Triple exposure: the geographic correlation between flood risk, climate skepticism, and social vulnerability in the United States

This study employs flood risk data generated by the First Street Foundation. Their flood model estimates flooding probabilities from fluvial, pluvial, coastal, and storm surge for approximately 142 million properties across the contiguous U.S. Using 1980–2010 as a baseline, the model examines various pathways under greenhouse gas emission scenarios. The output is derived from an ensemble of 21 Global Circulation Models to address uncertainty. Additionally, the dataset incorporates increased stormwater infiltration rates resulting from both gray and green infrastructure projects, where feasible. This supplements the hydraulic characteristics derived solely from natural streamflow, offering a more comprehensive depiction of local adaptation strategies. Recent flood risk assessments have utilized this flood model [16, 23], recognizing its broader coverage compared to the hydrologic and hydraulic models typically used to derive the FEMA FIRMs [5].

We accessed the parcel-level flood risk data (version 2.0) through First Street Foundation's Probability API (version 1.2.0) using Python. Each property is assigned an overall risk score ranging from 1 to 10, where 1 signifies minimal flood risk and 10 indicates extreme risk. This score is computed based on the cumulative risk over 30 years and flood depth, calculated at the lowest elevation of the building footprint. Subsequently, we aggregated the initial parcel-level data by state and computed the average flood risk at both census-tract (n = 72 337) and county (n = 3108) levels across the contiguous U.S.

To pinpoint disparities between the FEMA Special Flood Hazard Area (SFHA) and the 100-year floodplain according to the First Street model, we utilized a publicly available dataset from First Street's data repository, offering summary statistics for the model (version 1.3). This dataset includes counts of properties within specific flood zones and aggregates the data to various jurisdictional levels for comparison with FEMA's SFHA [9].

We used census tract-level socioeconomic data from the U.S. Census Bureau, specifically the CDC's Social Vulnerability Index (SVI) [10, 24] (SI table 1). The SVI integrates variables reflecting communities' capacity to prepare for and respond to natural hazards. Specifically, we included variables reflecting the socio-economic status of communities including income, poverty level and unemployment as well as mobile homes, educational attainment, and minority status as direct indicators of how well a community can endure and rebound from a disaster event. We also included vehicle ownership, disability, and health insurance as indicators of vulnerability at different phases of a flood event.

We assessed knowledge vulnerability by examining the correlation between flood risk and public opinions on climate change, using the Yale Climate Opinion Survey [25]. This survey aggregates data from national surveys and employs geo-statistical modeling to provide subnational estimates. It includes 62 geolocated questions related to climate change. Among these, 31 questions gauge positive attitudes towards climate change, while the remaining 31 gauge opposing views or uncertainty. The survey encompasses inquiries about beliefs in climate change, levels of concern, and frequency of discussions with family and friends. It also includes questions on personal experiences with the impacts of climate change, perceived potential harm to others, and the frequency of media coverage. Additionally, the survey addresses beliefs about whether climate change is anthropogenically induced and opinions on whether citizens should take more or less action to address it. Questions specifically related to hazards include perceptions of risk from climate-related events and support for policies aimed at mitigating these risks.

Our initial research sought to identify clusters of counties with high flood risk and a significant proportion of properties outside FEMA's SFHA boundaries, referred to hereafter as FEMA undercounts. To pinpoint these clusters, we compiled average flood risk scores and FEMA undercounts per county. Employing the bivariate local indicators of spatial association (LISA) spatial clustering model, we parameterized it with neighbors weighted according to the first-order Queen's rule of adjacency [26]. For this analysis, we set the significance level at p > 0.05 and adjacency to one county.

2.4.1. Selecting subregions

We employed multinomial ordinary least squares (OLS) regression models to identify social vulnerability variables that explain flood risk at the census-tract level. We assessed multicollinearity among the predictors using the Variance Inflation Factor (VIF) with the regclass package in R [27], ensuring all covariates had VIF scores below 5 [28]. We visualized significant variables using boxplots and by comparing to national averages. We then applied the same methodology and ran OLS models for each region separately.

We focused on the 31 survey questions capturing opposing views or uncertainty about climate change and computed bivariate Pearson correlations between these questions and flood risk, at the county level. We conducted this analysis at multiple scales: national (n = 3108), high-risk clusters with FEMA undercounts (n = 250), and each subregion individually: Appalachia (n = 129), Northwest (n = 93), and New England (n = 28).

Our exploratory analysis and modeling highlighted Appalachia as exhibiting the most pronounced social vulnerability, alongside flood risk. To further investigate this relationship in Appalachia, we employed bivariate LISA clustering models to identify areas with high projected FDP and high percentage of poverty, unemployment, disability, and mobile homes. We used FDP data from Collins et al [29] and calculated the average value at the census-tract level. This dataset extends beyond flooding vulnerability or risk, encompassing the likelihood of floods causing direct impacts on communities, the economy, and the built environment. Given our focus on a region characterized by high flood risk and FEMA underestimation, and therefore ill-prepared to offer support, this dataset is particularly pertinent for identifying communities expected to bear substantial costs due to this underestimation.

