Project URL: http://climateprediction.net

Summary:

What is climateprediction.net?
Climateprediction.net is the largest experiment to try and produce a forecast of the climate in the 21st century.
To do this, we need people around the world to give us time on their computers - time when they have their computers switched on, but are not using them to their full capacity.
Read more about the experiment

Why?
Climate change, and our response to it, are issues of global importance, affecting food production, water resources, ecosystems, energy demand, insurance costs and much else.
There is a broad scientific consensus that the Earth will probably warm over the coming century; climateprediction.net should, for the first time, tell us what is most likely to happen.
Read more about climate science

What do we want you to do?
You can download a climate model from this website. It will run automatically as a background process on your computer whenever you switch your computer on.
It should not affect any other tasks you use your computer for.
As the model runs, you can watch the weather patterns on your, unique, version of the world evolve.
The results are sent back to us via the internet, and you will be able to see a summary of your results on this web site.
Climateprediction.net uses the same underlying software, BOINC, as many other distributed computing projects.
If you like, you can participate in more than one project at a time.
Go to Boinc / climateprediction.net download page

If you are in the climate research community and are interested in participating in the experiment in a research capacity, the research pages provide some basic background material.

CPDN FAQs
News and Announcements
News feed
'ReadMe' postings

---

The following is an excellent post from Crandles which describes the various types of climate model available from CPDN.

Science info for selection of tasks
"Crandles" wrote:

This thread deals with science information about the applications available though your preferences. If you are more interested in computing information (and the way the preferences work), see http://www.climateprediction.net/board/viewtopic.php?p=66108#66108

Science Information

General Why

Climate change, and our response to it, is an issue of global importance, affecting food production, water resources, ecosystems, energy demand, insurance costs and much else. There is a broad scientific consensus that the Earth will probably warm over the coming century; climateprediction.net should, for the first time, tell us what is most likely to happen.

The project is therefore aimed more at producing probabilistic estimates than a single best estimate model. To do this, the project needs to run lots of slightly different versions of possible future climates. It does this by varying Parameters of the model.

There is an index of much more science information on the CPDN website at Climate science explained including objectives, strategy etc. For more general information about climate science see the climate science README.

After this brief introduction this page will now look at the three applications available on your preferences page. (Pre 21 December 2007 there were only two options. The third model was previously available from the Seasonal attribution site only but has now been moved to here.):

The three models are:

1. Equilibrium Slab model (UK Met Office HADSM3) - 45 model years

2. Transient (or control) Coupled model (UK Met Office HADCM3) - 160 or 80 model years
3. Seasonal Attribution Model (UK Met Office HADAM3) - 1 year high resolution



1. Equilibrium Slab model (UK Met Office HADSM3) - 45 model years

Explanation of terms:

Follow links for meanings of Equilibrium and Slab.

HADSM3 stands for Hadley Centre (The climate division of the UK Met Office) Slab Model version 3. So this is the name of the model climateprediction.net are using. The Hadley centre is responsible for developing the model, climateprediction.net is just using it.

Why this type of Model?

Climateprediction.net wanted to start with an equilibrium slab model to:

1. Investigate climate sensitivity. This has led to a few papers including

Stainforth et al. Uncertainty in predictions of the climate response to rising levels of greenhouse gases, Nature, 433, pp.403-406, January 2005.

Piani et al. Constraints on climate change from a multi-thousand member ensemble of simulations, Geophysical Review Letters, 32, L23825, December 2005.

For a fuller list see http://www.climateprediction.net/science/scientific_papers.php

2. It would help gather information to reduce the computing time needed to set up the Transient coupled models.

Sufficient of these model runs for the above purposes had been received when the Sulphur Cycle Model was launched. Hence it was withdrawn (Except for classic which was kept open for the Open University course).

