JFSP_research_product_search

Project ID:

24-1-01-10

Year:

2024

Date Started:

09/01/2024

Date Completed:

Title:

Machine Learning Classification of Unknown Fire Causes in Western US

Project Proposal Abstract:

1. Problem Statement: Wildfire, hereafter fire, activity in the western United States has increased substantially in the past five decades, due to a variety of factors including longer and more intense dry-hot seasons. Nearly 60% of fire ignitions across the western United States are caused by humans, and are potentially preventable. Human-caused fires generally occur close to human settlements and infrastructure and account for a large fraction of fires' death toll and structure loss. A variety of engineering, education and enforcement strategies are available to prevent human-caused fires, which are specific to the ignition cause. Predictive understanding of fire causes and their drivers is indispensable in reducing fire burdens on management agencies and the public. However, the number of fires associated with an unknown or missing cause have increased in the past decade, accounting for nearly 50% of all fire incidents in recent years. While these fires are mostly considered as human-started, lack of understanding of their specific cause inhibits more effective prevention and response efforts.

2. Objectives: To inform effective fire prevention and response strategies, this project will:
- Develop a state-of-the-art Machine Learning model to probabilistically predict the cause of fires associated with an unknown or missing cause: I will develop an advanced ML model using physical, biological, social and administrative attributes associated with historical fires to classify fire causes, and use it to predict the cause of unknown fires from 1992-2020 and their associated probability/confidence.
- Advance management-relevant, fundamental understanding of factors that drive fire causes across environmental gradients: I will use explainable Artificial Intelligence techniques to reveal the relative importance of various factors in predicting fire causes across the western United States, each western state and each pyrome.

3. Benefits: Improved understanding of fire causes and their drivers is indispensable to inform fire prevention and management efforts, and to save lives, safeguard resources-at-risk, and avert millions-to-billions of dollars in fire direct and indirect costs. This project will advance predictive understanding of fire causes, and will probabilistically assign a cause to fires associated with unknown or missing causes based on a variety of covariates and drivers (e.g., land cover, distance to infrastructure, weather), and thereby inform effective fire prevention. In doing so, this project will leverage the advances in Machine Learning and big data approaches, and help land and fire managers better plan for prevention strategies specific to regional and environmental characteristics, time of the year, and weather conditions, among other factors. The proposed methodology capitalizes on the recently developed FPA FOD-Attributes dataset that includes >300 attributes associated with ~2.3 million fires across the United States between 1992-2020 to develop this model, and draws important inferences about the environmental and anthropogenic factors that shape human and lightning ignitions. This fundamental understanding can inform more effective fire prevention efforts.

Principal Investigator:

Sadegh Mojtaba

Agency/Organization:

Boise State University

Branch or Dept:

Department of Civil Engineering

Other Project Collaborators

Type	Name	Agency/Organization	Branch or Dept
Agreements Contact	Erin Keen	Boise State University	Office of Sponsored Programs
Budget Contact	Erin Keen	Boise State University	Office of Sponsored Programs
Student Investigator	yavar Pourmohamad	Boise State University	Department of Civil Engineering