Financial losses associated with US floods occur with surprisingly frequent, low return period precipitation

Financial losses associated with US floods occur with surprisingly frequent, low return period precipitation | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Financial losses associated with US floods occur with surprisingly frequent, low return period precipitation Adam Nayak, Pierre Gentine, Upmanu Lall This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6025742/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 19 Sep, 2025 Read the published version in Nature Water → Version 1 posted You are reading this latest preprint version Abstract Flooding in the U.S. leads to billions of dollars in financial losses annually, with projected increases due to escalating extreme precipitation, population growth, and deteriorating flood infrastructure. While federal regulation mandates flood insurance purchase within 100-year floodplains, analysis of millions of federal insurance claims reveals that most flood losses arise from frequent, low-intensity precipitation events relative to regional climatology, with average regional precipitation return periods of under five years. Similarly, precipitation linked to disaster aid and property buyouts has return periods averaging less than 20 years. Using unsupervised learning, we identify that space-time precipitation clusters associated with major storms dominate losses, emphasizing the need for flood risk assessments and mitigation strategies that account for recurrent spatiotemporal compound events. The findings bring the putative 100-year flood protection strategy into question and provide a focal point for the ongoing national discussions that underscore systemic challenges in U.S. flood preparedness. Earth and environmental sciences/Hydrology Earth and environmental sciences/Natural hazards Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Main Flooding leads to billions of dollars in financial losses annually in the United States (NCEI 2024). Between 1954 and 2014, 80% of US disaster declarations were related to flood or storm events (Lindsay and McCarthy 2015). The number of disaster declarations in the US has been increasing over time (C. Kousky and Shabman 2017 ). Financial losses due to flooding are also increasing, and this trend is projected to continue in the near term (Hallegatte et al. 2013 ; Paprotny et al. 2018 ). Systemic intervention in the form of preventative infrastructure, policy reform, and/or relocation is necessary to reverse this trend. Extreme precipitation is a primary driver of flood damage (Davenport, Burke, and Diffenbaugh 2021 ). Under global climate change, precipitation extremes are expected to be exacerbated as a warmer atmosphere is expected to hold more water (Tabari 2020 ; Pfahl, O’Gorman, and Fischer 2017 ). Extreme precipitation is not isolated, but rather typically clustered in space and time due to large scale synoptic patterns. Spatiotemporally clustered flood risks have long been cited to pose damage concerns (Jain and Lall 2001; Carolyn Kousky and Cooke 2012; Zscheischler et al. 2020 ; Bonnafous and Lall 2021). Under climate change, these effects are likely to worsen, with projections of more frequent long-duration and intense storms (Hirabayashi et al. 2013 ; Panagoulia and Dimou 1997). Attribution of causal mechanisms for flood losses is further complicated due to interactions between extreme events, infrastructure health and population exposure. Flood infrastructure in the US is rapidly deteriorating, aging levees and floodwalls sparking national concern (Tobin 1995 ). Although flood protection measures are typically designed for 100-year return periods (Bell and Tobin 2007 ), recent analysis shows that recent U.S. dam failures due to overtopping have predominantly occurred under precipitation with much lower return periods (Hwang and Lall 2024). Recent federal legislation such as the Bipartisan Infrastructure Law (BIL) seeks to rebuild deteriorating critical infrastructure (The White House 2021 ), but many efforts are overdue and repairs cannot keep pace with failures (Petroski 2016 ). Further, the "levee effect" feedback loop exacerbates risk: structural protection spurs urban development in areas nominally protected from a 100-year flood, increasing pressure to sustain and expand these protections (Tobin 1995 ). Recent studies document that population growth and development in flood zones are leading to increased flood risk exposure (Neumann et al. 2015 ; Tellman et al. 2021 ; Rentschler et al. 2023 ). U.S. federal financial preparedness and recovery mechanisms for flooding are under stress from these natural and human-made compounded risks. Financial mechanisms intended to buffer the impact of financial flood losses at the household level include flood insurance, individual disaster aid, and grants for managed retreat – the acquisition of severely damaged properties (FEMA 2024a; Carolyn Kousky 2018 ). Historically, private insurers withdrew from flood insurance markets in the US due to the concentration of risk, leading to the creation of the National Flood Insurance Program (NFIP) in 1968 (Institutes for Research 2005 ). NFIP insurance covers over 95% of flood insurance policies in the U.S., and is available to any household (C. Kousky et al. 2018 ); however, the vast majority of policyholders are government-mandated due to being located within a FEMA 100-year flood zone under a federally-backed mortgage (Carolyn Kousky 2018 ). Since inception, the NFIP has struggled to recover losses: currently the program is over $ 20B in debt even after $ 16B of forgiveness was granted by Congress in 2017 (FEMA 2024c). In order to improve the financial balance of the program in addition to past debt relief, a new risk-based premium methodology (Risk Rating 2.0) was introduced in 2023 for pricing premiums more accurately to reflect risk (FEMA 2023 ). Similarly, US disaster assistance has frequently come up short of funds. The Stafford Act was passed in 1988 by Congress as a systematic effort to disburse federal natural disaster assistance (Moss, Schellhamer, and Berman 2009 ; Lindsay and McCarthy 2015; FEMA 2024a). Disaster aid in the form of Individual Assistance (IA) and Public Assistance (PA) and ex-post event buyout programs (primarily funded through Hazard Mitigation Grant Program, HMGP) require the declaration of a major disaster by the President to be issued (FEMA 2024a; Moss, Schellhamer, and Berman 2009 ; Lindsay and McCarthy 2015). However, due to an uptick in declarations in the 21st Century (NCEI 2024), spending restrictions on FEMA’s Disaster Relief Fund (DRF) have been imposed nine separate years since 2003 due to lack of available funds to meet needs assessments (FEMA 2024b). Although recent studies discuss economic limitations of the NFIP (Michel-Kerjan 2010 ; Carolyn Kousky and Kunreuther 2014; Carolyn Kousky 2018 ; Kunreuther 2021 ; de Ruig et al. 2022 ), interactions between disaster aid and flood insurance (Carolyn Kousky, Michel-Kerjan, and Raschky 2018; C. Kousky and Shabman 2017 ), or climatological changes in extreme precipitation (Hirabayashi et al. 2013 ; Pfahl, O’Gorman, and Fischer 2017 ; Rahmani and Harrington 2019; Visser et al. 2023 ), to our knowledge no recent work provides a systematic evaluation of the characteristics of precipitation events that lead to federal financial losses. In this work, we explore this aim, conducting a first assessment of precipitation return periods associated with financial losses across three major sources of federal funding. Specifically, we define federal financial losses to include 1) insurance claims, 2) disaster aid, and 3) property buyouts for severely damaged residences. To achieve our primary objective of evaluating the regional return periods of precipitation associated with damage losses nationwide, we examine several key questions: 1. How do evaluations of loss-inducing precipitation events vary temporally (with event duration), spatially (across counties), and by event type (severe storms, hurricanes, etc)? 2. How does spatiotemporal clustering impact the evaluation of loss-inducing precipitation? 3. What do nonstationary trends and changes over time in precipitation indicate for future flood-related losses? Our evaluation covers the reported history of flood-related, household-level national financial losses and their associated precipitation events through 2020. We conduct a thorough evaluation of millions of federal financial support records across insurance, disaster aid, and property buyouts dating back to 1978[1] . Using traditional statistical methods of flood risk evaluation paired with unsupervised machine learning, we provide a robust evaluation of loss attribution, and evaluate sensitivities to varying precipitation datasets, clustering parameterizations, and methods of nonstationary analysis. Results We find that flood-related financial losses are predominantly associated with low return period precipitation events with sub-decadal recurrence, indicating that the vast majority of financial flood losses in the United States are associated with low-intensity, high-frequency events relative to regional climatology. We define a loss event based upon household-level federal financial support mechanisms for ex-post flood recovery, which include 1) an insurance claim, 2) an individual disaster aid payment, and/or 3) a property buyout. A return period is defined as the expected time between occurrences of an event at or above a given magnitude. Precipitation event return periods are calculated as the inverse of the event probability in any given year. A summary of our major findings includes: 1. Return periods for precipitation associated with insurance claims average under 5 years, while for household disaster aid and property buyouts return periods average under 20 years. 2. Return periods vary spatially, by event type, and event duration. We find significantly lower return periods associated with losses in rural communities than urban communities, and surprisingly low return periods in areas with high average precipitation and coastal regions. Loss-attributed precipitation tends to exhibit higher return periods when examining short duration (daily to three-day) events. Return periods tend to be higher for dam/levee break, tropical cyclone, and flood disaster declarations than for coastal and severe storm declarations. 3. Over 90% of losses are associated with spatiotemporal clusters. While the maximum return periods of events within clusters are on average decadal to multi-decadal, the vast majority of loss clusters are attributed to events with return periods averaging under 20 years, i.e., the space-time organization that is associated with high loss clusters is often more frequent than the extremes locally embedded in the cluster. 4. Nonstationary analysis predominantly indicates significant increasing trends in return periods over time for losses, potentially indicating that the intensity of loss-inducing events is increasing, possibly due to a changing climate and human exposure factors. In our assessment, we evaluate multiple datasets to address potential data biases (Kang et al. 2024; NCAR 2025), examining Multi-Source Weighted-Ensemble Precipitation (MSWEP) daily gridded reanalysis data, European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis Version 5 (ERA5) daily data and Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly data (see SI sections 5 and 6). Unless otherwise specified, we present results for the MSWEP maximum daily precipitation preceding a loss event within 30 days. We filter out insurance claims below $1,000 and any losses associated with zero monthly precipitation (assumed likely erroneous due to human activity, snowmelt, or other anomalies). We provide a full sensitivity across different return period metrics and datasets in the SI Sections 5–7. 2.1) Variation by Time, Space, and Event Type In Fig. 1, we present the associated precipitation return periods for event durations of maximum daily, three-day, weekly, and monthly sums preceding the loss event. Event return periods are computed using the Generalized Extreme Value (GEV) distribution with the annual maxima of the precipitation for the corresponding duration for the county that experienced the loss. Low, frequent return periods under 20 years are noted across all financial loss types and event durations. Higher return periods for shorter durations (daily and three-day) for insurance claims and disaster aid are expected[2] . Similar return periods across weekly, three-day, and daily maximum events for property buyouts could indicate that heightened levels of extreme precipitation persistence are associated with severely damaged properties for buyout. ERA5 estimates for return periods are largely consistently lower than those produced by MSWEP (see SI, Section 4). Sensitivity to return periods using monthly PRISM data is illustrated in SI, Section 6. For each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range. We also examine spatial variations in return periods across counties (Fig. 2). Similar to previous studies (Yin et al. 2018), we find that in aggregate, rural and suburban counties tend to exhibit losses linked to insurance claims (panel a), disaster aid (panel b), and property buyouts (panel c) at lower return periods than urban counties (p < 0.001). Similar patterns of heightened return periods in southern regions frequently prone to tropical cyclones (such as lower Texas, Louisiana, Florida, and North Carolina) emerge across all three loss datasets. In snowmelt-driven regions such as North Dakota, Maine, Montana, and upstate New York, low return periods are largely anticipated, as any snowmelt flooding events should exhibit a lower correlation between floods and rainfall. However, across most counties, even coastal and inland regions that are rainfall-dominated, return period averages are predominantly under 20 years. The prevalence of low-return-period precipitation linked to financial support across the U.S. may be due to: 1) a large number of insured risk-prone properties encountering repeated losses, 2) deterioration/failure of flood control infrastructure under nonstationary climate conditions, and/or 3) wind damage being mistakenly filed under flood damage claims. Analysis using ERA5 data corroborates the short return periods across financial losses (see SI, Section 4). Next, we evaluate precipitation return period across event types by Presidential declaration. Figure 3 displays the spatial distribution of most common types of flood disaster declarations (panel a), the count of county-level declarations of major disasters (panel b), and the distribution of average precipitation return period by disaster type (panel c) across the contiguous United States (CONUS). Categories of flood-related major disaster declarations include hurricane, tropical storm, flood, severe storm, coastal storm and dam/levee break (full descriptions of flood-related disaster declarations are found in the SI, Section 2).