Introduction

Flooding is a pervasive and impactful hazard, responsible for damages exceeding one trillion dollars globally since 19801. Climate change and rapid urbanization are projected to increase flood exposure in the future2,3. Social experiences of flooding are often inequitable and disproportionately impact socially vulnerable communities. Addressing flood risk inequities, especially in areas lacking detailed flood maps, is critical for informing research, policy, and advancing environmental justice.

In the United States, the Federal Emergency Management Agency (FEMA) classifies 100-year floodplains via Flood Insurance Rate Maps (FIRMs). These maps play a crucial role in federal initiatives for identifying, mitigating, and insuring against floods through the National Flood Insurance Program (NFIP). FEMA creates FIRMs for localities using spatially discontinuous streamflow data and historical flood records, but these methods fail to account for built environmental and climate changes4. Many FIRMs are outdated, and their coverage is incomplete. FIRMs entirely overlook pluvial flood risks (i.e., extreme rainfall-induced flooding that occurs independent of an overflowing water body)5. FEMA FIRMs can also be influenced by financial interests. This occurs as real estate developers pressure local officials to omit certain areas from 100-year flood risk zoning to facilitate continued development6,7.

Consequently, FIRMs inadequately predict location-specific risks linked to real flood occurrences8,9,10. Notably, 27% of flood damage claims via the NFIP originate from properties situated beyond FEMA 100-year floodplains where flood insurance uptake rates are quite low11, and research has identified socially inequitable impacts beyond FEMA floodplains during actual flood events based on minority race/ethnicity10.

Flood hazard modeling has progressed with advanced computational techniques that account for complex hydrological processes, as well as land use and land cover changes. One such model, referred to here as the Bates et al. model, provides fine-scale flood risk estimates for the entire CONUS. This model has undergone rigorous validation12,13. Bates et al. validated the model by comparing it to 100-year flood maps generated by the Iowa Flood Center, and it achieved a critical success index (CSI) of 0.87, meaning that the model correctly predicted inundation over 87% of the floodplain. That CSI value is within the performance range for high-quality flood models12. Wing et al. further validated the model against historical flood events—using interpolated high-water mark data from the United States Geological Survey (USGS) as benchmarks—and also achieved a CSI value of 0.8713. While the Bates et al. model is not without limitations, it provides flood risk information of well-documented validity, which is particularly useful for assessing risk in areas where existing FEMA FIRMs are outdated, incomplete, or not available.

Flooding may have environmental justice (EJ) implications as social hierarchies produce unequal distributions of environmental burdens and benefits14. A U.S. Government Accountability Office report highlighted FEMA’s underinvestment in flood risk mapping and management for EJ communities where socially vulnerable residents concentrate15. EJ research has assessed both FEMA-delineated and Bates et al. delineated floodplains. At the national level, lower socioeconomic status was associated with an increased likelihood of experiencing a 100-year flood, as classified by FEMA16,17, while race/ethnicity was not16. When looking specifically at inland vs. coastal locations, racial/ethnic minority populations concentrated in FEMA-delineated inland flood risk areas in Miami (Florida), whereas White and higher socioeconomic status populations were overrepresented in coastal floodplains18. In Washington state, research found an elevated concentration of Latinx residents within FEMA-delineated inland flood risk areas19. When examining flood risks assessed using the Bates et al. model, prior national and regional level studies have identified socioeconomic and racial inequities, although they have not specifically focused on areas included in Bates et al.’s 100-year floodplains but omitted from FEMA 100-year flood maps. For example, one study linked lower socioeconomic status with higher 100-year Bates et al. flood risk3. Another identified hot spots of overlap between high Black composition and elevated 100-year flood risk (according to the Bates et al. model)20. Similarly, Galster et al. analyzed nationwide inequities in flood risk using annualized flood probabilities from the Bates et al. model aggregated to the census tract level21, but they did not examine areas at risk that are excluded from FEMA 100-year flood maps. Their study specifically focused on the probability of experiencing a flood of at least one centimeter of depth in 2020, and found that Native Americans and lower-income residents experienced higher flood exposure in inland areas, while Latinx Americans experienced higher exposure in coastal areas21. In the U.S. Southeast region, Noonan et al. compared FEMA 100-year floodplains with First Street Foundation (FSF) flood risk, which uses the Bates et al. model as the basis for their property-level risk assessments22 (e.g., Flood Factor23). They applied a ‘Risk Difference’ metric, which subtracts the percentage of land area in a FEMA 100-year floodplain from the percentage of properties classified as at-risk by the FSF, and found that FSF identifies greater flood risk in census tracts with higher Black and Latinx composition, and lower flood risk in high-income and high-property-value areas compared to FEMA22. This suggests that FEMA underestimates flood risk in socially vulnerable areas and overestimates it in socially advantaged areas. However, because their calculation did not exclude properties in both floodplains, it captured FSF risk already recognized by FEMA, rather than isolating flood risk excluded from FEMA flood maps. Additionally, these related prior studies have not explored how inequities may differ based on flood type (with the exception of Tate et al. 2021), or urban-rural status, like we do here. These studies also have not adjusted for potentially confounding effects of water-related (dis)amenties, as we do here.

