Urban areas in the United States experience substantial wildfire impacts

Abstract

How common are urban wildfires relative to those occurring in less populated places? Although they are not a new phenomenon, the prevalence of urban wildfires has been underreported in prior demographic analyses due to historic data limitations. Here we investigate a range of wildfire impacts in the United States, demonstrating how linked administrative and spatial data sources can enhance our understanding of human exposure to wildfire, especially in urbanized settings. From 1999 to 2020, wildfires consistently occurred across the rural-urban continuum, but with fatalities and structure losses concentrated in more urbanized places. Nearly three quarters of all structures destroyed by wildfires were in metropolitan or micropolitan census tracts. In contrast, wildfires burned nearly 2.5 times as much land area in rural census tracts as in metropolitan census tracts. We conclude that whether wildfire exposure is understood as more urban or more rural depends on the measurement of impact used.

From 1999 to 2020, wildfires consistently occurred across the rural-urban continuum, but fatalities and structural losses were concentrated in more urbanized areas in the United States, according to an analysis of data on wildfire incidents and management records.

Introduction

In recent years, disasters, such as the 2023 wildfires in Maui, Hawaii and the 2025 wildfires that burned across Los Angeles, California have drawn attention to the impacts of wildfires in urban areas. During these events, fires sparked in wildland vegetation have ignited ‘urban conflagrations,’ in which structures themselves become a primary source of fuel that drives fire spread^1,2. Outside of the U.S., urban wildfires have also ignited in places, such as Alberta, Canada³, Athens, Greece⁴, Valparaíso, Chile⁵, and Cape Town, South Africa⁶. These high-profile incidents have highlighted the fact that wildfire is not an exclusively rural phenomenon. Rather, wildfire occurs across the rural-urban continuum, or, the spectrum of population sizes and economic structures that ranges from dense urban cores to sparsely populated rural areas⁷.

As wildfires pose increasing threats to human communities, fire-focused social science research has emerged to investigate the demographic characteristics of exposed populations⁸. But, despite substantial growth in this field, we still lack a comprehensive understanding of the relative urbanicity versus rurality of wildfire impacts. How commonly do wildfire impacts occur in urbanized versus less populated places? Do these trends vary across geographic regions? And are wildfires occurring in urban areas more frequently now than in the past?

Alongside the growing attention to urban wildfires, physical scientists are making increasingly nuanced distinctions between qualitatively different wildfire types. On one hand are fires, such as those in Maui and Los Angeles - fast-moving, causing fatalities and widespread destruction^1,9,10. But on the other hand, many wildfires are considered beneficial for the ecosystems in which they occur, leading land managers and firefighters to embrace fire management rather than full suppression^11,12. In fact, over the past few decades, our understanding of fire’s crucial role in preserving ecosystem health has undergone a considerable revival. This has been made possible by increased recognition of the historical and contemporary practices of Indigenous peoples’ fire stewardship and cultural burning^13,14. Taken together, these two lines of research suggest that generalized measures of wildfire presence on a landscape collapse important distinctions around fires that have the potential to harm human communities and ecosystems, and fires that support improved ecosystem health.

Population research on wildfires has not kept pace with these important biophysical science distinctions. Much existing research on wildfires and population exposure relies predominantly on off-the-shelf wildfire data which measure wildfire risk or occurrence, overlooking important differences in types of wildfire impacts. As a result, we argue, wildfire impacts in more urbanized places have been underexamined and, in many cases, categorically excluded from demographic analyses of wildfire exposure. Existing demographic knowledge of wildfire exposure in the U.S. is therefore incomplete. While we focus our inquiry on the U.S., this issue carries international relevance, given recent wildfire impacts in urbanized places across the globe.

We utilize a relatively new wildfire dataset, the ICS-209-PLUS, to analyze a range of wildfire impacts across the rural-urban continuum in the U.S. from 1999 to 2020. The ICS-209-PLUS combines a comprehensive set of administrative reports on wildfire management with wildfire burn footprints and points of origin that delineate precise locations of each event¹⁵. These data enable the spatial precision needed to identify smaller events and further document different types of wildfire characteristics (e.g. size and speed) and impacts (e.g. structures destroyed, fatalities). Our analysis makes two contributions to the growing body of demographic research on population exposure to wildfire. We present a systematic evaluation of the relative urbanicity versus rurality of wildfires across the contiguous U.S., drawing on integrated administrative and spatial wildfire data sources that address a rural bias in other, more commonly-used wildfire datasets. Second, in line with biophysical wildfire science’s emphasis on important distinctions among wildfire characteristics, we analyze distinct measures of wildfire impact: total wildfire occurrence, area burned, destructive wildfires (1 or more building destroyed), total structures destroyed, fast-moving wildfires (daily maximum fire spread rate of 1,620 hectares or more), and fatal wildfires (1 or more fatality). By evaluating this range of different wildfire measures, we illustrate that the relative impact of wildfire across the rural-urban continuum depends on the outcome that is measured, suggesting nuanced experiences of wildfire across different communities.

Limitations of existing wildfire data

An emerging body of research investigates the relationships between wildfire exposure and population dynamics⁸. This literature primarily investigates two areas. First, it considers differential wildfire exposure and risk among demographic subpopulations, with a special emphasis on measures of social vulnerability^16–19. Social vulnerability is an analytical concept that refers to an individual or community’s exposure to a hazard, their susceptibility to harm when exposed, and their capacity to adapt to the exposure’s impacts²⁰. Second, this literature investigates human development patterns within the “wildland-urban interface” (WUI), the area where housing and wildlands meet^21–25. While both offer measures of human presence on fire-prone landscapes, the WUI is distinct from the rural-urban continuum measure used in this study. The WUI measures development morphology, or, the spatial form of buildings amidst wildlands, while the rural-urban continuum measures a combination of population density, economic structure, and residential commuting patterns. Existing literature has examined wildfires and a broad range of demographic characteristics across the U.S., however, to our knowledge, there has been limited systematic evaluation of wildfire impacts across the rural-urban continuum. We build on the work of Winkler and Mockrin and Kramer et al., who are among the few to include measures of urbanicity in their analysis of wildfire risk in the U.S^26,27.

