I have a paper [PDF] in this week’s Science discussing the water vapor feedback. It is a Perspective, meaning that it is a summary of the existing literature rather than new scientific results. In it, my co-author Steve Sherwood and I discuss the mountain of evidence in support of a strong and positive water vapor feedback.

Interestingly, it seems that just about everybody now agrees water vapor provides a robustly strong and positive feedback. Roy Spencer even sent me email saying that he agrees.

What I want to focus on here is model verification. If you read the blogs, you’ll often see people say things like “the models are completely unvalidated.” What they mean is that no one has produced a 100-year climate run with a model, then waited a hundred years, and evaluated how the model did. There are many practical problems with doing this, but the biggest is that by the time you determine if your model was right or not, it would be too late to take any meaningful action to head off the problem.

Of course, we could compare the last 150 years of observations to climate model runs for that same period. This has been done, and the models do a pretty good job. But because the models are constructed with knowledge of the present climate, this is clearly a weaker test than one in which you do not know what the answer is in advance.

What about comparing the climate models with shorter temperature records? Like 10 or 15 years? This will not work. In order for the models to correctly simulate short timescales like this, the models have to be initialized with the atmosphere’s present state. This is not done, so short-term fluctuations in the model and in the atmosphere may be completely uncorrelated, and over short times the temperature record may diverge significantly.

So how do we get a sense of the reliability of climate models? The approach taken by the scientific community is really the only one available: analyze and validate the individual processes in the climate models. Thus, the focus of my work has been to study the water vapor feedback and validate its incorporation into climate models. What we have found is that the models appear to be doing a pretty good job at getting this key parameter correct.

If you can validate enough processes of the model (water vapor, clouds, ice, oceans, etc.) then you generate confidence that your model is probably making predictions that are at least in the ballpark. In addition, it gives you a good feeling of where the weak points in the model all are. For example, this type of analysis has demonstrated that most of the uncertainty in the model predictions arises due to clouds.

And having identified the most uncertain processes in the model, you can use other physical constraints to bound their magnitude. For example, a strong argument can be made that the cloud feedback is unlikely to be large and negative.

Once you have a good handle on the individual processes that are operating in the climate, then you can actually estimate future warming without a climate model.

Thus, saying that climate models are “unvalidated” sells them short. They may not have been validated in all the ways we would like to validate them, but scientists have indeed spent a lot of time studying and validating the individual physical processes within the models. The result of this is that we can have a reasonably high confidence that we will get a few degrees of warming by the end of the century if emissions are not reduced.