While others have examined social inequities in measures of flood risk produced by FEMA and Bates et al., little research has examined federally-overlooked flood risks. One exception is a study of Greater Houston (Texas)24, which focused on floodplains identified by the Bates et al. model not included within 100-year floodplains classified by FEMA FIRMs. That study found that Greater Houston neighborhoods with higher Black and Asian composition faced disproportionately high risks of federally-overlooked fluvial and pluvial flooding24. Also, a higher proportion of Latinx residents was associated with significantly greater risks of coastal, pluvial, and fluvial federally-overlooked flooding24. In terms of socioeconomic status, middle (vs. lower and higher) income neighborhoods exhibited increased risks to 100-year pluvial and coastal federally-overlooked flooding24.

In summary, prior national studies have linked lower socioeconomic status with higher FEMA-delineated flood risk but found inconsistent evidence of racial/ethnic inequities. Importantly, no prior national-level studies have investigated social inequities in terms of 100-year flood risks that are federally-overlooked. Knowing who is at risk beyond the coverage of FEMA DFIRMs is critically important for informing local, state and federal governments as well as at-risk communities. Estimates from the Bates et al. model indicate that nearly 6 million properties in the CONUS are within 100-year floodplains that FEMA does not recognize23, which suggests that the overall scope of federally-overlooked flood risk nationally is enormous. When existing flood risks are not delineated in FEMA maps, which guide funding and incentives for flood risk mitigation nationwide, communities lack awareness of their risks and may be less likely to take protective actions, thereby exacerbating their flood vulnerability. Very few residents outside FEMA-delineated floodplains purchase flood insurance while local and regional planning and infrastructure investment for flood protection beyond FEMA floodplains is largely nonexistent25,26,27. In sum, flooding beyond FEMA’s 100-year floodplains raises critically important policy and planning challenges, which have yet to be fully comprehended.

To address these gaps, we leverage data from the Bates et al. model, FEMA FIRMs, and the American Community Survey (ACS) as well as dasymetric mapping techniques to evaluate the extent and socially uneven distribution of population exposure to federally-overlooked 100-year flood risk in the CONUS. Our nationwide assessment examines coastal, pluvial and fluvial federally-overlooked flooding, and inequities by race/ethnicity and socioeconomic status. We explore four research questions: (1) What is the population count residing in federally-overlooked 100-year floodplains across the CONUS? (2) Which factors reflecting social disadvantage in neighborhoods are associated with increased risk of federally-overlooked 100-year flooding? (3) Do social inequities in federally-overlooked flood risk differ based on the type of flooding? (4) Do social inequities in risk to federally-overlooked flooding differ by urban-rural location?

Results

Univariate analysis

Figure 1 illustrates Bates et al. and FEMA 100-year floodplains in the CONUS, and Fig. 2 displays federally-overlooked 100-year floodplains across the CONUS, which are 100-year floodplains identified by the Bates et al. model that fall outside of FEMA classified floodplains. Using dasymetric mapping techniques to improve the accuracy of risk assessment, we estimated the population residing in these federally-overlooked floodplains. Figure 3 depicts the population distribution in federally-overlooked floodplains, with darker blue indicating higher population risk quintiles. A map of the census tract percentages of residents at risk is included in the Supplementary Information (see Figure S1). Census tracts in New York City (New York), Miami (Florida), Los Angeles (California), and Houston (Texas) contain the largest populations at risk to all types of flooding (coastal, pluvial, and fluvial) (Fig. 3A). In terms of federally-overlooked coastal floodplains, tracts along the Texas, Louisiana and Florida Gulf Coast, as well as the Atlantic Coast of Georgia and the Carolinas, exhibit the highest population risk (Fig. 3B). Regarding federally-overlooked pluvial floodplains, tracts in Houston (Texas), New York City (New York), and Fort Myers (Florida) exhibit the highest population risk (Fig. 3C). As for federally-overlooked fluvial floodplains, tracts in the Pacific Northwest, Appalachia, and along the San Joaquin River (California) and Chicago River (Illinois) exhibit the greatest population risk (Fig. 3D).