Why is it being re-released? / Number of runs needed

The initial release only set each parameter at one of three values. If the effect of changes in the parameters had only simple linear effects then it would be possible to interpolate for parameter values between the ones handed out easily. As expected, the relationships have been shown to be nonlinear. Ben Sanderson has developed a neural net to estimate these nonlinear interactions. They now want to run some more models using hopefully better parameter values as estimated by the neural net to see if these do produce better models and see how well the neural net manages to do at estimating the nonlinear relationships. If successful, this will be used as an active parameter selection technique for future ensembles using HadGEM (a higher resolution model).

Also Falk Niehörster (Free University of Berlin) is interested in doing comparisons of the hopefully optimised models with a different model (the German ECHAM5 model).

Just 1000 are needed.

Further details of the model

The model is split into 3 phases of 15 years each. These are a calibration phase to teach the slab ocean what values to take to get a realistic and steady climate in the absence of changing forcings. A control phase checks that the values obtained from the calibration phase do result in a steady climate. The final phase introduces a forcing of doubling the CO2 level.

The resolution of the model is 96 longitude by 73 latitude by 19 atmospheric layers.

More advanced information:
Hadley Centre Types of Climate Models
Hadley Centre QUMP project using HADSM3. This is a sister project to CPDN.




2. Transient (or control) Coupled model (UK Met Office HADCM3) - 160 or 80 model years

Explanation of terms:

Follow links for meanings of Transient and Coupled and control.

HadCM3 stands for
Hadley Centre (The climate division of the UK Met Office) Coupled Model version 3. So this is the name of the model climateprediction.net are using. The Hadley centre is responsible for developing the model, climateprediction.net is just using it.

You may get a transient model or you may get a control model. One is of no use without the other as the scientists want to see what the differences are. This is because coupled models take a very long time for the ocean to get into an equilibrium state. If this has not been fully done the model may drift. Comparing to a control model is a way of eliminating this drift and reduces the total computation effort required.

Why this type of Model?

After climateprediction.net had investigated climate sensitivity with an equilibrium slab model, they want to move on to exploring how climate would really evolve with time (The instantaneous doubling of CO2 carried out in the slab model is not realistic). This requires a transient model and more realistic forcings. It also requires a more realistic ocean that can change dynamically hence a coupled model was needed.

This model has stayed at the same resolution. For some impact studies a higher resolution model might be needed. However, halving the length of a cell takes 16 times as much computing time (2*2*2 for 3 spatial dimensions and the timesteps need to be half as long as well). Such high resolution models and high computing requirements are not needed for all purposes. Also, it may well be best to study impacts by a large number of one year models starting from data for 2040 and/or 2080 provided by one of these lower resolution models.

Number of runs needed

A few thousand are wanted to explore a limited set of forcing scenarios and parameters. However, it is easily possible to explore more scenarios and/or more parameters and/or more values for some of the parameters to add many more models that could usefully be run.

Further details of the model

The model starts in 1920 running a hindcast for 80 years. It then continues to run a forecast. The resolution of the model is 96 longitude by 73 latitude by 19 atmospheric layers. There are 20 ocean layers.

The 80 year model starts in 2000 so it does not do a hindcast. This allows different forcing scenarios to be used with model physics that have already been run to 2000. There is no point repeating work that has already been done.

More advanced information:
Hadley Centre Types of Climate Models


3. Seasonal Attribution Model (UK Met Office HADAM3) - 1 year high resolution

Draft info (ie we don't think this had expert review)

The main source for info on this model included here is
this attribution site thread post by Pardeep Pall.

Explanation of terms:

Follow links for meanings of Atmosphere Model.

HadAM3 stands for
Hadley Centre (The climate division of the UK Met Office) Atmosphere Model version 3. So this is the name of the model climateprediction.net are using. The Hadley centre is responsible for developing the model, climateprediction.net is just using it.

You may get an Industrial or Non-Industrial model. For the first experiment (investigating the UK autumn floods in 2000), The Industrial climate models the year 2000. The year starts in March well before the UK autumn floods. Slight changes in the initial conditions change the way the model develops and only rarely will an extreme event like the UK autumn floods appear in the models.

The Non-Industrial models are not 'Pre-Industrial' meaning what the climate would look like in 1750. The models are of the year 2000 but dehumanised. This means how it would be expected that the climate would be if we went back to 1900 and allowed the climate to develop for 100 years with non human influences only.

