Traditionally the production of climate and earth system models has been the preserve of large modelling centres whose calibration process is opaque to outsiders and appears to be very manually intensive. Computational methods to calibrate models to observations would have many applications:
Generating multiple models consistent with observations. This then allows uncertainties on climate projections constrained by agreement with observations to be produced without the need for artificial flux corrections.
Adding new processes to, or changing the resolution of, a model and then recalibrating it.
Recalibrating the model to a different set of observations.
In this talk I will show that it is possible to automatically calibrate the HadAM3 and HadAM3P atmospheric models using short simulations of 5 years or so. Coupled atmosphere/ocean models produced from these calibrated HadAM3 models are often, but not always, stable and broadly consistent with observed estimates of pre-industrial temperature. For some model configurations we computed their equilibrium climate sensitivity (ECS) and transient climate response (TCR). We find a narrow range of roughly 3.5+/-0.5 for ECS and 2.25+/-0.25 for TCR. In general uncertainties on other aspects of climate system change also appear relatively narrow from this ensemble of perturbed models. One exception appears to be the Atlantic Meridional Overturning Circulation.
We also investigated calibrating the sea-ice climatology of HadCM3 through changing sea-ice properties and ocean diffusion. We found that the calibrated model has sea-ice extent changes much closer to observed losses in the Northern Hemisphere. However the model fails to reproduce the observed changes in the Southern Hemisphere.
Overall, we conclude that automatic calibration of climate models is possible and computationally feasible.