Dr Jess Robertson (left) with Science, Innovation and Technology Minister Dr Shane Reti. Photo / Earth Sciences New Zealand
By Kate Green of RNZ
A new supercomputer is expected to vastly improve the accuracy of daily weather forecasting around the country, with the help of artificial intelligence.
Dr Jess Robertson, chief scientist for advanced technology at Earth Sciences New Zealand, says the three-fold increase in computing power would make a five-day weather forecast as accurate as a two-day forecast.
Created by Hewlett Packard Enterprises and known as Cascade, the new computer allowed forecasters to run multiple simulations simultaneously.
And it was already online, based in a CDC data centre in Auckland, replacing the “blue box” at Wellington’s Greta Point.
Where the old computer measured 20m across and sucked up more power and needed more cooling than any co-hosting space could accommodate, the new supercomputer was far more economical.
It measured about 3m across and looked, Robertson said, just like any other computer – “a lot of wires, a lot of hoses carrying water into the chips so that they can be cooled, and just a big bunch of non-descript boxes”.
Having it hosted in a data centre, rather than on-site, meant they could focus on doing the science rather than looking after the machine itself, he said.
“We’re benefiting from the overseas investment into these really, really big data centres for cloud vendors and so on, to be able to host our machines.”
Its capabilities were a big step up, capable of processing 19 petabytes of data. For comparison, Earth Sciences had been gathering climate data for the past 30 years every 15 minutes – and that data set measured about half a petabyte, Robertson said.
It meant more efficient and accurate forecasting.
Day to day, members of the public would have more certainty about whether to host a family barbecue on the weekend or go tramping in the Tararua Range, for example.
But it would also allow energy companies to maximise power production from solar and wind and give emergency management an earlier heads-up on potentially catastrophic events.
The computer was undoubtedly still power-hungry, but the data centre ran on renewable energy and the water used to cool the machine was recycled, Robertson said.
“The idea here is that we are not making the climate issues worse as we model them.”
Hewlett Packard vice-president and managing director for the South Pacific, Chris Weber, explained the company was able to maximise energy efficiency through direct liquid cooling and leveraging renewable energy.
“We’re empowering New Zealand’s leading climate and environmental researchers with the tools they need to drive scientific discovery and tackle some of the most pressing challenges of our time,” Weber said.
Robertson said that using artificial intelligence (AI) in the environmental space was about noticing patterns and predicting future weather, without needing to analyse the science behind it every time.
“Sometimes you just need the pattern,” he said.
Cascade was capable of using AI to predict short-term, highly-localised weather – something energy companies and emergency management alike would be able to leverage.
Scientists called it “nowcasting”, with AI making the forecasting process sleeker and faster.
“Our big physical models will take about two hours to run for about a week’s worth of data, or a week-long forecast,” Robertson said. “And what these AI models can do is, rather than wait that two hours, it can take exactly what it’s seeing on the ground now and effectively learn to blend between what are the conditions right now and what is my model saying is going to happen.”
That cut out errors by about half, he said.
It also applied to flood models. Forecasts which would take a day would take two minutes using AI – something which could really help to make fast decisions on evacuations or directing first responders in a storm.
National Emergency Management Agency (Nema) chief science adviser Tom Wilson told RNZ weather forecasts were essential for emergency management – particularly when it came to storms or droughts.
“Having greater accuracy of any of those forecasts is always going to be extremely useful,” Wilson said.
During rapidly developing severe weather events, from tornadoes to thunderstorms, the greater the accuracy and the longer the lead time, the faster a warning could be sent to communities or to critical infrastructure, such as the national power grid.
New Zealand was in a “peculiar part of the world”, he said, exposed to ex-tropical cyclones and big storms coming up from the south, and the geography of its land masses made things particularly difficult to forecast – the Auckland area in particular.
“The new advanced technology that Earth Sciences New Zealand is gaining with this new supercomputer represents a really important step forward,” Wilson said.
New technology, coupled with people’s expertise, kept the country ahead of the curve when it came to weather-related risks.
MetService, the country’s other key weather forecaster, was due to merge into Earth Sciences in the next few years, meaning it, too, would have access to the benefits of the new supercomputer.
And what of the old 20m computer at Greta Point? Already gone, Robertson said, taken apart and many of its parts recycled, with processing units and memory components being sent overseas to replace bits in similar machines.
There was now a “very, very big empty blue box” at the facility at Greta Point, Robertson said – but likely, too, some savings on the power bill.
– RNZ