Can AI help firefighters manage wildfires? | FireSat

Summary

The script discusses the increasing threat of wildfires and introduces FireSat, a Google-led initiative to combat this issue. FireSat utilizes satellites equipped with machine learning and AI to detect fires in real-time, providing critical data to first responders and fire scientists. The initiative aims to enhance early detection, improve firefighting strategies, and contribute to community safety. The collaboration includes experts from various fields, emphasizing the importance of accessible, high-resolution data for effective fire management.

Takeaways

• 🔥 Kate Dargain's first experience with a massive fire led her to seek better ways to combat fires.
• 🌍 Juliet Rothenberg highlights the global impact and unprecedented human cost of increased wildfires over the past 20 years.
• 🤖 Chris Van Arsdle and Google's research group are leveraging technology, specifically AI and machine learning, to address fire management and detection.
• 🛰️ The development of a specialized camera for machine learning-based fire detection is underway to optimize satellite imagery analysis.
• 🚀 FireSat, a constellation of satellites, is being designed for rapid, high-resolution detection of fires worldwide.
• 🔎 FireSat aims to provide real-time data to first responders and incident commanders, filling information gaps in firefighting.
• 🌳 The project also targets fire scientists to enhance fire behavior models, crucial for predicting and managing wildfires.
• 🧠 Google's computational power and expertise are being used to create a faster, deep neural network model for fire spread prediction.
• 📊 The new AI model can make thousands of fire predictions in milliseconds, a significant improvement over traditional methods.
• 🌐 Google's commitment to providing wildfire information during crises is exemplified by the integration of wildfire boundaries in Google Maps and Search.
• 🌟 The ultimate goal of FireSat is to democratize access to fire data, enhancing global health and safety through early detection and understanding of fire behavior.

Q & A

What was Kate Dargan’s first experience with large-scale fire, and how did it affect her perspective?

Kate Dargan's first experience with a large-scale fire was during the Cedar Fire in 2003, where she witnessed chaos, destruction, and rows of houses burning. This experience made her realize that there must be a better way to handle wildfires.

How have wildfires changed over the past 20 years, according to Juliet Rothenberg?

Wildfires have dramatically increased over the past 20 years, which was anticipated by climate scientists. However, the extent of the human cost and the global impact has been unprecedented and underestimated by many.

What was one of the main challenges faced by Google’s research group when trying to address wildfires?

One of the main challenges was the lack of reliable data on how wildfires spread and where they were located. Satellite images often mistook other elements like clouds, ponds, or smokestacks for fires, making it difficult to track fires accurately.

How is AI and machine learning used in Google’s approach to detecting wildfires?

AI and machine learning are used to analyze satellite images and detect fires in real time. By combining thousands of past images of the same area, the AI can determine whether the current images show signs of fire. This system enables more accurate and timely fire detection.

What is FireSat, and how does it improve wildfire detection?

FireSat is a satellite-based program designed to detect wildfires. It uses a constellation of satellites in low Earth orbit to monitor the entire globe every 15 to 20 minutes, providing rapid, high-resolution images to spot fires as small as a classroom. This significantly improves the current detection systems, which revisit areas less frequently and at lower resolutions.

What role does Cathy Olkin and her team at Muon play in the FireSat project?

Cathy Olkin and her team at Muon are responsible for designing, building, and operating the FireSat mission. Their role is crucial in making sure the satellite system functions optimally to provide accurate wildfire data.

How does the Rothermel fire spread equation contribute to current firefighting efforts, and what are its limitations?

The Rothermel fire spread equation, developed in the 1960s, is the foundation for current fire behavior modeling and firefighter training. However, it does not account for changing fire conditions like wind, moisture, or vegetation variability, making it less adaptable to real-time fire behavior changes.

How has Google's partnership with fire scientists like Mark Finney enhanced fire behavior modeling?

By leveraging Google’s massive computing power, scientists like Mark Finney have been able to run millions of simulations and generate data points for fire behavior. Google then used this data to train deep neural networks, allowing for much faster and more accurate fire predictions, reducing prediction times from a minute to 20-30 milliseconds.

What advantages does FireSat offer to firefighters and first responders?

FireSat provides real-time data that helps first responders detect and respond to fires quickly, improving their ability to protect communities and ecosystems. The system also aids in training by allowing simulations of different fire scenarios, improving preparedness for unexpected fire behavior.