[3] Spatial distributions of declarations reveal the dominant hydroclimatic forcings by region across CONUS with tropical storms dominating the South through most of Texas and the East Coast; severe storms dominating the Midwest, Pacific Northwest, and Northeastern U.S.; and other floods dominating most other snow-prone and temperate counties (Fig. 3, panel a). As seen earlier, the median values for the distributions of the maximum precipitation return period by county associated with each disaster declaration are estimated to be lower than 20 years across all major categories with the exception of dam/levee breaks (panel c), and the 100-year level is identified as an outlier. Return periods for severe and coastal storms are shown to be much lower on average than tropical cyclones, floods, and dam/levee breaks. 2.2) Spatiotemporal Clustering Using unsupervised learning, we also extract spatiotemporal clusters of flood-related losses and analyze their corresponding precipitation event return periods. Clustering losses allows for the evaluations of the movement of a storm and the spatiotemporal cascading effect of multiple related events (Fig. 4). As exhibited by clustering of disaster declarations and insurance claims for Hurricane Katrina (Fig. 4 panels a, c, e, g, i) and Hurricane Ike (Fig. 4 panels b, d, f, h, j), clustered losses result from a series of storms connected by large scale synoptic patterns. Spatiotemporal clustering allows us to group these loss events (panels a, b), and examine their return periods through space and time, visualizing the event trajectory (panels c, d). We can then examine the associated total financial losses across insurance claims (panels e, f), disaster aid (panels g, h), and property buyouts (panels i, j) in relation to the spatiotemporally clustered events. Spatiotemporal clustering of loss events reveals that over 90% of losses belong to space-time clusters. On average, regional return periods across clusters look similar (Fig. 5, panels c and d) to those in aggregate (Fig. 1). However, when we examine clustered event maximums, these events exhibit a much larger range in return periods, reaching up to 54.9 years on average across clusters for disaster aid (Fig. 5, panel b). This signifies that within a given cluster, there are typically “hotspots” of extreme antecedent precipitation, although the majority of clustered damages are induced under less extreme regional conditions. The higher return period for the maximum in a cluster in comparison to most individual loss instances is notable, since the trigger for a given declaration for the event is most likely its hotspot, but a number of proximal locations will experience impact. In aggregate, mean clustering results affirm the predominance of low-return period precipitation associated with losses. Even when considering disaster aid (which shows the greatest clustered maximum return periods), the first quantile for clustered maxima return periods falls just above a 10-year mark. This indicates that 25% of spatiotemporally clustered declarations exhibit a maximum regional precipitation recurrence period of a decade or less. When we examine the averages across these same clusters, we see the majority of payouts occurring within a given storm or cluster at a much lower regional return level, all under 15 years. 2.3) Nonstationary Trends Lastly, we examine sensitivity to monotonic trends in return periods over time (Fig. 6). To do so, we evaluate 1) the ratio of expected 100-year events by county under a nonstationary GEV (panels a and b), and 2) county-level monotonic trends in stationary return periods using the Mann-Kendall test and Sen’s Slope to evaluate significant trends in insurance claims with both MSWEP and ERA5 precipitation datasets (panels c and d). Nonstationary GEV results show discrepancies in trends between datasets (panels a and b), which is largely reflective of the difficulties of multi-decadal trend analysis across precipitation data with only a handful of decades on record. While ERA5 data implies a dominant decreasing trend in return periods across the United States, MSWEP implies dominant increases. Mann-Kendall tests indicate a dominance of significant (p < 0.05) increasing trends in return periods associated with claims across both datasets. Significantly increasing return periods could indicate increasing intensity of claim-triggering events. Mann Kendall tests for cluster return periods also indicate significant increasing return periods (p < 0.001). Our evaluation of regional trends reveals the importance of considering climatological shifts in our understanding of return periods over time. 2.4) Summary and Aggregate Costs In summary, examination of precipitation data at the time of insurance claims, disaster aid, and property buyouts suggests a high degree of losses result from relatively low-intensity, high-frequency precipitation across all three federal financial loss datasets. In aggregate, we find the median county-level return period of daily maximum precipitation linked to claims to be 3.9 years, disaster aid to be 13.2 years, and buyouts to be 8.1 years under traditional return period evaluation using the unclustered loss datasets and stationary GEV (Fig. 7, panel a). Average precipitation return periods for disaster aid and property buyouts by county tend to be higher than those associated with insurance claims (Fig. 7, panels b and c). This is expected since disaster aid and property buyouts require the issuance of a Presidential major disaster. In all cases, the 100-year return period that is nominally the criteria for mandated insurance, is associated with very few of the claims (scored as an outlier in the box plot, i.e. more than 1.5 standard deviations beyond the 75th percentile, which would correspond to less than 1% of the claims if they were normally distributed). Analysis using ERA5 data corroborates short return periods across financial metrics (see SI, Section 4). Figure 7 also depicts the scale at which federal expenditures occur across varying forms of financial support. Expenditures by program type are displayed at a national aggregate in panel b and a household level in panel c. Since insurance claims require policyholder premiums to offset claims, we show the historic annualized sum of premiums and 2024 annual sum of risk-based premiums (under FEMA Risk Rating 2.0) with black horizontal bars in panel b. Results reveal that disaster aid (only considering individual assistance, not public assistance) is the largest financial loss in aggregate. Household-level aid grants are small in comparison to insurance claim dollars and property buyout costs (panels b and c). Discussion Our comprehensive evaluation of the history of federal financial flood losses and the return periods of associated precipitation events reveals a dominance of frequent, low-intensity events linked to damages. Recent investigation in the Northeastern United States found that the majority of damaging floods were unassociated with extreme precipitation, corroborating a trend we find nationally (Teale and Winter 2024). Across modes of federal assistance in insurance, disaster aid, and property buyouts, median associated precipitation events are shown to be dominated by return periods lower than 20 years. With a history of flood preparedness measures focusing around the 100-year flood event in the United States and beyond, our analysis calls into question mitigation and preparedness design strategies for flood management that may be relying too heavily on outdated metrics that do not reflect the realities of repeated losses occurring nationally at lower severities. We evaluated spatially heterogeneous precipitation return periods, nonstationary distributions over returns, spatiotemporally clustered extremes, and a variety of event durations to capture temporally cascading precipitation. There are many ways to calculate spatially-heterogeneous return periods for precipitation events, and we present a county-level approach here. We highlight that this method will not fully capture stream network flooding effects, snowmelt, tidal flooding, storm surge, or sub-daily intense rainfall, yet the prevalence of linkages between damages and short return period precipitation events is concerning. Our method is spatially limited to county-level analyses, which are non-homogenous in size, presenting limitations for larger counties. Higher resolution financial loss data could allow for higher precision in regionally-specific return periods for each event. However, our work in spatiotemporal clustering alludes to the need for more robust measures of disaster-inducing storms that better capture the multidimensional distributions of clustered regional effects. We find that damage is frequently induced in spatiotemporal clusters, which have varying regional characteristics. To enhance preparedness, engineers must better quantify the compound return periods of these damage-driving events, and shift to an event-driven preparedness strategy that considers clustered hydroclimatic dynamics rather than the traditional 100-year return. A new approach to spatiotemporal flood recurrence analysis and prediction could inform flood risk at a portfolio level, noting the multiple geographies that may have near simultaneous exposure. Linking such exposure directly to the types of climate events responsible for clustered or non-clustered outcomes would permit improved projections of future flood risks, as well as of the associated financial implications for asset holders and the government programs charged with flood protection and response. Our work suggests that current structural approaches to flood risk management need to be critically examined. The NFIP and DRF are sources of billions of dollars of national financial debt, consistently overdrawing public funds. Methods of intervention such as risk-based pricing of NFIP premiums have been suggested and recently implemented (de Ruig et al. 2022 ; FEMA 2023 ). However, our findings showing the frequency of these losses begs the question of the effectiveness of pricing intervention if mass losses are occurring repeatedly. Additionally, evaluating the extent to which our current systems of flood preparedness are effectively reducing realized risks will be crucial to understanding how we can more proactively use federal resources to protect against anticipated hazards. Further, work showing low-return period failures from dam breaks (Hwang 2024 ) begs the question if traditional metrics for infrastructural design (such as the univariate 100-year flood) are effective enough to prepare for spatiotemporally clustered hydroclimatic pressures that are likely to be exacerbated under global climate change (Hirabayashi et al. 2013 ; Bonnafous and Lall 2021). Solving future flood risk management problems will require interdisciplinary approaches to systems engineering that consider variability in both our stochastic natural systems and human behavioral intervention points. In lieu of considering 100-year events in isolation, a dynamic approach for planning that considers sequences of more frequent extremes and quantifying characteristics of damage-inducing storms is necessary to holistically capture associated losses that are occurring at lower precipitation severities. Recent studies have considered methods of computing high dimensional return periods of compound events (Del, Urrea, and V. 2024) and modern stochastic simulators are attempting to capture varied cascading dynamics in flood frequency, intensity, and duration to propel more creative adaptive planning solutions (Nayak, Gentine, and Lall 2024 ). Future work should consider more robust assessment of vulnerabilities within the financial system to identify strategic points for future adaptive intervention. In order to enable proactive adaptive practices, planning must consider both the endogenous, institutional failure points within the national financial systems of natural disaster insurance and aid in relation to the exogenous nonstationary extreme hydroclimatic forcings. Data and Methods 4.1) Precipitation Data To assess the precipitation associated with flood losses, we assess the record of historical county flood loss records in relation to gridded precipitation history and reported dates of losses. In order to validate our findings, we use two daily gridded precipitation reanalysis datasets and one monthly gridded precipitation dataset: Multi-Source Weighted-Ensemble Precipitation (MSWEP) daily precipitation at 0.1 degree resolution from 1979 to 2020, ECMWF Reanalysis 5th Generation (ERA5) single-levels daily precipitation at 0.25 degree resolution from 1940 to 2023 , and Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly precipitation at 4 km resolution from 1923 to 2023 . MSWEP uses a weighted-ensemble method for daily precipitation estimates that sources gridded observations from gauge, satellite, radar, and bias-corrected reanalysis data providing the finest spatial resolution of the daily datasets but the shortest time horizon. ERA5 is developed using the European Centre for Medium-Range Weather Forecasts (ECMWF) model paired with data assimilation techniques, sourcing from gauge, satellite, lidar, and radar providing the longest temporal record for daily observations at a daily resolution. PRISM is developed through Oregon State University’s PRISM Climate Group and uses a combination of gauge data, topographic data, geographic features, and climatological normals to develop the longest history of monthly precipitation datasets for the continental US dating back to 1923. Gridded precipitation data was obtained directly from GloH2O for MSWEP daily data, ERA5 single levels since 1940 daily at 0.25 degrees were provided by the Copernicus Data Store and implemented by ECMWF, and PRISM monthly 4km precipitation was sourced from the PRISM Climate Group . We provide a table of all datasets used, their period of record, and their sourcing location in the SI, Section 1 . 