A snowy country scene

Met Office supercomputer: the future of weather forecasting

Will the new Met Office supercomputer, up and running later this year, improve the accuracy of weather forecasting in Britain's fickle climatic conditions?

Talking about the weather is easily the most common conversational gambit among British people, if only because it is so reliably unreliable. However, the easy filling of awkward silences may be about to become a thing of the past if a new supercomputer brings hoped-for transformation to the Met Office's powers of prediction.

True enough, many a weather computer has foundered on the rocks of the British Isles: small islands with the Atlantic ocean on one side and the European landmass on the other, not to mention the Arctic air from the north, and tropical air from the south. Even subtle changes in wind direction can create a major change in the weather.

And changes in the weather make for changes in behaviour - from people's travel plans and heat budgeting to the basic inclinations of shoppers: many a chief executive has pinned poor quarterly results on inclement weather. Accurate forecasts - made as far in advance as possible - are vital to manage such changes.

When the forecasts are wrong, the forecasters get it in the neck. There is simply no way that a feature article about the technology of weather prediction could slip by without a mention of weather presenter Michael Fish and his notorious forecast of 1987, when he reassured a viewer who had apparently phoned in concerned that a hurricane was on the way ("don't worry; there isn't"). The hurricane that followed killed 19 people and has become known as the Great Storm. The April 2009 prediction (at a specially-convened press conference, no less) of a "barbeque summer" is quickly gaining traction as legendarily wide-of-the-mark; the July in question was one of the wettest on record.

"The weather people experience is very localised," says David Underwood, the Met Office's deputy director of technology and information services. "It's impossible to give a precise local forecast for all 60 million people in the UK."

Yet that is what people want and expect, and according to Professor Peter Clark from Reading University, they'll often look to unreliable sources to get it - such as newspaper headlines, which are more about selling newspapers than about reporting the weather. "It's easier to predict that we'll have gale-force winds somewhere in the North than specifically over York or Newcastle," Clark says. "It's an enormous challenge to be able to tell people what will be happening in their backyard."

Collating the data

We have some idea about what sort of weather is imminent: data from satellites and ground stations all over the world is constantly being fed into supercomputers that predict what will happen in the following 24 hours and beyond.

The more data there is, the more computing power is needed to make the prediction. The Met Office currently takes in 106 million observations a day from weather balloons, satellites and commercial aeroplanes around the world. Its existing supercomputer then integrates the weather observations into modelling software to predict the world's weather.

Paul Sellwood, the Met Office's lead for high-performance computing, explains that one of the problems with providing a precise local weather forecast is that the information coming in is changing all the time. "If a farmer ploughs a field, that changes the relationship between the moisture and surface heat in that area," he says. "It's impossible to model a system perfectly."

Professor Clark explains that we have made enormous progress in forecasting over the last century, especially since computer-based systems became the norm in the 1970s. However, weather systems are full of incredibly complex phenomena, some of which scientists cannot explain. "For instance, there are things going on within fog that we just don't understand," he says. "Like how it interacts with urban environments and heat sources.

"It's called chaos. We just don't know what the atmospheric state is at any one time." He adds that differences or occurrences beyond what is possible to measure can eventually create huge changes in the weather, and those changes are impossible to predict precisely.

Clark likens it to a coin toss. "We can use the laws of physics to predict the path a coin takes, but we would need to know the initial velocity, spin and so forth, very precisely indeed to predict heads or tails.You can say with accuracy that there is a 50 per cent probability of heads, but it is very hard to predict which side the coin will come down at any one time," he says.

He points to the devastating floods in Boscastle, Cornwall in 2004: "A five-degree change in wind direction and the storm would have happened out at sea. With a thunderstorm, if you have one-tenth of a degree less, there's nothing; one-tenth more and you have a flash flood. All you can do is predict the probability of an event. The problem is that people want specifics from their weather forecast, not probabilities."

David Underwood adds that the Met Office can give a reasonably accurate indication of what the weather will be like over London in two hours' time. "We'd just need weather information from the rest of the UK for that," he says. "For a seven-day forecast we'd need information from all over the globe and from this calculate how the world's weather would affect London."