Fig. 1
figure 1

CONUS 100-year floodplains: (Top) Bates et al. combined (coastal, pluvial, fluvial), (Bottom) FEMA.

Full size image
Fig. 2
figure 2

Federally-overlooked (A) combined, (B) coastal, (C) pluvial, and (D) fluvial 100-year floodplains in the CONUS.

Full size image
Fig. 3
figure 3

Population of census tracts (shown as centroids) within federally-overlooked 100-year floodplains, classified into quintiles and represented as (A) combined, (B) coastal, (C) pluvial, (D) fluvial.

Full size image

Table 1 presents descriptive statistics for each of the analysis variables, which are detailed in the materials and methods section. We analyzed sociodemographic variables derived from the ACS, including Racial/ethnic composition and median household income as a measure of socioeconomic status. Our models include control variables representing environmental (dis)amenities (including seasonal/recreational housing composition, tree canopy cover, urban imperviousness, and industrial facility concentration), which influence residential decision-making and are potential confounders of associations between social characteristics and flood risk. Models also include variables (rural-urban commuting area codes) that classify census tracts within metropolitan, suburban, and rural areas to control for and stratify by urban-rural context.

Table 1 Overview of descriptive statistics for each analysis variable at the census tract level.
Full size table

We estimate that over 26 million CONUS residents live in federally-overlooked floodplains. State- and county-level univariate statistics identifying areas ranking highest in terms of risk are included in Supplementary Information (Tables S3, S4). An estimated 3.1 million residents in the CONUS reside in federally-overlooked coastal floodplains, with Florida, New York, and Louisiana being states with the highest counts: 1.3 million, 400,000, and 320,000, respectively. Federally-overlooked pluvial floodplains include the highest population at risk among the flood types, with approximately 18 million in the CONUS. California has over 2.4 million residents in these floodplain, while Florida and Texas have 1.3 and 1.2 million, respectively. An estimated 5.2 million CONUS residents live in federally-overlooked fluvial floodplains, with California, Texas, and Illinois having 1.3 million, 527,000, and 447,000, respectively.

Multivariable models

Table 2 presents multivariable generalized estimating equation model results for federally-overlooked flood risk. In interpreting our results, each increase or decrease corresponds to one standard deviation and its associated percent change in the count of residents at risk. In the model for combined federally-overlooked flood risk, a decrease in median household income was associated with a 5.2% rise in the count of residents at risk (p < 0.001). Racial/ethnic minority variables were not statistically significant.

In the coastal federally-overlooked flood risk model, increases in the proportions of non-Latinx Black, Latinx, non-Latinx Asian, and non-Latinx other/multiple race residents were linked to reductions in federally-overlooked flood risk by 10.1% (p < 0.01), 16.7% (p < 0.001), 18.6% (p < 0.001), and 6.2% (p < 0.001), respectively. Conversely, a decrease in median household income was linked to a 16.7% risk increase (p < 0.001).

In the pluvial federally-overlooked flood model, an increase in non-Latinx Black composition was linked to a 3% risk increase (p < 0.05), while increases in Latinx composition were associated with a risk reduction of 3.9% (p < 0.01). A decrease in median household income was associated with a 6% risk increase (p < 0.001).

For the fluvial federally-overlooked flood model, an increase in non-Latinx Black composition correlated with a 13.4% risk reduction (p < 0.001) and an increase in Latinx composition was associated with a 6.8% risk increase (p < 0.05). A decrease in median household income was linked to a 16.7% risk increase (p < 0.001).

Stratified multivariable models

Table 3 presents results for models stratified by urban-rural context. In the combined flood risk model for census tracts in metropolitan core tracts (RUCA code 1), racial/ethnic minority variables were not significant, while a decrease in median household income was linked to a 4.4% risk increase (p < 0.001). In suburban (RUCA codes 2–3) and rural (RUCA codes 4–10) contexts, respectively, an increase in non-Latinx Black composition was associated with combined flood risk reductions of 6.5% (p < 0.001) and 11.4% (p < 0.001), whereas a decrease in median household income correlated with 4.7% (p < 0.01) and 3% (p < 0.01) risk increases.

