The instructions post about the CPDN HADCM 6.02 model with the bug has been edited. If you have already deleted .nc files for your model, please read the modified red section of the instructions.

On the CPDN website index page there's a link to download a lecture by Myles Allen, CPDN's principal researcher.

The RSS news feed from the index page was broken but should now be working properly.

Some links of interest:

Hiro Yamazaki of the CPDN research team has given details in the Climate Science section of the website about the Millennium Experiment.

Real Climate recently posted a comprehensive FAQ on climate models (of all types, not only the Met Office Unified Model that CPDN uses).

In March 2007 the organisation Sense about Science held a very informative symposium about climate change at St John's College, Oxford. Milo demonstrated CPDN models there. The talks have been made available for download here. (The files are rather large!)

Dave Frame and his colleagues at CPDN have published a paper titled 'The climateprediction.net BBC climate change experiment: design of the coupled model ensemble' in the Philosophical Transactions of the Royal Society which is in London.

http://journals.royalsociety.org/content/w214633520523134/fulltext.pdf

These 160- and 80-year models were run initially by BBC participants and later also by CPDN members. Other types of CPDN model are also discussed in the paper. Most of the authors of the paper can be seen here. The last reference in the paper shows more research by the CPDN team in progress.

The paper is a 16-page pdf download, not too large for most of us.

A very small number of brave and loyal BBC members are still uploading trickles from their Climate Change Experiment models more than two years after starting them. The top BBC project participants can be seen here.

A CPDN credit-calculating script hasn't been working properly. Milo's already isolated one problem. There may also be delays in the transfer of Beta members' credits to our CPDN accounts. So for the next few days our recent CPDN credits will be temporarily missing or delayed on the stats sites.

Milo has updated most of the non-BOINC pages on the CPDN website to use a new system based on the Drupal content management system.

This will make it much easier to keep the site up to date, add new information about experiments, and so on. Inevitably there will be some bugs, broken links etc., but finding and fixing these is Milo's top priority. Please note that the BOINC pages should be unaffected, other than a colour and logo change to match the new pages.

If you discover any problems please post the details on the phpBB forum (separate registration required) here.

---

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

CPDN all projects

We have now completed over 30,000 HadAM and 21,000 HadSMMH models. Our crunching statistics can be seen on the CPDN home page here.

At the beginning of March we reached the impressive total of 300,000 completed HadSM models. These models have been used for several research projects. Ben Sanderson, who has now left Oxford, analysed them using neural nets. Dan Rowlands, who can be seen here on the research team's new web page, is building on Ben's work by using emulators (techniques to perform regression) to fit a surface to the slab model response in parameter space.

Dan says 'While Ben used neural nets, I am using a technique called Random Forests as an emulator, which benefits from requiring very little tuning to get good results. It also provides a measure of uncertainty in each prediction. I am using the slab models as a test bed for this technique, and so have been sending out some verification runs on CPDN - i.e. train the emulator on this part of the current CPDN ensemble, then use the emulator to predict the output for new parameter combinations then distribute these models on CPDN and compare the actual model response to what was predicted. I distributed around 10,000 models (each with an initial condition ensemble of 4 members) based on a continuous sampling of parameter space (rather than the discrete sampling used in the original ensemble), and am currently analysing the results. I plan to compare different emulation techniques, namely neural nets v random forest v gaussian process emulators and highlight the strengths and weaknesses of each.'

Random Forests

Reminder: You may need to edit your model selection in the climateprediction section of account to tell the server whether you want to run the new short HadAM3P models. If you run too many of these side-by-side on a multi-core computer they will slow each other down, so let them run alongside other model types.

The current issue of Nature is about climate change. It includes two letters, Greenhouse-gas emission targets for limiting global warming to 2 °C and Warming caused by cumulative carbon emissions towards the trillionth tonne, which CPDN researchers contributed to.

CPDN main project

Please see the CPDN forum News threads for detailed updates about the recent server problems affecting uploads of HadAM3P zip files. The News threads are here and here. We advise members to enable emails from the project in their accounts and to subscribe to one of those threads.

CPDN main project

There's good news for Mac users. HadAM3P ran much more slowly on Mac than on Windows and Linux. Tolu has released a new 6.07 version of this model for Mac. It uses a different compiler and should run faster than version 6.06. It will still be helpful to CPDN if members can complete their current HadAM3P models before downloading new ones as they all produce good data for the researchers.

Thanks to Billy and Zombie67 for their reports on the old and new versions and to Tolu for improving the compiler for this model.