As there is uncertainty about what this Non-Industrial climate would be like there are a few (at least 5?) different sets of ocean temperature patterns used, based on simulations on many different types of climate model.

Subsequent experiments will be for different time periods and regions.

See also
http://attribution.cpdn.org/about.php and
http://attribution.cpdn.org/extension.php



Why this type of Model?

The aim is to determine the extent to which the risk of occurrence of extreme weather events is attributable to human-induced climate change. The first analysis is directed at the flooding in the UK in Autumn 2000.

The atmosphere has a high spatial resolution. This is necessary when looking at impacts as larger cells may well average out detail that is important. This is probably even more important when looking at extreme events and also when looking at precipitation rather than temperature.

There is a slab ocean which is not allowed to vary from provided sea surface temperatures provided. As we are only modelling one year ocean responses are not wanted. This also makes it easier to specify sea surface temperatures and get the model to react sensibly and not drift away from these specified temperatures.


Number of runs needed

The aim is to complete around 10,000 climate model simulations of each of the Autumn 2000 climates. There is one Industrial and at least 5? such Non-Industrial climates. (Perhaps more can usefully be added as has previously happened with the Jan 07 Extension.) The industrial (A2000) models are identified by _a_ in the model name, whereas the nonindustrial (NIA2000) models have an _n_ in the model name.

Further details of the model

The first experiment's model starts in March 00 and runs for 1 model year. It has a horizontal resolution of approximately 100km2 at mid-latitudes enabling it to reasonably capture the storms and weather patterns associated with the Autumn 2000 UK flooding. There is a single slab ocean layer. This just maintains the sea surface temperatures it has been told to maintain in the same way as phase 1 of the slab model.

The sea surface temperatures have been found to be important to extreme precipitation in the climate model, and this has lead to extra model (such as the Japanese MIROC 3.2) being used to generate different sea surface temperature patterns to further test this dependance.

This first experiment is most interested in certain regions of the globe (Western Europe, India, Southern Africa, and Northwestern USA), so when the climate is uploaded at the end of the model, the climate for these areas is uploaded to the server in greater detail).

ReadMe: Running the model
ReadMe: Climate Science
ReadMe: Crashes and other problems
ReadMe: Backup and restore
ReadMe: Community

---

More on the hi-resolution climate model available from Climateprediction.net (cpdn).

Tolu ( Chief Software Architect) says

The HADAM seasonal models now being sent out to CPDN members are a new type.

This new experiment supposes that in 1999 we had suddenly found a very, very cheap and fast way of removing all of the anthropogenic greenhouse gases from the atmosphere and that this method was adopted in early 2000. Because the ocean responds slowly, we can suppose that this did not affect the ocean over the 2000-2001 period of Pardeep's simulations. But what affect would it have had on the atmosphere? We are expecting to find that the land would have cooled nicely (away from the coasts anyway), which seems ideal. But we also think that the warm ocean/cool land contrast would have produced more severe storms and rainfall events, which is not so ideal.

These new HADAM models are named hadam3h_c_*. The previous workunits were called hadam3h_n_* and hadam3h_a_*.

"Pardeep's simulations" refer to the Seasonal Attribution Project, which finished last year. (Although it's still slowly collecting data from a few hundred computers.)

The server only sends HADAM models to computers with at least 1.5Gb RAM. However, that might not be enough for computers with Vista, and dual or quad cores also need more.

In Spring 2007 Carl, who was then CPDN's chief programmer, outlined new CPDN projects that were at the proposal or planning stage. I've added comments in green to show what has been implemented since then. Carl said

'Here is an update on developments for modelling done via CPDN, in the order they will happen over the next two years. Some are at the "proposal" stage so may not come to anything if the reviewers & funding bodies don't fund them. I'll put a (P) for those:

1) optimized (CPU & file I/O) HadCM3L model to come out of beta testing soon (1-2 months). Implemented Summer 2007

