DNDC Model Calibration, Validation and Quantification of Structural Uncertainty to Support Rice Methane Offset Protocols

Abstract:

Agriculture represents an important near-term option for GHG offsets. Currently, the most widely accepted low-cost approaches to quantify N2O and CH4 emissions are based on emission factors. Given that N2O and CH4 emissions from agricultural practices exhibit high spatial and temporal variability, emission factors are not very sensitive to estimate this variability in emissions at the farm level, even when the emission factors are regional. It is clear that if agricultural offset projects are going to include N2O and CH4 reductions, then process-based biogeochemical models are potentially important tools to quantify emission reductions within offset protocols. The question remains how good a model’s performance is with respect to emission reductions. As PBM, are integrated into protocols for agricultural GHG offsets, comprehensive and systematic validation is needed to statistically quantify uncertainties in model-based estimates of GHG emission reductions that are obtained by standardized approach to parameterization and calibration that can be applied across a whole region.

The DNDC model was validated against 88 datasets of rice methane emissions. Data were collected at sites in California and MidSouth. In addition to examining the magnitude of the measured versus modeled emissions, we analyzed model performance for estimating the changes in emissions associated with a change in management practices (e.g. dry versus wet seeded rice, different fertilizer rates, etc.). We analyzed 100 pairs of modeled and measured emission reductions. DNDC model performance and uncertainty was quantified using a suite of statistical measures. First, we examined how well the modeled emissions differences match the field-measured differences on a case-by-case basis and also on average, using a combination of Monte Carlo approaches and equivalence testing. Although modeled emissions for individual fields show a slight bias, emissions reductions for baseline:treatment pairs fall close to the 1:1 line. We illustrate how these results can be used to drive a Margin Of Safety (MOS) uncertainty deduction. The magnitude and impacts of the MOS deduction depend on how emissions are bundled at the field, project, or market levels.