Which of the models is the most useful for the researchers here? Which gives the most contribution to the science? I would guess that it is not the slab model since it is not very realistic?

The slab models have turned out to be very useful.
Those being run now are for a different purpose to the original slabs. There\'s a research group in Germany that\'s <using them>/<going to use them> to test a neural network. I forget where the message from the project people was posted, possibly on the original beta site.

And there were some 4 phase models tested on the new beta site earlier this year. I don\'t know what has happened to these, but this item is what they\'re about.

And there\'s likely to be more research groups around the world that want to use them for whatever they\'re working on.

Of course, it depends on what you mean by \"here\", in

most useful for the researchers here?

And currently the most valuable type of model are the \'spinups\' currently being created on the new beta site. These will be used to start some new model types over the next few years.
Just as 61 spinups that were created a couple of years back are being used to start the Coupled Ocean models being run today.
You can see which of the spinups is used for a model in the parameter list of the model: Ocean Start Dump.

I finally found the post from the original beta site about the new slabs:

News

23 April, 2007
We have a new beta for running the \'old\' 3-phase slab experiments (hadsm3). Ben Sanderson (Oxford) Falk Niehörster (Free University of Berlin) are interested in \'optimising\' slab runs for a neural net prediction (Ben) and for cross comparison with the ECHAM5 German model (Falk). There will just be 1000 of these workunits put on the main site shortly (it\'s just reusing the version 4.14 hadsm3 which was our first BOINC job). In Ben\'s words:

We are releasing a small number of slab experiments to test a non-linear model emulator which has been built using the results of the slab ensemble. If successful, this will be used as an active parameter selection technique for future ensembles using HadGEM. The principle of the emulator is to make a prediction of the climates which would result from experiments that we have not yet run, including some theoretical experiments which do not use the discrete parameter values which have been used until now in climateprediction.net.

These 80 simulations should span the climate sensitivity range from 2 to 10K - and each run should be as close as possible to observations in its control stage. So in theory, we should find out the limit of how believable a high-sensitivity model can be...

If the test verifies the emulators ability to interpolate between known parameter values - it provides a very useful tool for further ensemble analysis. In the future, we could use the model emulator to find poorly-sampled regions of parameter space and then conduct more experiments in that region. This will enable to sample model behaviour more efficiently and make every run count.

If you look at the News thread posts of 21 Dec, 7 May and 1 June, you\'ll see that all three types of model are currently being used for research projects.

---

Cpdn news

There seem to be zero HADCM3 models available for download, yet over a quarter of a million SM3s. Is this the Preferred Crunch now, or will there be further CM3s later?

John

Problem solved. Thanks.

John