2) 80-year workunits to redo some of the more interesting results of the BBC experiment with different forcings (1920-2000 and 2000-2080). These workunits would use (1) and probably happen in 1-3 months' time. Implemented 2007

3) I think we'll be doing another experiment or two with the HadAM3 model (9-times higher resolution model from the Seasonal Attribution Experiment).
New experiment using HadAM3 models launched on CPDN in March 2008

4) regional model (UK MetOffice PRECIS Programme HadRM3 model, see http://www.precis.org.uk) embedded with the HadCM3L global model (which generates the boundary conditions for the region). We'll probably do regions of southern Africa and US Pacific Northwest and work with scientists from those areas ((P) -- Microsoft is hopefully going to fund this. I'm going to Seattle with Richard Jones of the MetOffice for meetings next week).

I find this project very exciting (and I hope Microsoft does :-) as it will allow scientists from countries that don't have a lot of computer resources to take advantage of setting up experiments with CPDN, having the huge CPU power thanks to the participants, and then being able to use data they would otherwise never have been able to get.

Spinup HadCM3L global models currently being run by CPDN beta2 members in preparation for this new work. Funding received.

5) high-resolution global modelling, with the latest generation MetOffice model (HadGAM1). This is at 4 times the resolution (i.e. "half" the hadcm3 resolution, but not quite as high as HadAM3)

This experiment is led by William Ingram of the MetOffice (and Oxford too now) to do an in-depth study of cloud feedbacks within the model (i.e. the positive & negative feedbacks of clouds on the climate etc). The modelling code is quite a bit different, and we are keeping things in 64-bit for the first time, so it could take a gigabyte of RAM to run, as well as a 64-bit PC!

Not yet implemented

6) HadCM3 but use a higher resolution ocean with hopefully no flux corrections (P) Not yet implemented.

7) in addition to the modelling, a big part of the project work will be getting the results online for scientists'

Implemented Summer 2007 in Milo's Results website

There is a discussion thread for this topic here.

Peter Darch is a doctoral student based in the Computing Laboratory at Oxford University. He's doing research on how scientists, computer experts and members of the public perceive each other and work together in volunteer computing projects, and has been given permission to use climateprediction.net as a case study for his research.

He's trying to get in touch with as many climateprediction.net volunteers as possible to share their perspectives and experiences. If you would like to help out, he has a questionnaire at the following address:

http://users.ox.ac.uk/~lina1211/

It consists of several open-ended questions and you're free to make your answers as short or long as you wish. Your anonymity will be completed respected and maintained throughout.

---

"The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer

CPDN main project

Tolu has released a new version 5.06 of the HADSM model for Intel Macs to fix problems. It's here on the CPDN Applications page. Your comments on the new version will be welcome for example in this thread.

Reminder: All CPDN project forums except BBC have private messaging. You will notice private messages you have received more quickly if you enable email notification in the project section of your accounts.

CPDN Beta2 project needs new members!

CPDN Beta2 project members are crunching 160-year spinup models of the HADCM type. They run very stably at about the same speed as HADCM models from the main CPDN project. The minimum specifications are listed here, though more than 256MB RAM would be preferable, and at least 1GB if the computer has Vista. A fairly fast computer running 24/7 could complete a spinup in about 3½ months. But there's no race to complete them. The first to finish will be used first and others will be used later. Spinups are available for Windows, Linux and Intel Mac.

CPDN Beta2 members have also started testing the first BOINC v.6-compliant version of HADSM3 slab models. HADAM3 and HADCM3 models will also need to be tested for BOINC6 compliancy very soon. The first BOINC6-compliant models are not yet running stably. Testers of these models will need to upgrade to BOINC6 and be prepared for model crashes, or to abort them if Tolu says so.

Tolu later expects to test sulphur slab models for a new experiment. So there's a lot of testing to be done.

Beta2 members can select the model types they want to run in their accounts.

Beta2 crunching earns credits at about the same rate as on the other CPDN projects. Beta2 credits will be transferred to our CPDN main project accounts and aggregated with our CPDN credits. This credit transfer has not yet been properly implemented. At present it only works erratically for some testers, so new members will need to be very patient. It may be some time before their testing credits appear on the BOINC stats sites.