4.2) Flood Loss Data All FEMA flood loss datasets are open access through OpenFEMA . For flood insurance claims, we analyzed over 2,000,000 claim records and over 80,000,000 policy records from the NFIP from its genesis in 1978 to May 2024 including records of current Risk Rating 2.0 policies (FEMA 2023 ). For property buyouts, we analyzed over 68,000 records of hazard mitigation assistance since its inception in 1979 to examine flood-related property buyouts across FEMA’s HMGP. For disaster aid, we examined over 250,000 county-level records of all Individual Assistance grants on file for both property owners and renters across the entire history of presidential disaster declarations, amounting to millions of federal aid household grants. Urban-rural classification information was obtained using the National Center for Health Statistics’ scheme provided by the CDC (See SI, Section 3 for more details of classifications). 4.3) Precipitation Event Extraction and Loss Attribution In order to attribute precipitation events to losses we evaluate a number of potential event characteristics. The date of the loss event is reported on NFIP claims (distinct from the date of claim filing), but NFIP claims can be filed up to 60 days after the incident which leaves room for human error in reporting. Thus, due largely to reporting error (for instance claims attributed to Hurricane Ian reported a date of loss up to 12 days after the latest incidence of precipitation), we examine the maximum continuous sums of 1, 3, 7, and 30 days within the preceding month of the reported date of loss for both MSWEP and ERA5 daily gridded datasets. We also tested lags of 5 days and 14 days within preceding windows of one week, two weeks, and one month, but found distributions to be largely encompassed by the daily, three-day, weekly and monthly values within a 30 day window of the reported event date. Similarly, disaster declarations list a begin date, but this can be up to a few days before the maximum extent of damage, which varies by event. Thus, we take a similar approach for disaster aid and relocation event attributions: extracting the maximum continuous sums of 1, 3, 7, and 30 days within one month of the reported date of loss for both MSWEP and ERA5 daily gridded datasets. In subsequent analysis, we filter all losses that are attributed with zero monthly precipitation (likely due to errors in date reporting, human activity, snowmelt-induced flooding, or other anomalies). We also compare this with the associated monthly values of precipitation in relation to claims using gridded PRISM reanalysis data in the SI, Section 6. 4.4) Calculation of Stationary and Nonstationary Return Periods Precipitation return periods of loss events are expected to vary spatially and temporally. In order to best characterize the local distribution of return periods for precipitation for county-level financial loss data, we extract a time series of weighted average precipitation values by county, considering the intersection of all gridded data pixels whose centroid is within county boundaries. If a given county is too small to intersect fully with any given pixel’s centroid, we source data from the pixel with the minimum Euclidean distance between the county’s centroid and surrounding pixels. We use this method for all three precipitation datasets, and fit separate generalized extreme value (GEV) probability distributions to each county-level annual maxima timeseries for the annual block maxima for each event duration. We use the GEV distribution to model return periods over the county annual maxima time series due to its widespread application in hydrometeorological settings and common applicability for return period calculation (Coles 2001 ). Under stationary conditions, for a given county \(\:c\) , we can consider probability distribution \(\:{P}_{c}\) as: $$\:{P}_{c}\:\sim\:GEV({\mu\:}_{{P}_{c}},\:{\sigma\:}_{{P}_{c}},\:{\gamma\:}_{{P}_{c}})$$ 1 In which \(\:{\mu\:}_{{P}_{c}},\:{\sigma\:}_{{P}_{c}},\) and \(\:{\gamma\:}_{{P}_{c}}\) denote the location, scale and shape parameters. The cumulative distribution function for the stationary GEV is given by: \(\:{F}_{{P}_{c}}({P}_{c}\) ) = \(\:{e}^{-{[1+{\gamma\:}_{{P}_{c}}(\frac{{P}_{c}-{\mu\:}_{{P}_{c}}}{{\sigma\:}_{{P}_{c}}}\left)\right]}^{-1/{\gamma\:}_{{P}_{c}}}}\) (2) We also fit the nonstationary GEV to each county \(\:c\) as seen in (Katz 2013 ) to consider linear trends in the return periods in which the location and scale parameters of the GEV become time varying by year \(\:t\) : $$\:{P}_{c,t}\:\sim\:GEV({\mu\:}_{{P}_{c,t}},\:{\sigma\:}_{{P}_{c,t}},\:{\gamma\:}_{{P}_{c,t}})$$ $$\:{\mu\:}_{{P}_{c,t}}=\:{\alpha\:}_{1c}+{\beta\:}_{1c}t$$ 3 $$\:{\sigma\:}_{{P}_{c,t}}=\:{\alpha\:}_{2c}+{\beta\:}_{2c}t$$ We test three configurations of the nonstationary GEV for each county \(\:c\) : 1) time-varying scale and time-invariant location, 2) time-invariant scale and time-varying location, and 3) both time-varying scale and time-varying location. We choose the distribution fit that best minimizes the Bayesian Information Criterion (BIC). All nonstationary distributions are fit using the extRemes package in R. Stationary distributions are fit using the SciPy package in Python. For each flood loss occurrence, the associated precipitation return period for a given GEV configuration at time \(\:t\) is found given: $$\:{T}_{{P}_{c,t}}\:=\:\frac{1}{1-{F}_{{P}_{c,t}}\left({P}_{c,t}\right|{\mu\:}_{{P}_{c,t}},\:{\sigma\:}_{{P}_{c,t}},\:{\gamma\:}_{{P}_{c,t}})}$$ 4 Lastly, for further statistical evaluations of significance, to test for significance across varying groups of comparisons, we performed Welch’s t-test for unequal variances, and three-way ANOVA for multi-group comparisons using the Python SciPy package. To evaluate monotonic trends in precipitation return periods for insurance claims, we use the Mann-Kendall test at a significance level of 0.05 as found in (Mann 1945 ), and extract the Sen’s Slope values to assess trends as detailed in (Sen 1968 ) using the Python package pymannkendall. 4.5) Spatiotemporal Clustering A longstanding method of clustering is DBSCAN (Ester et al. 1996 ) and its spatiotemporal extension ST-DBSCAN (Birant and Kut 2007). The premise of the original algorithm involves setting a distance threshold \(\:ϵ\) and a minimum number of points \(\:MinPts\) that must be within the Euclidean distance of a “core point”. The algorithm functions by starting with a random point \(\:p\) , iterating through its nearest neighbors, and clustering points that fall within the specified Euclidean distance threshold \(\:ϵ\) . It then continues to each clustered neighbor \(\:n\) and clusters the points within \(\:ϵ\) from \(\:n\) as long as the total is at least \(\:MinPts\) . If a point continues the iteration (by clustering with at least \(\:MinPts\) within \(\:ϵ\) ) it is a “core point”; if a point is clustered but does not continue the iteration (by clustering with less than \(\:MinPts\) within \(\:ϵ\) ) it is a “border point”; if a point does not fit to any cluster, it is labeled as “noise” and remains unclustered. ST-DBSCAN extends this approach to include both \(\:{ϵ}_{space}\) and \(\:{ϵ}_{time}\) , defining a “core point” such that it falls within both the time and space thresholds \(\:{ϵ}_{space}\) and \(\:{ϵ}_{time}\) set by the user. A major limitation in the use of DBSCAN and ST-DBSCAN is the ability to validate the choice of one set of threshold parameters against another. While other traditional clustering techniques, such as k-means, optimize clusters based upon a desired number of clusters, DBSCAN’s set of thresholding parameters can make its results sensitive, and the technique has met criticism due to its difficulty to parameterize (Schubert et al. 2017 ). Recent scholars have underscored its practical application for appropriate use cases (Gan and Tao 2015 ). One of these in recent years has emerged in the case of spatiotemporal clustering for extreme storm events, such as with lightning clusters (Augenstein, Mohr, and Kunz 2024 ; Shi et al. 2022 ). In our case, we similarly need to cluster extreme disaster-inducing events that have clear spatial and temporal clustering tendencies, but the number of such occurrences is unknown. With our data, there is a specific advantage for validation of the parameterization of the spatiotemporal clustering. For our disaster aid and property buyouts datasets, all presidential disaster declarations have an associated number for the same event and must specify the counties eligible for aid and their declaration date. Thus, there are natural “clusters” of aid and buyouts that spur at a county level by a given disaster declaration number. However, sometimes multiple disaster numbers are declared within a short period of time that are inherently related (eg. Hurricane Ike becoming a Tropical Storm Ike then a series of severe storms). Thus, we wanted to cluster our aid data in a way that did not split disaster declaration numbers, but rather encompassed them as a potential subset to the clusters extracted. More specifically, we aimed to choose the minimum space, time, and number threshold parameters that also minimized the number of presidential disaster declaration numbers split between multiple clusters. This gave us a strong validation metric for our clustering parameterization. As the claims dataset was not clustered in declarations, we first performed traditional DBSCAN with a temporal threshold by county, then spatiotemporally clustered each temporal cluster using ST-DBSCAN. We then performed a thorough sensitivity analysis on the characteristics of the extracted cluster duration, latitude and longitude span, and percent of data clustered across a range of thresholds for all three datasets (see SI Section 8). Declarations Data Availability All raw data that form the basis of this article are freely available. Financial loss records for insurance claims, disaster aid, and property buyouts can be obtained through OpenFEMA. ERA5 single-levels precipitation data can be obtained through the Copernicus Data Store, and MSWEP precipitation sourced from GloH2O. PRISM data can be accessed through Oregon State’s PRISM Group website. Records of urban-rural classification are obtained directly from the CDC’s National Center for Health Statistics. Processed datasets for financial losses and associated return periods can be accessed directly from this project’s GitHub. Code Availability All associated project code in Python and RStudio for data preprocessing, analysis, and evaluation can be accessed through this project’s public GitHub repository. Acknowledgements Financial support for this research was provided by the National Science Foundation Graduate Research Fellowship Program, and the Columbia Presidential Distinguished Fellowship from the Fu Foundation School of Engineering and Applied Sciences. Cloud computing resources were provided by the National Science Foundation’s Science and Technology Center for Learning the Earth with Artificial Intelligence and Physics (LEAP) at Columbia University (grant number 2019625). References Augenstein, M., S. Mohr, and M. Kunz. 2024. “Influence of the North Atlantic Oscillation on Annual Spatiotemporal Lightning Clusters in Western and Central Europe.” EGUsphere 2024:1–30. Bell, H., and G. Tobin. 2007. “Efficient and Effective? The 100-Year Flood in the Communication and Perception of Flood Risk.” Environmental Hazards 7 (4): 302–11. Birant, Derya, and Alp Kut. 2007. “ST-DBSCAN: An Algorithm for Clustering Spatial–temporal Data.” Data & Knowledge Engineering 60 (1): 208–21. Bonnafous, Luc, and Upmanu Lall. 2021. “Space-Time Clustering of Climate Extremes Amplify Global Climate Impacts, Leading to Fat-Tailed Risk.” Natural Hazards and Earth System Sciences 21 (8): 2277–84. Coles, Stuart. 2001. “Classical Extreme Value Theory and Models.” In An Introduction to Statistical Modeling of Extreme Values , 45–73. Springer Series in Statistics. London: Springer London. Congressional Research Service. 2023. “FEMA Pre-Disaster Mitigation: The Building Resilient Infrastructure and Communities (BRIC) Program.” https://crsreports.congress.gov/product/pdf/IN/IN11515. Davenport, Frances V., Marshall Burke, and Noah S. Diffenbaugh. 2021. “Contribution of Historical Precipitation Change to US Flood Damages.” Proceedings of the National Academy of Sciences of the United States of America 118 (4): e2017524118. Del, Jesus M., Méndez D. Urrea, and Gomez Rave D. V. 2024. “Return Period of High-Dimensional Compound Events. Part I: Conceptual Framework.” Hydrology and Earth System Sciences Discussions 2024:1–27. Ester, M., H. Kriegel, J. Sander, and Xiaowei Xu. 1996. “A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise.” Knowledge Discovery and Data Mining , August, 226–31. FEMA. 2021. “Flood Mitigation Assistance Community Flood Mitigation,” August. https://www.fema.gov/sites/default/files/documents/fema_fy21-fma-community-flood-mitigation_fact-sheet.pdf. ———. 2023. “NFIP’s Pricing Approach.” 2023. https://www.fema.gov/flood-insurance/risk-rating. ———. 2024a. “How a Disaster Gets Declared.” 2024. https://www.fema.gov/disaster/how-declared. ———. 2024b. “Immediate Needs Funding Fact Sheet.” FEMA Has Implemented Immediate Needs Funding Nine Times : , August. https://www.fema.gov/sites/default/files/documents/fema_ocfo-inf-fact-sheet.pdf. ———. 2024c. “NFIP Debt.” 2024. https://www.fema.gov/case-study/nfip-debt. Gan, Junhao, and Yufei Tao. 2015. “DBSCAN Revisited: Mis-Claim, Un-Fixability, and Approximation.” In Proceedings of the 2015 ACM SIGMOD International Conference on Management of Data . New York, NY, USA: ACM. https://doi.org/10.1145/2723372.2737792. Hallegatte, Stephane, Colin Green, Robert J. Nicholls, and Jan Corfee-Morlot. 