Regional scale cropland carbon budgets: Evaluating a geospatial agricultural modeling system using inventory data

Accurate quantification and clear understanding of regional scale cropland carbon (C) cycling is critical for designing effective policies and management practices that can contribute toward stabilizing atmospheric CO₂ concentrations. However, extrapolating site-scale observations to regional scales represents a major challenge confronting the agricultural modeling community. This study introduces a novel geospatial agricultural modeling system (GAMS) exploring the integration of the mechanistic Environmental Policy Integrated Climate model, spatially-resolved data, surveyed management data, and supercomputing functions for cropland C budgets estimates. This modeling system creates spatially-explicit modeling units at a spatial resolution consistent with remotely-sensed crop identification and assigns cropping systems to each of them by geo-referencing surveyed crop management information at the county or state level. A parallel computing algorithm was also developed to facilitate the computationally intensive model runs and output post-processing and visualization. We evaluated GAMS against National Agricultural Statistics Service (NASS) reported crop yields and inventory estimated county-scale cropland C budgets averaged over 2000–2008. We observed good overall agreement, with spatial correlation of 0.89, 0.90, 0.41, and 0.87, for crop yields, Net Primary Production (NPP), Soil Organic C (SOC) change, and Net Ecosystem Exchange (NEE), respectively. However, we also detected notable differences in the magnitude of NPP and NEE, as well as in the spatial pattern of SOC change. By performing crop-specific annual comparisons, we discuss possible explanations for the discrepancies between GAMS and the inventory method, such as data requirements, representation of agroecosystem processes, completeness and accuracy of crop management data, and accuracy of crop area representation. Based on these analyses, we further discuss strategies to improve GAMS by updating input data and by designing more efficient parallel computing capability to quantitatively assess errors associated with the simulation of C budget components. The modularized design of the GAMS makes it flexible to be updated and adapted for different agricultural models so long as they require similar input data, and to be linked with socio-economic models to understand the effectiveness and implications of diverse C management practices and policies.