Experiment 2 will be starting in a few months, from the <a href="http://www.climateprediction.net/science/strategy_adv.php">description</a>

1. It seems that we will be using a hindcast between 1950 and 2000. Why not use 1954 - 2004, the data for 2004 is almost certainly more accurate and extensive than in the 1950s due to the presence of satellites. It would also cover the period of greatest CO2 concentration so should be better at sorting out those parameter sets that model the climate caused by CO2 rise.

2. experiment 1 was about finding good regions in the parameter space that give stable solutions. If I understand the reasons behind the cold equator problem then they should not occur in a coupled model. Most of the unstable parameter sets seem due to the cold equator problem (though other like <a href="http://climateapps2.oucs.ox.ac.uk/cpdnboinc/result.php?field=Temperature&amp;resultid=391576&amp;phase=AT#graph">this</a> one seem to have a mildly unstable phase 2). If this is the case how is experiment 1 used to cull regions of the parameter space?

3. Experiment 3 is a forcast for 2000-2050, why not 2000-2100? Most of the other input for IPCC AR4 seems to be over that longer period.

4. It seems that the different forcing scenarios give similar results for the first 30-50 years. This is because of the long time constant of the climate to CO2 change and the slow divergence of CO2 emission under the various scenarios. What do you expect to learn from the scenario ensemble given the short 2000-2050 timescale?

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

Interesting points.

1. I wonder when the content of that strategy page was first written. Could it have been nearer 2000 or 2001 rather than at launch in 2003?

2. If I have gathered correctly, only a few oceans will be spun up (takes a long time). They will then be paired with appropriate atmospheres. The different oceans will have different heat uptake rates. To get a realistic hindcast they will need to be paired with an atmosphere that has an appropriate climate sensitivity. Thus I do not think that a region of parameter space will be culled for all oceans. Instead each region is likely to be appropriate to be tried with at least one of the oceans but we avoid having to try each atmosphere with all of the oceans.

3 &amp; 4. Did you notice 2100 was mentioned in the table on the page you linked?

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

&gt; Interesting points.
&gt;
&gt; 3 &amp; 4. Did you notice 2100 was mentioned in the table on the page you
&gt; linked?
&gt;

No.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

&gt; Interesting points.
&gt;
&gt; 2. If I have gathered correctly, only a few oceans will be spun up (takes a
&gt; long time). They will then be paired with appropriate atmospheres. The
&gt; different oceans will have different heat uptake rates. To get a realistic
&gt; hindcast they will need to be paired with an atmosphere that has an
&gt; appropriate climate sensitivity. Thus I do not think that a region of
&gt; parameter space will be culled for all oceans. Instead each region is likely
&gt; to be appropriate to be tried with at least one of the oceans but we avoid
&gt; having to try each atmosphere with all of the oceans.

This is similar to my understanding, I assumed that each parameter set would be matched up with only a single ocean.

My question was because of this from the experiment 2 description:
<b>"The perturbed physics ensemble will comprise those models that are viable, stable climate models, that span the interesting regions of parameter space we have identified in experiment 1. We will be looking to include as many of these as possible, and will really only be throwing out those models which have gone unstable because of the choice of parameter sets. We imagine that there will be regions of parameter space (families of closesly related models with similar parameter settings) that are viable (it's theoretically possible that some will even be better than the standard model!), and regions that are not. We will find those regions in experiment 1 and use the good ones in experiment 2."</b>

How will the regions be found from experiment 1 as there seem to be large numbers of parameter sets that are sometimes (but not always) unstable?

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

&gt;How will the regions be found from experiment 1 as there seem to be large &gt;numbers of parameter sets that are sometimes (but not always) unstable?

It would take a lot more analysis of a lot more data than I have to provide a good indication of the proportion of models. However as a very rough stab from what little I know:

Perhaps 12% of models develop cold equators. Suppose this divides into 3% of models that are always cold equator and 18% of models that sometimes develop a cold equator.

Assuming these figures, is the teams problem relating to just the 3% or does it apply to the 21%?
I would hope that for the 18% when a cold equator sometime develops and sometimes doesn't, then the models where a cold equator does not develop would be representative of what happens in the coupled model. (If it isn't, I suspect this would also be a problem for the majority of models as well.) Therefore the problem seems to me to be largely limited to the 3%. I am only guessing but 3% sounds small enough to cope with to me.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

The problem, it seems to me, is identifying whether a parameter set is in the 3% or the 18% (or whatever the percentages really are).

