In October 2006 we stopped issuing our original model simulations for download, and began analysis of the simulations uploaded back to us so far. This analysis has lead to some very interesting results and implications for attribution of UK flooding to human-induced climate change -- so much so that we've added small extensions to this project!

Here's an outline of why we are doing this, and we hope you be able to help us complete a few more simulations..

Why we are extending this project (January 2007)

The main aim of the project is to investigate the change in risk (if any) of UK Autumn 2000-like flooding attributable to human-induced climate change over the 20th century (see Project Overview for more details). We do this by comparing the number of occurrence of such a flood event in climate model simulations of two different climates:

• An "Industrial Autumn 2000 (A2000)" climate that includes the influence of the human-induced climate change on the Autumn 2000 period.
• A hypothetical "Non-Industrial Autumn 2000 (NIA2000)" climate during the same period that had not been any such influence.

Then if, for example, there are significantly more occurrences of UK Autumn 2000-like flooding in the A2000 climate than in the NIA2000 climate, we can say that 20th century human-induced climate change has significantly increased the risk of such flooding.

We've now compared all completed simulations returned so far, and one of the very interesting findings we've made is that the frequency of occurrence of UK Autumn 2000-like flooding seems very dependent on the type of sea surface temperature (SST) patterns that we input into our model -- particularly for the NIA2000 climate.

In more detail: Our climate model is only of the Earth's atmosphere. So the effect of the ocean is input to our model via SST patterns. For the A2000 climate we simply input the SSTs patterns observed at the time which is relatively easy to do. But for the hypothetical NIA2000 climate we have to input estimates of SSTs patterns present had there not been any human-induced climate change, which is more difficult to do since we have never observed them in reality. So, as an alternative, we use estimates of these NIA2000 SSTs that have been derived from a range of independent coupled atmosphere-ocean climate models which are generally believed to simulate SST patterns reasonably well. This derivation process is complicated, but some technical details can be found here for those interested.

There were 3 such coupled models: HadCM3 (which is used in the main climateprediction.net project) and two American models. Hence we input 3 different NIA2000 SST patterns into our atmosphere model, that are derived from these 3 different coupled models. These have led to 3 differing results when counting the frequency of occurrence of UK Autumn 2000-like flooding in the NIA2000 climate, which are very intriguing if not entirely clear. We would like to investigate this further, so as to make our results clearer.

Specifically, we want to see what the effect of including an additional (fourth) NIA2000 SST pattern would have on the frequency of occurrence of UK Autumn 2000-like flooding in the NIA2000 climate. For those interested in the technical details, this pattern is derived from the coupled Japanese MIROC 3.2 model. We believe that including the effect of this pattern in our analysis would add to the weight of our scientific arguments when publishing papers on our results.

What we are doing in this extension (January 2007)

We already have a mechanism for including SST patterns as input files to the model -- they are contained in the initial package that participants download. In fact that's exactly what was done in our original downloads, which included input files of NIA2000 SSTs patterns derived from the 3 coupled models mentioned above. So all we will do swap the existing NIA2000 SST input files with our new Japanese-model derived ones.

So far, we have collected approximately 1500 completed NIA2000 simulations, comprising of 500 simulations for each of the 3 coupled model derived SSTs patterns used. Hence in this extension we are only planning to collect approximately 500 new completed simulations, and that's why its only a small extension.

To achieve this target we will, however, issue many more than 500 simulations in order to account for redundancy, failures, aborts etc (just like we did for the original simulations). In fact, there will probably be a few thousand. Also, the nature of the extension's aim means that all new simulations issued will only be of the NIA2000 climate.

The second extension (May 2007)

We've now collected and analysed most of the results from the first extension to this project, described above. This analysis has confirmed our impression that the frequency of occurrence of UK Autumn 2000-like flooding seems very dependent on the type of sea surface temperature (SST) patterns that we input into our model.

In fact the graph below summarises one of our recent analyses. It shows the Return Period for Autumn rainfall (i.e. how often that rainfall occurs; horizontal axis) according to all our completed model simulations for both Industrial and Non-industrial climates. Each circle is an individual simulation. The different colours are for the different climates modelled, with the smaller dots representing different types of Non-Industrial climates according to the four different types of SSTs used there (see above sections for details). We don't want to say just yet exactly which colour is for which climate, or give exact values for the rainfall (vertical axis) -- at least not before we publish!.

The main point we want to make, in the context of this extension, is that we have relatively few simulations that capture the really extreme events (although we are capturing them, which is more that previous studies have achieved). To be more specific, for the really extreme events that occur less than once every 100 years there are relatively few simulations that robustly capture this. I.e. the circles on the graph get fewer and fewer and more "wobbly" toward the right hand side -- especially for the smaller circles representing the representing different types of SST-dependent Non-Industrial climates. This is not unexpected since, by definition, these events are rare and so hard to capture. But we would have more confidence in assessing their rarity if we had a few hundred more simulations for each of these climates.

So in this second extension all we are doing is simply distributing more of the same type of simulations as we had in our original release, in order to firm up our results. In fact, we will concentrate on distributing only more of the original type of Non-industrial simulations since it's the robustness ("wobbliness") of the small circles we really want to firm up.

As before, and as explained below, this is just a short extension with a few thousand simulations sent out.

What we would like you to do

Simple:
Download some new simulations and start crunching just like before!
Or, if you are still running our previous simulations then keep going with those!

The new simulations should work just as before, taking both a similar amount of time to complete and resource requirement (since all we are doing is swapping old input files with new ones that are the same size).

As explained above, this is a short extension as we are only planning to distribute a few thousand simulations. We will keep the project going until a few weeks after all new simulations have been downloaded. We also stress that all previously completed simulations returned to us thus far will still be included in results published in journals, and the new simulations from these extensions will simply add to those results.

The need for such extensions to our project reflects the nature of active scientific research -- there's always more questions to be answered, and we hope you'll help us once again to answer them!