Spatiotemporal Heterogeneous Change of Travel Behavior during Wildfires in California

Term Start:

July 1, 2025

Term End:

May 31, 2026

Budget:

$130,000

Keywords:

Geospatial Machine Learning, Spatiotemporal Travel Behavior, Wildfire impacts

Thrust Area(s):

Data Modeling and Analytic Tools, Understanding User Needs

University Lead:

California State Polytechnic University, Pomona

Researcher(s):

Yangsong Gu

California suffers from frequent and intense wildfires every year due to its unique weather patterns and dense, flammable vegetation. Wildfires often cause significant infrastructure damage, disrupt roadway networks, interrupt business operations, and generate hazardous smoke, all of which can deteriorate the driving environment, elevate perceived safety risks, limit individuals’ ability to meet daily work-life demands, and ultimately reshape individuals’ travel behavior. Current research related to travel behavior only limits the tourist and evacuee behavior during wildfires. For example, a couple of surveys conducted in Florida revealed that wildfires can lead to changes in destination plans, particularly if smoke is present or road closures are threatened. By contrast, evacuee behavior occurs in areas affected by wildfires. Several studies highlight the factors influencing individuals’ leave, stay, or defending decisions, as well as evacuation timing, destination choice, and route choice during wildfires.  However, those travel behavior studies were only concentrated within specific timeframes and individual groups in reaction to wildfire events. Most wildfire-affected individuals are residents who live in or downwind of the burn zones. They may experience varying impacts across phases of wildfires and respond differently due to their socio-economic and demographic conditions. Unlike tourists engaging in discretionary travel or evacuees responding to mandatory orders, resident travel is often necessary to meet daily work-life demands. For example, the Palisades and Eaton fires that occurred in Los Angeles in January of 2025 affected over 1,800 businesses, with most impacted industries closely associated with residents’ daily lives including retail trade, health care and social assistance, technical services, and construction. Due to widespread wildfire smoke, the air quality in devastated areas ever reached hazardous levels, and in downwind zones like Huntington Park, air quality consistently ranked “unhealthy for sensitive groups” or worse nearly 98% of the time. The wildfire smoke exposure could cause a series of adverse health and social outcomes. As of now, no attempt has been made to capture the spatiotemporal heterogeneous change of travel behavior in response to wildfires and their interplay with sociodemographic, built environment and smoke exposure remains largely underexplored. Understanding this change in travel behavior and heterogenous effect is critically important for effective wildfire management and policymaking. Insights from analyzing travel behavior can be used to optimize resource allocation and recovery efforts for affected communities.

In this context, we examine the spatiotemporal dynamics of human travel behavior by analyzing trip purposes and frequencies, as reflected in destination types and visitations, trip start times, and dwell times at destinations from the onset of the wildfire through potential returns to pre-wildfire patterns.  We use the foot traffic data provided by Advan Research and Point of Interest (POI) data from SafeGraph to measure the weekday trips within and near the wildfire areas. Foot traffic data are passively and unobtrusively collected through mobile devices and used as a proxy of travel demand at point-of-interest (POI) and census block group levels. We begin with the analysis of the Los Angeles wildfires that occurred on January 7-31, 2025, and subsequently apply the methodologies to retrospect wildfires that took place across California between 2018 and 2024. The 2025 Los Angeles wildfires have resulted in total property damages estimated between $28 billion and $53.8 billion. This case is particularly unique as it is one of the most destructive wildfires that devastated multiple communities with contrasting socio-demographic and economic conditions. Besides, ancillary data including wildfire perimeter data from CalFire, sociodemographic data from the 2019-2023 American Community Survey (ACS), air quality data from CARB, and traffic data from INRIX will be gathered to support the research tasks, which are as follows: First, we quantify the change of travel behavior at census block group and weekly level since the onset of wildfires. A robust pre-wildfire baseline is established by excluding the effects of traffic seasonal patterns and similar disasters. Travel behavior is categorized by the frequency of trips to work locations, grocery stores, restaurants, and tourist attractions. Trip start times and dwell times at destinations are also concurrently analyzed to capture behavioral changes related to perceived exposure to wildfires smoke. Second, by combining the sociodemographic, built environment, and wildfire smoke exposure information of affected communities, an explainable geospatial machine learning model (e.g., geographically weighted random forest with SHAP) is developed to tackle the heterogeneous impacts of wildfires across communities. This research ultimately aims to inform localized wildfire management strategies that mitigate the spatiotemporal heterogeneity of wildfire impacts on travel behavior.

Project scope

Scroll to Top