraska; Reddy, Vangimalla

Submitted to: Smart Agricultural Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/6/2025
Publication Date: 8/12/2025
Citation: Mitra, A., Karki, S., Fleisher, D.H., Ray, C., Reddy, V. 2025. A dynamic approach for corn yield prediction to ensure agricultural resilience in the U.S. Midwest. Smart Agricultural Technology. https://doi.org/10.1016/j.atech.2025.101295.
DOI: https://doi.org/10.1016/j.atech.2025.101295

Interpretive Summary: We need accurate crop production estimations to make sure there’s enough food, especially with different climate and weather, soils, and farming methods across the U.S. In this study, a novel AI-based algorithm for corn yield estimation was developed for the Midwest corn belt while considering varying weather conditions. The important factor is that this method is not crop specific and can be applied to any other crop and region. Scientists, farm managers, and land-use policy developers can benefit from the use of this predictive tool to estimate impacts of weather factors on crop productivity and food security.

Technical Abstract: Accurate prediction of crop yield is essential for food security. However, regional weather influences agricultural production. Machine learning (ML) and deep learning (DL) have been extensively used to estimate agricultural yields due to their precision. However, retraining these data-centric models with site-specific information is required to improve their scope and accuracy on a new site. This research aimed to incorporate spatial effects into our previously proposed ML-based methodologies to expand the method in the Midwest corn belt of the U.S. This paper presents a robust and adaptive neural network-based model to forecast future yields using dynamic integration of new data and addressing the unprecedented effects of climate change in the modeling. We considered more than $1000$ counties in the Midwest. Fourteen years of daily county average weather data and county-level yield data were used to build the model. A geo-spatial module was proposed to spread the model spatially. Using Bayesian Optimization, we efficiently navigated the hyperparameter space, ensuring the model's precision to capture the intricate patterns within the data, thereby improving the prediction precision. This comprehensive approach underlines our commitment to developing predictive models that are accurate and adaptable to the multifaceted influences of agriculture.