Wildfires are often implicitly modeled as a rural phenomenon, with a number of studies relying on wildfire data sources that exclude small fires or exclude urbanized places. Modeled wildfire risk grids, in particular, the U.S. Forest Service’s Wildfire Hazard Potential (WHP), are some of the longest-used sources of wildfire data in national demographic analyses of wildfire exposure in the U.S^16,17,28. The WHP is a 270 m resolution raster risk grid produced by inputs from the U.S. Forest Service’s Large Fire Simulator (Fsim) and fuels models²⁹. Drawing on WHP data for their examination of social vulnerability to wildfire risk, Davies et al. note that WHP values are generally lower in urban areas¹⁷. The authors’ exclusion criteria of census tracts that have only moderate to high WHP values therefore “effectively eliminates the most urbanized areas that lack fuels and other conditions necessary for wildfire”¹⁷. Yet, recent urban conflagrations, such as those across Maui and Los Angeles demonstrate clearly the ways that structures themselves can readily function as fuels². The WHP was designed to capture the potential for large wildfires¹⁸ and relies heavily on underlying fuels models that emphasize levels of flammable vegetation²⁹. It does not include buildings as potential fuel sources³⁰. While newer U.S. Forest Service data products do more directly model for potential structure loss (e.g. the Risk to Potential Structures database), the WHP dataset that has been used in prior social science research was not designed to capture burning in more built-up areas. The use of WHP wildfire data in demographic analyses will therefore—by design—show much stronger risk values in less densely populated, more rural places.

More recent research on wildfires and social vulnerability has shifted away from using modeled wildfire risk and towards measures of actual wildfire occurrence^28,31–34. Many of these studies rely on the Monitoring Trends in Burn Severity (MTBS) database, which provides spatial data of wildfire burn footprints from 1984 to the present³⁵. While measures of actual wildfire occurrence avoid some of the limitations associated with using modeled risk data, such as the WHP, the MTBS also uses an event exclusion criterion that risks biasing results away from more urbanized areas. The MTBS sets wildfire burned area size thresholds for inclusion at 1000 or more acres in the western U.S. and 500 or more acres in the eastern U.S³⁵. In short, smaller wildfires are categorically excluded from the MTBS database.

Yet, there is reason to believe that wildfires in close proximity to denser built environments are more likely to be quickly suppressed due to the risks they pose, meaning that these events would likely be smaller in size. Limited research in this area has found, for example, that in the more densely populated eastern U.S., small wildfires dominate the landscape³⁶, and in a nationwide study, researchers found that fires in the wildland-urban interface tend to be smaller than in wildlands³⁷. The importance and challenges of adequately documenting small wildfires extends to Europe as well; in a recent study in Sweden, researchers showed that small fires were more likely to cause structure loss or human injuries³⁸. And in the Netherlands, most wildfires are smaller in burn area than the size threshold established for inclusion in Europe’s primary wildfire information system, raising concerns about under-reporting of wildfire incidents³⁹.

In the U.S., predominant measures of wildfire occurrence used in research on wildfires and population exposure overlook small area events^35,37. As we begin to disentangle different types of wildfires - for instance, differentiating between events that result in structure loss or fatalities from those that do not—what if the small wildfires are of societal importance? In our final analysis, we compare wildfire impacts from the full population of wildfires included in the ICS-209-PLUS database to the subset of larger area wildfires included in the MTBS database. We show that small wildfires in the U.S. are societally important in that they make up nearly half of all destructive events over the two decades examined. Our analysis suggests that much existing research has mischaracterized the geography of wildfire exposure as categorically non-urban. We demonstrate a new approach to documenting different types of wildfire impacts and conclude that population scientists working in this area ought to consider which types of wildfire exposure should be measured when answering the specific research question at hand.

Results

Aggregate wildfire impacts across the rural-urban continuum

We first present descriptive measures of wildfire impacts from 1999 to 2020, disaggregated by level of urbanicity and measured at the census tract scale. U.S. census tracts are official spatial units that generally encompass between 1200 and 8000 residents and nest cleanly into counties, with an ‘optimal’ size of 4000 residents⁴⁰. For this and all subsequent analyses, we allocate census tracts into four categories based on the U.S. Department of Agriculture’s Rural-Urban Commuting Area (RUCA) codes, which include metropolitan, micropolitan, small town, and rural designations. RUCA categories account for tracts’ population size as well as their residents’ commuting flows (see Methods Table 1 for additional details on RUCA categories). In line with scholarship from rural sociology, we thus conceptualize rurality as a two-dimensional construct represented by not just low population density, but also the degree of connectivity a geographic unit has to urban services and amenities^7,41. The use of these four categories highlights the fact that, while many researchers rely on a rural-urban binary, the distribution of human settlement is better described as a spectrum, with urbanicity increasing as settlements range in density and commuting patterns from rural up to metropolitan.

Table 1.

