Wildfires

Using AI to provide accurate wildfire information to affected communities and fire authorities

Using AI to provide accurate wildfire information to affected communities and fire authorities

Wildfires around the world are becoming more frequent and more dangerous, especially with the rise of global temperatures. Their effects are felt by many communities as people evacuate their homes or suffer harm even from proximity to the fire and smoke. 

Google Research is applying AI and ML to better understand the Earth's physical processes and enhance our ability to react to natural disasters such as wildfires. To this end, Research teams are advancing a number of projects and efforts with diverse approaches in collaboration with wildfire experts - by using satellite imagery and ML to detect and track wildfires, by informing affected communities and helping fire authorities to take action, and by developing a simulator to generate data in a range of wildfire scenarios. 

This is part of Google's broad effort to use AI to address the growing threat of climate change, in our mission to apply AI to improve lives globally and help people access trusted information in critical moments.

Wildfire Boundary Tracking

We have developed a wildfire boundary tracker that uses AI and satellite imagery to map the boundaries of large wildfires and display it on Google Maps and Google Search. This wildfire tracker is available in parts of the US, Canada, Mexico and Australia, helping people to stay informed about potential dangers near them or their loved ones. We are advancing our research to expand coverage to more regions and countries.

Providing people with critical information during active wildfires: Google Search & Maps

We make the wildfire information accessible to users via SOS Alerts on Search and Maps.

How it works

  1. Satellite imagery is gathered: Using geostationary, lower earth orbit satellites and other data sources that provide continual coverage for a portion of the Earth. Satellites include GOES-16 and GOES-18 for North and South America, and Himawari-9 and GK2A for Australia, Suomi NPP and NOAA-20 satellites with the VIIRS imager in other locations.

  2. An ML-powered system analyzes the continent-scale images: By receiving a sequence of the three most recent images to compensate for temporary obstructions and inputs from two satellites

  3. An ML system detects wildfire boundaries: By using a neural network to identify the total burnt area of the fire 

Read our blog post to learn more.

Wildfire simulation  

We are also leveraging Google’s Compute, TPUs, and ML to bring improvements into wildfire prediction. Our Google Research team has developed a large-scale high-fidelity wildfire simulator (to appear in IJWF) that can be used to generate large quantities of data in a wide range of wildfire scenarios. 

This effectively addresses the data sparsity issue and allows for better ML being developed to address various fire predictions, such as early warning for extreme fire development.

A controlled burn field experiment (FireFlux II) simulated using Google’s high fidelity fire simulator(to appear in IJWF).

Fire spread and propagation models

In a related effort, we are working in close collaboration with the US Forest Service Fire Lab, using Google parallel computing and ML technologies, to develop the next generation of operational fires spread models. A Google-US Forest Service joint presentation in ICFFR 2022 and the USFS Fire Lab's project page provides a glimpse into the ongoing work.  

Our next challenge is to look at use cases and train ML-based models to incorporate weather and satellite data into forecasts of fire propagation, and to be able to predict fire propagation.

Latest publications

A High-resolution Large-eddy Simulation Framework for Wildland Fire Predictions using TensorFlow, Qing Wang, Matthias Ihme, Rod R. Linn, Yi-Fan Chen, Vivian Yang, Fei Sha, Craig Clements, Jenna S. McDanold, John Anderson

"Deep Learning for High-Resolution Wildfire Modeling” ICFFR 2022,  Mark A. Finney, Jason M. Forthofer, Xinle Liu, John Burge, Matthias Ihme, Fei Sha, Yi-fan Chen, Jason Hickey, John Anderson

"Towards real-time predictions of large-scale wildfire scenarios…” ICFFR 2022, Qing Wang, Matthias Ihme, Yi-fan Chen, Vivian Yang, Fei Sha, John Anderson

"Recurrent Convolutional Deep Neural Networks for Modeling Time-resolved Wildfire Spread Behavior” Fire Technology 2023, John Burge, Matthew Bonanni, Matthias Ihme, Lily Hu