2013. “Future Flood Losses in Major Coastal Cities.” Nature Climate Change 3 (9): 802–6. Hirabayashi, Yukiko, Roobavannan Mahendran, Sujan Koirala, Lisako Konoshima, Dai Yamazaki, Satoshi Watanabe, Hyungjun Kim, and Shinjiro Kanae. 2013. “Global Flood Risk under Climate Change.” Nature Climate Change 3 (9): 816–21. Hwang, Jeongwoo, and Upmanu Lall. 2024. “Increasing Risk of Dam Failures in the United States due to Compound Risk of Rainfall Clusters as Climate Changes,” February. https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=JA0o2TUAAAAJ:k1VWYO0b_yEC. Institutes for Research, American. 2005. “A Chronology of Major Events Affecting the National Flood Insurance Program.” Jain, Shaleen, and Upmanu Lall. 2001. “Floods in a Changing Climate: Does the Past Represent the Future?” Water Resources Research 37 (12). https://doi.org/10.1029/2001WR000495. Kang, Xiaoqi, Jianzhi Dong, Wade T. Crow, Lingna Wei, and Huiwen Zhang. 2024. “The Conditional Bias of Extreme Precipitation in Multi‐source Merged Data Sets.” Geophysical Research Letters 51 (22): e2024GL111378. Katz, Richard W. 2013. “Statistical Methods for Nonstationary Extremes.” In Extremes in a Changing Climate , 15–37. Dordrecht: Springer Netherlands. Kousky, Carolyn. 2018. “Financing Flood Losses: A Discussion of the National Flood Insurance Program.” Risk Management and Insurance Review 21 (1): 11–32. Kousky, Carolyn, and Roger Cooke. 2012. “Explaining the Failure to Insure Catastrophic Risks.” The Geneva Papers on Risk and Insurance Issues and Practice 37 (2): 206–27. Kousky, Carolyn, and Howard Kunreuther. 2014. “Addressing Affordability in the National Flood Insurance Program.” Journal of Extreme Events 01 (01): 1450001. Kousky, Carolyn, Erwann O. Michel-Kerjan, and Paul A. Raschky. 2018. “Does Federal Disaster Assistance Crowd out Flood Insurance?” Journal of Environmental Economics and Management 87 (January):150–64. Kousky, C., H. Kunreuther, B. Lingle, and L. Shabman. 2018. “The Emerging Private Residential Flood Insurance Market in the United States.” Wharton Risk Management and Decision Processes Center . https://riskcenter.wharton.upenn.edu/wp-content/uploads/2018/07/The-Emerging-Flood-Insurance-Market-Report.pdf. Kousky, C., and L. Shabman. 2017. “The Realities of Federal Disaster Aid.” https://media.rff.org/archive/files/sharepoint/WorkImages/Download/RFF-IB-12-02.pdf. Kunreuther, Howard. 2021. “Improving the National Flood Insurance Program.” Behavioural Public Policy 5 (3): 318–32. Lindsay, Bruce R., and Francis X. McCarthy. 2015. “Stafford Act Declarations 1953-2014: Trends, Analyses, and Implications for Congress.” Congressional Research Service 42702. Mann, Henry B. 1945. “Nonparametric Tests Against Trend.” Econometrica: Journal of the Econometric Society 13 (3): 245. Michel-Kerjan, Erwann O. 2010. “Catastrophe Economics: The National Flood Insurance Program.” The Journal of Economic Perspectives: A Journal of the American Economic Association 24 (4): 165–86. Moss, Mitchell, Charles Schellhamer, and David A. Berman. 2009. “The Stafford Act and Priorities for Reform.” Journal of Homeland Security and Emergency Management 6 (1). https://doi.org/10.2202/1547-7355.1538. Nayak, Adam, Pierre Gentine, and Upmanu Lall. 2024. “A Nonstationary Stochastic Simulator for Clustered Regional Hydroclimatic Extremes to Characterize Compound Flood Risk.” Journal of Hydrology: X 25 (100189): 100189. NCAR. 2025. “Precipitation Data Sets: Overview & Comparison Table.” 2025. https://climatedataguide.ucar.edu/climate-data/precipitation-data-sets-overview-comparison-table. NCEI. 2024. “Billion-Dollar Weather and Climate Disasters.” 2024. https://www.ncei.noaa.gov/access/billions/state-summary/US. Neumann, Barbara, Athanasios T. Vafeidis, Juliane Zimmermann, and Robert J. Nicholls. 2015. “Future Coastal Population Growth and Exposure to Sea-Level Rise and Coastal Flooding--a Global Assessment.” PloS One 10 (3): e0118571. Panagoulia, Dionysia, and George Dimou. 1997. “Sensitivity of Flood Events to Global Climate Change.” Journal of Hydrology 191 (1): 208–22. Paprotny, Dominik, Antonia Sebastian, Oswaldo Morales-Nápoles, and Sebastiaan N. Jonkman. 2018. “Trends in Flood Losses in Europe over the Past 150 Years.” Nature Communications 9 (1): 1985. Petroski, Henry. 2016. The Road Taken: The History and Future of America’s Infrastructure . Bloomsbury Publishing. Pfahl, S., P. O’Gorman, and E. Fischer. 2017. “Understanding the Regional Pattern of Projected Future Changes in Extreme Precipitation.” Nature Climate Change 7 (June):423–27. Rahmani, Vahid, and John Harrington Jr. 2019. “Assessment of Climate Change for Extreme Precipitation Indices: A Case Study from the Central United States.” International Journal of Climatology: A Journal of the Royal Meteorological Society 39 (2): 1013–25. Rentschler, Jun, Paolo Avner, Mattia Marconcini, Rui Su, Emanuele Strano, Michalis Vousdoukas, and Stéphane Hallegatte. 2023. “Global Evidence of Rapid Urban Growth in Flood Zones since 1985.” Nature 622 (7981): 87–92. Ruig, Lars T. de, Toon Haer, Hans de Moel, Samuel D. Brody, W. J. Wouter Botzen, Jeffrey Czajkowski, and Jeroen C. J. H. Aerts. 2022. “How the USA Can Benefit from Risk-Based Premiums Combined with Flood Protection.” Nature Climate Change 12 (11): 995–98. Schubert, Erich, Jörg Sander, Martin Ester, Hans Peter Kriegel, and Xiaowei Xu. 2017. “DBSCAN Revisited, Revisited: Why and How You Should (still) Use DBSCAN.” ACM Transactions on Database Systems 42 (3): 1–21. Sen, Pranab Kumar. 1968. “Estimates of the Regression Coefficient Based on Kendall’s Tau.” Journal of the American Statistical Association 63 (324): 1379–89. Shi, M., W. Zhang, P. Fan, Q. Chen, Z. Liu, Q. Li, and X. Liu. 2022. “Modelling Deep Convective Activity Using Lightning Clusters and Machine Learning.” Int. J. Climatol 42:952–73. Tabari, Hossein. 2020. “Climate Change Impact on Flood and Extreme Precipitation Increases with Water Availability.” Scientific Reports 10 (1): 13768. Teale, Natalie, and Jonathan M. Winter. 2024. “The Relationship between Extreme Precipitation and Damaging Floods in the Northeastern United States.” Journal of Applied Meteorology and Climatology 63 (9): 1035–47. Tellman, B., J. A. Sullivan, C. Kuhn, A. J. Kettner, C. S. Doyle, G. R. Brakenridge, T. A. Erickson, and D. A. Slayback. 2021. “Satellite Imaging Reveals Increased Proportion of Population Exposed to Floods.” Nature 596 (7870): 80–86. The White House. 2021. “Fact Sheet: The Bipartisan Infrastructure Deal.” The White House. November 6, 2021. https://www.whitehouse.gov/briefing-room/statements-releases/2021/11/06/fact-sheet-the-bipartisan-infrastructure-deal/. Tobin, Graham A. 1995. “THE LEVEE LOVE AFFAIR: A STORMY RELATIONSHIP? 1 .” Journal of the American Water Resources Association 31 (3): 359–67. Visser, Johan B., Conrad Wasko, Ashish Sharma, and Rory Nathan. 2023. “Changing Storm Temporal Patterns with Increasing Temperatures across Australia.” Journal of Climate 36 (18): 6247–59. Yin, Jiabo, Pierre Gentine, Sha Zhou, Sylvia C. Sullivan, Ren Wang, Yao Zhang, and Shenglian Guo. 2018. “Large Increase in Global Storm Runoff Extremes Driven by Climate and Anthropogenic Changes.” Nature Communications 9 (1): 4389. Zscheischler, Jakob, Olivia Martius, Seth Westra, Emanuele Bevacqua, Colin Raymond, Radley M. Horton, Bart van den Hurk, et al. 2020. “A Typology of Compound Weather and Climate Events.” Nature Reviews Earth & Environment 1 (7): 333–47. Footnotes We note that FEMA also provides pre-disaster mitigation grants that are not contingent upon presidential disaster declaration through two programs: the Building Resilient Infrastructure and Communities (BRIC) grants and Flood Mitigation Assistance (FMA) grants for NFIP policyholders (Congressional Research Service 2023 ; FEMA 2021). However, since these funds are non-event contingent and can be requested at any time, not all grants have associated extreme events with a return period, so we do not analyze these records here. Since daily duration maximum precipitation returns were most consistently the highest across all three modes of loss events, we report daily durations for the remainder of the results, but sensitivity to varying durations are presented in the SI. Note that we simplify these categories into four major groupings: hurricane/tropical cyclone, flood, storm, and dam/levee break for clarity. We do not consider snowfall-related declarations here due to the inconsistent time lags associated with snowmelt-related flooding, and neglect tsunamis due to their seismologic forcing. Additional Declarations There is NO Competing Interest. Supplementary Files SINayakGentineLallNW2.13.25.docx Cite Share Download PDF Status: Published Journal Publication published 19 Sep, 2025 Read the published version in Nature Water → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6025742","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":418079455,"identity":"fe135818-ec8d-452c-bfa5-e4c413c607d0","order_by":0,"name":"Adam Nayak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYFAC9oMPP1TYAGkGBmYGA5BIAiEtPMnGEmfSQOqJ1sJgJsHbdhiqhYEILfyzGxIkJNjOy/Mz85huLiiwY+BnzzHAq0XizsEDBgU8tw1nNvOY3Z5hkMwg2fMGvxaGGwkJQGtuJxgcBmrhMQB65wYBW+RvJBgc4DE4B9NSz2BPSIvBjQTDBp6EAzAthxkMJAhoMbyRk8wscSAZ6Be2MqBfjvNInHlWgFeL3I304z8//rOT52dv3na74E+1HH978ga8WjAAD2nKR8EoGAWjYBRgBQCWcEND0qzjRwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0003-8401-4599","institution":"Columbia University","correspondingAuthor":true,"prefix":"","firstName":"Adam","middleName":"","lastName":"Nayak","suffix":""},{"id":418079456,"identity":"4fc7d9b9-9944-4af3-8bbf-6d98a00c9406","order_by":1,"name":"Pierre Gentine","email":"","orcid":"https://orcid.org/0000-0002-0845-8345","institution":"Columbia University","correspondingAuthor":false,"prefix":"","firstName":"Pierre","middleName":"","lastName":"Gentine","suffix":""},{"id":418079457,"identity":"79246fda-7a57-4f88-aafd-4f509ff7bbaf","order_by":2,"name":"Upmanu Lall","email":"","orcid":"https://orcid.org/0000-0003-0529-8128","institution":"Columbia University","correspondingAuthor":false,"prefix":"","firstName":"Upmanu","middleName":"","lastName":"Lall","suffix":""}],"badges":[],"createdAt":"2025-02-13 20:25:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6025742/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6025742/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s44221-025-00506-8","type":"published","date":"2025-09-19T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":79242819,"identity":"da4003ed-aae1-4271-95d8-7af2474a3e30","added_by":"auto","created_at":"2025-03-26 06:22:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":167163,"visible":true,"origin":"","legend":"\u003cp\u003eReturn Period Sensitivity to Event Duration\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSensitivity of return period distribution calculations for financial loss data under varying precipitation event durations (maximum daily, three-day, weekly, and 30-day precipitation event within 30 days of declaration for the county) across the MSWEP dataset. The identical figure for ERA5 data is found in the SI, Section 4. We provide further sensitivity to event duration across all figures with MSWEP data in SI, Section 5.\u0026nbsp;\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/0a150ed386547354a8259eef.png"},{"id":79242820,"identity":"9a2daf84-0715-4a74-a75d-0c4b9ed8d520","added_by":"auto","created_at":"2025-03-26 06:22:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":791652,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial Distribution of Return Periods by Financing Type\u003c/p\u003e\n\u003cp\u003eSpatial distribution and rural-urban classification of mean county-level precipitation return periods associated with varying modes of federal flood loss financing: insurance claims (panel a, b), disaster aid (panel c, d), property buyouts (panel e, f). Rural-urban classifications are derived from the US CDC National Center for Health Statistics’ Urban-Rural Classification Scheme (See SI, Section 3 for more details). Return periods (in years) reflect the maximum daily precipitation event within 30 days preceding declaration or date of loss modeled by the county-fitted stationary GEV distribution using MSWEP daily precipitation data. One-way ANOVA tests across all three forms of financial support show statistically significant differences between rural, suburban, and urban return periods (p\u0026lt;0.001, degrees of freedom \u0026gt; 1.67million, subgroup sizes each larger than 200,000). Full details on ANOVA results for each subplot shown here and in the supplement with degrees of freedom, test statistics, effect sizes and p-values are provided in the supplementary data. For each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/e6bdb4b9b5a7b108e71b10ba.png"},{"id":79244288,"identity":"3f2b5534-9c7d-48ad-9b20-2353904cafe0","added_by":"auto","created_at":"2025-03-26 06:38:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":496644,"visible":true,"origin":"","legend":"\u003cp\u003eTypes of Flood-Related Presidential Declarations\u003c/p\u003e\n\u003cp\u003eOverview of flood-related presidentially declared disasters by type. Panel a) reflects the mode of most commonly-declared event by CONUS county. Panel b) reflects the raw counts by event type per county of declaration. Panel c) shows the distribution of the return period in years of the maximum daily precipitation event within 30 days of declaration for the county by disaster declaration type modeled by county-fitted stationary GEV distribution using MSWEP daily precipitation data. For each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/96041c7734781f733e81a430.png"},{"id":79242821,"identity":"c4554877-4970-4740-8fd0-6bbf9b28312d","added_by":"auto","created_at":"2025-03-26 06:22:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":571882,"visible":true,"origin":"","legend":"\u003cp\u003eSpatiotemporal Storm Clustering\u003c/p\u003e\n\u003cp\u003eVisualization of spatiotemporal clustering for Hurricane Katrina and Hurricane Ike. The figure progresses from disaster declaration event types (panels a, b), their associated county-level precipitation return periods (panels c, d) to associated total counts of insurance claims (panels e, f), disaster aid (panels g, h), and property buyouts (panels i, j) with total damage costs in the lower right of each panel (CPI-adjusted). Clustering is conducted using ST-DBSCAN with a three degree latitude/longitude space threshold, a time threshold of five days, and number threshold of seven datapoints. Return periods reflect MSWEP daily maximum preceding precipitation within 30 days of the reported loss. Sensitivity analysis for clustering is provided in the SI, Section 7.