College of Food, Ag & Nat Res Sci
Spatially and temporally explicit data is necessary for determining the environmental impacts of agriculture across landscapes and under future conditions of climate change, and for efficiently targeting policies and practices. For many agricultural systems, spatial and temporal data is still lacking; however, methods of developing this data are helping address this issue (Keating and Thorburn 2018). Biogeochemical models, in particular, are a key tool for developing spatial and temporal data because they overcome the lack of scalability of measurements and the lack of granularity of methods like emission factors. Leveraging Ecosys, a detailed mechanistic model that contains mathematical representations of the transport and transformation of heat, water, O, C, N, P, and ionic solutes through the soil-plant-atmosphere system, using processes that range in scale from organ to plant community, highly granular simulated agricultural data is possible. One drawback of complex models like Ecosys, however, is that they take considerable data and time to run. This makes it difficult to apply the model across large areas at fine spatial granularity. Using machine-learning-based metamodels, and through advanced high-performance computing infrastructure, this project seeks to connect biogeochemical modeling that captures the variability of environmental impacts due to climate and soil factors with a framework that translates physical variability into costs and benefits, which can more easily inform policymakers about potential benefits of changes to management practices and where they might be targeted.