An international intercomparison and benchmarking of crop and pasture models simulating GHG emissions and C sequestration

The development of climate mitigation services partly depends on our ability to simulate, with confidence, agricultural production and greenhouse gas (GHG) emissions so as to understand the effectiveness of the mitigation approach on both gas emissions and food production. The Soil C-N Group of the Global Research Alliance (GRA) on GHG has initiated an international model benchmarking and inter-comparison that will assess GHG balance and soil C sequestration of arable crops and grasslands as affected by agricultural practices. The inter-comparison arises from collaborations between GRA, AgMIP and four FACCE-JPI projects to lead to the largest exercise in this domain. An initial stock take has been conducted, resulting in the selection of datasets from five grasslands and five crop sites worldwide. A total of 28 models used in 11countries for the prediction of GHG emissions in crop and grassland systems are contributing, ranging from process-oriented models to simpler models. The study has been set up with five successive steps that gradually release information to the modeling groups ranging from fully-blind application of the models to complete availability of the experimental measurements. Model simulations are compared to experimental measurements for crop yield and grassland dry-matter production, N₂O emissions, soil C stocks and net CO₂ exchanges. The precision and accuracy of the predictions are evaluated at each step of the inter-comparison with statistical methods, facilitating quantification of projection uncertainties. Results from the first step on N₂O emissions with no prior information show variability between model predictions for any site and that model error tends to be conserved across sites. Moreover, the frequency distribution of N₂O emissions already provides an understanding of model functioning in terms of N₂O peak prediction. Further steps will allow for improved site-specific prediction and, as a final step, will expose the measured GHG emissions for model improvement.