Reducing Uncertainty in Cradle-to-Farmgate Carbon Intensity for Feedstock Using Model-Data Fusion

Abstract

Biofuel carbon intensity (CI) quantifies greenhouse gas (GHG) emissions across the biofuel supply chain, providing a transparent metric guiding renewable fuel policies and enabling market-based mechanisms to reduce GHG emissions. In this study, we focus on the cradle-to-farmgate feedstock segment of biofuel CI, the portion attributable to feedstock production from cradle to farm gate and directly influenced by agricultural management decisions. We quantify field-level cradle-to-farmgate CI of corn and soybean feedstocks by integrating a process-based model (ecosys) with a life cycle assessment (LCA) tool (GREET R&D FD-CIC). We first identified CI uncertainty sources at a “supersite,” a representative farm in Illinois with comprehensive GHG measurements, using a Monte Carlo sampling framework that varied soil properties, management practices, and plant trait parameters across plausible ranges to quantify each source's contribution to CI uncertainty. Unknown plant trait parameters were the largest CI uncertainty source for both crops, with fertilizer information also important for corn because these inputs affect yield, N₂O emissions, and soil organic carbon (SOC) stocks. We revealed widespread field-level variability in cradle-to-farmgate CI within the same county, demonstrating the need for field-level assessment. This variability was most strongly associated with SOC change for soybean (total-effect index, 38%) and N₂O emissions for corn (43%). While inherent uncertainty is inevitable, we demonstrated that a Model-Data Fusion (MDF) approach, which integrates observational data including in situ flux measurements and satellite-based remote sensing to constrain model parameters with known processes, can substantially reduce cradle-to-farmgate CI uncertainty. Specifically, we used crop photosynthesis and yield data for plant trait parameter calibration, and incorporated tillage and cover crop information as model inputs, achieving a 41%–52% CI uncertainty reduction. This work establishes a CI quantification framework composed of process-based models, an LCA tool, MDF, and satellite data, enabling scalable, cost-effective, field-level cradle-to-farmgate CI quantification across regions.