Optimal tuning of a GCM using modern and glacial constraints

In climate models, many parameters used to resolve subgrid scale processes can be adjusted through a tuning exercise to fit the model's output to target climatologies. We present an objective tuning of a low resolution Atmosphere-Ocean General Circulation Model (GCM) called FAMOUS where ten model parameters are varied together using a Latin hypercube sampling method to create an ensemble of 100 models. The target of the tuning consists of a wide range of modern climate diagnostics and also includes glacial tropical sea surface temperature. The ensemble of models created is compared to the target using an Arcsin Mielke score. We investigate how the tuning method used and the addition of glacial constraints impact on the present day and glacial climates of the chosen models. Rather than selecting a single configuration which optimises the metric in all the diagnostics, we obtain a subset of nine 'good' models which display great differences in their climate but which, in some sense, are all better than the original configuration. In those simulations, the global temperature response to last glacial maximum forcings is enhanced compared to the control simulation and the glacial Atlantic Ocean circulation is more in agreement with observations. Our study demonstrates that selecting a single 'optimal' configuration, relying only on present day constraints may lead to misrepresenting climates different to that of today.