For the coastal flood risk model for metropolitan core areas, increases in non-Latinx Black, Latinx, non-Latinx Asian, and non-Latinx other/multiple race composition were associated with risk reductions of 10.7% (p < 0.01), 16.7% (p < 0.001), 19.6% (p < 0.001), and 6.8% (p < 0.001), respectively. The racial/ethnic minority variables were not significant in suburban and rural contexts. A decrease in median household income was linked to coastal risk increases in metropolitan core and suburban tracts of 17.9% (p < 0.001) and 29.9% (p < 0.001), respectively.

For the pluvial flood risk model, an increase in non-Latinx Black composition was linked to a 4.6% risk increase (p < 0.001) in metropolitan core tracts, but it was associated with 5.3% (p < 0.001) and 8.7% (p < 0.001) risk decreases in suburban and rural contexts, respectively. An increase in Latinx composition correlated with respective risk reductions of 3.7% (p < 0.01) and 9.5% (p < 0.001) in metropolitan core and rural contexts. A decrease in median household income was associated with pluvial risk increases in metropolitan core and rural tracts of 4.9% (p < 0.001) and 4.6% (p < 0.001), respectively.

For fluvial flood risk model, increases in non-Latinx Black composition were associated with risk reductions of 13.6% (p < 0.001) and 20.6% (p < 0.001) in metropolitan core and suburban contexts, respectively. In suburban and rural contexts, an increase in Latinx composition was correlated with fluvial flood risk increases of 14.2% (p < 0.05) and 20.8% (p < 0.001), respectively. A decrease in median household income was associated with respective fluvial flood risk increases of 16% (p < 0.001), 26.4% (p < 0.001), and 11.9% (p < 0.001) in metropolitan core, suburban, and rural contexts.

Table 2 GEEs were utilized to predict federally-overlooked flood risk, considering each flood type individually and their combined impact for census tracts in the CONUS.
Full size table
Table 3 GEEs were utilized to predict federally-overlooked flood risk, considering each flood type individually and combined, in Metro, suburban, and rural census tracts in the CONUS.
Full size table

Discussion

This is the first nationwide assessment of inequities in 100-year flood risk outside of zones delineated by the U.S. federal government. In answer to the first research question, we found that over 26 million U.S residents—nearly one-tenth of the CONUS population—live in floodplains not recognized by federal flood maps. This underscores a pressing issue for planning and policy in the U.S. at all levels and exposes shortcomings in federal flood risk assessment that likely mislead local and regional planners in ways that undermine flood protection and disproportionately impact lower-income communities and, in particular contexts, racial/ethnic minority subpopulations. Widespread population exposure to these risks underscores the urgent need for interventions to address federally-overlooked flooding amid ongoing landscape and climatic changes.

Regarding research questions 2, 3 and 4, we observed a pattern of increased flood risk in areas overlooked by FEMA flood maps, specifically in lower-income census tracts across urban-rural contexts, flood types, and sensitivity analyses. Our findings align with prior national assessments of FEMA-delineated16,17 and Bates et al. delineated flood risk3,20,21, indicating that regardless of the specific flood risk metric employed, lower socioeconomic status is broadly linked to increased flood risk in the U.S.

In terms of race/ethnicity, we found an association between increasing Black composition and greater population risk to federally-overlooked pluvial flooding across the CONUS. While Tate et al. was not focused on federally-overlooked flood risk or multivariable models, these findings align with their results from a nationwide assessment of flood risk utilizing an earlier iteration of the Bates et al. model20. Our findings for federally-overlooked pluvial floodplains align with a study in Greater Houston, as do our results indicating that Black composition is negatively associated with federally-overlooked 100-year coastal flood risk24. Findings that Black Americans face disparate risks to federally-overlooked pluvial flooding, particularly within metropolitan core contexts, speak to environmental justice research spanning decades that has documented the disparate exposure of Black communities to urban hazards nationwide28,29. Our analysis reveals that approximately 85% of the CONUS Black population is concentrated in metropolitan core areas. These urban areas are where the pluvial flood risk problem is most acute, underscoring the inequity Black communities face regarding pluvial flooding. Ongoing anthropogenic changes to land use and cover, along with climate change leading to more frequent and intense extreme precipitation events, are likely to exacerbate these risks3.