BBC Climate Change Experiment

The project is being shut down. All models which are still running will be sent a killer trickle next time they send a message to the scheduler and you should not attempt to restore from backup. The scheduler and upload servers will be disabled at the end of this week and you will not be able to return any more trickles or upload files after that happens.

The results which have been returned will be used as part of a submission to the IPCC AR5 (Intergovernmental Panel on Climate Change Fifth Assessment Report). Carl Christensen is working part-time for CPDN on this and is adapting the model to collect some further data for the submission. This will be released on the main CPDN project.

Any BBC CCE users who have yet to sign up for the main project should follow the instructions here

CPDN main project

You may have noticed that Carl is back working for CPDN in Oxford with Tolu and Milo. He is still working part-time for the Quake Catcher Network Boinc project at Stanford University. The extra manpower is long overdue and very welcome.

He has been optimising the credit scripts which were taking up to 8 hours to run each day and causing serious database slowdowns which you will have noticed when you attempted to access the CPDN-Boinc forum and your model web pages or edit your account preferences at certain times of day. This big job has coincided with the installation of the new server and the closing of the BBC project.

Carl thinks he will have the credit scripts running again 'early next week'. Our credits should begin to appear on the stats sites a day after that depending on how quickly the new scripts process the acccumulated backlog.

SAP

Any unfinished Seasonal Attribution Project models should be completed as quickly as possible. The moderators would not be surprised if this project is also closed soon to optimise the use of server resources.

The online magazine International Science Grid This Week has published an article called How green is my grid? which mentions the role of CPDN.

CPDN credits

Carl rewrote the CPDN credit scripts as part of the recent big database and server upgrade. His new scripts run much faster than the old ones. There are now, however, some credit anomalies.

* There is an inexplicable loss of a small % of our total credits. Different percentages for different members but in no case very large. This is the credit loss showing today on the stats sites. There are several possible causes. For example, Carl has archived a lot of our old data because the database had become gigantic; this archiving may have resulted in the loss of some credits. Carl has been working on the server upgrade since Sunday and does not intend to spend more time looking for these lost credits!

* The Boinc cross-project comparison table shows that CPDN has been awarding credits at a lower rate than most other projects including Seti. CPDN also has a 'reference computer' used to compare credit given by Seti and CPDN. Our programmers can calculate credit comparisons per processing hour on the same computer.

Why compare with Seti? Because 1) it's the original Boinc project 2) when CPDN migrated to Boinc our credits were set at levels similar to Seti 3) the Seti programmers have taken many steps to avoid inflation or deflation of the value of their credits.

So Carl has decided to increase the credits awarded for all models by 5%. This increase has already been implemented retrospectively, which should compensate for the inexplicable loss. (So today's negative figures on the stats sites should be reversed tomorrow or on Saturday.) And in future we'll earn 5% more than before.

* Credits for each completed model type have been increased as follows:

HadAM3 was 5,184.00, now 5,443.20, +5%
HadAM3P was 1,982.64, now 2,081.77, +5%
HadCM3 (80 year) was 24,883.20, now 49,766.40, +5%
HadCM3 (160 year) was 49,766.40, now 52,254.72, +5%
HadSM3 was 6,805.26, now 7,145.52, +5%
HadSM3MH was 9,073.68, now 9,527.36, +5%

Credits per trickle will of course also increase by 5%.

The 4 degrees climate conference is taking place this week in Oxford. Dr Myles Allen, ClimatePrediction's chief researcher, will be speaking on Wednesday. The presentations about the impacts of climate change should be reported in many press articles during the next few days.

Our models have contributed to this research.

CPDN's chief scientist, Myles Allen, was mentioned in this Financial Times article.

Myles, Dave Frame also of Oxford University and Charles Mason of the University of Wyoming have published an article called The case for mandatory sequestration in Nature.

You may wish to see how much CO2 we've already emitted on the Oxford University website TrillionthTonne.org.

David Archer who lectures in climate and global warming at the University of Chicago has posted about his current course on the RealClimate website. We can download and watch the course, called Global Warming: Understanding the Forecast. The lectures are very clear but do include some physics and chemistry. One needs Quick Time player. (Not for members with dialup.)

Several publications by the ClimatePrediction researchers have been added to the Publications page.

Knutti, R., G.A. Meehl, M.R. Allen and D. A. Stainforth, 2006: Constraining climate sensitivity from the seasonal cycle in surface temperature, Journal of Climate, 19 (17), 4224-4233. A 10-page pdf.

Knutti, R., S.Krähenmann, D. J. Frame and M. R. Allen, 2008: Comment on 'Heat capacity, time constant and sensitivity of Earth's climate system' by S. E. Schwartz, JGR, 113, D15103, doi:10.1029/2007JD009473. A 6-page pdf.