How does Google incorporate wildfire detection data into its existing services?

Google uses wildfire boundaries detected through satellite imagery and AI in its Google Maps and Google Search services, providing people with real-time information during crises to help them stay safe and make informed decisions.

Outlines

00:00

🔥 Fire Detection Innovations

Kate Dargan recounts her first experience with a massive fire in 2003, highlighting the need for better fire management strategies. Juliet Rothenberg discusses the increase in fires over the past 20 years and their unprecedented global impacts. Chris Van Arsdale from Google's research group explains their initiative to use technology to address fire management, focusing on the lack of accurate fire data. They developed a camera optimized for machine learning-based fire detection and tested it on aerial platforms. The goal is to use AI and ML to analyze historical data and predict current fire occurrences in real-time. The Muon team, led by Cathy Olkin, is collaborating with Google to develop the FireSat Mission, a constellation of satellites for rapid, high-resolution fire detection. FireSat aims to provide real-time data to first responders and fire scientists to improve firefighting and fire behavior models.

05:01

🌐 Advancing Fire Modeling with AI

Mark Finney discusses the limitations of the Rothermel fire spread equation, which is foundational for firefighter training and decision-making but does not account for changing conditions. Google Research partners have collaborated to overcome these limitations by using their computational resources and expertise. They ran Mark's model to generate a vast amount of data, training a deep neural network to approximate the model's predictions much faster. This new AI model can make thousands of predictions in milliseconds, compared to the original model's minute-long predictions. Jason Forthoffer explains how these advancements can be used in planning phases to simulate fires under various conditions, identifying areas of significance for firefighting. Juliet Rothenberg emphasizes Google's role in providing crisis information and the integration of wildfire boundaries in Google Maps and Search. Chris Van Arsdale expresses hope that FireSat will provide equitable access to data, improving health and safety. The script concludes with a positive response from the International Association of Fire Fighters to the potential of FireSat for their community.

Keywords

💡Cedar Fire

The Cedar Fire is a specific historical wildfire event mentioned in the script, signifying the scale and chaos of wildfires. It serves as a personal experience for Kate Dargain, illustrating the destructive power of wildfires and the need for better management and response systems.

💡Climate scientists

Climate scientists are experts in the field of climate change, and their anticipation of increased fires is referenced to emphasize the scientific basis for the growing wildfire problem. Their role highlights the connection between climate change and the frequency and intensity of wildfires.

💡Machine learning

Machine learning is a subset of artificial intelligence that enables computers to learn and improve from experience without being explicitly programmed. In the context of the video, it is used to enhance fire detection capabilities by analyzing satellite imagery, which is crucial for the development of the FireSat program.

💡Artificial intelligence (AI)

AI refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The script discusses how AI can be paired with satellite data to generate real-time operational intelligence on fires, which is a key component of the FireSat initiative.

💡FireSat

FireSat is a program mentioned in the script that aims to create a constellation of satellites for detecting wildfires. It represents a technological solution to the problem of wildfire detection and monitoring, with the goal of providing rapid and high-resolution data to aid in firefighting and prevention efforts.

💡Low Earth orbit

Low Earth orbit (LEO) refers to the region of space around the Earth that is close to the planet's surface, typically up to a few thousand kilometers in altitude. In the script, FireSat satellites are in LEO to enable more frequent observations of the Earth's surface for early fire detection.

💡Rapid temporal revisit

Rapid temporal revisit is a term used to describe the ability of satellites to frequently revisit the same location on Earth to capture updated imagery. The script highlights the importance of this capability for early and accurate fire detection, which is a significant improvement over current satellite technologies.

💡First responders

First responders are individuals who are trained to respond to emergencies and provide immediate assistance, such as firefighters and emergency medical personnel. The script mentions how FireSat data will be distributed for the use of first responders to protect communities and ecosystems quickly, emphasizing the practical application of the technology.

💡Fire behavior modeling

Fire behavior modeling is the process of predicting how fires will spread and behave based on various factors like wind, moisture, and fuel type. The script discusses the limitations of current models and how AI can be used to create faster and more accurate predictions, which is vital for planning firefighting strategies and training.

💡Google Maps and Google Search

Google Maps and Google Search are mentioned as platforms where wildfire boundaries can be displayed using satellite imagery and AI. This showcases how technology and data can be integrated into widely used services to provide critical information to the public during emergencies, enhancing safety and awareness.

