NCI in the news: How supercomputers predict the weather
Originally published in The Canberra Times, Thursday 20 March 2014
by Iain Gillespie
The spectacular power of supercomputers, which are helping the world’s leading climatologists to wrestle with the vexed issue of global warming, is increasingly being used to benefit everyday users of mobile weather apps.
For instance, the perpetual question, ”What’s the weather going to be like today”, can now be extended beyond the Australian Bureau of Meteorology’s popular seven-day forecasts to cover entire seasons. Ask instead, ”What will autumn be like?”
Such projections are becoming more accurate, because the bureau is engaged in some of the world’s most sophisticated climate modelling, and the way it plans to provide the public with a new generation of weather information is changing too.
Strangely, there is no mention of an Android version in the tender for design and development of the apps issued earlier this year.
Tenders have now closed, but when approached by Fairfax Media, the bureau’s media office declined to discuss what services the new versions might provide, calling the issue ”commercial in confidence”.
A senior media officer was also silent when asked for specifics on the bureau’s new supercomputer, hinting darkly that bad people might target it for all sorts of nasty reasons. This is probably a valid concern, but what the computer is capable of makes a much more interesting story. It can be used to save farmers, manufacturers, retailers, firefighters and, ultimately, the nation, billions of dollars, by giving them a better idea of what to plant, market or invest in for the coming season.
The computer’s physics-based seasonal predictions are already featured on the bureau’s new-look website, which, incidentally, is the first major government internet portal to include commercial advertising.
The bureau has dubbed it Ngamai, a 100-teraflop (1 trillion floating point operations a second) monster installed in its Melbourne data centre. The amount of information it can happily chew on is so large and the algorithms it uses are so complicated that they can be hard for the average human mind to contemplate.
Consider those videos of a beautiful spinning planet captured during space-shuttle missions, showing an intricate biosphere of oceans, landmasses and cloud patterns, all interconnected and constantly changing. Now imagine the amount of data required to continuously record in detail, not only what is happening on the entire surface of the planet, but down into the depths of all the oceans and 50 different levels up into the atmosphere.
That global data is collected around the clock from satellites, radar weather stations, aircraft, ships and meteorological balloons. It incorporates the Earth’s rotation, temperatures, air pressure, solar radiation, humidity, wind direction and speed.
Ngamai continuously munches on all that information as it flows in, producing seven-day forecasts several times a day, and providing twice-weekly projections of what the world’s climate could look like in nine months.
”It’s not something you could do on your iPad,” says the bureau’s manager of climate prediction services, Dr Andrew Watkins.
”The climate model actually calculates weather, and says if you move these things around on a spinning planet, this is what you get. It’s pretty amazing.
”We know we will never be 100 per cent accurate. You just can’t do that with climate. There will always be random factors – chaos basically. But when you’re planning something over a number of seasons that will be affected by weather, it means that like a casino, you’ll have the odds in your favour and come out ahead overall.
”If we say there’s an 80 per cent chance of something happening in the coming season – whether it be above-average temperatures or below-average rainfall – those predictions will reliably occur about 80 per cent of the time.
”We get an average accuracy around the 60 per cent mark. Some times of the year, it’s far higher that that. It can get above the 75 per cent accuracy range, and accuracy is obviously higher when forecasting the first month of a three-month season.”
The bureau has moved from a statistical seasonal forecasting scheme to a new physics-based model, he says. The advance is comparable to jumping from analog to digital. ”As the science and technology moves forward, it will naturally become more accurate over time.”
Supercomputing has smashed traditional time barriers in climate modelling, progressing through improved weekly and seasonal forecasting to more confident predictions of what the weather is likely to be like in 50 years.
As well as Ngamai, the bureau has access to data produced by Australia’s most powerful supercomputer, Raijin, at the Australian National University in Canberra. It can run at a peak speed of 1.2 petaflops, equal to 170,000 calculations for every human on Earth, every second.
Andy Hogg, of the ANU’s Research School of Earth Sciences, uses Ngamai to run models that incorporate even small-scale ocean currents and eddies to project how global climate will be affected in decades.
”Our work is underpinning the improvements in climate models that we’ll see over the next five years,” he says.
Michael Hutchinson, professor of spatial and temporal analysis at the Fenner School of Environment and Society, has created landscape models that led to the discovery of hidden streams 35 metres under sand dunes in the Simpson Desert.
He is using the ANU’s computer to predict the destructive effects of long-term climate change on landscape, plants, animals and human infrastructure.
Asked if he believes in global warming, he says. ”I don’t believe it. I know it. It’s an irrefutable scientific fact.”