Classification of USDA Rural-Urban Commuting Area (RUCA) codes used for analysis

RUCA Codes	RUCA Classification	Description
1, 2, 3	Metropolitan	Primary commuting flow is within an urbanized area (1), 30% or more to an urbanized area (2), or 10-30% to an urbanized area (3)
4, 5, 6	Micropolitan	Primary commuting flow is within a large urban cluster of 10,000-49,999 (4), 30% or more to a large urban cluster (5), or 10-30% to a large urban cluster (6)
7, 8, 9	Small town	Primary commuting flow within a small urban cluster of 2500–9999 (7), 30% or more to a small urban cluster (8), or 10–30% to a small urban cluster
10, 99	Rural or zero population	Primary commuting flow to a tract outside of an urban area or urban cluster (10), tract has zero population (99)

Source: U.S. Department of Agriculture⁵⁶.

Our findings illustrate that the measurement selected to describe wildfire impacts will substantively influence whether wildfires are understood to be a more rural or urban phenomenon. In Fig.1 (top row), counts of census tracts burned by any wildfire or by fast-moving wildfires have an approximately bimodal distribution, with high counts of both metropolitan and rural tracts, and lower counts of micropolitan and small town tracts. The measure of total square miles burned shows a decidedly rural pattern, with nearly two and a half times as many square miles burning in rural tracts than in metropolitan tracts. In contrast, counts of tracts burned by fatal wildfires, destructive wildfires, and total structures destroyed show a far more metropolitan pattern (Fig.1, bottom row). Between 1999 and 2020, there were 2.3 times as many metropolitan tracts affected by fatal wildfires as there were rural tracts that experienced fatal fires and 63% more metropolitan tracts than there were rural tracts affected by destructive fires. Out of over 106,000 total structures destroyed within the study period, 74% of these losses occurred in metropolitan or micropolitan tracts compared to 26% that occurred in tracts designated as small town or rural.

Fig. 1. Aggregate wildfire impacts measured at the tract level across the rural-urban continuum, 1999–2020.

Tracts that burned multiple times within the study period are counted at each burn. All underlying point estimates are available in Supplementary Table 1. Data sources: USDA Rural-Urban Commuting Area (RUCA) codes, U.S. Incident Command System 209-PLUS database, Monitoring Trends in Burn Severity (MTBS) database, Fire Events Delineation (FIRED) database, and U.S. Census Bureau Population and Housing Unit Counts^{15,35,50,56,61}.

Geographic trends in wildfire impacts across the rural-urban continuum

Next, we disaggregate our six wildfire measures across geographies, drawing on the U.S. National Climate Assessment’s designation of geographic regions within the lower 48 states (Fig. 2, Supplementary Table 2). These regions, especially when they are further disaggregated by RUCA codes, differ widely in their total land area (see Supplementary Tables 2 and 3 for measures of land area across geographic regions and RUCA categories). We therefore present results in both counts of affected tracts (Fig. 3) as well as in rates of affected tracts per thousand square miles (Fig. 4). The latter approach captures a normalized measure of an area’s propensity to experience wildfire impacts across geographic regions and RUCA categories.

Fig. 2. Geographic regions used in analysis.

Region boundaries are adopted from the U.S. National Climate Assessment⁵⁸. Regions are indicated by color.

Fig. 3.

Wildfire impacts measured at the tract level across geographic regions and the rural-urban continuum, 1999–2020. Darker shades of brown indicate higher numeric values. Tracts that burned multiple times within the study period are counted at each burn. Data sources: USDA Rural-Urban Commuting Area (RUCA) codes, U.S. Incident Command System 209-PLUS database, Monitoring Trends in Burn Severity (MTBS) database, Fire Events Delineation (FIRED) database, and U.S. Census Bureau Population and Housing Unit Counts^{15,35,50,56,61}.

Fig. 4. Wildfire impact rates measured at the tract level across geographic regions and the rural-urban continuum, 1999–2020.

Darker shades of green indicate higher numeric values and lighter shades of yellow indicate smaller numeric values. Tracts that burned multiple times within the study period are counted at each burn. Data sources: USDA Rural-Urban Commuting Area (RUCA) codes, U.S. Incident Command System 209-PLUS database, Monitoring Trends in Burn Severity (MTBS) database, Fire Events Delineation (FIRED) database, and U.S. Census Bureau Population and Housing Unit Counts^{15,35,50,56,61}.

Wildfire characteristics vary widely across the U.S., shaped by regional differences in vegetation, fuel structure, climate, topography, and land use⁴². In the western U.S., fire activity is influenced by periods of prolonged drought, high interannual climate variability, and expansive fire-adapted vegetation, resulting in large, fast-spreading fires occurring in remote or mountainous terrain^43,44. In contrast, fires in the eastern and southeastern U.S. burn in more humid climates, in ecosystems where fuel accumulation is slower and fire return intervals are shorter⁴⁵. Across the Great Plains and Intermountain West, grass- and shrub-dominated landscapes experience frequent, rapidly spreading fires driven by wind and fine fuels⁴⁶, while forested regions of the Southwest and Rockies tend towards high-severity, stand-replacing events under extreme drought conditions^44,47.

In this analysis, regional wildfire hotspots become evident and, like aggregate measures, also vary based on metric reported. Across the Southwest, Southern Great Plains, Southeast, and Northwest, rates of tracts affected by wildfire are roughly comparable across the rural-urban continuum, ranging from 10.8 to 26.8 total wildfires per thousand square miles (Fig. 3). Wildfire burn area rates are highest in the Southwest and Northwest, and elevated across metropolitan, micropolitan, small town, and rural places in these regions compared to others (Fig. 3). The rural Northwest stands out as having the highest burned area rate, at 227.7 square miles burned per thousand square miles (Fig. 3). When examining raw counts of miles burned, we see a similarly strong rural trend in the Northwest as well as Southwest (Fig. 3). For instance, the rural Northwest experienced nearly five times the spatial area burned compared to the metropolitan Northwest (Fig. 3). These regional differences across RUCA categories are most pronounced for these two regions.

