Transient Coupled Model

From Unofficial BOINC Wiki

The contents of the following article relates to Climateprediction.net (CPDN) only.

[edit] General

A Transient Coupled Model is a mode of running a Global Climate Model (GCM) in which a period of time (in this case 1920-2080) is simulated with time-varying concentrations of greenhouse gases, anthropogenic sulphates, solar and volcanic forcings. The time-series of these climate forcings represent past historical forcings (plus associated uncertainties) for the period 1920-2000, and a small Ensemble of possible future forcings based on the IPCC A1B and B1 scenarios.

In climate change research, the use of such time-varying forcings allows us to examine both the warming due to different processes during the 20th century and the transient climate response: the actual path of warming and associated climate change we expect for a given scenario over the next 80 or so years. The transient climate response may be contrasted with equilibrium climate sensitivity in which GHG concentrations are doubled from pre-industrial values and the model then allowed to come into a new equilibrium (usually at higher temperature).

The time period indicates there is both a Hindcast and a Forecast period in the Climateprediction.net (CPDN) model.

Early Global Climate Moldels' used "swamp" or "slab" ocean models for reasons of computational simplicity. Climateprediction.net (CPDN) used Slab Models for its earlier climate sensitivity work. Such models cannot simulate the heat take up that occurs in the real ocean and were thus unable to run transient simulations: instead, the response to equilibrium change was investigated. There are problems with this approach, which were well appreciated at the time: the deep ocean has a very long time constant and in some areas (most notably the southern ocean) the transient and equilibrium responses are very different. In an equilibrium simulation in the Climateprediction.net (CPDN) project, time is merely a label and a given year or decade does not represent the simulation of a calendar year or decade. Thus a portion of a simulation labelled "1810-1825" represents 15 years, but not any particular 15 years.

In a Transient Coupled Model, instead of a sudden change in greenhouse gases and other Forcings, the Forcings are changed gradually, realistically for the Hindcast and either in an idealised way (1% CO2 increase, for example) or a more realistic fashion (one of the SRES scenarios) for the Forecast. Climateprediction.net (CPDN) uses the A1b and B1 SRES secenarios. If the simulation is partly of the past, observed greenhouse gases, sulphate and aerosol loadings and solar radiation levels will be used. The Transient Coupled Model is intended to be a physically plausible path for the climate system to follow. Although (given natural variability) even a perfect model would not simulate the year-to-year variations seen in the real world, in an ideal model the variation from decade to decade would track that of the real world.

The coupled part of the name indicates there is a dynamic ocean rather than a Slab Ocean. Including 'coupled' in the name is a little redundant as a transient model can only be done with a dynamic ocean. An alternative name would be Transient Climate Simulation and wikipedia uses this name (Wikipedia: Transient climate simulation).

Climateprediction.net (CPDN) uses the HadCM3L Model developed by the Hadley Centre part of the UK Met Office.

[edit] More about running the model

The Climateprediction.net (CPDN) Science Application places heavy demands on personal computers. This was demonstrated in Classic CPDN, in HadSM3 running on the BOINC Client Software and the Sulphur Cycle Model. The Transient Coupled Model continues the demand and adds significant new considerations.

Transient Coupled Model atmospheric timesteps are mainly the same as the sulphur timesteps and require around 70% more CPU Time as compared with HadSM3 Time Steps.
In addition to the atmosphere timesteps there are also ocean timesteps. Atmosphere timesteps represent half an hour of modeled time; ocean timesteps represent an hour. In the spinup there were 48 atmosphere timesteps followed by 24 ocean timesteps. (Not sure if this pattern will continue.) Update me. Ocean timestep process about 5 times faster than atmosphere timesteps so the extra ocean timesteps only add about 10% to the processing time.
160 years are modeled in the Transient Coupled Model. 80 Hindcast and 80 Forecast so this is a little over twice the number of modeled years. Total effect is that TCM is about 2.3 times longer than the Sulphur Cycle. That is about 6.6 times as long as a slab model.
For CPDN-dedicated machines:
- On an Athlon 64 3200+, in WinXP, running stand-alone, one Transient Coupled Model requires approximately Update me.(@24/7).
- On an Athlon 64 3400+, in WinXP, running stand-alone, one Transient Coupled Model requires approximately Update me.(@24/7).
- On an Athlon 64 3800+, in WinXP, running stand-alone, one Transient Coupled Model requires approximately Update me.(@24/7).
- On an Athlon 64 X2 3600+, in WinXP, running stand-alone, one Transient Coupled Model requires approximately 140 days. (@24/7).
- On an Athlon 64 X2 6000+, in WinXP, running two Transient Coupled Model in parallel require approximately 90 days. (@24/7).
- On a P4 3.0 in WinXP, running stand-alone, one Transient Coupled Model requires approximately 120 days. (@24/7).
- On a P4 3.0 in Linux, running stand-alone, one Transient Coupled Model requires approximately Update me. (@24/7).
- On a P4 3.0 in Linux, running two Transient Coupled Model in parallel require approximately Update me. (@24/7).
- On a P4 3.4 in Linux, running stand-alone, one Transient Coupled Model requires approximately Update me. (@24/7).
Transient Coupled Model are as computional demanding as SpinUp models.
Transient Coupled Model requires less disk space than Sulphur cycle version. Transient Coupled Model requires about 400MB to start and about 600MB Update me in total??? .
At the end of a successful Run, there are Update me zipped files consuming Update me of Disk space. With other models, unless off-loaded, large amounts of disk space per finished Model is added to the Hard Disk requirement for the next Run. With Transient Coupled Model there may be an option to have the model tidy up or this may happen automatically Update me.
Trickles send back more data and there are also uploads every 10 model years. So if something goes wrong at least the information up to the last 10 year interval has been uploaded so the whole of the processing does not get wasted.
Given the length of the WU, backups are even more important for the Transient Coupled Model . The BOINC directory should be backed-up periodically.
- Note: backups take less time because of decreased Folder size.
The application files consume +/- 35MB Update me of download bandwidth. Modem users be aware...
The deadline is set at 1 year Update me. Results will be accepted after this time.

This could cause some slower computers with a few projects to show as overcommitted and start crunching only the Transient Coupled Model until it is finished. If this happens your set resource share should still eventually be honoured but it may take a very long time to even out.