A Note on the Climate Model Software

The scientific method requires that a theory make a documented prediction and then an experiment (or observation) performed that tests the prediction. Bayesian inference rules are then used to condition belief in the theory based on the test's results.

Assuming the climate models are the embodiment of scientific theory, which of the climate models predicted the current 10 year (rather flat) temperature trend or the winter of 2010? The answer is none. But that's fine. The models predict climate and not weather, and since climate is usually defined as at least a 30 year record of climate, the climate models should not yet be used to "scientifically" condition our climate priors. (Either direction -- confirm or falsify.)

The alternative, assuming the science is settled and the climate models are engineering works, means that consensus software engineering quality assurance processes must be followed before the results can be used directly (without experiment) as evidence for Bayesian inference. IMHO, such SQA has not yet been adequately performed on the climate model software. A terrible shortcoming, since I think this alternative has the potential to allow us to rationally reach an earlier consensus.

No comments:

Post a Comment