The new hardware

The Met Office says that its new supercomputer will work 13 times faster than the existing one. That's more than twice as powerful as the next most powerful climate-dedicated supercomputer, the Hornet Machine, currently being built at the University of Stuttgart.

The Met Office currently uses an IBM POWER7, a 1.2-petaflop machine. The new supercomputer is a 16-petaflop Cray XC40. It will have two million gigabytes of memory available to run calculations - the equivalent of 120,000 smartphones. There will be over 17 million gigabytes of storage capacity. It will run at speeds of more than 1.5 terabytes bandwidth and use 480,000 cores - that's more processing power than 100,000 PlayStation 4 games consoles. The machine will weigh 140 tonnes, as much as eleven double-decker buses.

A supercomputer's effectiveness depends on its simulation's resolution. Like the pixels in a digital camera, the smaller the pixels, the more detailed the picture. But the smaller the pixels, the more computing power is needed to perform the calculation.

With the Met Office's current system, the pixels correspond to a 12km square on the ground. This hampers their ability to provide flood warnings because the heaviest rainfall usually occurs on much smaller scales.

The Met Office's existing computer can produce higher-resolution pixels of just 1.5km across. But these forecasts are less frequent. With the new computer, the Met Office hopes to run 1.5km forecasts every hour. Forecasters will be able to zoom in on particular regions to produce detailed local forecasts down to 300m resolution, which is useful for predicting localised weather conditions like fog at airports, strong winds, and heavy snowfall.

Met Office experts will also be able to analyse data once an hour rather than the current six hours, and give forecasts six days ahead instead of four. The computer will enable them to calculate temperatures for the next 24 hours with up to 90 per cent accuracy. Currently, they can do this for only 12 hours.

Underwood says that by the time the new supercomputer is fully functional by mid-2017, it will have accelerated the quality of service the Met Office provides by five years. "Where we can currently give 12 hours' warning we'll be able to give 24 hours' notice," he says. "72 hours will become five days."

He adds that as a result the Met Office's business customers will be able to make better infrastructure preparation for weather and reduce delays and down-times associated with severe weather events. "An energy company could work out in advance how many engineers it needed to get services up and running during and after severe winds," he says. "An aviation firm could predict how much de-icing equipment it needed at a certain airport to keep ahead of its competitors."

The Met Office also believes the computer will help scientists make more accurate climate change predictions and will facilitate forward-looking climate simulations and models that better assess how climate change will affect weather systems.

The future's a blur

Currently, the only way to predict weather patterns years, decades and centuries in advance is to reduce the resolution of the simulation. The new supercomputer, the Met Office says, will allow better resolution in the climate predictions. Experts will be able to program more complex science into their simulations and test if they more accurately reflect the data.

The Met Office plans to work with the pan-European Primavera study, which will examine the likely effects of oceans, aerosols and greenhouse gases on climate patterns and extreme weather events over the next 50 years.

The Office is also planning a UK-wide research project to create a next-generation climate model called an Earth System Model, which captures all major aspects of the Earth's climate system: oceans, atmosphere, atmospheric chemistry, terrestrial carbon cycle and ocean biogeochemistry.

Peter Clark is sure the new computer will have an effect on the quality of weather forecasting, but he doesn't think it will revolutionise everything overnight. "It'won't solve all the problems," he says. "It's a major step, but not the final step."

Clark explains that forecasters need better observation data and more sophisticated algorithms. "These will need even greater processing power," he says. "This means more sophisticated and expensive systems such as satellites and radar, but there's also the possibility of using large quantities of more cheaply available, lower-quality data such as weather stations in people's gardens as well as mobile phone and Internet technology.

"The question is, could a supercomputer quality-control the data to make it useful, or would the lower quality data undermine the computer's ability to predict accurately?" he wonders.

Clark adds that one thing the Met Office could do is think more carefully about how it communicates its forecasts. "People expect a lot from their weather forecasts," he says; it could be more than it is possible to get, however super the computer.

The problem is that it is better business for an organisation like the Met Office to get things occasionally wrong than to admit that it's just not possible for them to get it right all the time.

After all, the British people would rather complain about unpredictable weather than admit that all weather is, to some degree, unpredictable.

Further information

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close