Sunday, September 20, 2009

Model Misfits

The WCC-3 conference took place in Geneva, Switzerland. One of the session, on climate prediction was quite interesting. I have downloaded the powerpoints and listened to the audio. The powerpoints can be found at Meehl

The audios can be found at this place in the Climate prediction section.

I am going to talk about Palmer and Latif's talk. Palmer discussed the resolution issue and the problems global climate models have. The first one is one of bias. He said this about the biases in the climate models.

"Now these biases , unfortunately can be as large as the signal. We can take these biases out with these empirical post processing corrections. But anyone who knows anything about nonlinear dynamics and the climate system is a nonlinear system will know that this kind of post processing correction methodology is not a reliable procedure. So this raises some of the difficulties we have." T.N. Palmer, Seasonal to Interannual Prediction" WCC-3

So, we use an admittedly unreliable procedure to make the models output meaningful???? That sounds like mental mas..., well you know.

Palmer then speaks about the failure of the climatological predictions to predict the horribly heat wave and drought Europe experienced in 2003. The models now, he says have been entirely revamped and can now retrodict the summer of 2003. How wonderful that is. We were told how good the models were in 2001, 2002, 2003 and they failed to predict the event. But with about 4 years of revamping the models they can now retrodict what happened! He lists all the things that were not in the 2003 model which are now there. What he doesn't do is predict what will need to be revamped in the models to make it work in the future.

He then speaks about the grid size problem of models. The best climate model has a grid size of 20 km shown below. This is good but even this scale can't handle local cloud conditions.

Most models are around 80 km grid spacing and the world looks like this.

The models they began to scare us with are actually about 4x wose than the above.

Mojib Latif then gave a talk about decadal predictions by climate models. Afterall, if a model is correct, it should predict things on a decade level. They don't.

Latif compared two models with the observations. You can see that the errors in the models are bad. Some models, Latif says, have errors of up to 10 degrees and are not very representative or reality.

For the Atlantic dipole, the models predict too high of a temperature.

And models can't predict the rainfall any better than your local weatherman. On the right is the observed rainfall along the Gulf Stream. In the middle is a global climate model with a 50 km grid size--near state of the art. You can see that the rainfall amount is wrong and the place where the heaviest rainfall falls is also wrong. The last view is of the smoothed model, which loses all resolution and really isn't predicting anything--yet it is a 2009 model.

All the doom and gloom forecasts of climate heating come from the models. And the models can't get much really right when looking backwards. That should cause one to pause when one wants to predict forward.

One final picture from the conference shows the sunspot irradiance which is now lower than it has been for a long time. This effect as well should cool the earth unless the sun starts making sunspots.

Can anyone say "Maunder Minimum"?

No comments:

Post a Comment