Currently, a fast-moving wildfire is destroying part of an old mountain town called Jasper in the Canadian Rockies. The damage in Jasper is described as “beyond description and comprehension,” and 25,000 people have already been forced to flee.
Huge wildfires have destroyed up to half of the historic Canadian town of Jasper. At the same time, hundreds of blazes are active in neighboring British Columbia as well as in California and Utah.
Every year, thousands of people bear the brunt of wildfires, with towns getting incinerated by fire. In the US, over the last five years, 344 significant wildfires have killed 178 people. Meanwhile, since the 1980s, the country has had an average of about 70K total fires yearly. According to the National Interagency Fire Center (NIFC), a total of 2.7 million wildfires have occurred since 1983.
As the number of fires continues to increase, the cost to fight them has also been on the rise. The US government spent almost $4.4 bln fighting wildfires in 2021 alone. On average, it costs around $74K to put out each wildfire. But this is not all; homeowners insurance rates are also going up.
Wildfires have been on the rise worldwide in not just their frequency but also their severity and duration. The risk of wildfire is mostly in extremely dry conditions like high winds, drought, and heat waves. This means a growing adverse impact on human and wildlife health.
Besides the destruction caused by the fire itself, the smoke from wildfires consists of hazardous air pollutants such as NO2, ozone, aromatic hydrocarbons, lead, or PM2.5, which are particles with a diameter of 2.5 microns or less. PM2.5 is so fine that it can travel deeper into our respiratory tract and cause health effects.
PM2.5 from wildfire smoke is actually associated with premature deaths in the general population. It can cause and exacerbate diseases of the lungs, kidneys, heart, skin, gut, eyes, nose, and liver. It has further been shown to lead to cognitive impairment and memory loss.
Not only do wildfires contaminate the air with toxic pollutants, but they also impact the climate by releasing large quantities of CO2 and other greenhouse gases into the atmosphere. By blanking the Earth, greenhouse gas emissions actually trap the sun’s heat and make it inhabitable. Too much of it leads to climate change, which leads to drier conditions, further contributing to longer durations of fire seasons. Hence creating a feedback loop.
With wildfires becoming increasingly extreme in intensity and damage, they can disrupt the infrastructure in terms of communications, transportation, power supply, gas services, and water supply. Moreover, near populated areas, they significantly impact property, livestock, human mortality, and the environment.
Last year, a study by researchers at the University of Maryland calculated that forest fires now result in 3 million more hectares of tree cover loss per year compared to 2001. By causing 9.3 million hectares of tree cover loss globally, 2021 became one of the worst years.
It’s not that fires aren’t a natural part of how forests function. Actually, boreal forests, which cover vast tracts of land across northern Europe, Russia, Alaska, and Canada, are adapted to fire. Fire creates a variety of landscapes and removes old forests, giving birth to new ones.
However, fire-related tree cover loss, even in these areas, is happening at an accelerated rate. With boreal forests being among the world’s leading suppliers of clean water, harboring significant wildlife populations, and an important carbon sink, the ongoing trend of wildfires is a worrying one and needs to be addressed.
Click here for a list of biotech companies working on solutions to tackle global warming.
Leveraging AI Models in Combination with Satellite Images
One of the ways the issue of wildfires is being tackled is through the use of artificial intelligence (AI). This new tech is not only used for chatbots and virtual assistants but also to help the environment through advanced remote sensing.
Recently, AI helped firefighters respond quickly to a wildfire in a remote area of the Bay Area, California. The wildfire was stopped before it spread with the help of an AI-powered camera, which detected smoke and informed firefighting crews before the blaze was able to grow and became hard to control.
Even insurance providers are turning to AI. The San Francisco-based Delos Insurance Solutions has sold over 17K policies to homeowners, which other firms have rejected. This has been made possible through the company’s wildfire models, for which Delos works with university professors to get proprietary datasets on wildfire modeling and then adds an AI algorithm to incorporate more data. According to co-founder Shanna McIntyre, Delos considers over 200 variables covering vegetation, topography, and temperature.
There is clearly a growing focus on the development of high-resolution wildfire behavior models to forecast the fire spread.
This is being done in conjunction with satellites to detect fire locations. While satellites like LANDSAT, VHRS, and MODIS are already used for monitoring the distribution and disturbance of vegetation, predicting wildfires remains a challenging task. AI and machine learning can help provide a reliable way to predict and manage wildfires.
For instance, AI can be linked with imaging satellites that analyze different factors such as smoke emergence, fire incidences, and vegetation disturbance. By combining all the data together, AI algorithms can help predict wildfires through continuous learning. AI tech can further allow for the management of fire extinguishers and patrol vehicles with video cameras.
A year ago, a team of researchers from Aalto University in Finland began developing an AI model called FireCNN that can predict the spread of wildfires. The model is trained on satellite images and weather data recorded between 2002 and 2019 to identify high-risk areas. FireCNN analyzes 31 variables, including factors like vegetation and land cover.
However, there is a lack of AI algorithms that can correlate all the available data and realistic modeling methodologies that suit the needs of different landscapes across the world.
According to a new study, a new model can accurately predict wildfire spread. The model combines satellite imagery and AI to offer a major advancement in wildfire management and emergency response.
The study, funded by NASA, the Army Research Office, and the Viterbi CURVE program, was conducted by USC researchers and published in Earth System Science Data.
Under this new model, satellite data is utilized to track the progression of wildfires in real time. This information is fed into a sophisticated computer algorithm that can accurately forecast the likely path of the fire as well as its intensity and rate of growth.
