Rare and highly destructive wildfires drive human migration in the U.S.

In recent decades, wildfire destruction of the built environment has grown dramatically, creating a growing threat for human settlements across the U.S. This trend is driven in part by changes in wildfire patterns, which were shown to increase in total acres burned, number of large fires, and length of the fire weather season. Models project that, under climate change, the potential for very large fires will increase in the coming decades. Concurrent with the rise in wildfire frequency and severity, the number of people living in high-risk regions has increased, exposing more dwellings and residents to wildfires. The growing scale of wildfire destruction to buildings has the potential to impact human mobility patterns, yet little is known about the relationship between wildfire destruction and human migration. Wildfire-related mobility is notably absent in systematic reviews of environmental migration literature, a gap highlighted by the IPCC. Prior studies of other hazards show climate–migration relationships vary widely in direction and magnitude by hazard type and by geographic, social, and economic context. This study asks whether wildfires influence human mobility and identifies the destruction threshold at which migration effects become apparent, situating the analysis within the context of increasing wildfire risk and expanding populations in fire-prone areas.

Scholarship on environmental migration documents wide variability in migration responses across hazards and contexts, including non-linear effects. Emerging work emphasizes immobility as a common outcome in the face of environmental hazards due to factors such as place-based social and economic networks, amenity preferences, mitigation capacity, and housing affordability constraints. In wildfire contexts, rising housing costs in urban cores can push households toward more affordable, fire-prone areas, potentially limiting their ability to relocate away from hazard. Amenity-driven preferences also attract some residents to fire-prone places. Studies of extreme disasters (e.g., Hurricanes Katrina and Maria, the Indian Ocean tsunami) show heightened post-disaster out-migration and displacement following very severe events, suggesting a continuum from immobility to large-scale displacement. Recent wildfire studies have examined migration intentions due to wildfire and smoke exposure and relocation decisions in response to fire risk, but often lack integration of simultaneous drivers and civic decision-making dynamics. This study builds on that literature by testing two pathways: (1) direct, damage-driven displacement via structure loss; and (2) indirect effects that alter residential preferences and/or capabilities (via changes in amenities, air quality, economic conditions, risk perceptions, finances, or insurance access).

The study analyzes migration effects of the top decile (N = 519) of the most destructive wildfires that destroyed structures in the contiguous U.S. between 1999 and 2020. Wildfire destruction measures come from the U.S. National Incident Command System/Mitigant Status Summary Forms (ICS), linked to wildfire perimeters from the Monitoring Trends in Burn Severity (MTBS) and FIRE databases to create tract-by-quarter exposure metrics using 2010 census tract boundaries. The ICS provides counts of structures destroyed (residential, commercial, outbuildings, mixed-use), a direct impact measure preferred over monetary damage estimates. Migration measures are quarterly probabilities of out- and in-migration derived from the Federal Reserve Bank of New York/Equifax Consumer Credit Panel (CCP), an anonymized panel of credit-visible adults; movers are identified by address changes and aggregated to tract-level probabilities. The analytic design compares burned (treated) tracts to matched unburned (control) tracts using rings to address potential spatial spillovers: 0–5 miles, 5–25 miles, and 25–50 miles from the burn perimeter. Control tracts with destructive wildfire exposure within the 17-quarter observation window were excluded. To reduce outlier influence, observations with extreme migration probabilities were trimmed (high in-migration outliers; out-migration exceeding the maximum observed after the 2018 Camp Fire). Covariates used for matching captured wildfire-related landscape and settlement characteristics: elevation, slope (NASA SRTM), land cover shares (forest, shrub/scrub, developed; NLCD 2019), tract land area, and 2010 county population; processing used Google Earth Engine. Matching employed coarsened exact matching (CEM) to balance treated and control tracts within each wildfire subset, with post-match standardized mean differences typically ≤0.1. The Camp Fire model used a reduced covariate set due to sample constraints and achieved improved, though not uniformly ≤0.1, balance. Wildfire severity was stratified due to heavy right-skew in destruction: (1) full top decile (n = 519; 14–18,504 structures destroyed), (2) less destructive portion (n = 463; approximately 14–257 structures), (3) more destructive portion (n = 55; approximately 258–700 structures), and (4) the single most destructive event, the 2018 Camp Fire (n = 1; 18,504 structures destroyed). Event-time windows included the event quarter, the first year post-event, and the second year post-event. Difference-in-differences models estimated average treatment effects by comparing migration probabilities in treated versus matched control tracts across these periods, repeated for each control-ring set to assess spatial spillover and robustness. Sensitivity analyses with tract, quarter, and two-way fixed effects and with alternative pre/post period specifications yielded substantively similar results. The analysis was restricted to the contiguous U.S.

