Dissertation Defense - Mengfei Xin - Department of Geography and Geoinformation Science

PhD Candidate: Mengfei Xin
PhD of Science, Earth Systems and Geoinformation Sciences
Department of Geography and Geoinformation Science

Dissertation Title: Comparing Deep Learning Architectures for Corn Yield Prediction: Predictive Performance and Sensitivity to Weather Extremes

Dissertation Chair: Ruixin Yang, PhD, Department of Geography and Geoinformation Science

Committee Members: 

Dr. Chaowei Yang (George Mason University, Department of Geography and Geoinformation Science)

Dr.  Wenying Ji (George Mason University, Department of Civil, Environmental, and Infrastructure Engineering)

Dr. Qian Liu (University of Missouri, College of Agriculture, Food and Natural Resources)

Dr. Sun Ling Wang (U.S. Department of Agriculture, Former ERS Staff)

Abstract: 
Accurate prediction of crop yield is important for food security, resource planning, and climate adaptation. Traditional statistical models and early machine learning methods often struggle with complex spatial and temporal patterns in agricultural systems. Deep learning models such as Long Short Term Memory (LSTM) and Gated Recurrent Units (GRU) have improved yield prediction by learning nonlinear relationships from daily climate and soil data. Transformer models, which use self attention to capture long range dependencies, have recently been applied in some agricultural studies, but their performance for county level corn yield prediction and their behavior under extreme weather remain less well understood.

This dissertation compares three deep learning models for county level corn yield prediction under strictly identical data and experimental settings: a bidirectional LSTM, a bidirectional GRU, and a transformer based PatchTST model. The models are trained on daily climate, vegetation, and soil data from 1980 to 2014 and evaluated on a test period from 2017 to 2020 for 943 counties in the United States Corn Belt. Yields are detrended with piecewise linear regression to separate technological trends from weather driven variability, and each model processes 183 day growing season sequences.

The baseline comparison shows that the transformer provides the highest predictive accuracy, with an overall test R² of 0.805, compared to 0.773 for the LSTM and 0.758 for the GRU. A three stage, percentile based perturbation framework is then used to test model responses to drought, heat, compound hot and dry scenarios, and a reconstructed 2019 wet event. Under the most severe heat and compound scenarios, the transformer predicts substantially larger yield losses than the recurrent models, while the LSTM and GRU match the observed 2019 wet anomaly more closely. These results indicate that transformer models are attractive for accurate baseline forecasts, while recurrent models may be preferable when stable behavior is required under extreme climate conditions.