Agricultural nitrous oxide (N₂O) emission accounts for a non-trivial fraction of global greenhouse gas (GHG) budget. To date, estimating N₂O fluxes from cropland remains a challenging task because the related microbial processes (e.g., nitrification and denitrification) are controlled by complex interactions among climate, soil, plant and human activities. Existing approaches such as process-based (PB) models have well-known limitations due to insufficient representations of the processes or uncertainties of model parameters, and due to leverage recent advances in machine learning (ML) a new method is needed to unlock the “black box” to overcome its limitations such as low interpretability, out-of-sample failure and massive data demand. In this study, we developed a first-of-its-kind knowledge-guided machine learning model for agroecosystems (KGML-ag) by incorporating biogeophysical and chemical domain knowledge from an advanced PB model, ecosys, and tested it by comparing simulating daily N₂O fluxes with real observed data from mesocosm experiments. The gated recurrent unit (GRU) was used as the basis to build the model structure. To optimize the model performance, we have investigated a range of ideas, including (1) using initial values of intermediate variables (IMVs) instead of time series as model input to reduce data demand; (2) building hierarchical structures to explicitly estimate IMVs for further N₂O prediction; (3) using multi-task learning to balance the simultaneous training on multiple variables; and (4) pre-training with millions of synthetic data generated from ecosys and fine-tuning with mesocosm observations. Six other pure ML models were developed using the same mesocosm data to serve as the benchmark for the KGML-ag model. Results show that KGML-ag did an excellent job in reproducing the mesocosm N₂O fluxes (overall r²=0.81, and RMSE=3.6 $\mathrm{mg}\mathrm{N}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{d}}^{-\mathrm{1}}$ from cross validation). Importantly, KGML-ag always outperforms the PB model and ML models in predicting N₂O fluxes, especially for complex temporal dynamics and emission peaks. Besides, KGML-ag goes beyond the pure ML models by providing more interpretable predictions as well as pinpointing desired new knowledge and data to further empower the current KGML-ag. We believe the KGML-ag development in this study will stimulate a new body of research on interpretable ML for biogeochemistry and other related geoscience processes.

Corn production, and its associated inputs, is a relatively large source of greenhouse gas emissions and uses significant amounts of water and land, thus contributing to climate change, fossil fuel depletion, local air pollutants, and local water scarcity. As large consumers of this corn, corporations in the ethanol and animal protein industries are increasingly assessing and reporting sustainability impacts across their supply chains to identify, prioritize, and communicate sustainability risks and opportunities material to their operations. In doing so, many have discovered that the direct impacts of their owned operations are dwarfed by those upstream in the supply chain, requiring transparency and knowledge about environmental impacts along the supply chains. Life cycle assessments (LCAs) have been used to identify hotspots of environmental impacts at national levels, yet these provide little subnational information necessary for guiding firms’ specific supply networks. In this paper, our Food System Supply-Chain Sustainability (FoodS³) model connects spatial, firm-specific demand of corn purchasers with upstream corn production in the United States through a cost minimization transport model. This provides a means to link county-level corn production in the United States to firm-specific demand locations associated with downstream processing facilities. Our model substantially improves current LCA assessment efforts that are confined to broad national or state level impacts. In drilling down to subnational levels of environmental impacts that occur over heterogeneous areas and aggregating these landscape impacts by specific supply networks, targeted opportunities for improvements to the sustainability performance of supply chains are identified.

The aim of this study is to analyze the characteristics of drought, such as intensity and trends, based on SPEI (Standardized Precipitation Evapotranspiration Index) at 8 stations in South Korea from 1981 to 2010. The traditional SPEI is based on the Thornthwaite equation for estimating evapotranspiration; SPEI_th. However, a standard of agricultural water management in Korea suggests the FAO Penman-Monteith equation; SPEI_pm. Therefore, we analyzed the intensity, variability, and trends of drought using SPEI_th and SPEI_pm, respectively, and compared the results. SPEI_pm showed slightly more intensive drought rather than SPEI_th except for Chuncheon and Gwangju. In 5 stations—excluding Cheoncheon, Gwangju and Jinju—the cumulative probability that SPEI_pm was below −1.5 was significantly increased from 1981–1995 to 1996–2010. In addition, the northwest and southwest regions had higher intensity of 1-month droughts, and the central and southwest regions had a higher intensity of 3-month droughts. According to the Mann–Kendall test, there was a decreasing trend of 1-month SPEI during the fall season and 3-month SPEI during winter season.