Chris Brandolino: NIWA Principal Scientist on the Cascade supercomputer aiming to increase weather forecast accuracy

Unreliable weather forecasts could become a thing of the past. 

Earth Sciences NZ has unveiled a new supercomputer called Cascade, which uses AI to predict the weather patterns. 

It can now make a five-day forecast as reliable as a two-day one. 

Chris Brandolino, Principal Scientist at NIWA, told Mike Hosking this supercomputer is about three times more powerful than its predecessor, with computing speeds of 2.4 petaflops. 

He says they’ll be able to expand the area they monitor, and then over the next year or so they’ll be looking to grow longer range predictions from 5-days to 10-days. 

LISTEN ABOVE 

See omnystudio.com/listener for privacy information.

Episode Transcript

Speaker 1 (00:00):
For forecasting and improvement in the forecast. Earth Sciences they've
got a new supercomputer they've called it Cascade, uses AI
to predict the weather patterns. The claims it can produce
a five day forecast as good as a two day forecast.
But then the next question, obviously, is a two day
forecast really any good? Chris Brandolino as new as principal
scientist and as well us Chris, good.

Speaker 2 (00:18):
Morning, Hey, good morning, Mike.

Speaker 1 (00:20):
Is a two day any good?

Speaker 2 (00:24):
I think? So? Why not?

Speaker 1 (00:26):
Well, because sometimes it's wrong, that's a.

Speaker 2 (00:30):
Negative way of looking at things. A lot of times
it's right.

Speaker 1 (00:34):
So what do we base the ratio on? If it's
right more than fifty percent, then it's more right than
it's wrong. So that's fair enough as it.

Speaker 2 (00:43):
Well, look, the bottom line is their skill. There is
skill in weather forecasting. It's not perfect. I mean, part
of it is. You know, if you ask a person,
you know, what do you define a perfect forecast? If
I say it's going to be twenty one degrees in
Auckland in the heads of twenty two, is that accurate?
I would think that's pretty dark accurate, you know what
I mean?

Speaker 1 (00:58):
The scale up we've got here instication. Is this just
going to go exponential as II builds and grows?

Speaker 2 (01:05):
Yeah, I think the big thing I'll guess I'll answer
it this one. I'm not a supercomputer nerd. I wish
it was, but anonymal weather nerd. I think one of
the main things to kind of tell you and your
listeners is that this supercomputer Cascade is about three times
more powerful than its predecessor. So it does computing speeds
of two point four ready for this petaflops And you're
probably wondering, what the heck is a petaflop? So imagine

(01:28):
if the entire Earth, about eight billion people, everyone was
doing three hundred three hundred thousand calculations per second at
the same time. That's the capability of the supercomputer. And
because and the reason we need that MIC is because
data is become king or queen I suppose, and with
so much more data coming in, we need to leverage

(01:48):
that data. I think the analogy I've been using, which
may resonate, is that imagine a restaurant getting a brand
new kitchen, and that kitchen really increases the capability. The
auven cooks faster, so you get the dishes out quick
or you have better tools things like that, so you're
able to operate more effectively and do different things that
maybe you wouldn't be able to do before.

Speaker 1 (02:06):
Yeah, what's maximum value from a commercial scalable point of view?
Given what AI is going to do is a five
day forecast, seven to eighteen day, one month, team.

Speaker 2 (02:13):
Year, what I think it's all of that, So I
guess to take you on a bit of a journey.
So one of the things we can do now with
our high res model, which goes out to your favorite
time span. It sounds like two days. That used to
run on the old supercomputer, it will take one hundred
and eighty minutes. Now it takes eighty minutes, so that's better. Okay,
what about longer range predicting? So most times when people

(02:36):
go to an app, they you know, look at their
favorite weather app. That's what we call deterministic. That's simply
one outcome and that's okay, But in reality, there's a
lot of outcomes that could happen. So we have an
ensemble model that has eighteen different outcomes. We're going to
expand the rectangle it looks at over New Zealand to
go up to the Pacific to New Caledoni, out to
Aussie Melbourne and down to the southern Ocean, so better,

(02:58):
I guess, larger area where monitoring in terms of modeling
with our ensemble, and then over the next year or so,
we're going to be looking to increase that from five
days to ten days, so we're gaining not only areas
to get insight on, but also the time ten days
and then there's longer longer rains forecasting five weeks, six weeks,
and getting a better understanding of what themes may be

(03:18):
come in their way for I think we're planning and
getting some heads up as to hey, in three or
four weeks there's a signal that it could be really
dry or on the other end of the spectrum, really wet,
and then monitoring that and getting people prepared for potentially
some large scale big weather events or things like dryness
or drought.

Speaker 1 (03:37):
Fantastic like your passion. Chris Brandelina, who's the need of
principal scientists with us this morning.

Speaker 2 (03:41):
For more from the Mic Hosking Breakfast, listen live to
news talks it'd be from six am weekdays, or follow
the podcast on iHeartRadio.