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/342949607e3587be6aa50bfc.png"},{"id":79242824,"identity":"f81ab256-0191-4b40-8d6e-88303fdfa782","added_by":"auto","created_at":"2025-03-26 06:22:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1643592,"visible":true,"origin":"","legend":"\u003cp\u003eReturn Period Sensitivity to Spatiotemporal Clustering\u003c/p\u003e\n\u003cp\u003eSpatiotemporally clustered financial losses across insurance claims (panels a, b, c, d), as well as disaster aid and buyouts (panels b and d). Clusters in panels a) and c) vary in size on the basis of the median cluster size (15) and 90th percentile cluster size (170). Clusters are extracted using ST-DBSCAN with threshold parameters a space threshold of three degrees latitude/longitude, time threshold of five days, and a minimum cluster size of seven as optimized under validation and sensitivity, as detailed in Section 8 of the SI. Return periods are in years and reflect the maximum daily precipitation event within 30 days preceding declaration for each given loss modeled by county-fitted stationary GEV distribution using MSWEP daily precipitation data. For each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/acf6f6236796c182be24e995.png"},{"id":79242823,"identity":"df60921a-383e-483a-92bd-ed37db8bebbd","added_by":"auto","created_at":"2025-03-26 06:22:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":970808,"visible":true,"origin":"","legend":"\u003cp\u003eNonstationarity Sensitivity in Intensity Distributions\u003c/p\u003e\n\u003cp\u003eRatios of 100-year precipitation events under county-level heterogeneous nonstationary GEV distributions for the maximum daily precipitation event, and Sen’s Slope values for statistically significant trends in daily maximum precipitation return periods of claims using a Mann Kendall trend analysis by county (p \u0026lt; 0.05) for both MSWEP and ERA5 datasets for a sample size of over 1.67 million claim records. A full table of Mann Kendall results by county with degrees of freedom, test statistics, effect sizes and p-values are provided in the supplementary data. 100-year ratios are compared between years 1979 and 2020 (the temporal record of MSWEP for standard comparison between datasets). See SI, Section 8 for extended figures with three-day precipitation events.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/4102ea61b998e80816c594cc.png"},{"id":79243575,"identity":"971dbd05-57f5-4f74-a675-e27ace440d56","added_by":"auto","created_at":"2025-03-26 06:30:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":113721,"visible":true,"origin":"","legend":"\u003cp\u003eSummary of Costs and Return Periods by Financing Type\u003c/p\u003e\n\u003cp\u003eSummary of return periods, aggregate national annualized costs, and costs per household across various forms of federal financial flood support: insurance claims, disaster aid, and property buyouts. Panel a) shows the distribution of the return period in years of the maximum daily precipitation event within 30 days preceding declaration for each given loss modeled by county-fitted stationary GEV distribution using MSWEP daily precipitation data. Costs in panels b) and c) are inflation-adjusted using annual CPI-U values from the US Bureau of Labor Statistics based upon the reported event date. All records are obtained from OpenFEMA. For each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/28b82c7845c4f5d96f6de47e.png"},{"id":91757020,"identity":"8acfe3d6-47f4-47e6-abd8-06c61dfc6d29","added_by":"auto","created_at":"2025-09-20 07:06:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4797339,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/b8c1346a-8139-4137-85a3-f1da6509107c.pdf"},{"id":79242845,"identity":"f148fc1e-72bb-4ab5-b39f-8c8a66f07190","added_by":"auto","created_at":"2025-03-26 06:22:37","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":28256753,"visible":true,"origin":"","legend":"","description":"","filename":"SINayakGentineLallNW2.13.25.docx","url":"https://assets-eu.researchsquare.com/files/rs-6025742/v1/5af9fb09ddb8902b4ea5af3d.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Financial losses associated with US floods occur with surprisingly frequent, low return period precipitation","fulltext":[{"header":"Main","content":"\u003cp\u003eFlooding leads to billions of dollars in financial losses annually in the United States (NCEI 2024). Between 1954 and 2014, 80% of US disaster declarations were related to flood or storm events (Lindsay and McCarthy 2015). The number of disaster declarations in the US has been increasing over time (C. Kousky and Shabman \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Financial losses due to flooding are also increasing, and this trend is projected to continue in the near term (Hallegatte et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Paprotny et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Systemic intervention in the form of preventative infrastructure, policy reform, and/or relocation is necessary to reverse this trend.\u003c/p\u003e \u003cp\u003eExtreme precipitation is a primary driver of flood damage (Davenport, Burke, and Diffenbaugh \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Under global climate change, precipitation extremes are expected to be exacerbated as a warmer atmosphere is expected to hold more water (Tabari \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Pfahl, O\u0026rsquo;Gorman, and Fischer \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Extreme precipitation is not isolated, but rather typically clustered in space and time due to large scale synoptic patterns. Spatiotemporally clustered flood risks have long been cited to pose damage concerns (Jain and Lall 2001; Carolyn Kousky and Cooke 2012; Zscheischler et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Bonnafous and Lall 2021). Under climate change, these effects are likely to worsen, with projections of more frequent long-duration and intense storms (Hirabayashi et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Panagoulia and Dimou 1997).\u003c/p\u003e \u003cp\u003eAttribution of causal mechanisms for flood losses is further complicated due to interactions between extreme events, infrastructure health and population exposure. Flood infrastructure in the US is rapidly deteriorating, aging levees and floodwalls sparking national concern (Tobin \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e1995\u003c/span\u003e). Although flood protection measures are typically designed for 100-year return periods (Bell and Tobin \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2007\u003c/span\u003e), recent analysis shows that recent U.S. dam failures due to overtopping have predominantly occurred under precipitation with much lower return periods (Hwang and Lall 2024). Recent federal legislation such as the Bipartisan Infrastructure Law (BIL) seeks to rebuild deteriorating critical infrastructure (The White House \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), but many efforts are overdue and repairs cannot keep pace with failures (Petroski \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Further, the \"levee effect\" feedback loop exacerbates risk: structural protection spurs urban development in areas nominally protected from a 100-year flood, increasing pressure to sustain and expand these protections (Tobin \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e1995\u003c/span\u003e). Recent studies document that population growth and development in flood zones are leading to increased flood risk exposure (Neumann et al. \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Tellman et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Rentschler et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eU.S. federal financial preparedness and recovery mechanisms for flooding are under stress from these natural and human-made compounded risks. Financial mechanisms intended to buffer the impact of financial flood losses at the household level include flood insurance, individual disaster aid, and grants for managed retreat \u0026ndash; the acquisition of severely damaged properties (FEMA 2024a; Carolyn Kousky \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Historically, private insurers withdrew from flood insurance markets in the US due to the concentration of risk, leading to the creation of the National Flood Insurance Program (NFIP) in 1968 (Institutes for Research \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). NFIP insurance covers over 95% of flood insurance policies in the U.S., and is available to any household (C. Kousky et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2018\u003c/span\u003e); however, the vast majority of policyholders are government-mandated due to being located within a FEMA 100-year flood zone under a federally-backed mortgage (Carolyn Kousky \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Since inception, the NFIP has struggled to recover losses: currently the program is over \u003cspan\u003e$\u003c/span\u003e20B in debt even after \u003cspan\u003e$\u003c/span\u003e16B of forgiveness was granted by Congress in 2017 (FEMA 2024c). In order to improve the financial balance of the program in addition to past debt relief, a new risk-based premium methodology (Risk Rating 2.0) was introduced in \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e for pricing premiums more accurately to reflect risk (FEMA \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSimilarly, US disaster assistance has frequently come up short of funds. The Stafford Act was passed in 1988 by Congress as a systematic effort to disburse federal natural disaster assistance (Moss, Schellhamer, and Berman \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Lindsay and McCarthy 2015; FEMA 2024a). Disaster aid in the form of Individual Assistance (IA) and Public Assistance (PA) and ex-post event buyout programs (primarily funded through Hazard Mitigation Grant Program, HMGP) require the declaration of a major disaster by the President to be issued (FEMA 2024a; Moss, Schellhamer, and Berman \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Lindsay and McCarthy 2015). However, due to an uptick in declarations in the 21st Century (NCEI 2024), spending restrictions on FEMA\u0026rsquo;s Disaster Relief Fund (DRF) have been imposed nine separate years since 2003 due to lack of available funds to meet needs assessments (FEMA 2024b).\u003c/p\u003e \u003cp\u003eAlthough recent studies discuss economic limitations of the NFIP (Michel-Kerjan \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Carolyn Kousky and Kunreuther 2014; Carolyn Kousky \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Kunreuther \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; de Ruig et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), interactions between disaster aid and flood insurance (Carolyn Kousky, Michel-Kerjan, and Raschky 2018; C. Kousky and Shabman \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), or climatological changes in extreme precipitation (Hirabayashi et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Pfahl, O\u0026rsquo;Gorman, and Fischer \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Rahmani and Harrington 2019; Visser et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), to our knowledge no recent work provides a systematic evaluation of the characteristics of precipitation events that lead to federal financial losses. In this work, we explore this aim, conducting a first assessment of precipitation return periods associated with financial losses across three major sources of federal funding. Specifically, we define federal financial losses to include 1) insurance claims, 2) disaster aid, and 3) property buyouts for severely damaged residences. To achieve our primary objective of evaluating the regional return periods of precipitation associated with damage losses nationwide, we examine several key questions:\u003c/p\u003e \u003cp\u003e1. How do evaluations of loss-inducing precipitation events vary temporally (with event duration), spatially (across counties), and by event type (severe storms, hurricanes, etc)?\u003c/p\u003e \u003cp\u003e2. How does spatiotemporal clustering impact the evaluation of loss-inducing precipitation?\u003c/p\u003e \u003cp\u003e3. What do nonstationary trends and changes over time in precipitation indicate for future flood-related losses?\u003c/p\u003e \u003cp\u003eOur evaluation covers the reported history of flood-related, household-level national financial losses and their associated precipitation events through 2020. We conduct a thorough evaluation of millions of federal financial support records across insurance, disaster aid, and property buyouts dating back to 1978[1]\u003ca class=\"FNLink\" href=\"#Fn1\" id=\"#FNLinkFn1\"\u003e\u003c/a\u003e. Using traditional statistical methods of flood risk evaluation paired with unsupervised machine learning, we provide a robust evaluation of loss attribution, and evaluate sensitivities to varying precipitation datasets, clustering parameterizations, and methods of nonstationary analysis.