My opinion on the Global Climate Model clique feedback loop was requested by not one but two people, and how could I resist?
The text starts well, by assuring readers of the most important point, which is you don’t know enough to intelligently comment on the code itself which is true, certainly for readers of WUWT, and quite likely for many of my readers too. But already at that point its gone wrong by assuring readers this is because they need to know about computational fluid dynamics and some other stuff. This is about the most common mistake people make about GCMs. Of course they do have a fluid dynamics equation core, but that bit you can almost lay to one side, unless it happens to be the bit that really interest you, because its a fairly well known problem. Some GCMs, after all, come from or are linked to Numerical Weather Prediction (NWP) models, and if they got the basic fluid dynamics wrong, the forecasts would be wrong, and they aren’t, provably (interestingly, Easterbrook and Johns say that the codebase-sharing between the UKMO’s NWP model and the Hadley Centre’s GCM is probably unique). But there are vast piles of other stuff, that doesn’t appear directly as CFD, and it this stuff – the parameterisations, like convective cloud, evapotranspiration, etc. etc. – that’ actually more interesting, because less certain.
The post then disappears down the usual septic rabbits holes about how no-one will give them grants, which is dull. The author then decides to demonstrate that he is clueless about the models and their discretisations by saying the programs literally cannot be made to run at a finer resolution without basically rewriting the whole thing, and any such rewrite would only make the problem at the poles worse — quadrature on a spherical surface using a rectilinear lat/long grid is long known to be enormously difficult and to give rise to artifacts and nearly uncontrollable error estimates. There’s lots wrong with this – for one, the spectral models (i.e., those that do their dynamics in spectral rather than grid space) don’t suffer from this problem. For another, Fourier filtering at the poles keeps HadAM3 quite happy, and if you increase the spatial resolution you have to increase the temporal resolution. You don’t need to re-write the model, which is just as well, as it wouldn’t help. I’m assuming this is about the atmospheric models; I’m given to understand that modern forward-looking ocean models tend to use a modified grid which in the northern hemisphere maps the pole onto Greenland, thereby avoiding the problem. The author seems terribly interested in dynamically adaptive grids(he gets more and more obsessed with this the further you descend into the comments), which GCMs don’t use because they are too much trouble and too expensive and wouldn’t help, but never mind.
I don’t think the bit about treating the models as independent is very interesting. I’m more interested in whether some of them are a bit rubbish, which is indeed true; the IPCC as it seems to me decided to treat all as equal because doing otherwise would have been too embarrassing for those dissed.
Looking at the comments: Roy Spencer is wrong and he’s wrong because he is lonely because he has no-one to talk to. Nick Stokes mostly makes sense, as ever. I stopped when someone linked to http://www.cs.toronto.edu/~sme/papers/2008/Easterbrook-Johns-2008.pdf because that was more interesting.
What is wrong with models?
Ha, well, anything I can say is seven years out of date, so treat any of this with caution.
My own feeling (having ended up working with sea ice models) was that it was possible to make an atmospheric model that was sane, and an oceanic one, and a sea ice one; but putting them all together and keeping the coupled composite sane was much harder (I’m not claiming any credit for having done this – I was living off the efforts of others). HadCM3 achieved this, at some level, and as I was leaving I think they were finally beating the (much more computationally expensive) HadGEM into something almost as good :-)
The other main thing “wrong” with models isn’t a thing wrong with models at all, but in the way they are used. The GCMs are most useful for exploring and understanding the climate, and its response to changes in, say, GHGs. But as a means for detailed predictions of the future, I’m less convinced. Part of the “problem” is the way GCMs naturally end up producing maps of change at whatever their native resolution is. But that doesn’t mean you can believe it at that level of detail, however pretty it might look.