Fast wildfires are geographically clustered in the rural Northwest (591 tracts affected) and the rural and metropolitan Southwest (647 and 1000 tracts affected respectively) (Fig. 3). The metropolitan Southwest has both the highest overall count of tracts affected by fast wildfires, as well as the highest rate of tracts affected by fast wildfires (6.3 per thousand square miles), followed by the rural Northwest (4.6 per thousand square miles) (Fig. 4).

Turning to measures of tracts affected by fatal wildfires, destructive wildfires, and total structures destroyed, there is a clear urban and regional trend in both rates and raw counts. Fatal wildfires occurred primarily in the metropolitan Southwest, with 285 total affected tracts compared to the next highest total count, 80 tracts within the rural Southwest (Fig. 3). The highest number of tracts affected by destructive wildfires occurred in the metropolitan Southwest (1362 total) and the metropolitan Southeast (1153) (Fig. 3), which is reflected in the highest destructive wildfire rates across all geographic regions (Fig. 3). When examining total structures destroyed by wildfires, the metropolitan and micropolitan Southwest stand out as having far higher total structure loss counts and structure loss rates than any other geographic region. The micropolitan Southwest experienced 226.5 structures destroyed per thousand square miles, followed closely by the metropolitan Southwest’s structure loss rate of 216.8 (Fig. 3). These reflect a nearly three-fold greater structure loss rate than the third highest rate, illustrating how distinct the Southwest is in its experience of wildfire-related destruction. The geographic differences between measures of tracts affected by destructive wildfires versus total structures destroyed by wildfires reflect the high volume of relatively less destructive incidents in the Southeast and the concentration of extremely high loss events in California⁴⁸.

Temporal trends in wildfire impacts across the rural-urban continuum

After describing wildfire impacts across geographic regions, we next disaggregate our primary wildfire measures annually for all years available in the ICS-209-PLUS database, 1999 through 2020. Our previous analyses showed substantial concentrations of fast-moving wildfires, fatal wildfires, and wildfire-related structure loss in metropolitan census tracts. Here, we investigate whether the relative urbanicity versus rurality of wildfire impacts has changed over time. We test for statistical significance of temporal trends using a non-parametric Theil-Sen (TS) estimator.

Our analysis supports the notion that wildfire impacts in urban areas have increased over the past two decades. However, this growth generally occurred in step with rising wildfire impacts across other RUCA categories. We first describe changes over time in total wildfire impacts, measured in counts of affected tracts, square miles burned, and buildings destroyed (Fig. 5). Results show that there tends to be high variability from year to year in aggregated wildfire impacts, reflecting season-to-season changes in wildfire severity as well as the impacts of extreme outlying events^48,49.

Fig. 5. Temporal trends in wildfire impacts across the rural-urban continuum measured annually at the tract level.

Dark blue indicates rural, light blue indicates small town, tan indicates micropolitan, and dark brown indicates metropolitan. Tracts that burned multiple times within the study period are counted at each burn. Data sources: USDA Rural-Urban Commuting Area (RUCA) codes, U.S. Incident Command System 209-PLUS database, Monitoring Trends in Burn Severity (MTBS) database, Fire Events Delineation (FIRED) database, and U.S. Census Bureau Population and Housing Unit Counts^{15,35,50,56,61}.

TS tests show broadly increasing wildfire impact trends across distinct impact measures and across most RUCA categories (full results are reported in Supplementary Tables 5 and 6). Overall, wildfire occurrence grew significantly and to the greatest degree in rural tracts ($\widehat{\beta }$ = 11.25, p = 0.0047), followed by micropolitan ($\widehat{\beta }$ = 9.60, p = 0.0052) and small town tracts ($\widehat{\beta }$ = 8.12, p = 0.0025). However, there was no significant increase in overall wildfire incidents in metropolitan tracts. When examining our two measures of structure impact, we first observe that the total number of tracts that experienced destructive wildfires grew across all RUCA categories, with the largest magnitude of increases in both metropolitan ($\widehat{\beta }$= 3.27, p = 0.0052) and rural tracts ($\widehat{\beta }$ = 3.91, p = 0.0007). When examining changes in total structures destroyed, we see a very pronounced metropolitan trend, in which the magnitude of the TS coefficient ($\widehat{\beta }$ = 88.76, p = 0.0058) is nearly double the size of the second largest coefficient, among rural tracts ($\widehat{\beta }$ = 48.47, p = 0.0001). Increases in total area burned were significant only among metropolitan tracts ($\widehat{\beta }$ = 38.34, p =  0.0011). Significant increases in tracts that experienced fatal fires were observed across all RUCA categories, but with the most pronounced trend in metropolitan tracts ($\widehat{\beta }$ = 1, p = 0.0034). Lastly, we observed a significant and positive trend in tracts that experienced fast fires among all RUCA categories except for metropolitan tracts.