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eWe find that flood-related financial losses are predominantly associated with low return period precipitation events with sub-decadal recurrence, indicating that the vast majority of financial flood losses in the United States are associated with low-intensity, high-frequency events relative to regional climatology. We define a loss event based upon household-level federal financial support mechanisms for ex-post flood recovery, which include 1) an insurance claim, 2) an individual disaster aid payment, and/or 3) a property buyout. A return period is defined as the expected time between occurrences of an event at or above a given magnitude. Precipitation event return periods are calculated as the inverse of the event probability in any given year. A summary of our major findings includes:\u003c/p\u003e\n\u003cp\u003e1. Return periods for precipitation associated with insurance claims average under 5 years, while for household disaster aid and property buyouts return periods average under 20 years.\u003c/p\u003e\n\u003cp\u003e2. Return periods vary spatially, by event type, and event duration. We find significantly lower return periods associated with losses in rural communities than urban communities, and surprisingly low return periods in areas with high average precipitation and coastal regions. Loss-attributed precipitation tends to exhibit higher return periods when examining short duration (daily to three-day) events. Return periods tend to be higher for dam/levee break, tropical cyclone, and flood disaster declarations than for coastal and severe storm declarations.\u003c/p\u003e\n\u003cp\u003e3. Over 90% of losses are associated with spatiotemporal clusters. While the maximum return periods of events within clusters are on average decadal to multi-decadal, the vast majority of loss clusters are attributed to events with return periods averaging under 20 years, i.e., the space-time organization that is associated with high loss clusters is often more frequent than the extremes locally embedded in the cluster.\u003c/p\u003e\n\u003cp\u003e4. Nonstationary analysis predominantly indicates significant increasing trends in return periods over time for losses, potentially indicating that the intensity of loss-inducing events is increasing, possibly due to a changing climate and human exposure factors.\u003c/p\u003e\n\u003cp\u003eIn our assessment, we evaluate multiple datasets to address potential data biases (Kang et al. 2024; NCAR 2025), examining Multi-Source Weighted-Ensemble Precipitation (MSWEP) daily gridded reanalysis data, European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis Version 5 (ERA5) daily data and Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly data (see SI sections 5 and 6). Unless otherwise specified, we present results for the MSWEP maximum daily precipitation preceding a loss event within 30 days. We filter out insurance claims below $1,000 and any losses associated with zero monthly precipitation (assumed likely erroneous due to human activity, snowmelt, or other anomalies). We provide a full sensitivity across different return period metrics and datasets in the SI Sections 5–7.\u003c/p\u003e\n\u003cdiv id=\"Sec3\"\u003e\n \u003ch2\u003e2.1) Variation by Time, Space, and Event Type\u003c/h2\u003e\n \u003cp\u003eIn Fig. 1, we present the associated precipitation return periods for event durations of maximum daily, three-day, weekly, and monthly sums preceding the loss event. Event return periods are computed using the Generalized Extreme Value (GEV) distribution with the annual maxima of the precipitation for the corresponding duration for the county that experienced the loss.\u003c/p\u003e\n \u003cp\u003eLow, frequent return periods under 20 years are noted across all financial loss types and event durations. Higher return periods for shorter durations (daily and three-day) for insurance claims and disaster aid are expected[2]\u003ca href=\"#Fn2\" id=\"#FNLinkFn2\"\u003e\u003c/a\u003e. Similar return periods across weekly, three-day, and daily maximum events for property buyouts could indicate that heightened levels of extreme precipitation persistence are associated with severely damaged properties for buyout. ERA5 estimates for return periods are largely consistently lower than those produced by MSWEP (see SI, Section 4). Sensitivity to return periods using monthly PRISM data is illustrated in SI, Section 6.\u003c/p\u003e\n \u003cp\u003eFor each boxplot, median values, and interquartile ranges are displayed with outliers displayed as points at 1.5 times the interquartile range.\u003c/p\u003e\n \u003cp\u003eWe also examine spatial variations in return periods across counties (Fig. 2). Similar to previous studies (Yin et al. 2018), we find that in aggregate, rural and suburban counties tend to exhibit losses linked to insurance claims (panel a), disaster aid (panel b), and property buyouts (panel c) at lower return periods than urban counties (p \u0026lt; 0.001). Similar patterns of heightened return periods in southern regions frequently prone to tropical cyclones (such as lower Texas, Louisiana, Florida, and North Carolina) emerge across all three loss datasets. In snowmelt-driven regions such as North Dakota, Maine, Montana, and upstate New York, low return periods are largely anticipated, as any snowmelt flooding events should exhibit a lower correlation between floods and rainfall. However, across most counties, even coastal and inland regions that are rainfall-dominated, return period averages are predominantly under 20 years. The prevalence of low-return-period precipitation linked to financial support across the U.S. may be due to: 1) a large number of insured risk-prone properties encountering repeated losses, 2) deterioration/failure of flood control infrastructure under nonstationary climate conditions, and/or 3) wind damage being mistakenly filed under flood damage claims. Analysis using ERA5 data corroborates the short return periods across financial losses (see SI, Section 4).\u003c/p\u003e\n \u003cp\u003eNext, we evaluate precipitation return period across event types by Presidential declaration. Figure 3 displays the spatial distribution of most common types of flood disaster declarations (panel a), the count of county-level declarations of major disasters (panel b), and the distribution of average precipitation return period by disaster type (panel c) across the contiguous United States (CONUS). Categories of flood-related major disaster declarations include hurricane, tropical storm, flood, severe storm, coastal storm and dam/levee break (full descriptions of flood-related disaster declarations are found in the SI, Section 2).[3]\u003ca href=\"#Fn3\" id=\"#FNLinkFn3\"\u003e\u003c/a\u003e Spatial distributions of declarations reveal the dominant hydroclimatic forcings by region across CONUS with tropical storms dominating the South through most of Texas and the East Coast; severe storms dominating the Midwest, Pacific Northwest, and Northeastern U.S.; and other floods dominating most other snow-prone and temperate counties (Fig. 3, panel a). As seen earlier, the median values for the distributions of the maximum precipitation return period by county associated with each disaster declaration are estimated to be lower than 20 years across all major categories with the exception of dam/levee breaks (panel c), and the 100-year level is identified as an outlier. Return periods for severe and coastal storms are shown to be much lower on average than tropical cyclones, floods, and dam/levee breaks.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\"\u003e\n \u003ch2\u003e2.2) Spatiotemporal Clustering\u003c/h2\u003e\n \u003cp\u003eUsing unsupervised learning, we also extract spatiotemporal clusters of flood-related losses and analyze their corresponding precipitation event return periods. Clustering losses allows for the evaluations of the movement of a storm and the spatiotemporal cascading effect of multiple related events (Fig. 4). As exhibited by clustering of disaster declarations and insurance claims for Hurricane Katrina (Fig. 4 panels a, c, e, g, i) and Hurricane Ike (Fig. 4 panels b, d, f, h, j), clustered losses result from a series of storms connected by large scale synoptic patterns. Spatiotemporal clustering allows us to group these loss events (panels a, b), and examine their return periods through space and time, visualizing the event trajectory (panels c, d). We can then examine the associated total financial losses across insurance claims (panels e, f), disaster aid (panels g, h), and property buyouts (panels i, j) in relation to the spatiotemporally clustered events.\u003c/p\u003e\n \u003cp\u003eSpatiotemporal clustering of loss events reveals that over 90% of losses belong to space-time clusters. On average, regional return periods across clusters look similar (Fig. 5, panels c and d) to those in aggregate (Fig. 1). However, when we examine clustered event maximums, these events exhibit a much larger range in return periods, reaching up to 54.9 years on average across clusters for disaster aid (Fig. 5, panel b). This signifies that within a given cluster, there are typically “hotspots” of extreme antecedent precipitation, although the majority of clustered damages are induced under less extreme regional conditions. The higher return period for the maximum in a cluster in comparison to most individual loss instances is notable, since the trigger for a given declaration for the event is most likely its hotspot, but a number of proximal locations will experience impact. In aggregate, mean clustering results affirm the predominance of low-return period precipitation associated with losses. Even when considering disaster aid (which shows the greatest clustered maximum return periods), the first quantile for clustered maxima return periods falls just above a 10-year mark. This indicates that 25% of spatiotemporally clustered declarations exhibit a maximum regional precipitation recurrence period of a decade or less. When we examine the averages across these same clusters, we see the majority of payouts occurring within a given storm or cluster at a much lower regional return level, all under 15 years.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\"\u003e\n \u003ch2\u003e2.3) Nonstationary Trends\u003c/h2\u003e\n \u003cp\u003eLastly, we examine sensitivity to monotonic trends in return periods over time (Fig. 6). To do so, we evaluate 1) the ratio of expected 100-year events by county under a nonstationary GEV (panels a and b), and 2) county-level monotonic trends in stationary return periods using the Mann-Kendall test and Sen’s Slope to evaluate significant trends in insurance claims with both MSWEP and ERA5 precipitation datasets (panels c and d). Nonstationary GEV results show discrepancies in trends between datasets (panels a and b), which is largely reflective of the difficulties of multi-decadal trend analysis across precipitation data with only a handful of decades on record. While ERA5 data implies a dominant decreasing trend in return periods across the United States, MSWEP implies dominant increases. Mann-Kendall tests indicate a dominance of significant (p \u0026lt; 0.05) increasing trends in return periods associated with claims across both datasets. Significantly increasing return periods could indicate increasing intensity of claim-triggering events. Mann Kendall tests for cluster return periods also indicate significant increasing return periods (p \u0026lt; 0.001). Our evaluation of regional trends reveals the importance of considering climatological shifts in our understanding of return periods over time.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec6\"\u003e\n \u003ch2\u003e2.4) Summary and Aggregate Costs\u003c/h2\u003e\n \u003cp\u003eIn summary, examination of precipitation data at the time of insurance claims, disaster aid, and property buyouts suggests a high degree of losses result from relatively low-intensity, high-frequency precipitation across all three federal financial loss datasets. In aggregate, we find the median county-level return period of daily maximum precipitation linked to claims to be 3.9 years, disaster aid to be 13.2 years, and buyouts to be 8.1 years under traditional return period evaluation using the unclustered loss datasets and stationary GEV (Fig. 7, panel a). Average precipitation return periods for disaster aid and property buyouts by county tend to be higher than those associated with insurance claims (Fig. 7, panels b and c). This is expected since disaster aid and property buyouts require the issuance of a Presidential major disaster. In all cases, the 100-year return period that is nominally the criteria for mandated insurance, is associated with very few of the claims (scored as an outlier in the box plot, i.e. more than 1.5 standard deviations beyond the 75th percentile, which would correspond to less than 1% of the claims if they were normally distributed). Analysis using ERA5 data corroborates short return periods across financial metrics (see SI, Section 4). Figure 7 also depicts the scale at which federal expenditures occur across varying forms of financial support. Expenditures by program type are displayed at a national aggregate in panel b and a household level in panel c. Since insurance claims require policyholder premiums to offset claims, we show the historic annualized sum of premiums and 2024 annual sum of risk-based premiums (under FEMA Risk Rating 2.0) with black horizontal bars in panel b. Results reveal that disaster aid (only considering individual assistance, not public assistance) is the largest financial loss in aggregate. Household-level aid grants are small in comparison to insurance claim dollars and property buyout costs (panels b and c).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur comprehensive evaluation of the history of federal financial flood losses and the return periods of associated precipitation events reveals a dominance of frequent, low-intensity events linked to damages. Recent investigation in the Northeastern United States found that the majority of damaging floods were unassociated with extreme precipitation, corroborating a trend we find nationally (Teale and Winter 2024). Across modes of federal assistance in insurance, disaster aid, and property buyouts, median associated precipitation events are shown to be dominated by return periods lower than 20 years. With a history of flood preparedness measures focusing around the 100-year flood event in the United States and beyond, our analysis calls into question mitigation and preparedness design strategies for flood management that may be relying too heavily on outdated metrics that do not reflect the realities of repeated losses occurring nationally at lower severities.