💡Equitable access

Equitable access refers to the fair and just access to resources and information. In the context of the video, it is mentioned as a goal for the FireSat program, ensuring that data on wildfires is available to everyone, regardless of their location or resources, to improve health and safety outcomes.

Highlights

Kate Dargan recalls her first experience with a massive fire and the chaos it caused, highlighting the need for better fire management solutions.

Juliet Rothenberg discusses the dramatic increase in fires over the past 20 years and the unprecedented human and global impacts.

Chris Van Arsdle explains Google's research efforts to use technology for fire management and the challenges in obtaining accurate fire data.

The Google team realized the potential of combining satellite imagery with machine learning for fire detection.

A camera optimized for machine learning-based fire detection was designed and tested over controlled burns.

AI and ML are used to analyze historical satellite data to predict current fire conditions in real-time.

The Muon team is collaborating with Google to bring the FireSat mission to life, focusing on wildfire detection.

FireSat aims to provide global fire detection with high-resolution imagery every 15 to 20 minutes.

Current satellite technology for fire detection has limitations in frequency and resolution.

FireSat will offer rapid detection of fires at a very small size, approximately the size of a classroom.

Cathy Olkin emphasizes the importance of understanding the needs of first responders and incident commanders through interviews.

Kate Dargan, with decades of firefighting experience, discusses the information gaps in firefighting and the potential of FireSat to fill them.

FireSat will provide a global data stream for first responders to protect communities and ecosystems quickly.

Data from FireSat will also help fire scientists improve fire behavior models.

Mark Finney discusses the limitations of the Rothermel fire spread equation and the need for a more accurate and faster model.

Google's computational resources and expertise are used to create a faster, deep neural network model for fire behavior prediction.

The new AI model can make thousands of predictions in milliseconds, compared to the traditional model's one-minute prediction time.

The new modeling system can simulate thousands of fires to identify areas of high significance for firefighting strategies.

Juliet Rothenberg highlights Google's role in providing crisis information and the potential of FireSat to improve early fire detection.

Chris Van Arsdle expresses hope that FireSat will bring equitable access to fire data, improving health and safety globally.

The International Association of Fire Fighters shows enthusiasm for FireSat, recognizing its potential benefit for firefighters.

Transcripts

KATE DARGAN: In 2003, we were called down to the Cedar Fire.

It's orange everywhere, and smoke is blowing.

And I just see the rows of houses burning.

I see horses running wild trying to escape the flames.

That was my first experience in fire

at that pace and scale of chaos and destruction.

And what I walked away with is

there has got to be a better way.

JULIET ROTHENBERG: Fires have dramatically

increased over the past 20 years in a way

that climate scientists anticipated.

But I don't think any of the rest of us really understood

what the cost ultimately would be.

We're seeing a human cost to communities and impacts

around the world that are really unprecedented.

CHRIS VAN ARSDALE: So the research group at Google

was trying to find ways of using technology to help address fires

and to help manage natural ecosystems.

We realized one of the main issues

was that there just wasn't a lot of good data for how fires

spread, or where they even were half the time.

Because in a satellite image of the Earth,

there's a lot of things that can be mistaken for a fire.

Clouds reflecting sunlight, a pond that's

reflecting light back into space,

something that's hot like a smokestack.

All of these things could look like a fire,

especially when you're looking at a relatively coarse pixel.

We realized that if we can pair satellites

with machine learning and artificial intelligence,

it was the perfect platform to generate

real-time operational intelligence on fires.

So we designed and built a camera

in order to optimize for machine learning-based fire detection.

We spent a long time testing it on an aerial platform,

flying over controlled burns, trying

to establish a baseline for what the machine learning algorithms

would have to pick out.

AI and ML will allow us to take the last 1,000 times we saw

a particular spot on the earth and estimate whether or not

the images that we're seeing now have a fire in them or don't.

So the Google team really focuses on the data.

The Muon team is helping bring that to the first protoflight

satellite.

CATHY OLKIN: At Muon, we are designing, building,

and we'll be operating the FireSat Mission.

CHRIS VAN ARSDALE: FireSat is a program

that we started with satellite experts

and a number of nonprofits.

It will be a new constellation of satellites in low Earth

orbit, and be able to see the whole globe every 15

to 20 minutes and spot a fire, both

when it starts and for its full evolution.

