EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Soil-weather-yield dynamics: Contextualized LSTM-DNN model for county-level corn yield estimation from weather and soil

M.S. Basir, Dennis R. Buckmaster, Yue Shirley Li, Ningning Nikole Kong, James V. Krogmeier

Smart Agricultural Technology · 2026-12 · 15 · 102660

DOI: 10.1016/j.atech.2026.102660 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von M.S. Basir und 4 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate crop yield prediction is essential for agricultural insurance, improved risk management, and the development of efficient production strategies. In this study, we investigate the spatial and temporal variability of weather, soil, and corn yield in the state of Indiana to understand the relationship between these factors and corn yield. Heterogeneous spatial and temporal granularity complicates the modeling process. We propose a novel machine-learning framework for county-level corn yield prediction that integrates daily weather data, weather-derived persistent-homology features, and static physical soil information. A Long Short-Term Memory (LSTM) network captures sequential weather patterns, and a shallow feed-forward network handles yearly weather-derived persistent-homology features. Two network outputs are concatenated with soil data and fed into a deep feed-forward network optimized to predict the corn yield at a county level for each year. For the predefined holdout experiment, the model was developed using 1,300 county-year observations and evaluated on 50 predefined holdout observations. Performance was compared to a Convolutional Graph Neural Network (CGNN), a Transformer, Support Vector Regression (SVR) with a radial basis function (SVR-RBF), Extreme Gradient Boosting (XGBoost), and combined Vector Autoregression with SVR (VAR-SVR) models. On the predefined 50 county-year holdout set, the LSTM-DNN model achieved R 2 = 0.77 , RMSE = 0.735 t/ha, MAE = 0.590 t/ha, and MAPE = 5.18%. The nearest competitor, the CGNN model, achieved corresponding values of 0.76, 0.746 t/ha, 0.587 t/ha, and 5.23%, respectively. The LSTM-DNN model was computationally simpler than the CGNN model. Adding topology features increased holdout R 2 from 0.581 to 0.766 (absolute increase of 0.185) and reduced RMSE from 0.980 to 0.735 t/ha (25% reduction). Under blocked validation, pooled R 2 was 0.565 for transfer to unseen counties and 0.255 for forward prediction to future years, indicating stronger spatial than temporal generalization. Perturbation analysis showed that the model was sensitive to changes in the seasonal weather representation and weather timing. The approximately 7–8% prediction changes associated with 1 ∘ F temperature or 1% rainfall perturbations represent model sensitivity propagated through the LSTM weather sequence, recalculated GDD, and weather-derived topological features. The Contextualized LSTM-DNN model also shows sensitivity to shifting weather by one week, with 3.5-4.5% reduced average yield prediction. As the development and evaluation of the model were done using data from Indiana only, its transferability to regions with different soil, climate, and production conditions remains to be evaluated.

Förderung

  • National Science Foundation

Zitieren

APA

Basir, M., Buckmaster, D. R., Li, Y. S., Kong, N. N., & Krogmeier, J. V. (2026). Soil-weather-yield dynamics: Contextualized LSTM-DNN model for county-level corn yield estimation from weather and soil. Smart Agricultural Technology, 15, 102660. https://doi.org/10.1016/j.atech.2026.102660

Vancouver

Basir M, Buckmaster DR, Li YS, Kong NN, Krogmeier JV. Soil-weather-yield dynamics: Contextualized LSTM-DNN model for county-level corn yield estimation from weather and soil. Smart Agricultural Technology. 2026;15:102660. doi:10.1016/j.atech.2026.102660

BibTeX
@article{Basir2026,
  author = {Basir, M.S. and Buckmaster, Dennis R. and Li, Yue Shirley and Kong, Ningning Nikole and Krogmeier, James V.},
  title = {Soil-weather-yield dynamics: Contextualized LSTM-DNN model for county-level corn yield estimation from weather and soil},
  journal = {Smart Agricultural Technology},
  year = {2026},
  volume = {15},
  pages = {102660},
  doi = {10.1016/j.atech.2026.102660}
}
Kostenlos starten →