While the trends above indicate growing wildfire impacts across the rural-urban continuum, to understand whether the overall share of annual wildfire impacts has become more urban versus rural over time, we use TS tests to evaluate year-over-year trends in the proportion of all wildfire impact measures that occurred in metropolitan, micropolitan, small town, or rural tracts (proportions shown in Supplementary Fig. 1 and full TS tests reported in Supplementary Table 6). Here, we observe far more stability in temporal trends. For example, there were no statistically significant changes in the relative proportion of buildings destroyed across RUCA categories over time; the larger share of total structures destroyed tends to occur in metropolitan and micropolitan tracts (Supplementary Fig. 1), and this proportion was not monotonically increasing or decreasing. Across other measures of wildfire impact, the larger share of area burned consistently occurs in rural places and, to a lesser extent, small towns (Supplementary Fig. 1). We observe a slight decline in the proportion of burned area in small town tracts ($\widehat{\beta }$ = -0.0022, p = 0.0124) and a slight rise in fast wildfires in micropolitan tracts ($\widehat{\beta }$ = 0.0016, p = 0.0251). The proportion of fatal wildfires increased in metropolitan ($\widehat{\beta }$ = 0.0175, p = 0.0046) and decreased in rural tracts ($\widehat{\beta }$ = -0.0110, p =  0.0046). While metropolitan and rural tracts both experienced overall increases in the total count of destructive incidents, the proportion of those incidents occurring in each RUCA category did not significantly change over time. The proportion of overall incidents experienced the most changes over time, with a decline in the proportion in metropolitan tracts ($\widehat{\beta }$ = -0.0018, p = 0.0052) compared to an increase in the proportion of micropolitan ($\widehat{\beta }$ = 0.0006, p = 0.0251) and small town tracts ($\widehat{\beta }$ = 0.0011, p = 0.0015). These fairly minimal changes in proportional temporal trends suggest that, even as wildfire impacts are intensifying on aggregate across the rural-urban continuum, their relative distribution across the rural-urban continuum has not changed dramatically. Thus, we are not revealing a new urban impact of wildfire. Rather, wildfires have long affected urbanized places at a degree understated by prior data.

Comparing measures of wildfire impact between the ICS-209-PLUS and the MTBS databases

Finally, we compare the distributions of wildfire impacts between the ICS-209-PLUS database and the smaller set of fires included in the longstanding MTBS database. Because MTBS uses acreage thresholds for inclusion (500 acres for the eastern U.S. and 1000 acres for the western U.S.), this database by design does not include smaller wildfires. Does this exclusion result in substantively different measures of wildfire impacts?

As expected, the MTBS database includes a much smaller share of overall wildfires compared to the ICS-209-PLUS database, including only one third as many tracts burned. These differences are approximately evenly distributed across the rural-urban continuum. However, the wildfire incidents excluded were geographically clustered in the Northeast, Southeast, and Midwest, where fewer than 17% of burned tracts included in the ICS-209-PLUS were covered by the MTBS database.

Also as expected, given its emphasis on large wildfires, MTBS captures the majority of area burned (89.2%) reported in the ICS-209-PLUS. Similarly, MTBS captures 96.1% of tracts burned by fast-moving wildfires included in the ICS-209-PLUS. These findings suggest that, for researchers conceptualizing wildfire impacts in terms of size or the presence of fast-moving fires, analysis drawing on MTBS data will not yield substantially different results than analysis based on ICS-209-PLUS data.

However, we are especially interested in understanding the prevalence of wildfire-induced fatalities and structure loss, as these impacts reflect instances in which a common environmental phenomenon becomes a disaster, and is thus of major societal concern¹. We find that the MTBS set of wildfires includes 77.2% of tracts burned by fatal wildfires included in the ICS-209-PLUS. Those fires that caused fatalities but were too small to meet the MTBS size threshold were most common in micropolitan and small town tracts.

MTBS’s coverage of wildfire-induced structure loss can be approached in two different ways. On the one hand, a small number of highly destructive and generally large wildfires have caused an outsized share of all building destruction in the U.S⁴⁸. As a result, we find that the set of fires included in MTBS covers the majority (84.9%) of all structure loss reported in the ICS-209-PLUS. However, the majority of destructive wildfires do not destroy many buildings; McConnell et al. report that 90% of destructive wildfires during the same period destroyed only 13 or fewer structures⁴⁸. Are these less destructive but more common wildfires well-captured by the MTBS database? We find that no, only 49.9% of tracts burned by destructive wildfires are included. Further, MTBS is especially limited in its ability to capture destructive wildfires in the Northeast, Southeast, and Midwest.

Examining wildfires reported within the ICS-209-PLUS, we find that destructive wildfires are generally fairly small in acreage, with a median of 179 acres burned. Out of 5385 total destructive wildfires reported, 3198 (59.4%) burned fewer than 500 acres and 3536 (65.7%) burned fewer than 1000 acres. In essence, over half of destructive wildfires that occurred in the U.S. over two decades were too small to be documented in the MTBS data. This means that over half of wildfires that had substantial impacts to people via destruction of the built environment have been effectively excluded in analyses that rely primarily on MTBS burn footprints to measure wildfire exposure.

Discussion

Recent wildfire disasters, such as those in Maui, Los Angeles, and other cities around the world, have highlighted the threat of wildfire in urban areas. This study offers an assessment of the relative urbanicity versus rurality of wildfires in the U.S., examining differentiation across geographic regions and evaluating the extent to which the urbanicity of wildfires has changed over time. Using a combination of administrative details about wildfires taken from incident status reports and wildfire burn footprints derived from multiple satellite missions, we document a range of wildfire impacts between 1999 and 2020^15,35,50.