\u003c/p\u003e \u003cp\u003eWe evaluated spatially heterogeneous precipitation return periods, nonstationary distributions over returns, spatiotemporally clustered extremes, and a variety of event durations to capture temporally cascading precipitation. There are many ways to calculate spatially-heterogeneous return periods for precipitation events, and we present a county-level approach here. We highlight that this method will not fully capture stream network flooding effects, snowmelt, tidal flooding, storm surge, or sub-daily intense rainfall, yet the prevalence of linkages between damages and short return period precipitation events is concerning. Our method is spatially limited to county-level analyses, which are non-homogenous in size, presenting limitations for larger counties. Higher resolution financial loss data could allow for higher precision in regionally-specific return periods for each event. However, our work in spatiotemporal clustering alludes to the need for more robust measures of disaster-inducing storms that better capture the multidimensional distributions of clustered regional effects.\u003c/p\u003e \u003cp\u003eWe find that damage is frequently induced in spatiotemporal clusters, which have varying regional characteristics. To enhance preparedness, engineers must better quantify the compound return periods of these damage-driving events, and shift to an event-driven preparedness strategy that considers clustered hydroclimatic dynamics rather than the traditional 100-year return. A new approach to spatiotemporal flood recurrence analysis and prediction could inform flood risk at a portfolio level, noting the multiple geographies that may have near simultaneous exposure. Linking such exposure directly to the types of climate events responsible for clustered or non-clustered outcomes would permit improved projections of future flood risks, as well as of the associated financial implications for asset holders and the government programs charged with flood protection and response.\u003c/p\u003e \u003cp\u003eOur work suggests that current structural approaches to flood risk management need to be critically examined. The NFIP and DRF are sources of billions of dollars of national financial debt, consistently overdrawing public funds. Methods of intervention such as risk-based pricing of NFIP premiums have been suggested and recently implemented (de Ruig et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; FEMA \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, our findings showing the frequency of these losses begs the question of the effectiveness of pricing intervention if mass losses are occurring repeatedly. Additionally, evaluating the extent to which our current systems of flood preparedness are effectively reducing realized risks will be crucial to understanding how we can more proactively use federal resources to protect against anticipated hazards. Further, work showing low-return period failures from dam breaks (Hwang \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) begs the question if traditional metrics for infrastructural design (such as the univariate 100-year flood) are effective enough to prepare for spatiotemporally clustered hydroclimatic pressures that are likely to be exacerbated under global climate change (Hirabayashi et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Bonnafous and Lall 2021).\u003c/p\u003e \u003cp\u003eSolving future flood risk management problems will require interdisciplinary approaches to systems engineering that consider variability in both our stochastic natural systems and human behavioral intervention points. In lieu of considering 100-year events in isolation, a dynamic approach for planning that considers sequences of more frequent extremes and quantifying characteristics of damage-inducing storms is necessary to holistically capture associated losses that are occurring at lower precipitation severities. Recent studies have considered methods of computing high dimensional return periods of compound events (Del, Urrea, and V. 2024) and modern stochastic simulators are attempting to capture varied cascading dynamics in flood frequency, intensity, and duration to propel more creative adaptive planning solutions (Nayak, Gentine, and Lall \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Future work should consider more robust assessment of vulnerabilities within the financial system to identify strategic points for future adaptive intervention. In order to enable proactive adaptive practices, planning must consider both the endogenous, institutional failure points within the national financial systems of natural disaster insurance and aid in relation to the exogenous nonstationary extreme hydroclimatic forcings.\u003c/p\u003e"},{"header":"Data and Methods","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e4.1) Precipitation Data\u003c/h2\u003e \u003cp\u003eTo assess the precipitation associated with flood losses, we assess the record of historical county flood loss records in relation to gridded precipitation history and reported dates of losses. In order to validate our findings, we use two daily gridded precipitation reanalysis datasets and one monthly gridded precipitation dataset: Multi-Source Weighted-Ensemble Precipitation (MSWEP) daily precipitation at 0.1 degree resolution from 1979 to 2020, ECMWF Reanalysis 5th Generation (ERA5) single-levels daily precipitation at 0.25 degree resolution from 1940 to \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e, and Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly precipitation at 4 km resolution from 1923 to \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e. MSWEP uses a weighted-ensemble method for daily precipitation estimates that sources gridded observations from gauge, satellite, radar, and bias-corrected reanalysis data providing the finest spatial resolution of the daily datasets but the shortest time horizon. ERA5 is developed using the European Centre for Medium-Range Weather Forecasts (ECMWF) model paired with data assimilation techniques, sourcing from gauge, satellite, lidar, and radar providing the longest temporal record for daily observations at a daily resolution. PRISM is developed through Oregon State University\u0026rsquo;s PRISM Climate Group and uses a combination of gauge data, topographic data, geographic features, and climatological normals to develop the longest history of monthly precipitation datasets for the continental US dating back to 1923. Gridded precipitation data was obtained directly from \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eGloH2O\u003c/span\u003e for MSWEP daily data, \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eERA5 single levels since 1940\u003c/span\u003e daily at 0.25 degrees were provided by the Copernicus Data Store and implemented by ECMWF, and PRISM monthly 4km precipitation was sourced from the \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ePRISM Climate Group\u003c/span\u003e. We provide a table of all datasets used, their period of record, and their sourcing location in the SI, Section \u003cspan refid=\"Sec1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e4.2) Flood Loss Data\u003c/h2\u003e \u003cp\u003eAll FEMA flood loss datasets are open access through \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eOpenFEMA\u003c/span\u003e. For flood insurance claims, we analyzed over 2,000,000 claim records and over 80,000,000 policy records from the NFIP from its genesis in 1978 to May 2024 including records of current Risk Rating 2.0 policies (FEMA \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For property buyouts, we analyzed over 68,000 records of hazard mitigation assistance since its inception in 1979 to examine flood-related property buyouts across FEMA\u0026rsquo;s HMGP. For disaster aid, we examined over 250,000 county-level records of all Individual Assistance grants on file for both property owners and renters across the entire history of presidential disaster declarations, amounting to millions of federal aid household grants. Urban-rural classification information was obtained using the \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eNational Center for Health Statistics\u0026rsquo; scheme\u003c/span\u003e provided by the CDC (See SI, Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e for more details of classifications).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e4.3) Precipitation Event Extraction and Loss Attribution\u003c/h2\u003e \u003cp\u003eIn order to attribute precipitation events to losses we evaluate a number of potential event characteristics. The date of the loss event is reported on NFIP claims (distinct from the date of claim filing), but NFIP claims can be filed up to 60 days after the incident which leaves room for human error in reporting. Thus, due largely to reporting error (for instance claims attributed to Hurricane Ian reported a date of loss up to 12 days after the latest incidence of precipitation), we examine the maximum continuous sums of 1, 3, 7, and 30 days within the preceding month of the reported date of loss for both MSWEP and ERA5 daily gridded datasets. We also tested lags of 5 days and 14 days within preceding windows of one week, two weeks, and one month, but found distributions to be largely encompassed by the daily, three-day, weekly and monthly values within a 30 day window of the reported event date. Similarly, disaster declarations list a begin date, but this can be up to a few days before the maximum extent of damage, which varies by event. Thus, we take a similar approach for disaster aid and relocation event attributions: extracting the maximum continuous sums of 1, 3, 7, and 30 days within one month of the reported date of loss for both MSWEP and ERA5 daily gridded datasets. In subsequent analysis, we filter all losses that are attributed with zero monthly precipitation (likely due to errors in date reporting, human activity, snowmelt-induced flooding, or other anomalies). We also compare this with the associated monthly values of precipitation in relation to claims using gridded PRISM reanalysis data in the SI, Section 6.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e4.4) Calculation of Stationary and Nonstationary Return Periods\u003c/h2\u003e \u003cp\u003ePrecipitation return periods of loss events are expected to vary spatially and temporally. In order to best characterize the local distribution of return periods for precipitation for county-level financial loss data, we extract a time series of weighted average precipitation values by county, considering the intersection of all gridded data pixels whose centroid is within county boundaries. If a given county is too small to intersect fully with any given pixel\u0026rsquo;s centroid, we source data from the pixel with the minimum Euclidean distance between the county\u0026rsquo;s centroid and surrounding pixels. We use this method for all three precipitation datasets, and fit separate generalized extreme value (GEV) probability distributions to each county-level annual maxima timeseries for the annual block maxima for each event duration.\u003c/p\u003e \u003cp\u003eWe use the GEV distribution to model return periods over the county annual maxima time series due to its widespread application in hydrometeorological settings and common applicability for return period calculation (Coles \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). Under stationary conditions, for a given county \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:c\\)\u003c/span\u003e\u003c/span\u003e, we can consider probability distribution \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{P}_{c}\\)\u003c/span\u003e\u003c/span\u003e as:\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:{P}_{c}\\:\\sim\\:GEV({\\mu\\:}_{{P}_{c}},\\:{\\sigma\\:}_{{P}_{c}},\\:{\\gamma\\:}_{{P}_{c}})$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eIn which \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\mu\\:}_{{P}_{c}},\\:{\\sigma\\:}_{{P}_{c}},\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\gamma\\:}_{{P}_{c}}\\)\u003c/span\u003e\u003c/span\u003edenote the location, scale and shape parameters. The cumulative distribution function for the stationary GEV is given by:\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{F}_{{P}_{c}}({P}_{c}\\)\u003c/span\u003e \u003c/span\u003e) = \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{e}^{-{[1+{\\gamma\\:}_{{P}_{c}}(\\frac{{P}_{c}-{\\mu\\:}_{{P}_{c}}}{{\\sigma\\:}_{{P}_{c}}}\\left)\\right]}^{-1/{\\gamma\\:}_{{P}_{c}}}}\\)\u003c/span\u003e\u003c/span\u003e \u003cem\u003e(2)\u003c/em\u003e\u003c/p\u003e \u003cp\u003eWe also fit the nonstationary GEV to each county \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:c\\)\u003c/span\u003e\u003c/span\u003e as seen in (Katz \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2013\u003c/span\u003e) to consider linear trends in the return periods in which the location and scale parameters of the GEV become time varying by year \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:t\\)\u003c/span\u003e\u003c/span\u003e:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{P}_{c,t}\\:\\sim\\:GEV({\\mu\\:}_{{P}_{c,t}},\\:{\\sigma\\:}_{{P}_{c,t}},\\:{\\gamma\\:}_{{P}_{c,t}})$$\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{\\mu\\:}_{{P}_{c,t}}=\\:{\\alpha\\:}_{1c}+{\\beta\\:}_{1c}t$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{\\sigma\\:}_{{P}_{c,t}}=\\:{\\alpha\\:}_{2c}+{\\beta\\:}_{2c}t$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWe test three configurations of the nonstationary GEV for each county \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:c\\)\u003c/span\u003e\u003c/span\u003e: 1) time-varying scale and time-invariant location, 2) time-invariant scale and time-varying location, and 3) both time-varying scale and time-varying location. We choose the distribution fit that best minimizes the Bayesian Information Criterion (BIC). All nonstationary distributions are fit using the extRemes package in R. Stationary distributions are fit using the SciPy package in Python.