Evaluating land cover influences on model uncertainties—A case study of cropland carbon dynamics in the Mid-Continent Intensive Campaign region

Quantifying spatial and temporal patterns of carbon sources and sinks and their uncertainties across agriculture-dominated areas remains challenging for understanding regional carbon cycles. Characteristics of local land cover inputs could impact the regional carbon estimates but the effect has not been fully evaluated in the past. Within the North American Carbon Program Mid-Continent Intensive (MCI) Campaign, three models were developed to estimate carbon fluxes on croplands: an inventory-based model, the Environmental Policy Integrated Climate (EPIC) model, and the General Ensemble biogeochemical Modeling System (GEMS) model. They all provided estimates of three major carbon fluxes on cropland: net primary production (NPP), net ecosystem production (NEP), and soil organic carbon (SOC) change. Using data mining and spatial statistics, we studied the spatial distribution of the carbon fluxes uncertainties and the relationships between the uncertainties and the land cover characteristics. Results indicated that uncertainties for all three carbon fluxes were not randomly distributed, but instead formed multiple clusters within the MCI region. We investigated the impacts of three land cover characteristics on the fluxes uncertainties: cropland percentage, cropland richness and cropland diversity. The results indicated that cropland percentage significantly influenced the uncertainties of NPP and NEP, but not on the uncertainties of SOC change. Greater uncertainties of NPP and NEP were found in counties with small cropland percentage than the counties with large cropland percentage. Cropland species richness and diversity also showed negative correlations with the model uncertainties. Our study demonstrated that the land cover characteristics contributed to the uncertainties of regional carbon fluxes estimates. The approaches we used in this study can be applied to other ecosystem models to identify the areas with high uncertainties and where models can be improved to reduce overall uncertainties for regional carbon flux estimates.