We show that, when evaluating the geographic distribution of wildfire impacts, the measurement matters. When using measures of census tracts impacted by fatal wildfires, destructive wildfires, and counts of total structures lost, a clear urban pattern emerges. For example, out of over 106,000 total structures destroyed within the study period, 74% of these losses occurred in metropolitan or micropolitan tracts compared to 26% that occurred in tracts designated as small town or rural. These findings are in line with prior research illustrating an expansion of human-caused wildfire ignitions during a similar period²¹ and showing that wildfires tend to be more destructive in the wildland-urban interface²³. Both of these prior studies illustrate the correlation between human presence on landscapes and wildfire prevalence. Our results build on this work by showing that specific impacts of wildfires on human life and on the built environment tend to be more urban—a distinct geographic measure from the wildland-urban interface, which describes development morphology but not urbanicity. Further, we argue that many of these metropolitan wildfires have elided prior population analyses due to their exclusion from underlying wildfire data, for instance due to burn area size thresholds (e.g., MTBS) or due to burning in places without traditional vegetative fuels (e.g. WHP). In contrast, counts of tracts affected by wildfire and by fast-moving wildfires show a bimodal metropolitan and rural trend, with far less exposure among micropolitan and small town tracts. Even more distinct, measures of total area burned show a strong rural trend, reflecting that some of the largest wildfires occurring in sparsely-populated areas may not be suppressed as quickly, potentially due to longer response times, difficult accessibility due to terrain, and a lower potential to impact built infrastructure.

When considering the distribution of wildfire impacts across the country, we observed strong regional clustering. The metropolitan Southwest (California, Nevada, Utah, Arizona, Colorado, and New Mexico) had the highest counts of census tracts impacted by fatal, fast-moving, and destructive wildfires as well as the highest count of structures destroyed by wildfires compared to any other region. These trends were also reflected in rate calculations of the same measures, which accounted for differences in land area across geographic regions and RUCA categories. In contrast, the rural Northwest (Idaho, Oregon, and Washington) stood out as having one of the largest amounts of burned area as well as the highest burned area rate out of any region.

Examining the relative urbanicity of wildfires over time, we find that a number of wildfire impacts have significantly increased in urban areas. Although we did not observe an increase in the number of metropolitan tracts impacted by wildfires, we did find that, across all rural-urban continuum categories, metropolitan tracts did experience the largest increase in structures destroyed. We also observed significant growth in the number of fatal wildfires that occurred in metropolitan tracts and the overall metropolitan area burned. When considering the relative proportion of wildfire impacts across the rural-urban continuum year over year, however, we find that urban wildfires are not a new phenomenon; wildfires have consistently been occurring in urbanized areas over the twenty-year period examined. These findings suggest two simultaneous trends: the extent of urban wildfire impacts has been overlooked in prior demographic analyses due to data limitations and urban wildfire impacts did indeed increase over time.

Finally, our analysis shows meaningful differences between the comprehensive set of wildfires included in the ICS-209-PLUS database and the more widely-used MTBS database, which includes only larger fires. In particular, we find that MTBS has limited coverage of wildfire incidents in the Northeast, Southeast, and Midwest, a finding that is in line with regionally-focused research on the predominance of small fires in the eastern U.S³⁶. Additionally, we find that the MTBS database does not include a large set of small-area but destructive wildfires; over half of wildfires that destroyed at least one building were not included in the MTBS database. This finding echoes results from wildfire scholarship conducted in Europe, in which small wildfires were found to be widely prevalent and impactful^38,39. While MTBS is a longstanding and invaluable resource, our analysis suggests that researchers should use caution when assuming that the larger events included in the data comprehensively reflect all forms of wildfire exposure.

The geographic variation in the type of wildfire impact being measured means that, when conducting analyses of population exposure, social scientists need to more critically evaluate what their wildfire data are actually measuring. Further, they should consider the extent to which a given data source does or does not accurately reflect how they are conceptualizing wildfire risk or exposure. Off-the-shelf measures of wildfire risk may be derived from models that are not designed to capture wildfire risk to humans and their built environments. And measures of generalized wildfire occurrence do not allow researchers to distinguish between types of wildfires, in particular fast-moving, fatal, and destructive wildfires. We show how the ICS-209-PLUS dataset can be leveraged to conduct more nuanced evaluations of wildfire impacts to human communities, bringing social sciences into conversation with biophysical scientists who are increasingly distinguishing between beneficial and harmful wildfire.

Our work points to a number of future directions for inquiry at the intersection of wildland fire science and demographic research. The ICS-209-PLUS database enables U.S.-focused researchers to move beyond measures of wildfire risk or general wildfire occurrence, specifying particular wildfire impacts and making finer distinctions regarding the types of wildfires that have the greatest impacts on human communities. Additional wildfire consequences, such as evacuations or impacts to natural resource-based and agricultural economies, could be measured to further hone our understanding of wildfire risk. Prior national-level assessments of wildfire and social vulnerability^16,17,31 could be productively updated with ICS-209-PLUS data to incorporate previously overlooked smaller and more urban wildfires. Finally, questions regarding the extent of wildfire impacts and the populations at greatest risk are relevant to many different regions around the world, and should continue to be explored with country-specific wildfire data.

Methods

ICS-209-PLUS wildfire data

We leverage a recently published database of wildfire incident response administrative records known as the ICS-209-PLUS¹⁵ linked to satellite-derived wildfire burn footprints^35,50 to analyze wildfire impacts along the rural-urban continuum between 1999 and 2020. These data are distinct in that they provide not only finely resolved spatial details of where wildfires occurred, but also specific measures of wildfire impacts beyond the often-used metric of area burned. These measures include: counts of structures destroyed, counts of casualties, and daily wildfire growth, an indication of fast-moving fires, which are considered especially dangerous⁹. In line with recent scholarship, we measure fast fires as incidents with a daily maximum fire spread rate of 1620 hectares (4003 acres) or greater⁹. The data include the full set of all wildfire events that required an incident command response—essentially, incidents that were large or otherwise societally important—and therefore represent one of the most comprehensive datasets of wildfires in the U.S.

