During the past couple of decades, Australia's weather forecasting ability has improved dramatically, thanks to advances in science and computational technology.
Now the Bureau of Meteorology (BoM) is collaborating with NCI to test and refine its next-generation weather forecasting models to provide even greater prediction power.
"Back in the 70s and 80s, with severe weather events like Tropical Cyclone Tracy or the Ash Wednesday bushfires, there was very little warning," says Dr Michael Naughton from the Bureau's Centre for Australian Weather & Climate Research partnership with CSIRO.
"But in recent times, for similar events like when Cyclone Yasi hit North Queensland or the Black Saturday bushfires in Victoria, there was far more warning time and that meant the possibility to prepare."
The ability to predict weather events a few days or even a week out also has major economic advantages, says Dr Naughton.
"Not only are forecasts useful for public safety – for example when planning large scale events like the Melbourne Cup or the Sydney to Hobart yacht race – they're also important for people protecting their cars if there's a hail storm coming, for deciding whether to divert aircraft because of storms, for farmers protecting their crops and herds, and families preparing for important occasions like weddings."
While BoM has been running weather models on their own supercomputing facilities since the 1960s, the advent of Raijin provided the Bureau with the opportunity to shift their research activities to NCI.
"Research is always one step ahead of operations in terms of compute requirements," explains Dr Naughton. "Our current computer wasn't growing fast enough to accommodate the growth in the forecasting model. So the expanded resources available through NCI are important for us."
In order to provide the most useful, up-to-the-minute information, the models also have to run very efficiently and quickly. As part of a formal collaboration, staff from BoM, Fujitsu and NCI are working together to improve the efficiency and increase the 'scalability' of the model; that is, to perform computations quickly, and to divide the calculations across multiple computer processing cores to save time and meet forecast deadlines.
This is a challenging task, explains Dr Naughton.
"Weather systems don't just stand still; they move across space. To get the calculations right, the forecast variables at each computer core have to know what's going on at adjacent cores.
"The models currently run on about 1,000 cores but one of the things we're exploring with NCI is scaling up to several thousand. Ultimately, we'll want our models scaling out to run on the order of hundreds of thousands of cores in a few years' time."