We found linkages between higher Latinx neighborhood composition and elevated federally-overlooked fluvial flood risk in suburban and rural contexts. These results align with a county-based national assessment indicating that higher Latinx composition was associated with increased flood risk as per FEMA FIRMs, which primarily gauge fluvial flood risks16. Suburban and rural census tracts within agricultural production areas of western Texas along the Rio Grande River and in California near the San Joaquin River—where the majority of residents identify as Latinx—may drive the pattern of federally-overlooked fluvial flood risk inequity we observed here. In contrast, we found that higher Latinx neighborhood composition was associated with reduced federally-overlooked pluvial flood risk in metropolitan core and rural contexts and reduced federally-overlooked coastal flood risk in metropolitan core contexts. That patterns deviates from a study in Greater Houston, which found that higher Latinx composition was associated with increased federally-overlooked flood risk across all flood types24. These results also align with regional level studies indicating Latinx populations experience increased risks19,22.

This study highlights how U.S. communities disadvantaged by low socioeconomic status, and in some contexts, minority racial/ethnic status, face disproportionate risks to federally-overlooked flooding. This is concerning because underinvestment in flood risk reduction strategies beyond the boundaries of FEMA 100-year floodplains exacerbates the vulnerability of residents there25,30. Catastrophic and inequitable flood impacts beyond FEMA 100-year floodplains are well-documented after major events9,10,31. Concentrating flood risk mitigation initiatives exclusively within FEMA 100-year floodplains and the neglecting to disseminate flood risk information and resources beyond those limits is likely to worsen uneven flood impacts25, especially for the at-risk groups we identified in this study.

Furthermore, populations residing within federally-overlooked floodplains almost certainly experience disproportionate long-term impacts from flooding, given their notably reduced rates of flood insurance coverage compared to residents of FEMA-delineated floodplains. Research shows that the prevalence of flood insurance is substantially higher within FEMA 100-year floodplains compared to areas outside of those boundaries26. Lower rates of coverage outside of FEMA floodplains is one of the pivotal factors undermining coping capacities, prolonging recovery times, and elevating the incidence of post-flood displacement26. The emergence of Risk Rating 2.0, a transformative U.S. federal policy shift in flood insurance that takes property characteristics into account, may exacerbate disparities in flood insurance coverage. While evidence remains preliminary, analyses suggest that the flood insurance gap is widening due to increasing premiums associated with the enactment of Risk Rating 2.032. This gap is manifesting as a growing number of individuals opt to relinquish coverage, leaving them even more vulnerable to the impacts of flooding. Given the relatively low socioeconomic status that generally characterizes federally-overlooked floodplain communities, we surmise that initiatives to increase flood insurance coverage in these areas will be undermined without the provision of financial need-based support through Risk Rating 2.0.

Limitations

While we used Bates et al. model to determine federally-overlooked flood risk due to its many advantages12,13, it too has limitations. As with other predictive flood models, the Bates et al. model is subject to uncertainties at fine spatial scales, due to the limits of input data on flood defenses, terrain, and infrastructure. These modeled scenarios cannot fully capture local hydrological and infrastructural complexities, which may introduce potential inaccuracies. However, the model demonstrates minimal error propagation compared to benchmark datasets, including local flood model predictions5,12. Importantly, the Bates et al. model complements FEMA’s floodplain maps by providing more comprehensive, current, and fine-scale estimates of risks from multiple types of flooding, particularly pluvial flood risk, which FEMA’s models do not incorporate. This model also fills critical gaps in unmapped areas, allowing examination of flood risk in areas not covered by FEMA’s maps. For these reasons, recent flood risk research is increasingly leveraging estimates derived from Bates et al. flood model8,20,21,22,24,33.

Additionally, we focused on areas within the CONUS covered by DFIRMs, excluding those with non-digitized FIRMs (paper maps), which are home to 3.4% of the total CONUS population. These areas are predominantly White, rural, and middle-income. While digitizing the non-digital FIRMs for study inclusion would be ideal, it was both impractical and unnecessary, as existing DFIRM coverage enabled comprehensive assessment of federally-overlooked flood risk across the vast majority of the CONUS population and land area.

Another limitation of our study is the assumption that socially vulnerable and advantaged groups are evenly distributed within census tracts. However, sorting effects may influence residential patterns, with some evidence suggesting that lower-income and minority households are more likely to reside in higher-risk areas due to affordability constraints34. If within-tract sorting follows this pattern, our approach may underestimate flood exposure for socially vulnerable populations and overestimate it for socially advantaged groups. Conversely, in high-amenity coastal areas, wealthier households may sort into high-risk but desirable locations. However, there is mixed evidence on whether socially vulnerable populations disproportionately reside in FEMA 100-year floodplains across the U.S. While some studies suggest this18,34, several others indicate that socially vulnerable populations are more likely to reside in areas at risk beyond FEMA floodplains, rather than within them10,24. Future research should examine sub-census tract or parcel-level flood risk disparities at the national level to clarify whether within-tract sorting biases flood risk estimates.

