Title: Artificial Intelligence in Severe Weather and Wildfire Modeling

Description: As weather-related events--like hurricanes and
wildfires--become more frequent and severe, meteorologists and others
are
looking for new ways to forecast these events. GAO's Brian Bothwell
joins us to share his new report on the role artificial intelligence may
play in predicting weather events. 

Related GAO Work: GAO-24-106213, Artificial Intelligence in Natural
Hazard Modeling: Wildfires, Storms, Hurricanes, and Floods 

Released: December 2023

[MUSIC]

[Brian Bothwell:] By using machine learning techniques, it can make the
traditional model of forecasting wildfires more accurate.

[Holly Hobbs:] Hi and welcome to GAO'S Watchdog Report. Your source for
fact-based, nonpartisan news and information from the U.S. Government
Accountability Office. I'm your host, Holly Hobbs. As weather-related
events, like hurricanes and wildfires, become more frequent and
severe--meteorologists and others are looking for new ways to predict or
forecast these events. In a new report, we take a look at the role
artificial intelligence may play in predicting weather events, and the
status of AI use in environmental modeling. Joining us to tell us more
is Brian Bothwell, an expert on AI technologies. Thanks for joining us,
Brian. 

[Brian Bothwell:] Thanks for having me, Holly. 

[Holly Hobbs:] So, Brian, how is AI currently being used to predict
weather events, and what's the benefits of using AI versus traditional
modeling?

[Brian Bothwell:] So we talk specifically about machine learning
techniques, which are branched by AI. They're mostly being used in the
research and development phase as relates to natural hazard modeling.
But there are some operational uses. For example, there's one model
that's being used to predict hurricanes. And the machine learning model
is better at predicting the intensification of hurricanes than
traditional models are. So that's one of the benefits of using AI. In
some situations, you can significantly speed up the forecasting. And
quicker predictions to these natural hazards could lead to increased
warning times and greater opportunities to protect people and property
from the effects of these hazards. 

[Holly Hobbs:] So how exactly does it work? Like, how would AI be used
to forecast a hurricane versus a wildfire?

[Brian Bothwell:] One thing that machine learning techniques need is a
lot of data. So, if you have enough data about a natural hazard and the
different aspects of it, then you uncover relationships in the data that
aren't represented by traditional models. So that can help you with your
forecasting. You can also use these techniques to improve the input data
for traditional models. So, we can use machine learning to provide
better data regarding the fuels data for wildfires. That's the trees,
the brush, all the other things that can burn in a wildfire. And if you
have improved data going in, it helps the traditional model forecast the
results better.

[Holly Hobbs:] So if AI has all this potential when it comes to weather,
why isn't it being used more widely already? 

[Brian Bothwell:] Well, there are some challenges. One is that machine
learning can be expensive. It's very computing intensive. Also, it needs
large amounts of historical data to create a model that will model these
hazards. For example, we have a category five hurricane, so they're very
destructive, but they are relatively rare. We don't have that many of
them. So it could be a situation where you don't really have enough data
to feed that machine learning model to give you accurate forecasting of
those types of events. Another challenge is just the acceptance by
forecasters. The forecaster has to be making high-risk decisions like
issuing evacuation orders, they're probably going to be more comfortable
with the traditional models that they've been working with and
understand better rather than newer AI enabled models.

{MUSIC}

[Holly Hobbs:] So Brian just told us that machine learning, a form of
artificial intelligence, has the potential to forecast weather events
more quickly and accurately, which could save lives. But that the cost
and other obstacles are preventing its broader use. So, Brian, our new
report includes some policy options or considerations for increasing the
use of AI in environmental modeling. What can you tell us about those
options? 

[Brian Bothwell:] Yeah, we developed four different policy options for
Congress and other decisionmakers to consider. They address data
collection, sharing, education and training in these topics, hiring and
retention issues in the workforce, and protecting against bias in the
data. And let me just give you one example--since I've mentioned data a
lot--but that first policy option is talking about enhancing data
sharing
and collection. We go into how you could potentially increase the amount
of data that you can collect. You may need to actually build out more
infrastructure for that. There's a lot of existing data out there that's
not necessarily shared. So some of those barriers can be broken down to
increase the amount of data that modelers could use. And there's also an
issue about making that data AI ready. It's all in different formats. So
creating a standard would also help to enhance the data collection
sharing.

[Holly Hobbs:] We've done a lot of work on AI use. How does this work
fit into the portfolio of work, and what other work do we have that is
sort of relevant to what we discussed today? 

[Brian Bothwell:] We've done a lot of work on the use of artificial
intelligence. Just a couple of years ago, we issued an AI framework that
identifies key practices for federal agencies. And we've also issued
several reports that assess artificial intelligence use in specific
areas like drug development, disease diagnosis--and now with this
report,
natural hazard forecasting. We've also issued to science and technology
spotlights on topics like generative AI and deepfakes. But these are
just a few of the reports we've published.

[Holly Hobbs:] Brian, last question. What's the bottom line of this
report? 

[Brian Bothwell:] So AI in particular, machine learning, can provide
benefits by reducing forecast times. It can make models more accurate.
It can reduce forecast uncertainty, which could result in potentially
saving lives and property. But there are still some challenges that have
to be overcome in implementing these techniques. 

[Holly Hobbs:] That was Brian Bothwell talking about AI environmental
modeling. Thanks for your time, Brian. 

[Brian Bothwell:] Thank you, Holly.

[Holly Hobbs:] And thank you for listening to the Watchdog Report. To
hear more podcasts, subscribe to us on Apple Podcasts, Spotify or
wherever you listen and make sure to leave a rating and review to let
others know about the work we're doing. For more from the congressional
watchdog, the US Government Accountability Office, visit us at GAO.gov.