OOS 82-5
Explaining the sources of variation in CMIP5 models by fitting reduced complexity models to their simulation outputs

Friday, August 14, 2015: 9:20 AM
310, Baltimore Convention Center
Forrest M. Hoffman, Department of Earth System Science, University of California, Irvine, CA
Matthew Smith, Computational Science Laboratory, Microsoft Research, Cambridge, United Kingdom
Katherine Todd-Brown, Microbiology Group, Pacific Northwest National Laboratory, Richland, WA
Yiqi Luo, Microbiology and Plant Biology, University of Oklahoma, Norman, OK
Yingping Wang, CSIRO Marine and Atmospheric Research, Victoria 3195, Australia

Use of Earth system models (ESMs) is critical for understanding climate responses to anthropogenic perturbations of global biogeochemical cycles. We explained primary carbon fluxes (gross primary productivity (GPP), autotrophic respiration (Ra), heterotrophic respiration (Rh), vegetation-to-soil flux (Fv)), net ecosystem exchange (NEE), and carbon stocks (in soils and vegetation) in ESMs participating in the Fifth Climate Model Intercomparison Project (CMIP5) by statistically parameterizing reduced-complexity process-based models to best match the ESM carbon cycle responses when driven by ESM physical climate variables. These reduced complexity models were fit at the grid level and then validated using both local and global totals and means. Parameter estimates were analysed to assess the relative contributions of different sources of model variation to overall predictions.


In general the primary fluxes were well described by our proposed model (R2 > 0.9), although Fv was only moderately described (R2 > 0.75). To explain GPP, all ESMs required shortwave radiation and leaf area index (LAI); most models required CO2; and some models were modestly improved with temperature and moisture. For Ra, all models required GPP; some models were modestly improved by temperature and moisture. Fvwas almost entirely explained by GPP alone across all models. Like previous work done with soil respiration, most of the models did not require direct consideration of moisture. This was a surprising result but may be due to the fact that moisture is frequently correlated with other driving variables.

When used to explain the carbon stocks, fluxes, and NEE together, our model captures primary fluxes with similar accuracy and also captures the dynamics of vegetation and soil carbon well (R2 > 0.9), but explains global net ecosystem exchange with lower accuracy (R2 > 0.4). In that regard, our study appears to be analogous to the process of predicting the carbon cycle with complex ESMs in that, in both cases, the simpler model appears to explain the primary fluxes well but cannot capture the net overall fluxes well. Our analysis reveals potentially important contributions to model variation caused by baseline residence time parameters for soil carbon, variation in the sensitivity of those residence times to climate change, the sensitivity of GPP to variation in leaf area, and the sensitivity of GPP to climate change. These results should be used to guide future efforts to reduce model uncertainty by pinpointing key fluxes where important uncertainties can be reduced by calibrating multiple models to observational data.