Furthermore, while our study provides a national-scale analysis of federally-overlooked flood risk, it does not examine how these risks vary by states or regions. Prior research has demonstrated state-level heterogeneity in the relationship between social demographics and flood risk in the U.S. Southeast22, suggesting that patterns of flood exposure and inequity may vary across states. Future research should explore these variations at the state or local level to better understand regional heterogeneity in federally-overlooked flood risks.

Implications

Findings suggest that FEMA should adopt a forward-looking flood risk assessment and management approach sensitized to emergent local risks in the context of ongoing land use/cover and climate changes. Relying on historical records to respond to current and future flood risks will likely exacerbate existing inequities, as socially disadvantaged populations experience flood risk disparities now and are projected to face worsening disparities in the future3. Revising FEMA’s approach to incorporate accurate flood risk information, however, is costly endeavor—estimated between $3 and $12 billion35. Despite the Biden administration’s initial proposal to itemize $600 million for FEMA flood risk mapping improvements, that item was dropped from the Inflation Reduction Act in August 2022. In sum, while there has been some political movement at the national level, it has thus far been inadequate to address the scale and scope of the nationwide problem of federally-overlooked flood risk.

Communities in floodplains not recognized by federal flood maps lack sufficient protection against flooding, highlighting the imperative for investments that enhance flood resilience. This need is particularly pronounced in metro areas vulnerable to pluvial flooding, as our findings revealed. Addressing the problem demands careful planning to guide the development of resilient and sustainable infrastructures, land-use and emergency response plans, and flood mitigation and recovery programs that reduce risks, especially for socially vulnerable people and their property. Without comprehensive investments in plans and policies to address federally-overlooked flood risks at all levels in the U.S., flooding will likely impact more Americans, inequitably so, in the coming years. While expanding federally recognized floodplain boundaries to include areas currently overlooked by FEMA could improve flood risk awareness and access to protective resources, such changes may also have significant environmental justice implications. For socially disadvantaged homeowners, inclusion in a newly mapped floodplain may result in increased financial strain due to mandatory participation in the NFIP31. This additional cost burden may be particularly challenging for lower-income and minority homeowners, who often have limited financial flexibility. To ensure that flood risk mapping updates do not inadvertently exacerbate existing inequities, future policy efforts should incorporate affordability measures, such as financial assistance programs, to mitigate the impact on socially vulnerable populations.

It is also possible that newly designated floodplains could experience declines in property values, leading to in-migration of lower-income residents, there is limited national evidence that this is a widespread pattern following FEMA floodplain designations. Furthermore, any price discount tied to FEMA-designation floodplains would not apply to federally-overlooked floodplains, as these area areas not officially recognized as high-risk at the time of purchase or rental. Without formal flood risk designation, potential buyers and renters in these areas may not be aware of the flood hazard, making it unlikely that housing markets would respond in the same way as they might for FEMA-designated floodplains.

Materials and methods

Dependent variables: census tract population in federally-overlooked 100-year floodplains

We focus on federally-overlooked 100-year floodplains at the level of census tracts for the entire CONUS (Fig. 4). To generate measures of the population residing within each type of federally-overlooked floodplain, we first obtained 10-meter resolution raster data from the Bates et al. model. This flood model provides 100-year flood risk estimates for the entire CONUS, categorized by coastal, pluvial and fluvial flood types. Subsequently, we retrieved FEMA digital Flood Insurance Rate Maps (DFIRMS) from the National Flood Hazard Layer. Next, we superimposed both sets of floodplain data, extracting all 100-year floodplains identified by the Bates et al. model that fall beyond FEMA classified 100-year floodplains. These areas represent the locations of 100-year floodplains that are federally-overlooked.

Fig. 4
figure 4

United States census tracts.

Full size image

Studies in the U.S. that investigate inequities in flood risk typically assume uniformly distributed populations within census units and floodplains, employing GIS techniques like polygon containment or areal apportionment, which introduces errors into flood risk estimates and weakens the basis for inferences36. To improve accuracy, we utilized dasymetric mapping, a technique that refines population distribution within census boundaries by integrating supplemental land classification data37,38. To redistribute census populations to specific land cover classes considered habitable, we used the Intelligent Dasymetric Mapping (IDM) Toolbox39. In this process, we designated 2010 census blocks as source zones and utilized the National Land Cover Database from 2016 as ancillary data, resulting in a 30-meter resolution population density raster. We overlaid that raster with each type of federally-overlooked floodplain to compute estimates of population risk at a fine scale. Subsequently, these estimates were aggregated to the census tract level, producing four outcomes representing counts of individuals within each tract at risk to each type of federally-overlooked flooding, including combined, coastal, pluvial, and fluvial.