The LISA spatial clustering analysis identified three primary regions with high flood risk and significant FEMA undercounts: the Northwest, Appalachia, and New England, along with several smaller clusters across the contiguous U.S. (figure 1). These regions are at high risk of flooding according to the First Street model, but FEMA's SFHA boundaries underestimate the number of properties at risk, leaving them less prepared for potential flooding. The Northwest region includes extensive areas of Washington, Oregon, and Idaho, eastern Montana, and two counties in eastern Wyoming. The Appalachia region covers most of West Virginia, southeastern Ohio, eastern Kentucky, eastern Tennessee, western North Carolina, western Virginia, and parts of northern Georgia. The New England region comprises counties in upstate New York, central Pennsylvania, and parts of Vermont and New Hampshire.

Zoom In Zoom Out Reset image size

After extracting the census tracts within the counties of the three major clusters where high flood risk aligns with high FEMA undercounts, we conducted an exploratory regression analysis to investigate the relationship between flood risk and proxies of social vulnerability. The results (SM-table 2) indicate that census tracts with high flood risk and FEMA undercounts are characterized by higher poverty, elevated unemployment rates, lower high school diploma attainment, increased disability rates, higher proportions of mobile homes, and more households lacking vehicle access. Per capita income was not a significant predictor, and fewer individuals identified as minorities in these regions. Health insurance rates and cost-burdened housing were significant but negative predictors.

Focusing on census tracts within the Appalachia cluster, we found that this region primarily drove the overall modeling results. In Appalachia, most social vulnerability indicators significantly correlated with flood risk, resulting in a model with good explanatory power (R²: 0.21) (figure 2 and SM-table 5). High poverty levels, elevated unemployment rates, lower educational attainment, increased disability rates, higher prevalence of mobile homes, and lack of vehicle access were all significant predictors of flood risk.

Zoom In Zoom Out Reset image size

In contrast, the social vulnerability indicators within the New England and Northwest clusters showed lower explanatory power for flood risk. In New England, vehicle availability was the only significant predictor (see SM-table 3) while in the Northwest, educational attainment, disability rates, and minority population were significant but accounted for a smaller fraction of flood risk variation (see SM-table 4). These findings highlight Appalachia's heightened social vulnerability and its substantial impact on flood risk and community resilience.

Our analysis reveals a concerning trend in regions facing high flood risk, significant FEMA undercounts, and high social vulnerability: elevated rates of climate change skepticism as compared to the national average (see figure 3; SM-table 7). We find that in these particular regions there is a notably higher correlation between flood risk and a substantial percentage of individuals who dismiss the reality of climate change or express minimal concern (see figure 4). This skepticism seems rooted in lack of awareness, as many residents seldom encounter discussions about climate change in the news or with family and friends.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Residents in these regions often mistakenly believe that climate change is a natural process with minimal impact on weather patterns. Paradoxically, despite living in high-risk zones, many believe they will be minimally affected, and that the adverse effects of climate change will take decades to manifest, if at all. We also find a strong correlation between flooding risk and disagreement with scientific evidence on climate change. This disconnect between scientific consensus and local understanding exacerbates vulnerability to climate change consequences.

Regarding regulation and mitigation, a significant portion of residents oppose measures such as adopting energy-efficient technologies and reducing carbon emissions. They also display little inclination to support individual or collective efforts to mitigate climate change, advocating for reduced responsibilities for both citizens and corporations. This widespread skepticism towards prioritizing climate action poses a significant challenge to fostering comprehensive and coordinated responses to the climate crisis in these vulnerable areas.

Given Appalachia's significant flood risk, FEMA undercounts, social vulnerability, and climate change skepticism (see figure 4), we focused on evaluating the expected flood damage in this region. Using spatial clustering models with FDP data, we identified regions where flooding damages are more likely to occur. Our analysis highlights Eastern Kentucky, Southern Ohio, and parts of West Virginia as areas where flood damage is anticipated to be more prevalent. These areas also exhibit high levels of unemployment, poverty, mobile home rates, and populations with disabilities, suggesting residents will face challenges in responding to and recovering from future flooding incidents. This analysis complements our previous regression findings on Appalachia (see figure 5 and SM-table 2), providing geographic specificity that can inform interventions aimed at mitigating potential flooding-related damage to individuals, the economy, and the built environment of the region.

Zoom In Zoom Out Reset image size

Our findings reveal that parts of Appalachia, New England, and the Northwest are hotspots of high flood risk, with many properties being excluded from FEMA's SFHA. These high-risk areas align with previous studies [16, 23]. Our study extends beyond pragmatic flood risk, highlighting that not only do current FEMA FIRM boundaries undercount at-risk communities, but also many communities identified in the 100-year floodplain do not participate in the NFIP. This finding presents an opportunity for FEMA to leverage its connections with state, local, and tribal authorities to increase program participation. Tools like the Headwater Economics' Rural Capacity Index indicate that communities in these areas may lack resources for climate adaptation planning and implementation. Additional research is needed to understand why some communities do not participate in this program.