• Destruction distribution: Most wildfires (84.4%, N = 29,216) caused no structural damage; among destructive wildfires (15.6%, N = 5406), damage was highly right-skewed. The top two fires accounted for 39.5% of all structure loss; the 2018 Camp Fire alone accounted for 17.2%.
• Threshold and non-linearity: Only the highest severity events showed significant out-migration effects, indicating a non-linear response and a destruction threshold around 258+ structures destroyed.
• Full top decile (14–18,504 structures destroyed; N = 519): Significant positive out-migration in the event quarter, increasing in magnitude in the first year after the event; effects correspond to roughly 4–5 additional movers per 1,000 residents in the first year (depending on control ring).
• Less destructive portion (≈14–257 structures; N = 463): No consistent significant changes in out-migration in the event quarter, first year, or second year, indicating no detectable indirect migration effects at these lower destruction levels.
• More destructive portion (≈258–700 structures; N = 55): Clear out-migration effects. Event quarter: about 4 additional out-migrants per 1,000 residents. First post-fire year: 4–5 additional out-migrants per 1,000 residents (depending on control ring). No significant effects in the second post-fire year.
• Camp Fire (18,504 structures; N = 1): Very large and temporally persistent effects. Event quarter: 53–69 additional out-migrants per 1,000 residents. First post-fire year: approximately 68–83 additional out-migrants per 1,000 residents per quarter (a more than threefold increase from pre-fire). Second post-fire year: still significant, with 13–26 additional out-migrants per 1,000 residents per quarter (depending on control ring).
• In-migration: Across full, less destructive, and more destructive subsets, no significant post-fire differences in in-migration relative to controls. For the Camp Fire, in-migration increased (e.g., ≈30 additional in-migrants per 1,000 residents in the event quarter relative to the 50-mile controls), and remained elevated in the first and second years under some control-ring comparisons, consistent with recovery migration.
• Overall: Migration responses were rare and concentrated in extreme, highly destructive events; most destructive wildfires did not produce population-level migration changes, indicating prevalent immobility.

The analysis shows that wildfire impacts on migration are non-linear and primarily damage-driven: only rare, highly destructive events surpass a threshold where housing and infrastructure loss produce measurable out-migration. For the majority of destructive wildfires, migration patterns remained stable, suggesting that residents’ preferences and capabilities generally favored remaining in place despite hazard exposure. The pronounced and persistent out-migration following the Camp Fire indicates both direct displacement and indirect effects that likely altered residential preferences and capabilities (e.g., through perceived risk, economic conditions, or insurance access). Minimal changes in in-migration for most fires further support the conclusion that indirect avoidance effects were limited over the study window, with the Camp Fire’s elevated in-migration reflecting recovery dynamics. These findings address the study’s hypotheses by confirming a threshold effect of structural loss on out-migration and by highlighting the rarity of wildfire-induced mobility at the population level. They underscore the importance of considering severity when assessing environmental migration and support emerging perspectives emphasizing immobility as a common response to environmental hazards.

The study demonstrates that only rare, highly destructive wildfires in the U.S. between 1999 and 2020 triggered detectable out-migration at the neighborhood scale, confirming a non-linear, threshold-driven relationship between structural loss and mobility. Most destructive wildfires did not alter out- or in-migration, indicating that immobility dominated wildfire responses during the period. Methodologically, the integration of structure-level wildfire impact data with tract-level migration probabilities and matched, multi-ring controls enables robust local-scale causal inference and identification of thresholds. Future research should: (1) investigate individual mechanisms and intentions underlying indirect effects; (2) examine demographic heterogeneity and equity implications with richer microdata; (3) explore within-tract and short-distance moves; (4) assess external validity across regions and events beyond the contiguous U.S. (e.g., Hawaii’s 2023 Maui fires); and (5) evaluate how adaptation, insurance markets, and housing conditions may shift mobility responses under increasing wildfire severity in a warming climate.

• Spatial resolution and scope: Migration was measured at the tract level; within-tract moves were not captured, potentially missing short-distance relocations. Treated tracts may include areas outside the burn footprint, diluting treatment and potentially underestimating effects.
• Data coverage and representation: The CCP includes only credit-visible individuals with SSNs, underrepresenting younger and financially disadvantaged populations and excluding some vulnerable groups (e.g., undocumented residents). Lack of demographic variables (sex, race, ethnicity, nativity) precludes analysis of distributional impacts. Estimates for events like the Camp Fire may be conservative if disproportionately affected groups are underrepresented.
• Wildfire impact measurement: ICS counts structures destroyed but not impacts to wildlands, agricultural lands, or livestock, which could influence migration where livelihoods depend on these sectors.
• Matching and balance: While CEM achieved good balance for most analyses, balance for the single-event Camp Fire model was more limited due to smaller sample sizes. Residual imbalance and unobserved confounding remain possible.
• Spillovers and controls: Spatial spillovers were addressed via multiple control rings, but unmeasured regional dynamics could still affect comparisons. Analyses were restricted to the contiguous U.S.
• Model specifications: Primary models avoided fixed effects based on methodological considerations; sensitivity checks with FE yielded similar results, but causal identification remains quasi-experimental.