Contact Us

Forest fire machine
learning: Enabling
faster response
and aiding forest
fire prediction

Author: Gary Hilson

Traditionally, forest rangers have kept watch over vast, wooded landscapes from the high perches of fire towers and radioed authorities when they spotted smoke plumes above the treetops—however, continuous advancements in forest management, advancements that include forest fire machine learning (ML), have helped bring awareness and improved methods to the cause.

Artificial intelligence (AI) for forest fire prediction plays a vital role in preparing for and responding to wildfires. A review of forest fire machine learning applications in wildfire science and management by the journal Environmental Reviews identifies the following key areas of study and problem domains:

  • Fuels characterization, fire detection and mapping
  • Fire weather and climate change
  • Fire occurrence, susceptibility and risk
  • Fire behavior prediction
  • Fire effects
  • Fire management

Machine learning isn't just changing the business world it is also helping public safety outcomes. Data gathered by Internet of Things (IoT) sensors, drones, satellites and historical data maps are helping to create new fire detection methodologies powered by ML which are helping to aid in both forest fire prediction and with boots on the ground fire responses. 

The cost of forest fires

Forest fires can have a massive and lasting impact on communities, both in economic and social terms. For example, the 2018 California forest fires are estimated to have cost the US economy $148.5 billion.  

The National Interagency Fire Center (NIFC) is the nation’s support center for wildland fires and other emergency situations and it is composed of partners that work together to compile information about the current wildfire situation along with statistics that encompass lands managed by federal, state, local, tribal, and private agencies.

Forests, shrubland and grassland cover more than half of the land area in the U.S. These ecosystems are important environmental and economic resources. Increased temperatures and drought due to climate change threaten to increase the frequency, extent and severity of fires. Multiple studies have found that climate change has already caused an increase in wildfire season length, frequency and burned area. Wildfires are estimated to cause more than 33,000 deaths each year globally.

As the impact of wildfires continues to rise, some public sector agencies are facing challenging budgetary circumstances. Technological advancements in forest fire prediction have the potential to help solve both issues.

How forest fire prediction works

Modern wildfire prediction goes far beyond simply stopping fires—it can also be vital in choosing where to begin and how to tackle fires, determining what vehicles and equipment should be needed, and how to best evacuate individuals in its path. To demonstrate the range of data available for forest fire prediction, consider the US Forest Service Predictive Services team uses of data such as:

  • Fire weather and climate data, including drought and precipitation indexes and climate modeling
  • National Interagency Fire Center statistics on historical wildfires
  • Fuels and fire danger
  • Mapping and satellite imagery
  • On-the-ground intelligence provided by firefighters and other local officials

Researchers working on predicting forest fires are using drones that carry instruments with the ability to capture and measure useful data such as temperature, humidity, location, wind speed and direction.

IoT sensors can also measure relevant metrics such as humidity, temperature, rainfall rates, soil moisture, wind speed and direction. This data is then aggregated to build models designed to determine where fire impacts are more likely to happen.

According to the publication Environmental Reviews, AI for forest fire prediction has traditionally used neural networks and expert systems. But as computational limits are reduced and more data becomes available, it is now possible to use more recent machine learning methods—including deep learning and agent-based learning.

Modern fire detection and the programs that drive them

Pilot programs are popping up all over the western half of the country to push the boundaries of modern fire detection, such as this system in northern California that uses cellular technology and rotating, high-definition cameras to snap 10 hi-def images of the surrounding landscape every 60 seconds to produce 360-degree views of up to 10 miles. Using a time-lapse video stream, this solution scours the landscapes for wisps of smoke or other potential signs of fires.

ALERTWildfire is a modern solution that delivers 24/7 monitoring, real-time data streams and automated email and text alerts for critical incidents. These tools will help firefighters and first responders to discover and locate fires, monitor fire behavior, scale resources, and help with evacuations. The monitoring system will back up the county’s existing network of wildfire detection cameras, which monitor the region around the clock, to not just help to spot wildfires but also to help predict how they will spread. County officials are happy with the upgraded system, although challenges remain, such as distinguishing between planned/permitted burns and wildfires.

In New Mexico, imagery from two U.S. weather satellites helped detect at least 6,000 fires. Images are searched for signs of smoke and hot spots, indicated by shifts in infrared data that could signal the outbreak of a fire. Several algorithms that ran reported that the AI could successfully identify fires in 97% of the images, analyzing factors such as:

  • Spatial context to determine if the fire is hotter than its surroundings
  • What the area looks like without a fire to see if it's changed
  • Areas that are unexpectedly hot compared to historical temperatures

If the algorithm detects a wildfire, it alerts the New Mexico Forestry Department, and the whole process from beginning to end takes approximately nine minutes. 

Fire and smoke maps

Much of the heat mapping and smoke analysis performed today is through infrared satellite imagery and AI. Drones have also helped to aid in the detection and mapping of wildfires. For example, emergency services in California have used thermal sensors to detect fires in early stages or in remote areas. Cameras are being used to help spot dry vegetation likely to be sparked by lightning and paired with AI, detailed wildfire maps can be created to help track and forecast fires. 

Forest fire machine learning technology

The rise of IoT sensors, drones, smart cameras, satellites, and more, has not just helped technology and fire experts converge, it has also created an enormous amount of data.

Improvements in computing power have been key reasons for the enhancements in forest fire machine learning and modeling. Virtual and augmented reality are other relevant technologies that can help create simulations for real-time decision-making and could soon allow firefighters and other officials to visualize machine learning predictions in relation to the relevant landscape.

While collecting data is crucial for forest fire machine learning purposes, quickly disseminating and assessing the data, for instance with edge computing, will be crucial to enabling rapid response, helping first responders to combat fires and evacuate those in the path of danger. Advances in connectivity, like the expansion of 5G coverage, will continue to power solutions that make it easier for public safety officials to do their primary job: saving lives.

Learn more about how Verizon helps respond to western wildfires.

The author of this content is a paid contributor for Verizon.