This dasymetric approach has been validated in previous research. For example, Baynes et al. assessed IDM-based population estimates against U.S. census block-level data, finding a normalized root mean square error ranging from 1.21 to 3.39 across states40. These values indicate low relative error, with better accuracy observed in urban areas. Their study also tested the accuracy of IDM by aggregating block-level population estimates back to census tracts and comparing them with census reported totals, demonstrating high accuracy40. Given its demonstrated accuracy, this dasymetric approach provides more accurate and spatially refined estimates of population exposure to federally-overlooked flood risk than reliance on areal based metrics.

Covariates

We utilized census tract-level sociodemographic data obtained from the American Community Survey (ACS) five-year estimates spanning 2015–2019. Variables in our analysis included race/ethnicity, with proportions calculated for non-Latinx Black, Latinx, non-Latinx Asian, and non-Latinx Other/Multiple race residents, using the proportion of non-Latinx White residents as the reference. Additionally, median household income was employed as a measure of socioeconomic status. In line with our focus on examining inequities, these variables enable us to assess differential risks of federally-overlooked flooding for various social groups.

We accounted for water-related environmental (dis)amenities that are known to influence residential choices, as they may confound associations between social characteristics and flood risk. We included four control variables: recreational housing41, tree canopy cover42, urban imperviousness43, and industrial facilities24. For recreational housing, which in the context of flood risk reflects water-based amenities, we used 2015–2019 ACS estimates to calculate the proportion of housing units in each census tract designated for seasonal or recreational purposes24,44. For tree canopy cover, we used a measure of the mean proportion of canopy cover in habitable areas within each census tract using 30-meter resolution data from the Multi-Resolution Land Characteristics (MRLC) Consortium (2016). We measured urban imperviousness by calculating the mean proportion of impervious surfaces in habitable areas for each census tract, utilizing 30-meter resolution data obtained from the MRLC dataset. For industrial facilities, we calculated each census tract’s proportion of population residing within 1-kilometer buffers of Toxic Release Inventory (TRI) facilities, with data obtained from the U.S. EPA TRI Explorer (2019). These factors are important to adjust for due to the influence of water-related (dis)amenities on the social patterning of flood risk. By including these variables, we aimed to control for potential confounding effects in our analysis of flood risk inequities.

To adjust for the influence of rural and urban contexts on flood risk inequities, we employed rural-urban commuting area (RUCA) codes from the U.S. Department of Agriculture (USDA), to categorize census tracts into metropolitan core, suburban, and rural classes. In our comprehensive multivariable models, we adjusted for urban-rural context by introducing dummy variables for suburban and rural classes. This approach enables the interpretation of results for these RUCA codes in comparison to RUCA code 1 (metropolitan area core), which encompasses the majority of census tracts. For our stratified multivariable models (research question 4), we separated tracts into metropolitan (RUCA code 1), suburban (RUCA codes 2–3), and rural (RUCA codes 4–10) subgroups as separate analyses, allowing us to investigate similarities and differences in federally-overlooked flood risk inequities across metropolitan core, suburban, and rural contexts.

Analysis approach

To address research question one, we summed the number of residents in census tracts within federally-overlooked floodplains, in terms of all flood types combined as well as by flood type. We then visualized these distributions across the CONUS and calculated state- and county-level statistics to identify highest-risk areas.

In conducting multivariable analyses, we employed generalized estimating equations (GEEs) to estimate the adjusted population count of residents within federally-overlooked floodplains. All analyses were conducted at the census tract level. To answer the second research question, we implemented a comprehensive model encompassing every census tract in the CONUS (n = 64,828). For the third research question, we modeled inequities by flood type. This resulted in 15,589 tracts in the coastal model, 64,820 in the pluvial model, and 56,728 in the fluvial model. To answer the fourth research question, we fit stratified models after separating census tracts into metropolitan (RUCA code 1), suburban (RUCA codes 2–3), and rural subgroups (RUCA codes 4–10). Sample sizes for each of these models are listed in the notes of Table 3. Census tracts were excluded from our analysis if they had missing data for the variables under analysis, populations below 500, and fewer than 200 housing units. Additionally, we omitted census tracts in counties without DFIRMs and those with completely unmapped areas. This exclusion was based on the unique processes governing how population risks to flooding are unaccounted for in these contexts compared to counties with FEMA DFIRMs. By our calculations, 3,391 tracts representing ~ 4.5% of the CONUS population lack FEMA FIRMs completely or have FIRMs in an undigitized format. For each multivariable analysis by flood type, we further excluded tracts in counties with fewer than 100 residents within each respective type of federally-overlooked floodplain due to the very low levels of risk in those contexts.