The foundation of the ICS-209-PLUS database is an historical archive of National Incident Management System (NIMS) Incident Status Summary (“ICS-209” for short) forms. An ICS-209 is completed when a wildland fire requires higher-level coordination, decision support, and resource allocation. National Wildfire Coordinating Group (NWCG) guidelines specify that it is required for all large wildland fires—defined as >100 acres in timber or >300 acres in grasslands—and for any incident that warrants special attention, requires additional resources, attracts media interest, or poses an elevated threat to public safety⁵¹. In practice, ICS-209 reporting is mandated by the U.S. federal government at all levels of incident response when a fire escapes initial attack, involves multiple jurisdictions, commits substantial resources, or threatens life, property, infrastructure, or communities⁵². As a result, the ICS-209 form captures both large wildland fires and smaller fires that pose meaningful risk to homes or communities, while excluding short-duration, low-impact, and low-complexity events. The ICS-209-PLUS does not have a minimum size threshold. Consequently, the historic ICS-209 archive represents a population-level record of meaningful wildfire events.

The ICS-209-PLUS defines a wildfire as an unplanned, unwanted wildland fire including unauthorized human-caused fires, escaped wildland fire use events, escaped prescribed fire projects, and all other wildland fires where the objective is suppression, which is the working definition we use in this study¹⁵. Wildfire scholars make the distinction between “wildfires,” defined as fires that burn in wildlands, and “urban fires,” “building fires,” or “urban conflagrations,” defined as fires that burn in built environments^1,53,54. Many recent, high-profile wildfire incidents, such as those in Maui and Los Angeles, blur the boundary between these categories by igniting in wildlands and subsequently spreading into built environments. The ICS-209-PLUS database captures both traditional wildfires that burn only in wildlands as well as incidents that burn across both wildlands and built environments. Our analysis defines urbanicity based on RUCA categories (i.e., the level of urbanicity of a given place) and does not use “urban” as some fire scientists do to refer to fuel source. For these reasons, when we refer to “urban wildfires,” we are referring to wildfires that occurred in urbanized places, as defined by RUCA categorization.

We process wildfire incidents from the ICS-209-PLUS database at the census tract scale following St. Denis et al.’s procedure to construct the ICS-209-PLUS spatiotemporal linkage⁴⁸. Through this process, each wildfire observed in the ICS-209-PLUS administrative records is then associated with the census tract or tracts in which the fire burned through either its MTBS burn footprint, Fire Events Delineation (FIRED) burn footprint, or latitude-longitude point of origin. The latter two sources of spatial details allow us to capture wildfires that are far smaller than the MTBS size threshold (some even just a few acres in size)¹⁵.

Rural-urban continuum classifications

We harmonize selected wildfire datasets with U.S. census tract boundaries⁵⁵, and, in turn, the U.S. Department of Agriculture’s 2010 Rural-Urban Commuting Area (RUCA) codes, which are widely used to differentiate between places based on their population density, degree of urbanization, and commuting patterns⁵⁶. We use the four primary RUCA classifications: metropolitan, micropolitan, small town, and rural (Table 1). RUCA codes are measured at the census tract scale, which we select as our primary spatial unit of analysis first because this unit is better-matched to the scale of wildfire burn footprints compared to larger units, such as counties⁴⁸. Additionally, tracts are the smallest spatial unit for which the USDA reports RUCA codes. We use RUCA codes that reflect conditions in 2010, the midpoint of our analysis period. We adopt this approach because the U.S. has consistently urbanized over time⁵⁷, meaning that the use of more contemporary data would likely be selecting on urbanity, and thus risk biasing our results towards more frequent urban wildfire occurrence.

Through linking wildfire impact measures to RUCA codes, we are able to evaluate a range of wildfire impacts across the rural-urban continuum, utilizing the RUCA codes’ classification of rural, small town, micropolitan, and metropolitan census tracts across ten primary codes (Table 1). Additionally, we group each state (and its corresponding census tracts) into one of nine geographic regions, utilizing the National Climate Assessment’s reported geographies⁵⁸ (see Supplementary Table 2 for details). This linkage allows us to evaluate the relative urbanicity versus rurality of wildfire impacts within specific geographic regions, which we expect to vary substantially as a result of varying landscape and climatic conditions. For measures of total wildfire occurrence, fast wildfires, fatal wildfires, and destructive wildfires, we report burned census tracts as our unit of analysis (rather than individual wildfires) to reflect the fact that a small proportion of fires burn across multiple tracts. When presenting regional statistics, we present both counts of affected tracts as well as rates of affected tracts divided by total land area.

Spatial unit selection

There are two spatial challenges to describing the geographic distribution of wildfire impacts. First, McConnell et al. note the wide range of land area burned by wildfires captured in the ICS-209-PLUS database, spanning from less than one acre at the smallest to over a million acres at the largest⁴⁸. In some cases, wildfire footprints are a small proportion of their spatial unit’s overall land area, while in other cases, footprints encompass multiple spatial units⁴⁸. St. Denis et al. estimate that 8.3% of wildfire incidents burned across more than one county and 14.4% burned across more than one census tract¹⁵. Second, spatial units, such as census tracts or counties that are commonly used in geographic analyses also vary substantially in their size. This is the case both for census tracts, whose size varies because tracts attempt to capture 4000 people⁴⁰, and for counties, which tend to be much smaller in the eastern U.S. compared to the western U.S. due to historical development patterns⁷. As a result of this dual variability in both wildfire size and in the land area of different spatial units, there is no one optimal spatial unit that would consistently be the equivalent size as the wildfire incident which it encompasses. This raises concerns regarding the Modifiable Areal Unit Problem, that the selection of one spatial unit or another may bias the analytical results⁵⁹.

