Deep neural networks autonomously learn and utilize critical environmental conditions to capture the impacts of weather on spring wheat performance

Authors: Benke, Ryan; HAN, LINQIAN - Washington State University; Garland Campbell, Kimberly; Li, Xianran

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/12/2025
Publication Date: 6/19/2025
Citation: Benke, R.L., Han, L., Garland Campbell, K.A., Li, X. 2025. Deep neural networks autonomously learn and utilize critical environmental conditions to capture the impacts of weather on spring wheat performance. Computers and Electronics in Agriculture. 237(B). Article 110665. https://doi.org/10.1016/j.compag.2025.110665.
DOI: https://doi.org/10.1016/j.compag.2025.110665

Interpretive Summary: Crop performance variations recorded over large spatiotemporal scale reflect the changing climatic conditions crops experienced. Deep neural networks (DNN) were applied to develop weather predictive models to capture this hidden relationship between weather conditions and spring wheat performance, including yield and other important traits. To further understand how these "black box"DNN works, a prioritize bulk-feature sensitivity approach was developed. The analysis showed that DNN autonomously learn and exploit critical environmental conditions to capture impacts from changing climate on wheat performance. This study represents an important step toward understandable AI in enviromic prediction.

Technical Abstract: Environmental conditions play a critical role in shaping crop performance. Enviromic prediction aims to develop models that learn the hidden relationship between environmental conditions and crop traits, enabling forecasts of performance under new conditions. Here, we explored applications of deep learning for enviromic prediction. Deep neural networks (DNN) were trained on over 100,000 environmental features to predict wheat yield and other important traits recorded from 322 environments spanning two decades across 20 locations. The trained DNN models demonstrated a high level of prediction accuracy in both cross-validation and forecast schemes. To shed insights into the inner workings of these “black box” DNNs, we devised a bulk-feature sensitivity analysis and revealed that DNN autonomously learned and utilized key environmental conditions associated with phenotypic variation. Furthermore, we showed that DNN-EP could be integrated with the Finlay-Wilkinson regression model to forecast variety-level performance in new environments. Our findings highlight the potential of deep learning to develop interpretable artificial intelligence models that accurately predict crop performance, offering guidance towards optimizing agricultural practices under changing climate conditions.