How Improving Aviation uses Gemini and Google Cloud to predict and combat wildfires

Key Concepts:

• Wildfire modeling
• Ember-caused spot fires
• Fire behavior determinants (topography, fuel models, weather)
• Rothermel equation
• Atmospheric boundary layers
• Machine learning for ignition probability
• Google Cloud (Cloud Run, Kubernetes, BigQuery, Vertex AI, Gemini)
• Real-time data analysis
• Computational power for predictions
• AI for natural disaster prediction

1. The Problem: Limitations of Traditional Wildfire Models

• Traditional wildfire models, some dating back to the 1970s, are effective at predicting fire spread on the ground but struggle with ember-caused spot fires.
• These models often rely on the Rothermel equation, which uses empirical evidence to predict surface fire spread.
• Factors like atmospheric boundary layers, winds, and ember transport are difficult to predict, yet crucial in rapidly evolving wildfires.
• Rocio Frej notes her past research lacked real-world impact until she focused on improving wildfire prediction.

2. Embers and Spot Fires: A Major Challenge

• Embers are red-hot pieces of material lifted by the fire that can start new spot fires.
• Jake Rosenzweig emphasizes the danger of embers drifting from house to house.

3. Factors Influencing Fire Behavior

• Fire behavior is determined by three main components:
  • Topography: Slope, elevation, and aspect of the landscape.
  • Fuel Models: Types and amount of nearby vegetation.
  • Weather: Wind, precipitation, humidity, and temperature.

4. Solution: Leveraging Machine Learning and Google Cloud

• The team is using machine learning to estimate the probability of ignition once embers land, considering weather and topographical conditions.
• Google Cloud is used to overcome computational limitations and deliver real-time predictions to firefighters in the field.
• Before Google Cloud, the code was thousands of lines long and ran on a single computer.
• Alex Rodriguez highlights the ease of using Google Cloud products like Cloud Run and Kubernetes for rapid deployment and visualization.

5. Google Cloud Technologies in Detail

• Gemini: Used to scour the web for historical fire data, collecting variables of interest into a database.
• BigQuery: Processes the collected data.
• Vertex AI: Used for post-analysis and training the machine learning models to improve efficiency and speed.
• Google Kubernetes Engine: Enables running thousands of instances on Google Cloud, providing near real-time predictions to users in the field via a browser-based device.

6. Real-Time Data and Prediction Pipeline

• Rocio Frej describes a pipeline of real-time data coming from drones, sensors, phones, and satellites, feeding into prediction models.
• This pipeline enables predictions in seconds.

7. Impact and Future Applications

• Kevin Speer emphasizes that this work is at the forefront of wildland fire modeling, combining traditional models with new computational techniques.
• The result is a rapid and useful risk assessment for users.
• Rocio Frej believes AI empowers smaller companies to challenge the status quo and build impactful solutions.
• The technology can potentially be expanded to predict other natural disasters.

8. Notable Quotes

• Rocio Frej: "I have been in research my whole life, and I have never done something that really had an impact."
• Kevin Speer: "The work here is at the forefront of wildland fire modeling by bringing together the traditional models with new computational techniques and putting it together in a package that produces, for a user, a rapid and useful assessment of risk."
• Rocio Frej: "AI gives companies like us, smaller companies, the opportunity to break the status quo and build something amazing."

9. Technical Terms

• Rothermel Equation: A mathematical model used to predict the rate of spread of a surface fire.
• Atmospheric Boundary Layer: The lowest part of the atmosphere, directly influenced by the Earth's surface.
• Fuel Models: Classifications of vegetation types and their characteristics that influence fire behavior.
• Machine Learning: A type of artificial intelligence that allows computer systems to learn from data without being explicitly programmed.
• Cloud Run: A managed compute platform that enables you to run stateless containers via web requests or Pub/Sub events.
• Kubernetes: An open-source container orchestration system for automating application deployment, scaling, and management.
• BigQuery: A fully-managed, serverless data warehouse that enables scalable analysis over petabytes of data.
• Vertex AI: A unified machine learning platform that allows you to build, deploy, and scale ML models.
• Gemini: A family of multimodal large language models developed by Google AI.

10. Synthesis/Conclusion

The video highlights the limitations of traditional wildfire models in predicting ember-caused spot fires and presents a solution leveraging machine learning and Google Cloud technologies. By integrating real-time data from various sources and utilizing the computational power of the cloud, the team has developed a system that provides near real-time predictions to firefighters in the field, enabling more effective response and mitigation efforts. The project demonstrates the potential of AI to revolutionize natural disaster prediction and response.