The Ohio State University Department of Food, Agricultural and Biological Engineering faculty Darren Drewry and Ph.D. student James Cross recently published an article in Ecological Informatics, a publication of high-quality, peer-reviewed articles on all aspects of computational ecology, data science, biogeography, and ecosystem analysis.
The article, titled "Ensemble machine learning for interpretable soil heat flux estimation" focuses on accurately estimating soil heat flux (SHF), crucial for understanding surface energy balance and soil processes. Existing methods have primarily targeted midday estimates, often overlooking diurnal and seasonal variability. Addressing this gap, the study introduces ensemble machine learning (ML) models to predict SHF at half-hourly intervals throughout growing seasons across various crops (soybean, corn, sorghum, and miscanthus). These ML models, compared to semi-empirical methods, consistently outperform in capturing SHF variability, achieving over 86% accuracy with minimal predictor variables. Importantly, the study utilizes Shapley additive explanations (SHAP) to enhance model interpretability, highlighting insights into predictor interactions crucial for model selection and application.
Overall, the research underscores ML's capability to effectively model SHF dynamics across diverse agricultural systems, emphasizing the importance of tailored predictor selection and interpretability in advancing predictive accuracy. This comprehensive evaluation sets a foundation for leveraging advanced satellite and in-situ observations to enhance future SHF estimations and applications in agricultural and environmental studies.
The article will appear in the September 2024 print issue of Ecological Informatics and is available online now.