The figure in the top panel opposite compares the distribution of Autumn 2000 precipitation totals for Industrial ("A2000") and Non-Industrial ("NIA2000") climate model simulations returned from the Beta Test.

Specifically, they are precipitation totals over the England & Wales region in the model (about 20 model gridsquares), and there are 21 Industrial (blue) versus 18 Non-Industrial (green) simulations compared.

The observed Autumn 2000 value of 502.7mm (Alexander & Jones, 2001) is also displayed as the red line, and the climatological mean value for 1961-1990 (a common base period for comparison) is also displayed at the right of the panel, along with the means of the respective model scenarios.

Immediately we can see that most of the simulations are not near to the observed value. This is to be expected, as the observed Autumn 2000 precipitation total was in itself a rare extreme event (the wettest autumn since records began in 1766), so that the majority of attempts at re-simulating a period like this may well fall below the observed - only a fraction of simulations may reach or exceed the observed total, and that's why we need so many!

Each and every simulation simulations, however, is still extremely useful because it's the comparison of the above fraction and how it changes between the Industrial Non-Industrial distributions that will be the most interesting result - and the ultimate aim of this project.

By obtaining large ensembles of simulations of the Industrial and Non-Industrial scenarios, we can better estimate their distribution and hence infer the "climates" ("average weather") of each of these scenarios.

For example, the A2000 simulations already seem to be exhibiting a tendency to be wetter than the NIA2000; the mean of the distribution is about 10.9% wetter, and there appears to be a skew towards higher precipitation totals. Both these means are also slightly lower than the observed 1961-1990 climatological average, suggesting that we may need to apply some bias correction to the model results.

Moreover, with a better estimate from more simulations, we can be more confident in our analyses when examining the upper tails of the distribution - i.e. the most extreme events.

At the moment, though, it's clear we still need a lot more simulations!

[The bottom panel is a normalized version of the top panel, so as to give the probability distribution, as opposed to the actual count.]