Sort Options:

• Sort By Last Name
• Sort By Title

Other:

• Print All

Sort by time-scale:

• Short
• Intermediate
• Long
• SHOW ALL

Impacts of Systematic Error Reduction on CAM3.1 Sensitivity to CO2 Forcing

Impacts of Systematic Error Reduction on CAM3.1 Sensitivity to CO2 Forcing

Charles Jackson, Institute for Geophysics, University of Texas at Austin
Yi Deng, Institute for Geophysics, University of Texas at Austin
Gabriel Huerta, Department of Statistics, University of New Mexico
Mrinal K. Sen, Institute for Geophysics, University of Texas at Austin

The current disparity that exists among models of the climate system in their response to projected increases in greenhouse gases provides a measure of uncertainty in the model development process. A large part of this uncertainty is likely related to specification of model parameters. We estimate a lower bound to this part of the uncertainty as may be inferred from an ensemble of model configurations made from a single model (the NCAR CAM3.1 atmospheric GCM). The choice of ensemble members is constrained by a stochastic, Bayesian based, importance sampling strategy whose likelihood function includes a normalized, multivariate measure of model skill that quantifies the distance among seasonal climatologies of model predictions and fifteen observational/reanalysis data products. We consider the effects of six parameters important to clouds and convection. The top six performing parameter sets improved model skill by 7% with nearly identical skill scores, but for different reasons related to the wide range of selected parameter values. These model configurations were chosen for estimating the effect of parametric uncertainties on the predicted global warming response to 2xCO2. Five of the six model configurations had a 2xCO2 near surface air temperature sensitivity of 3 or 3.1 degrees with the final member having a sensitivity of 3.4 as compared to the 2.4 degree sensitivity of the default model configuration. Although the range in sensitivities was quite narrow after parameter values have been systematically constrained by observations, the regional climate predictions exhibited significant uncertainties up to 25% of the climate change signal for predictions of surface air temperature and up to 160% of the signal for precipitation. This calculation demonstrates the potential of using observations to substantially reduce climate model prediction uncertainties with a more formal method of multivariate model tuning. It also provides an estimate of the upper bound for single-model prediction skill, particularly for regional climates.