Beta2 testers should refer to the CPDN README advice collections here, back up their BOINC folder contents regularly, read the Beta2 forum for news and developments, and post about problems on the Beta2 forum. New Beta2 testers should have previous experience of running CPDN models from the main project, BBC or SAP.

New members need to attach to Beta2 as a separate project through the BOINC manager Tools menu, enter http://cpdnbeta.oerc.ox.ac.uk/ when prompted, then join as a new member using the same email address as for CPDN.

If you wish to discuss joining Beta2 there's a thread here.

CPDN and original Beta project

The original CPDN Beta project has been closed for many months. But on Friday 27 June some of its former crunchers had all their original Beta credits transferred to CPDN. This is a mistake as these credits will all have been transferred before. So when Tolu corrects the error these extra credits will be taken away.

CPDN main project

On Sat 28 June the CPDN credits export to the stats sites did not take place and may not today. Milo will check this on Monday.

At the moment there are no HADCM models available for download. If you need a new model and cannot download one, you can either wait until Milo fills the HADCM work queue again or select HADSM/HADAM/both in the project preferences of your account.

New CPDN Moderators

Ananas and Gerald Davison have been appointed moderators on the CPDN BOINC message boards and phpBB forum.

---

"The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer

Thyme Lawn has updated the CPDN moderator list here.

James French who's a Geography student at Oxford University is collaborating with the Environmental Change Institute for his dissertation. You can help him by responding to his survey.

Peter Darch of Oxford University would still welcome extra responses to his survey.

You don't need to be registered on the CPDN independent forum to access the survey links. Peter's survey is for CPDN members but everyone is invited to respond to James's.

BOINC 6.2.14 has been released to the public. Congratulations to Rom, the Berkeley team and all the alpha testers! There are CPDN forum threads with more details and further links here and here. As this is a major new version it's important for CPDN members to read Mike's post and Jorden's FAQ before deciding whether and how to upgrade.

CPDN main project

The CPDN server program that exports credits to the stats sites is not working properly. I will not make repeated news posts about this problem which may continue until Milo, who knows about the situation, has time to fix it. CPDN and CPDN Beta2 are giving us our credits correctly and we will eventually receive them on the stats sites. Many thanks to all members for your patience.

On Monday 4 Aug we received (on the stats sites) our credits for Friday, Saturday and Sunday. On Tuesday 5 Aug there was no CPDN credits export.

SAP

There have also been long delays in the exports of SAP credits to the stats sites.

CPDN Beta

There's a big change of IP address in progress affecting OeRC which is the CPDN Beta address. This change is scheduled to take place at about 10.00 UTC on Friday 22 August. There may be some temporary outages as systems are changed over.

Ananas says

Under certain circumstances (not on all systems!), BOINC seems to cache the IPs forever.

So for some of us BOINC might require a restart after the changeover. Usually a PC restart is not necessary.

Apologies for the late notification of this event which Milo hopes will cause very little disruption.

Tolu has been very busy. He has released new BOINC6-compliant versions of the following CPDN models:

HADSM3 slab for Windows, Linux and Mac
HADAM3 for Windows and Linux

The BOINC6-compliant HADCM3 coupled models are still being beta-tested.

BOINC6-compliant means that when we upgrade to BOINC version 6, whether we install in 'protected' (service) mode or in 'unprotected' (screensaver) mode, we will be able to see the graphics and screensaver of these models.

While we finish our current BOINC5-compliant models, we can only see their graphics if we have BOINC version 5 or install BOINC6 in unprotected screensaver mode.

All currently available CPDN model version can be seen on the applications page. Versions numbered 5.** are not BOINC6-compliant, ie their graphics can only be seen in BOINC6 screensaver mode. Versions numbered 6.** are BOINC6-compliant ie their graphics are viewable in both BOINC6 installation modes.

The BOINC6-compliant model versions are backwards-compatible; this means that if we decide to keep BOINC version 5 we will be able to see the graphics of these new model versions. So we are not yet obliged to upgrade to BOINC6.