If the workunit result contains only one member, then if it a cold equator then it is impossible to assign it to the 3% or the 18%. Only when there are two or more members with at least one cold and one cold equator can a result be unequivically assigned to the 18%.

Its hard to picture an N dimensional parameter space :-) but I think you can imagine a clear cube (3 dimensional) with some points in it (which represent cominations of the descrete parameter values along each of the axes). These points are either red (unstable) or green (stable). Hopefully the green points form a continuous volume in the centre with the old red point surounding the green, or perhaps down one side of the cube. Extend this to 29? dimensions :-)

Now extend this picture to the current experiment situation, the 18% where a cold equator sometimes develop should orange. Hopefully they should lay between the green and the red points. However as workunits with one result which should be orange are either stable (green) or unstable (red), so islands of green may be produced cut off from the main green parameter space volume by a curtain of red. In this case the red points between the green volumes should probably be marked orange. If I'm correct in my analysis then all green and orange parameter sets should go forward to experiment 2.

A fly in the ointment of the above analysis is that we have got results for only about 60,000 of the possible 3,000,000 parameter sets in the ensemble. Most of the points are still black.

So all this points to a strategy of choosing parameter sets that are close to the boundary between the green and red areas, as the goal of experiment 1 is to find the boundary. There is not much point running parameter sets that are very likely to be stable because they are inside the green volume. It would be interesting to know whether such a strategy has been followed.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.

&gt; A fly in the ointment of the above analysis is that we have got results for
&gt; only about 60,000 of the possible 3,000,000 parameter sets in the ensemble.
&gt; Most of the points are still black.
&gt;
So far we only have 13122 (3^8*2) models * 10 initial conditions. So we should on average have already completed 5 different sets of initial conditions for each model. I presume there will be a lot more possible models when the sulphur cycle is added. I am concerned that we have thououghly sampled most of the runs possible without the sulphur cycle but there will be little time to do many sulphur cycle runs before experiment 2 is launched. Also the sulphur cycle will be more demanding so not all computers will run it. Of course experiment 2 might be delayed more than the sulphur cycle.

&gt;Its hard to picture an N dimensional parameter space :-) but I think you can &gt;imagine a clear cube (3 dimensional) with some points in it (which represent &gt;cominations of the descrete parameter values along each of the axes). These &gt;points are either red (unstable) or green (stable). Hopefully the green &gt;points form a continuous volume in the centre with the old red point &gt;surounding the green, or perhaps down one side of the cube. Extend this to &gt;29? dimensions :-)

I see the red points tending to be in one (of many) corner and extending particularly along 3 sides and a few other sides as well. An orange curtain separates the red from green. 29D objects eh fun :-) (over 500 million corners) ~) (actually there are to be 21 parameters but we only have 9 so far)

&gt; So all this points to a strategy of choosing parameter sets that are close to
&gt; the boundary between the green and red areas, as the goal of experiment 1 is
&gt; to find the boundary. There is not much point running parameter sets that are
&gt; very likely to be stable because they are inside the green volume. It would be
&gt; interesting to know whether such a strategy has been followed.
&gt;
I am not sure about the 'not much point', I think we do want to see how they turn out in order to decide which ocean to match them up with. Maybe such a strategy can and will be used for the sulphur cycle.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

&gt; I am concerned that we have thououghly sampled most of
&gt; the runs possible without the sulphur cycle but there will be little time to
&gt; do many sulphur cycle runs before experiment 2 is launched. Also the sulphur
&gt; cycle will be more demanding so not all computers will run it. Of course
&gt; experiment 2 might be delayed more than the sulphur cycle.

Perhaps some machines could be left running suphur cycle models even after experiment 2 is launched. Slower machines (either slow CPU or only intermittently on) may take too long to complete experiment 2 (and almost certainly too long to complete experiment 3) to be useful running them.

---

____________________________<br>
<a href="http://www.boincforum.info/boinc/">boinc forum</a> and <a href="http://www.uk4cp.co.uk/">United Kindom</a> team, my climate change <a href="http://www.livejournal.com/users/mike_atkinson/">blog</a>.