Predicting county-scale maize yields with publicly available data.

Area of Science:

• Agricultural Science
• Data Science
• Machine Learning

Background:

• Maize (corn) is a globally dominant grain, with 2018 production reaching 1.12 billion tons, primarily used for animal feed.
• Accurate corn yield prediction is vital for agricultural markets, influencing prices and addressing information asymmetry.
• The United States is the largest producer, accounting for 32% of global maize production.

Purpose of the Study:

• To develop a deep learning model for accurate, county-level corn yield prediction in the US Corn Belt.
• To provide pre-harvest yield predictions with monthly updates starting in August.
• To establish a foundation for a publicly accessible county yield prediction system.

Main Methods:

• Utilized a deep learning model architecture.
• Focused on county-level prediction across 10 states in the US Corn Belt.
• Incorporated monthly updates for pre-harvest predictions.

Main Results:

• The deep learning model demonstrated promising predictive power for corn yields.
• Results showed superiority or competitiveness compared to existing survey-based prediction methods.
• The model provides monthly pre-harvest updates, enhancing forecast timeliness.

Conclusions:

• The developed deep learning model offers a valuable tool for corn yield prediction.
• This approach can significantly improve the efficiency and transparency of agricultural markets.
• The study lays the groundwork for a public resource complementing existing government forecasts.