CATHY OLKIN: A number of satellites do many things,

but FireSat is focused on the problem of wildfire.

CHRIS VAN ARSDALE: Satellites today

that are used to do detection and tracking for fires,

they come by roughly once every 12 hours

or have very low resolutions.

At best, you're going to get something

that's two or three acres in size before you can actually

see a fire.

The main improvement FireSat is going to provide

is really this rapid temporal revisit,

especially at the really high resolution needed

to see the fires when they're very tiny in size,

roughly the size of a classroom.

CATHY OLKIN: We've had more than 200 interviews

with first responders and incident

commanders across the globe to understand the diverse needs

of people in this area.

KATE DARGAN: I was a line firefighter for decades,

and I finished as the state fire marshal of California.

So I was aware of some information gaps

in the firefighting space.

This is a global data stream that

will be distributed for the use of first responders

to get to places to protect communities

and ecosystems quickly.

And that's just the firefighting part.

CATHY OLKIN: We also want to provide data to fire scientists

so that they can improve models of fires.

MARK FINNEY: The foundation of all of our fire behavior

modeling systems is the Rothermel fire spread equation,

developed by Dick Rothermel in the 1960s based on experiments

at this very laboratory.

It's the foundation for firefighter training materials

and for all of the modeling systems

that people rely on to support decisions

associated with fighting fire and fire management.

One of the major limitations of it

is that it doesn't explain how fires spread.

It does not allow you to take into account

the changing of fire behavior that goes along

with changing wind conditions, or changing moisture conditions,

or the variability in the vegetation or the fuel.

We've been working towards replacing the Rothermel spread

equation for many years, but one of the challenges

is that the run times for this model are too slow.

Maybe that means I'm not very good at it.

I don't know. (laughs)

We began working with Google Research partners

a couple of years ago.

JOHN BURGE: The two advantages that we really bring at Google

are the large amount of compute that we

have access to, and the expertise in using it.

Mark and his team have run the model on the compute

that they have available to them, generated

about a million data points.

We were able to take Mark's model, run it

on Google's infrastructure, and we created hundreds

of millions of data points.

And then we've used that data to train up a deep neural network

to come up with an approximation for what Mark's model does

that is just much, much faster.

So Mark's model can make a fire prediction in about a minute.

Ours can make a prediction in about 20 to 30 milliseconds.

And due to how AI works, we're not

limited to just making a single prediction

at a time like Mark's model.

In that same 20 milliseconds, we can actually

make thousands of predictions.

Are these spotted, or are these--

JASON FORTHOFER: The fire behavior modeling is often

used in planning phases.

Maybe there's a community that we're trying to protect

and we want to do some kind of mechanical treatment

or a prescribed burn.

With these new modeling systems, we can simulate maybe 10,000

fires in a community under different weather and wind

conditions and different locations of where the fire

might start, and start to identify areas of high

significance.

Like, if we could change the fuel type in this area,

that would really help firefighters

when they're trying to suppress a fire.

From this new model, we'll be able to teach firefighters what

dryness does to a fire, or wind, or a change in fuel type,

because we have seen so many firefighters die on fires

when they don't anticipate what the fire's gonna do.

And there could be some huge gains from something

like FireSat, where we will get extremely good information

about how fires spread, and we can use

that to forecast fire behavior.

JULIET ROTHENBERG: People turn to Google in times of crisis

to get information about what to do next.

And that's part of the reason why

it's been so important for Google

to have wildfire boundaries as a part of Google Maps and Google

Search.

Our wildfire boundary product shows wildfire boundaries

using satellite imagery and AI.

And what's so exciting about FireSat

is that we can get a real step change in improvement

in our ability to detect fires early

and to detect small fires, which is

very important for helping people stay as safe as possible.

CHRIS VAN ARSDALE: Our hope is that FireSat

will shine a light where there wasn't one

before to really bring data to the world to make sure

that everybody has equitable access to it,

and is able to actually achieve improvements to people's health

and safety.

KATE DARGAN: We were doing a FireSat presentation back

in Washington DC, and the head of the International Association

of Fire Fighters, a Boston character,

was sitting in the audience and--

excuse me if I fumble my Boston accent--

but he said, (BOSTON ACCENT) "Do you

mean to tell me that we'll get satellites for just

the firefighters, just for us?

That's amazing.

I want that to happen."

[MUSIC PLAYING]