Estimating uncertainty in N2O emissions from US cropland soils

A Monte Carlo analysis was combined with an empirically based approach to quantify uncertainties in soil nitrous oxide (N2O) emissions from U.S. croplands estimated with the DAYCENT simulation model. Only a subset of croplands was simulated in the Monte Carlo analysis, which was used to infer uncertainties across the larger spatiotemporal domain. Specifically, one simulation representing dominant weather, soil type, and N inputs was performed for each major commodity crop in the 3000 counties occurring within the conterminous United States. We randomly selected 300 counties for the Monte Carlo analysis and randomly drew model inputs from probability distribution functions (100 iterations). A structural uncertainty estimator was developed by deriving a statistical equation from a comparison of DAYCENT-simulated N2O emissions with measured emissions from experiments in North America. We estimated soil N2O emission of 201 Gg N from major commodity crops in 2007, with a 95% confidence interval (CI) of 133-304 Gg N. This implies a relative CI of 34% below and 51% above the estimate at the national scale, but the CIs tended to be larger at the regional level, particularly in regions with low emissions. Spatial variability in emissions was driven primarily by differences in N inputs from fertilizer and manure, while temporal variability was driven more by N mineralization rates, which are correlated with weather patterns in DAYCENT. A higher portion of total uncertainty was due to model structure compared to model inputs, suggesting that improvements in model algorithms and parameterization are needed to produce results with higher precision and accuracy.