2019 CARD Award for Best Ph.D. Dissertation in Agricultural, Environmental, and Energy Economics and Policy

Predicting County Level Corn Yields Using Deep Long Short Term Memory Models In the Corn Belt

Zehui Jiang

Corn yield prediction is beneficial as it provides valuable information about production and prices prior to harvest. Publicly available high-quality corn yield prediction can help address emergent information asymmetry problems and, in doing so, improve price efficiency in futures markets. This paper is the first to predict corn yield using Long Short-Term Memory (LSTM), a special Recurrent Neural Network method. Our prediction is only 0.83 bushel/acre lower than actual corn yields in the Corn Belt, whose difference is lower than the prediction from USDA. Eighty percent of our LSTM county-level corn yield predictions fall in the +/-20 region. Results show our LSTM model can provide good early prediction and accurate Corn Belt county-level corn yield prediction without farm management and corn seed data. Our LSTM models for county level corn yield prediction in the Corn Belt are an improvement to the USDA prediction. And more importantly, our models provide a publicly available source that will contribute to eliminating the information asymmetry problem that arises from private company crop yield prediction.