Information about BOINC6

Further information about BOINC6 can be found in the BOINC FAQs. The most useful FAQ items for BOINC6 are The Big BOINC Answer thread, and here (to read after installation) and here. There is also a preview of the BOINC6 protected/service installation process here. If you wish to install in screensaver mode ie not as a service, the preview is also useful.

Many of these FAQs are also available in German, Dutch, Czech, French and Spanish by clicking on the flags at the top of the page.

Please note that BOINC6 installs itself in two folders in two different locations, one for BOINC itself and the other for Data, which includes all project and model/task data. The Big BOINC Answer Thread lists the default BOINC6 locations for Windows, Linux and Mac computers.

BOINC6 Backups

Before upgrading to BOINC6, make a backup of the complete contents of your Boinc folder! Here are the READMEs about backups.

After upgrading to BOINC6, when we back up our models, we only back up the complete contents of the Data folder, not the BOINC folder. The backup methods described in the READMEs also work for backing up this new Data folder. A Data folder backup must be restored to the Data folder in the same location as before.

If you make a backup when you have BOINC5 and need to restore a crashed model after upgrading to BOINC6, please ask for advice on the forum. To avoid this situation it is a good idea to back up the BOINC6 Data folder soon after your BOINC upgrade.

Please complete your current models before downloading new ones!

New experiment for CPDN!

Tolu has released BOINC6-compliant versions of HADSM3MH for Windows, Linux and Mac. The models are for a new research project to study the mid-Holocene climate. They are 4-phase 60-year slab models with one phase for mid-Holocene conditions. You can select these models in the CPDN preferences section of your acccount. They are 33% longer than the current HADSM slab models which are still available.

The researchers for this project are Helene Muri, Myles Allen and Hiro Yamazaki of the Dept of Atmospheric Physics in Oxford, Gideon Henderson from the Dept of Earth Sciences in Oxford and Paul Valdes from the Dept of Geographical Sciences at the University of Bristol.

We will add links to threads about the mid-Holocene models as more information becomes available.

Please see previous News posts on the CPDN forums about precautions to take with all slab models particularly at phase changes.

Crandles has posted in the News thread of the CPDN independent forum:

The CPDN team have published another paper entitled

Constraints on Model Response to Greenhouse Gas Forcing and the Role of Subgrid-Scale Processes

in the Journal of Climate.

The abstract is posted in this thread which would also be a good place for any discussion.

BOINC6 versions

Many computers with Windows and BOINC versions 6.2.14 and 6.2.16 are crashing models immediately after download with error code -185 and the message 'Can't get shared memory segment name: shmget() failed'. This problem is happening with HADAM, HADSM and HADCM models (and also with tasks from some other BOINC projects). You can see the code and messages for finished models on their web page; click + beside 'stderr out' to see these extra details.

Check your version of BOINC by right-clicking on the BOINC icon in the notification area, bottom right of the screen, and selecting About Boinc Manager.

Members with these earlier BOINC6 versions should upgrade as soon as possible to v. 6.2.18 which contains a fix. Download it from here. Please see previous posts in this news thread for links to information about BOINC6.

BOINC5 versions are not affected by this problem.

From the CPDN index page news, 'We thought our participants might be amused to see Myles Allen in action in a debate against Bjørn Lomborg, the well-known critic of mainstream opinion on climate change, at last week's Swiss Climate Forum in Thun, Switzerland. Myles would like to make clear the debate format pushed him into being unscientifically combative...

If any participants were at the forum, we'd be happy to hear from you.'

HADCM v.6.02 on CPDN main project

We have discovered a bug in these models which were available for Windows, Linux and Mac between 29 Sep and 22 Oct 2008. They have now been replaced by HADCM 6.04 in which the defect appears to be fixed. Version 6.02 will produce valid data for research on the Geoengineering Experiment, but these models' files will become abnormally large, up to 6.5GB by the time they complete. There is a risk that they will use all the disk space allocated to BOINC and crash.

Please see this thread for details about how you can identify and fix the problem. Apologies to members for the inconvenience caused.