Our understanding, planning, and response to wildfires benefit from connections with data and computing sciences, UC Berkeley researchers say.
Recent developments in machine learning and simulations can help first responders detect fires earlier, predict fires’ paths and limit blazes quickly. Through collaborations with practitioners in other fields like microbiology and forest management, these tools are answering previously intractable questions about fires that can inform policy and practice.
“Data science is this common language that we all need to speak in order to get this really full, rich picture,” said Ciera Martinez, research lead for biodiversity and environmental sciences at the Berkeley Institute for Data Science (BIDS). “The more information we have, the more accurate our predictions of the future will be and the more prepared we will be.”
This shift comes as climate change fuels more frequent and extreme natural disasters including wildfires. Four of California’s five largest wildfires ever occurred in 2020 alone. These events devastate property and people, exacerbate existing inequalities and harm human health.
Algorithms, Machine Learning, and Simulation for Fires
Between increasingly available imagery from satellites, footage from drones, and even posts on social media, more real-time information about fires exists now than ever before. Data scientists use machine learning to quickly combine, analyze and interpret that data for first responders.
Berkeley Professor of Mechanical Engineering and Fire Research Group Founding Director Tarek Zohdi is working to embed such data into so-called digital twins of fire scenarios. These "twins" are virtual replicas of the fire's behavior, showing factors like fire location, how intense it is and how fast it's spreading.The simulations then account for variables such as wind speed in the fire's current environment and predict the fire’s future behavior, rapidly advising responders on the safest and most effective flight paths for pilots fighting the flames.
“If you tell a person, ‘Look, here's where things are going to happen and this is what's going to occur,’ a few minutes or even hours before, that's pretty valuable,” said Zohdi, chair of Berkeley’s Designated Emphasis in Computational and Data Science and Engineering Program.
Researchers also use data science to understand the impact of wildfires over time. BIDS-Accenture Global Environmental Change Research Fellow Cody Markelz is looking at how fires have affected the biodiversity of certain ecological reserves in California.
He and his research partners are collecting DNA from those reserves, then sequencing and combining it with other data to model how the ecosystems changed and under what conditions. This work allows them to predict what kinds of ecosystems are most likely to recover from fires and why.
“Hopefully, those predictions can help us prioritize how we use our resources collectively as a society to manage these forests and grasslands in California,” Markelz said.
Shedding New Light on Existing Data
Data analysis can aid wildfire prevention efforts, too. Rob York, adjunct associate professor of forestry in the Department of Environmental Science, Policy, and Management, reviewed decades of weather records with data-focused researchers to identify which days had optimal prescribed burn conditions. These planned fires can limit the intensity of future unplanned fires.
In research that’s currently under peer review, they found there have been more winter days with good burn conditions than the typical burn seasons of spring or fall. That new information could help solve the long-standing dilemma of not having enough days for these burns, York said.
“This study did need a collaboration between a practitioner -- someone who could identify the practical conditions under which a burn is feasible -- and someone on the data side, who can take what the practitioner tells them is the reality, search these huge databases and hone in on when prescribed fires do work,” said York, a UC Cooperative Extension forestry specialist.
It’s relatively new to combine so many diverse types of data totackle wildfires. Martinez and Markelz help scaffold this work by building and sharing data science infrastructure from databases and documentation to standards for using tools and collaborating with others
It’s only within the last two decades that the data science and fire community realized how essential tools like simulation are to this work, Zohdi said. Progressive businesses and new climate analytics firms are now using them in the field.
“Simulation and data science play a key role in firefighting,” Zohdi said. “It’s just too unsafe to do it based solely upon real, physical events. You have to have some type of simulation capabilities to predict where things are going to occur and what to do about them beforehand.”