GEEs facilitate examination of flood risk inequities by accommodating variables that do not exhibit normal distributions and by adjusting for the presence of clustering, where observations within the dataset may exhibit correlation or grouping tendencies45. Clusters for GEEs were classified according to the median decade of housing construction (≤ 1939, 1940–1949, 1950–1959, 1960–1969, 1970–1979, 1980–1989, 1990–1999, 2000–2009, ≥ 2010) within each county. Defining clusters according to the median decade of housing construction by county corresponds to the varied contexts of residential development, including distinctions between older inner-city neighborhoods and more recent urban developments. This cluster definition has been widely used in environmental justice research24,46 and aligns with historical and spatial patterns of environmental inequality in the U.S47. For instance, systemic biases in housing and urban development have linked older inner-city neighborhoods with poverty and communities of color, contrasting with recent housing developments in outlying suburban areas associated with social privilege47.

Since our dependent variables represent counts of tract residents in federally-overlooked floodplains, Poisson and negative binomial distributions were suitable options, with the latter being more appropriate due to overdispersion. To select the best fitting model, we computed negative binomial models using exchangeable, independent, and unstructured correlation matrices, as well as logarithmic and identity link functions45. The models yielding the best fit (lowest QIC values) included the exchangeable correlation matrix and a logarithmic link function. To account for different population sizes, we included the natural log of each tract’s total population as an offset in the GEEs48. Before integrating them into the models, we transformed continuous covariates to z-scores. We evaluated potential multicollinearity using tolerance, variance inflation factor, and condition index statistics, and our models were not affected by multicollinearity. Additionally, we conducted three sensitivity analyses to ascertain the robustness of results under varying census tract inclusion criteria (see Table S1, Supplementary Information).

Sensitivity analyses with multivariable models

In addition to our main analysis, which includes census tracts in counties with 100 or more residents at risk to federally-overlooked flooding, we conducted three sensitivity analyses separately for each flood type to examine the robustness of the race/ethnicity and income findings to alternative tract inclusion criteria. Specifically, we examined the effects of including all census tracts, tracts in counties with FEMA FIRMs, and tracts in counties with 1,000 residents or more at risk to federally-overlooked flooding. Results are provided in Table S2 of the Supplementary Information. All of our findings are robust to alternative tract inclusion criteria, with two exceptions: In the federally-overlooked coastal floodplain model, the non-Latinx other/multiple race variable lost statistical significance when restricting to tracts in counties with at least one resident at risk; and in the federally-overlooked fluvial floodplain model, the Latinx variable lost significance in each of the three sensitivity analyses.

We conducted an additional sensitivity analysis to assess how results might vary using interaction effects instead of the stratified (subgroup) approach to determine differences in associations between socio-demographic variables and flood risk by metro (RUCA code 1, excluded as reference category in interaction models), suburban (RUCA codes 2–3), and rural contexts (RUCA codes 4–10). Results are provided in Table S5 of the Supplementary Information. Findings indicate that the direction of associations between socio-demographics and flood risk in each context remained generally consistent between both approaches. A few additional statistically significant findings emerged with the interaction effects. While Asian was negatively and significantly related to coastal risk in the metro model, there was a positive, non-significant relationships between coastal risk and Asian composition in the suburban model (Table 3). The interaction between Asian and suburban was significant, indicating that coastal risk decreases in metro contexts but increases in suburban contexts based on greater Asian composition (Table S5). There was a positive, non-significant relationship between pluvial risk and Asian composition in the metro model and a negative and non-significant relationship in the rural model (Table 3). The interaction between Asian and rural was significant, indicating that pluvial risk decreases in rural contexts but increases in metro contexts based on greater Asian composition (Table S5). There was a positive, non-significant relationship between pluvial risk and Other race/ethnicity composition in the metro model; that association was negative and not significant in the suburban and rural models (Table 3). The interaction effects between Other and suburban and Other and rural were significant, indicating that pluvial risk decreases in suburban and rural contexts but increases in metro contexts based on greater Other race/ethnicity composition (Table S5).