Supercomputer Stokes soaks up statistics to predict the weather

The Met Éireann weather forecast sounds simple, but it is the result of extremely complex number crunching and modelling done on Ireland's most powerful computer, reports KARLIN LILLINGTON

NEXT TIME you check the weather forecast, take a moment to give special thanks to a supercomputer.

For although the brief daily prediction from Met Éireann that most of us catch on radio or television sounds simple, it is the result of extremely complex number crunching and modelling done on Ireland’s most powerful computer, a mammoth named Stokes.

That daily forecast from Stokes – or rather, four of them, as a full analysis is done every six hours – is the result of a collaboration between Met Éireann and ICHEC, the Irish Centre for High-End Computing, that began in 2007 and will officially run till 2011.

ICHEC computer scientists work with Met Éireann staff to fine-tune the complicated software needed to produce forecasts, and to advance research on climate change in Ireland to produce data for a European initiative on climate change called EC-Earth.

Like previous supercomputers operated by ICHEC, Stokes is named for a famed Irish mathematician, in this case, Sir George Gabriel Stokes (its predecessor, Walton – named for Trinity mathematician William Walton – was retired in December). Launched officially this week, Stokes – a Silicon Graphics Altix ICE 8200EX computer – has an extraordinary 2,560 “cores”, or processors (most home PCs have one or two processors), and 5,120GB of memory (RAM).

The reason weather prediction is so demanding of processor power is that weather – or rather, the numerous individual elements that come into play and cause a storm front to build, or sunny weather to stabilise – is very complex, says Ray McGrath, head of Met Éireann’s research and applications division.

Different software modules that are part of huge weather prediction applications each map different weather variables, such as wind speed, precipitation, atmospheric pressure and wave development. These are all then pulled together to model likely weather scenarios. Finally, those are analysed by the experienced meteorologists at Met Éireann to produce a forecast that is based on a combination of computing and human brainpower.

To analyse all those separate possibilities at once requires thousands of computer processors working separately but at the same time, in a process called parallel processing. As “massively parallel” supercomputers like Stokes have grown larger, cheaper and more efficient, software that can harness that power to give more accurate weather predictions has developed in tandem, says Prof Jim Slevin, director of ICHEC.

This reflects scientists’ growing understanding of how weather patterns emerge, which is also a result of having the computing power to understand how small changes – in wind speed or temperature – can have knock-on effects. “The models that are used are constantly evolving and becoming more and more complex,” notes McGrath.

Met Éireann is migrating from using an older software package that can analyse and make weather predictions for areas of five kilometres to a new package called Harmonie, that can bring resolution down to just one kilometre. Currently they are focusing on 2.5 kilometre areas.

“Harmonie is a very sophisticated weather forecasting program that will allow very high spatial resolution,” notes Dr J-C Desplat, ICHEC’s associate director. But this involves bumping up the processing power required to analyse data by two orders of magnitude – or 100 times the computing power.

This is largely because Harmonie integrates a physics engine – the ability to analyse and model dynamic changes in weather and climate, such as rising and falling wind speed or the effects of melting sea ice. This type of complex programming is also used in computer animations and games, to give a real-life look to onscreen motion and special effects.

In order to produce a forecast, Met Éireann first receives a set of “boundary files” full of raw data for the area they wish to analyse.

The raw weather data takes 20 minutes to assimilate into Harmonie, using 48 of Stokes’ many processors, and then another 25 minutes is required to run the analysis that produces forecasts.

“All this is very intensive computing,” says Desplat.

Back in 2006 before Met Éireann started using ICHEC supercomputing power, says Desplat, its computers took 81 minutes to assimilate raw data into the forecasting program, and then an hour to analyse it, for 15-kilometre areas.

However, Harmonie is not yet being used to produce the Met’s daily weather forecasts, says McGrath, even though it accurately mapped the downpours that caused localised flooding in August, which Met Éireann was criticised for failing to predict.

But McGrath points out that while it might have got the downpours right, Harmonie still can get other forecasts wrong, and needs to be carefully tested before it will be put into general use later this year. “Harmonie is such a complex system that a lot of tuning has to be done for the system. One of the issues at the moment is surface wind projections with Harmonie are a little too strong,” he says.

As Harmonie is a co-ordinated, European-wide community programming effort, bugs like this will be assigned to a programming group somewhere to resolve.

“We’re always extra cautious, and we could always do with a better model,” laughs McGrath. “But you can’t rush these things either.”