In recent years, FEMA has been releasing updated guidance for the base-level engineering of two-dimensional flood models. In a one-dimensional model, flooding is modeled by placing cross sections at even spacing across a stream. For a two-dimensional model, a mesh is overlayed on the floodway which allows for flow to be calculated in multiple layers. These models also can utilize rain-on-grid scenarios in addition to riverine inflows to model the floodplain directions [30]. This added dimension of inflow information can help communities obtain a better understanding of their flood risk. As FEMA continues to support the development of these models, there is potential to reduce the 'undercounted' properites at risk in the U.S.

Our findings indicate that a 'one-size-fits-all' approach to mitigating flood risk is impractical for the U.S. For instance, in Appalachia, social vulnerability proxies such as poverty, living in mobile homes, and the percentage of people with disabilities explain a large part of the variation in flood risk. In contrast, the relationship is weaker in the Northwest and New England. The findings in Appalachia align with previous national-level studies identifying high poverty rates as a significant factor [14]. As Kinzer et al [31] point out, public perceptions about people living in floodplains have influenced NFIP policy enforcement and reform. To provide more equitable insurance premiums, the NFIP's Risk Rating 2.0 program was introduced in 2021. However, 77% of policyholders saw an increase in their premiums, indicating that the program does not adequately consider a resident's ability to afford flood insurance. Our analysis provides a baseline with geographic specificity and highlights where residents with the highest flood risk and financial barriers exist.

The results of our study align with previous findings that the prevalence of mobile homes in Appalachia coincides with high flood risk, underscoring the need to tailor flood mitigation efforts for these residents. Past analyses have linked mobile home sitting within floodplains to increased flood vulnerability [32, 33]. Although federal flood insurance is available for mobile homes, our findings emphasize the need for state and county-level mitigation efforts. For example, a study of post-disaster recovery from a 2013 flood in Colorado found that mobile home park residents received buyouts based on post-flood property values, unlike homeowners who were compensated based on pre-flood values [33]. Our geographically specific findings provide the necessary nuance for targeted interventions to enhance preparedness and adaptation for these communities.

Furthermore, we found that areas in Appalachia with higher populations of people with disabilities are more likely to experience flooding. Although specific disabilities cannot be disentangled due to data aggregation and privacy protocols, individuals with disabilities generally have limited capacity to evacuate before a disaster or access medical care [34]. This hurdle, paired with our findings, highlights the need for emergency management professionals to incorporate specific strategies for people with disabilities in adaptation and mitigation strategies, and emergency response plans.

Our analysis reveals that people living in flood-prone areas tend to be skeptical of climage change, adding an additional layer of vulnerability. We argue that this extends to a reluctance to acknowledge individual or collective responsibility, posing direct challenges to creating and implementing adaptation and mitigation plans. This resistance manifests in four primary ways: lack of awareness, distrust in science, underestimation of societal damage, and opposition to mitigation and adaptation efforts. First, lack of awareness hampers the willingness to adopt mitigation strategies. Many individuals misapprehend the science behind climate change [35], subscribing to common refrains of climate change skepticism such as 'it is not real', 'humans are not the main cause', 'the impacts are not serious', 'the experts are unreliable' and 'solutions are inefficient' [36]. This leads to a discounting of the risks they face. Despite residing in flood-prone regions, a significant portion believe they will not be personally affected by climate change, nor do they anticipate harm to their communities or future generations. Our findings demonstrate that climate change is not a common topic of discussion among friends and family in these communities. This isolation leads to an echo-chamber effect, perpetuating opposition to mitigation and adaptation efforts and hindering climate change communication campaigns [21].

Without experiencing extreme events firsthand, skepticism often grows among individuals, as well as among local governments and policymakers who tend to prioritize short-term needs [37]. Ironically, flood mitigation infrastructure and social safety nets like flood insurance or recovery support programs, may provide a false sense of security, potentially influencing public perceptions and skepticism about climate change [38]. Despite the sentiment towards flood risk at the individual level, a growing number of states identify flooding mitigation as a high priority. This highlights a disconnect between mitigation and emergency management stakeholders and the public. For instance, 131 of the 156 mitigation actions outlined in the 2018 Kentucky Enhanced Multi-Hazard Mitigation Plan directly or indirectly address flooding [39]. Similarly, 49 of the 57 mitigation actions in the 2023 West Virginia State Hazard Mitigation Plan directly address flooding [40]. Future research should further investigate the connection between perceptions and past exposure to flooding and other hazards, as well as the role of measures in shaping these views.

Investigating the interplay between flooding risk, social vulnerability, and climate change skepticism builds upon the double exposure concept introduced by O'Brien and Leichenko [18], expanding their framework to encompass a 'triple exposure' condition. This reveals how these interconnected vulnerabilities exacerbate impacts due to flooding and create significant barriers to adaptation and mitigation. The added dimension of climate skepticism undermines trust in science-based conclusions and hinders community support for necessary policy interventions. Addressing these complex issues, therefore, requires integrated policy strategies that are attentive to intertwined socio-economic and environmental factors as well as to skepticism about climate change and its impacts.