“This model represents an important step forward in our ability to combat wildfires. By offering more precise and timely data, our tool strengthens the efforts of firefighters and evacuation teams battling wildfires on the front lines.
– Bryan Shaddy, a corresponding author of the study and a doctoral student at the USC Viterbi School of Engineering
The study began by collecting historical data from high-res satellite images. Then, through behavior analysis of past wildfires, the researchers tracked how the fire started, spread, and eventually was contained.
This careful study unraveled patterns that are influenced by factors such as terrain, weather, and fuel, like trees and brush.
The researchers then trained a gen AI-powered model to simulate how these factors impact the wildfires’ evolution over time. The model was then trained to recognize patterns in the satellite images matching the wildfire spread in their model.
The model is called a conditional Wasserstein Generative Adversarial Network or cWGAN, which has already been tested on four real wildfires in California over two years, between 2020 and 2022, to see just how effective it is in predicting the fire spread. Trained with WRF-SFIRE simulations, cWGAN is used to work out the arrival time of fire from satellite active fire data.
Results from cWGAN were impressive. Trained on simple data under ideal conditions, such as unidirectional wind and flat terrain, the model performed well in real California wildfires.
The model’s success is attributed to using the AI model along with satellite imagery, which provided actual wildfire data instead of on its own.
This hasn’t been easy, though. Study co-author Assad Oberai, a Professor of Aerospace and Mechanical Engineering at USC Viterbi, stated that modeling wildfires was among the most challenging due to their “intricate processes.”
Not only fuel, such as trees and shrubs, ignites, leading to complex chemical reactions, producing heat and wind currents, but even weather and topography influence fire behavior. Assad said:
“These are highly complex, chaotic, and nonlinear processes. To model them accurately, you need to account for all these different factors. You need advanced computing.”
Enhanced Data for AI Models
Besides advancements in AI tech, data to train these models are also being improved by covering hundreds of additional factors that impact the ignition and spread of fire. This will help wildlife managers and scientists make more accurate predictions about where and when wildfires may occur.
Recently, the Fire Program Analysis Fire-Occurrence Database (FPA FOD), which is the most extensive source of data on georeferenced fire ignition in the US, was enhanced significantly. First developed in 2013 by the US Forest Service, the model has been updated five times. It aggregates fire reports from local, state, and federal entities with fire protection and reporting responsibilities.
The model comprises basic information such as the location of ignition, date of discovery, and the size of the final wildfire. The revised database, however, will cover even more social and environmental factors. According to Erica Fleishman, an Oregon State University professor:
“There is a tremendous amount of interest in what enables wildfire ignitions and what can be done to prevent them. This database increases the ability to access relevant information and contribute to wildfire preparedness and prevention.”
This enhanced database will not just help on-the-ground firefighters and managers fight the fire, but it could also aid in assessing short-term power risk. Power companies can make more informed decisions regarding implementing a public safety power shutoff. Moreover, land management agencies can decide if they should reduce access to public lands during certain times of the year.
According to Fleishman, a lot of policies are guided by emotions instead of a large body of evidence. The database presents “one way to increase the objective evidence to consider when making those decisions,” she said.
Published in the journal Earth System Science Data, the research outlining the database was supported by the Joint Fire Science Program. It outlined biological, social, and physical attributes for gaining an improved understanding of wildfires as well as prediction.
In total, the research team, which included Fleishman, who directs the Oregon Climate Change Research Institute and was led by Mojtaba Sadegh and Yavar Pourmohamad, an associate professor and a doctoral student, respectively, at Boise State University, added almost 270 additional attributes. Now, the database includes information on 2.3 million fires in the US from 1992 to 2020.
Under the revised database, each wildfire now has physical attributes such as weather, climate, topography, and infrastructure; biological attributes like vegetation index and land coverage; administrative attributes covering national and regional preparedness level and jurisdiction; and social attributes including social vulnerability index and population density.
The sixth version of FPA FOD was further augmented with summary statistics within a temporal and spatial buffer around the point of ignition and vegetation indices obtained monthly from satellites during a year prior to the ignition. “This rich, tabular dataset can be used in a variety of hypothesis-driven or data-exploration applications,” the study said.
Such a massive database will help provide a significantly deeper understanding of the individual and compounded impact of the attributes on wildfire ignitions and size, said Pourmohamad, adding:
“It also identifies the unequal effects of wildfires on distinct human populations and ecosystems, which can, in turn, inform efforts to reduce inequities.”
The database can also be incorporated into AI and machine learning models that explain the factors driving past fires, the likelihood of future fires, and the effects of future fires. She said:
“It’s amazing what you can infer when you have the computational capacity and this much information. You can ask a lot of questions that inform different actions in different places and to understand what is associated with wildfire ignitions and fire effects.”
Final Thoughts
Wildlands such as forests or grasslands are integral to human prosperity. After all, they mitigate climate change by storing carbon, preserving wildlife habitat, and enhancing landscape resilience. These wildlands have always been subject to wildfires, but over the past few decades, they have increased so much that wildfires have become a threat.
These massive, unplanned fires are ravaging lands all across the world, calling for a better understanding, assessment, and analysis of past fires, their successful prevention, and mitigation strategies to not only prevent wildfires but also improve fire planning, response, adaptation, cost-effective mitigation, and limiting the adverse impacts.
As we noted above, authorities are constantly enhancing databases, which can then be incorporated into AI models to explain fire likelihoods and their future effects. These well-tested AI models can then be utilized in conjunction with satellite images to predict and prevent future wildfires accurately. This way, we can protect our wildlands, safeguard biodiversity, and enrich lives for future generations.
Click here to learn all about investing in artificial intelligence.