We argue that census tracts are the most appropriate spatial unit for our analysis for several reasons. First, tracts represent approximately neighborhood-level measures that, when merged with RUCA codes, best reflect residents’ lived experiences of a place’s relative urbanicity versus rurality. RUCA codes were specifically designed to address scaling problems with county-level rural-urban continuum data, which the U.S. Department of Agriculture notes are, “often too large to accurately delineate boundaries between rural and urban areas”⁵⁶. Tract-level RUCA codes are generally considered an improvement over county-level data in that they offer more geographically-specific information and also incorporate residential commuting patterns⁵⁶. Second, prior wildfire research further suggests that the proportion of a county’s total land area burned by destructive wildfires is relatively small; more than 90% of destructive wildfires in the ICS-209-PLUS database burned less than 10% of the total area of the county in which they occurred⁴⁸.

To address any lingering concerns about the Modifiable Areal Unit Problem, we take several steps. First, in Fig. 4 we present all of our wildfire impact variables as rates, normalized by the total land area of each RUCA categorization of a given geographic region (reported in Supplementary Table 4). These results account for the differences in tracts’ spatial size across different RUCA categories. Second, in Supplementary Fig. 2, we present an approximation of Fig. 1 based on county-level data, which we obtained from the U.S. Department of Agriculture’s Rural-Urban Continuum (RUC) codes⁶⁰. Exact one-to-one comparisons between tract-level RUCA codes and county-level RUC codes are not possible because the categories included in each are somewhat different. Given its coarser spatial scale, this county-level analysis is not a mere sensitivity test of the tract-level analysis; it is asking a distinct question about the larger spatial dynamics of wildfires, rather than the more localized, neighborhood-level dynamics presented in our main findings. Nevertheless, we observe broadly similar trends at the county scale to those presented in Fig. 1 (see Supplementary Fig. 2 and Supplementary Notes 1 for further details).

Structures destroyed per tract

Measuring structure loss across census tracts poses an additional challenge. The ICS-209-PLUS reports total structures destroyed per wildfire, but does not include spatial details on where within a wildfire burn footprint individual structures were destroyed. For cross-boundary wildfires, then, simply ascribing the total number of structures lost to each corresponding census tract would result in substantial double-counting of destroyed structures in cross-tract wildfires¹⁵. We allocate estimates of counts of structures destroyed per tract through the following procedure. First, following Palaiologou et al., we use U.S. Census housing unit counts as a proxy for the total number of structures located within each tract affected by a given wildfire^18,61. For each tract i burned by wildfire f, we multiply the count of housing units by the proportion of each tract that was burned, yielding a measure of structures exposed (Eq. 1). We then calculate the proportion of firewide exposed structures for each tract i burned by wildfire f by dividing the structures exposed per tract i by the total number of exposed structures across all tracts burned by the wildfire f (Eq. 2). Finally, we multiply this tract-level proportion of firewide structures exposed by the total number of structures reported destroyed by wildfire f in the ICS-209-PLUS (Eq. 3). This gives us a weighted estimate of total structures destroyed within each census tract burned by a given wildfire (see Supplementary Fig. 3 for an illustration of this procedure).

1. $${Structuresexposed}_{i,f}={Housingunits}_{i,f}X{Proportionoftractburned}_{i,f}$$
2. $${Proportionoffirewidestructuresexposed}_{i,f}=\frac{{Structuresexposed}_{i,f}}{{Structuresexposed}_f}$$
3. $${Weightedcountofstructuresdestroyed}_{i,f}={Proportionoffirewidestructuresexposed}_{i,f}X{Totalstructuresdestroyed}_f$$

Analysis

To evaluate temporal trends, we use the Theil-Sen (TS) estimator to test hypotheses that the magnitude of wildfire impacts has changed over time. The TS estimator is a non-parametric, median-based estimator that is robust to outlying points and does not require an assumption of distribution normality⁶². A statistically significant TS test indicates that the temporal trend is not zero; it is either increasing (positive slope) or decreasing (negative slope). We calculate TS slopes using the mblm package in R⁶³ and use p < 0.05 as our preferred significance threshold. The ICS-209-PLUS dataset encompasses the full population of consequential wildfires within the U.S., meaning that the descriptive findings presented in the study reflect population-level trends. For this reason, we do not conduct significance testing for non-temporal analyses.

We conducted all analysis in R statistical software version 4.3.3 using the tidyverse and sf packages^64,65. We used ChatGPT to support small code debugging tasks when creating data visualizations and tables⁶⁶. We did not use ChatGPT or any other AI tool for written text generation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

K.M., J.T.M., P.B., and L.S. conceptualized and designed the study. K.M. and L.S. curated the data. K.M. wrote software, performed formal analysis, and created visualizations. K.M. wrote the original draft and K.M., J.T.M., P.B., and L.S. contributed to review and editing of manuscript drafts.

Peer review

Peer review information

Communications Earth & Environment thanks Mark Billings and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Martina Grecequet. [A peer review file is available.

Data availability

All data used in this analysis are publicly available to download. The ICS-209-PLUS wildfire dataset is available at ref. ¹⁵ and linked to the MTBS and Fire Events Delineation (FIRED) wildfire burn footprints described and available in refs. ^35,50,67. USDA Rural-Urban Commuting Area (RUCA) codes are available at⁵⁶. Housing unit counts used to spatially weight structure loss estimates are available through the U.S. Census Bureau⁶¹. U.S. Census tract boundaries are available through the IPUMS National Historical Geographic Information System⁵⁵.

Code availability

Codes produced for this project are available in the following OSF repository: https://osf.io/grdm6.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43247-026-03336-y.