Sanderson, B., R. Knutti, T. Aina, C. Christensen, N. Faull, D. J. Frame, W. J. Ingram, C. Piani, D. A. Stainforth, D. A. Stone and M. R. Allen, 2008: Constraints on model response to greenhouse gas forcing and the role of subgrid scale processes, Journal of Climate, 21, 2384-2400. A 17-page pdf.

H. Fowler, D. Cooley, S. Sain and M. Thurston: Detecting change in UK extreme precipitation using results from the climateprediction.net BBC climate change experiment, 2010, Extremes DOI 10.1007/s10687-010-0101-y.

Millennium Project

CPDN has launched new models called FAMOUS which will contribute to the European multidisciplinary experiment involving 16 countries and many universities. The Millennium website explains how the climate of the last millennium will be analysed using climate models and proxy data. The CPDN modelling contribution is described here and its lead researcher is Dr Hiro Yamazaki at Oxford University.

The very long FAMOUS models are distributed in 200-year segments each of which at the end generates a restart dump from which the next 200-year period can begin. They are available on all platforms. At the moment the models run very fast, taking less than 130 hours on a C2D 6600. A slower version is, however, being beta-tested in order to improve model stability.

A small proportion of these models do crash and should not be restored from backup. Members will as usual receive credit for the trickles they upload at the end of each model year.

FAMOUS can only run on computers with SSE2 capability. SSE is not sufficient. You can check whether your processor has SSE2 by looking at your Boinc manager messages about 8 lines from the top. This also applies to HadAM3P models. Computers without SSE2 should select HadSM models.

As well as the globe graphics, Tolu is developing sophisticated graphs which will later become available on models' web pages.

If you wish to try FAMOUS you can select it in the ClimatePrediction section of your CPDN account.

Members who run these models should follow the News thread of either the CPDN independent forum (where separate registration is required) or the CPDN-Boinc forum in the Number Crunching section. That is where further detailed announcements will be posted.

FAMOUS v.6.10 MILLENNIUM models

• Available on all platforms.
  
• You may need to select it in the climateprediction.net preferences of your account.
  
• FAMOUS takes about 9 days crunching 24/7 on a fairly fast computer.
  
• 200 model years.
  
• These 200-year segments form part of models lasting 1600 years. After completion the segments are stitched together.
  
• The first 200 years, 599 - 799, are a spinup or settling-down period which allows climatic conditions to stabilise.
  
• The deadline for these models is three months which is much shorter than for other model types. This is because CPDN needs the results back as quickly as possible to stitch the segments together.
  
• Files are uploaded every five model years ie 20 files in total.
  
• Trickles are uploaded every model year ie 200 trickles.
  
• As v.6.10 processes more slowly than the previous CPDN version, credits have been increased to maintain parity with other CPDN models. (If you crunched the earlier version models you will also receive the increased credits for them.)
  
• At the moment CPDN FAMOUS displays the traditional graphics as for other model types. New graphics are being developed in conjunction with Sony Labs in Paris but are still being beta-tested. They will be released on CPDN in a future FAMOUS version.
  
• Each FAMOUS task web page, reached from our computer's task list in our account, displays new interactive graphs. They include clickable red dots for volcanic eruptions. Here is a historical list of Large Volcanic Eruptions.
  The graphs display in most browsers but not in Internet Explorer 7 (Windows). You can upgrade to IE 8 at no cost. If you have XP please upgrade to IE 8 in case you are not able to when Windows support for XP stops soon.
  
• Some FAMOUS models crash, typically with exit code 22 and messages including 'INVALID THETA' or 'NEGATIVE PRESSURE', because of instabilities that develop with some combinations of parameter values. Please do not restore backups of models and rerun them after model instability crashes. On the same platform with the same operating system the models would crash again at the same point. (Reminder: if your models crash with other exit codes eg -186 or -226 you should post on the forum for advice.)
  
• If your multicore computer is also running other model types you may still wish to make backups to help ensure completion of the other models.
  
• FAMOUS does not use enormous amounts of memory. Most multicore computers should be able to load all cores with FAMOUS and run them simultaneously. (Reminder: on a multicore computer with 4 or more cores don't fully load with HadAM3P as they will slow each other down, in a few cases disastrously.)
  
• A future FAMOUS version is expected to be hyperthreaded.
  
• You can read about the Millennium Experiment led by Dr Hiro Yamazaki at Oxford University here. Here is the website of The Millennium Project which is pan-European and multidisciplinary.
  

Thank you to all members who are helping to advance this large research project.