\u003c/p\u003e \u003cp\u003eFor each flood loss occurrence, the associated precipitation return period for a given GEV configuration at time \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:t\\)\u003c/span\u003e\u003c/span\u003e is found given:\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:{T}_{{P}_{c,t}}\\:=\\:\\frac{1}{1-{F}_{{P}_{c,t}}\\left({P}_{c,t}\\right|{\\mu\\:}_{{P}_{c,t}},\\:{\\sigma\\:}_{{P}_{c,t}},\\:{\\gamma\\:}_{{P}_{c,t}})}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eLastly, for further statistical evaluations of significance, to test for significance across varying groups of comparisons, we performed Welch\u0026rsquo;s t-test for unequal variances, and three-way ANOVA for multi-group comparisons using the Python SciPy package. To evaluate monotonic trends in precipitation return periods for insurance claims, we use the Mann-Kendall test at a significance level of 0.05 as found in (Mann \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e1945\u003c/span\u003e), and extract the Sen\u0026rsquo;s Slope values to assess trends as detailed in (Sen \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e1968\u003c/span\u003e) using the Python package pymannkendall.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.5) Spatiotemporal Clustering\u003c/h2\u003e \u003cp\u003eA longstanding method of clustering is DBSCAN (Ester et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e1996\u003c/span\u003e) and its spatiotemporal extension ST-DBSCAN (Birant and Kut 2007). The premise of the original algorithm involves setting a distance threshold \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ϵ\\)\u003c/span\u003e\u003c/span\u003e and a minimum number of points \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:MinPts\\)\u003c/span\u003e\u003c/span\u003e that must be within the Euclidean distance of a \u0026ldquo;core point\u0026rdquo;. The algorithm functions by starting with a random point \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e, iterating through its nearest neighbors, and clustering points that fall within the specified Euclidean distance threshold \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ϵ\\)\u003c/span\u003e\u003c/span\u003e. It then continues to each clustered neighbor \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e and clusters the points within \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ϵ\\)\u003c/span\u003e\u003c/span\u003e from \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:n\\)\u003c/span\u003e\u003c/span\u003e as long as the total is at least \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:MinPts\\)\u003c/span\u003e\u003c/span\u003e. If a point continues the iteration (by clustering with at least \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:MinPts\\)\u003c/span\u003e\u003c/span\u003e within \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ϵ\\)\u003c/span\u003e\u003c/span\u003e) it is a \u0026ldquo;core point\u0026rdquo;; if a point is clustered but does not continue the iteration (by clustering with less than \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:MinPts\\)\u003c/span\u003e\u003c/span\u003e within \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:ϵ\\)\u003c/span\u003e\u003c/span\u003e) it is a \u0026ldquo;border point\u0026rdquo;; if a point does not fit to any cluster, it is labeled as \u0026ldquo;noise\u0026rdquo; and remains unclustered. ST-DBSCAN extends this approach to include both \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ϵ}_{space}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ϵ}_{time}\\)\u003c/span\u003e\u003c/span\u003e, defining a \u0026ldquo;core point\u0026rdquo; such that it falls within both the time and space thresholds \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ϵ}_{space}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ϵ}_{time}\\)\u003c/span\u003e\u003c/span\u003e set by the user.\u003c/p\u003e \u003cp\u003eA major limitation in the use of DBSCAN and ST-DBSCAN is the ability to validate the choice of one set of threshold parameters against another. While other traditional clustering techniques, such as k-means, optimize clusters based upon a desired number of clusters, DBSCAN\u0026rsquo;s set of thresholding parameters can make its results sensitive, and the technique has met criticism due to its difficulty to parameterize (Schubert et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Recent scholars have underscored its practical application for appropriate use cases (Gan and Tao \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). One of these in recent years has emerged in the case of spatiotemporal clustering for extreme storm events, such as with lightning clusters (Augenstein, Mohr, and Kunz \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Shi et al. \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In our case, we similarly need to cluster extreme disaster-inducing events that have clear spatial and temporal clustering tendencies, but the number of such occurrences is unknown.\u003c/p\u003e \u003cp\u003eWith our data, there is a specific advantage for validation of the parameterization of the spatiotemporal clustering. For our disaster aid and property buyouts datasets, all presidential disaster declarations have an associated number for the same event and must specify the counties eligible for aid and their declaration date. Thus, there are natural \u0026ldquo;clusters\u0026rdquo; of aid and buyouts that spur at a county level by a given disaster declaration number. However, sometimes multiple disaster numbers are declared within a short period of time that are inherently related (eg. Hurricane Ike becoming a Tropical Storm Ike then a series of severe storms). Thus, we wanted to cluster our aid data in a way that did not split disaster declaration numbers, but rather encompassed them as a potential subset to the clusters extracted. More specifically, we aimed to choose the minimum space, time, and number threshold parameters that also minimized the number of presidential disaster declaration numbers split between multiple clusters. This gave us a strong validation metric for our clustering parameterization. As the claims dataset was not clustered in declarations, we first performed traditional DBSCAN with a temporal threshold by county, then spatiotemporally clustered each temporal cluster using ST-DBSCAN. We then performed a thorough sensitivity analysis on the characteristics of the extracted cluster duration, latitude and longitude span, and percent of data clustered across a range of thresholds for all three datasets (see SI Section 8).\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eData Availability\u003c/p\u003e\n\u003cp\u003eAll raw data that form the basis of this article are freely available. Financial loss records for insurance claims, disaster aid, and property buyouts can be obtained through OpenFEMA. ERA5 single-levels precipitation data can be obtained through the Copernicus Data Store, and MSWEP precipitation sourced from GloH2O. PRISM data can be accessed through Oregon State\u0026rsquo;s PRISM Group website. Records of urban-rural classification are obtained directly from the CDC\u0026rsquo;s National Center for Health Statistics. Processed datasets for financial losses and associated return periods can be accessed directly from this project\u0026rsquo;s GitHub. \u003c/p\u003e\n\u003cp\u003eCode Availability\u003c/p\u003e\n\u003cp\u003eAll associated project code in Python and RStudio for data preprocessing, analysis, and evaluation can be accessed through this project\u0026rsquo;s public GitHub repository.\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eFinancial support for this research was provided by the National Science Foundation Graduate Research Fellowship Program, and the Columbia Presidential Distinguished Fellowship from the Fu Foundation School of Engineering and Applied Sciences. Cloud computing resources were provided by the National Science Foundation\u0026rsquo;s Science and Technology Center for Learning the Earth with Artificial Intelligence and Physics (LEAP) at Columbia University (grant number 2019625).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAugenstein, M., S. Mohr, and M. Kunz. 2024. \u0026ldquo;Influence of the North Atlantic Oscillation on Annual Spatiotemporal Lightning Clusters in Western and Central Europe.\u0026rdquo; \u003cem\u003eEGUsphere\u003c/em\u003e 2024:1\u0026ndash;30.\u003c/li\u003e\n\u003cli\u003eBell, H., and G. Tobin. 2007. \u0026ldquo;Efficient and Effective? The 100-Year Flood in the Communication and Perception of Flood Risk.\u0026rdquo; \u003cem\u003eEnvironmental Hazards\u003c/em\u003e 7 (4): 302\u0026ndash;11.\u003c/li\u003e\n\u003cli\u003eBirant, Derya, and Alp Kut. 2007. \u0026ldquo;ST-DBSCAN: An Algorithm for Clustering Spatial\u0026ndash;temporal Data.\u0026rdquo; \u003cem\u003eData \u0026amp; Knowledge Engineering\u003c/em\u003e 60 (1): 208\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eBonnafous, Luc, and Upmanu Lall. 2021. \u0026ldquo;Space-Time Clustering of Climate Extremes Amplify Global Climate Impacts, Leading to Fat-Tailed Risk.\u0026rdquo; \u003cem\u003eNatural Hazards and Earth System Sciences\u003c/em\u003e 21 (8): 2277\u0026ndash;84.\u003c/li\u003e\n\u003cli\u003eColes, Stuart. 2001. \u0026ldquo;Classical Extreme Value Theory and Models.\u0026rdquo; In \u003cem\u003eAn Introduction to Statistical Modeling of Extreme Values\u003c/em\u003e, 45\u0026ndash;73. Springer Series in Statistics. London: Springer London.\u003c/li\u003e\n\u003cli\u003eCongressional Research Service. 2023. \u0026ldquo;FEMA Pre-Disaster Mitigation: The Building Resilient Infrastructure and Communities (BRIC) Program.\u0026rdquo; https://crsreports.congress.gov/product/pdf/IN/IN11515.\u003c/li\u003e\n\u003cli\u003eDavenport, Frances V., Marshall Burke, and Noah S. Diffenbaugh. 2021. \u0026ldquo;Contribution of Historical Precipitation Change to US Flood Damages.\u0026rdquo; \u003cem\u003eProceedings of the National Academy of Sciences of the United States of America\u003c/em\u003e 118 (4): e2017524118.\u003c/li\u003e\n\u003cli\u003eDel, Jesus M., M\u0026eacute;ndez D. Urrea, and Gomez Rave D. V. 2024. \u0026ldquo;Return Period of High-Dimensional Compound Events. Part I: Conceptual Framework.\u0026rdquo; \u003cem\u003eHydrology and Earth System Sciences Discussions\u003c/em\u003e 2024:1\u0026ndash;27.\u003c/li\u003e\n\u003cli\u003eEster, M., H. Kriegel, J. Sander, and Xiaowei Xu. 1996. \u0026ldquo;A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise.\u0026rdquo; \u003cem\u003eKnowledge Discovery and Data Mining\u003c/em\u003e, August, 226\u0026ndash;31.\u003c/li\u003e\n\u003cli\u003eFEMA. 2021. \u0026ldquo;Flood Mitigation Assistance Community Flood Mitigation,\u0026rdquo; August. https://www.fema.gov/sites/default/files/documents/fema_fy21-fma-community-flood-mitigation_fact-sheet.pdf.\u003c/li\u003e\n\u003cli\u003e\u0026mdash;\u0026mdash;\u0026mdash;. 2023. \u0026ldquo;NFIP\u0026rsquo;s Pricing Approach.\u0026rdquo; 2023. https://www.fema.gov/flood-insurance/risk-rating.\u003c/li\u003e\n\u003cli\u003e\u0026mdash;\u0026mdash;\u0026mdash;. 2024a. \u0026ldquo;How a Disaster Gets Declared.\u0026rdquo; 2024. https://www.fema.gov/disaster/how-declared.\u003c/li\u003e\n\u003cli\u003e\u0026mdash;\u0026mdash;\u0026mdash;. 2024b. \u0026ldquo;Immediate Needs Funding Fact Sheet.\u0026rdquo; \u003cem\u003eFEMA Has Implemented Immediate Needs Funding Nine Times :\u003c/em\u003e, August. https://www.fema.gov/sites/default/files/documents/fema_ocfo-inf-fact-sheet.pdf.\u003c/li\u003e\n\u003cli\u003e\u0026mdash;\u0026mdash;\u0026mdash;. 2024c. \u0026ldquo;NFIP Debt.\u0026rdquo; 2024. https://www.fema.gov/case-study/nfip-debt.\u003c/li\u003e\n\u003cli\u003eGan, Junhao, and Yufei Tao. 2015. \u0026ldquo;DBSCAN Revisited: Mis-Claim, Un-Fixability, and Approximation.\u0026rdquo; In \u003cem\u003eProceedings of the 2015 ACM SIGMOD International Conference on Management of Data\u003c/em\u003e. New York, NY, USA: ACM. https://doi.org/10.1145/2723372.2737792.\u003c/li\u003e\n\u003cli\u003eHallegatte, Stephane, Colin Green, Robert J. Nicholls, and Jan Corfee-Morlot. 2013. \u0026ldquo;Future Flood Losses in Major Coastal Cities.\u0026rdquo; \u003cem\u003eNature Climate Change\u003c/em\u003e 3 (9): 802\u0026ndash;6.\u003c/li\u003e\n\u003cli\u003eHirabayashi, Yukiko, Roobavannan Mahendran, Sujan Koirala, Lisako Konoshima, Dai Yamazaki, Satoshi Watanabe, Hyungjun Kim, and Shinjiro Kanae. 2013. \u0026ldquo;Global Flood Risk under Climate Change.\u0026rdquo; \u003cem\u003eNature Climate Change\u003c/em\u003e 3 (9): 816\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eHwang, Jeongwoo, and Upmanu Lall. 2024. \u0026ldquo;Increasing Risk of Dam Failures in the United States due to Compound Risk of Rainfall Clusters as Climate Changes,\u0026rdquo; February. https://scholar.google.com/citations?view_op=view_citation\u0026amp;hl=en\u0026amp;citation_for_view=JA0o2TUAAAAJ:k1VWYO0b_yEC.\u003c/li\u003e\n\u003cli\u003eInstitutes for Research, American. 2005. \u0026ldquo;A Chronology of Major Events Affecting the National Flood Insurance Program.\u0026rdquo;\u003c/li\u003e\n\u003cli\u003eJain, Shaleen, and Upmanu Lall. 2001. \u0026ldquo;Floods in a Changing Climate: Does the Past Represent the Future?\u0026rdquo; \u003cem\u003eWater Resources Research\u003c/em\u003e 37 (12). https://doi.org/10.1029/2001WR000495.\u003c/li\u003e\n\u003cli\u003eKang, Xiaoqi, Jianzhi Dong, Wade T. Crow, Lingna Wei, and Huiwen Zhang. 2024. \u0026ldquo;The Conditional Bias of Extreme Precipitation in Multi‐source Merged Data Sets.\u0026rdquo; \u003cem\u003eGeophysical Research Letters\u003c/em\u003e 51 (22): e2024GL111378.\u003c/li\u003e\n\u003cli\u003eKatz, Richard W. 2013. \u0026ldquo;Statistical Methods for Nonstationary Extremes.\u0026rdquo; In \u003cem\u003eExtremes in a Changing Climate\u003c/em\u003e, 15\u0026ndash;37. Dordrecht: Springer Netherlands.\u003c/li\u003e\n\u003cli\u003eKousky, Carolyn. 2018. \u0026ldquo;Financing Flood Losses: A Discussion of the National Flood Insurance Program.\u0026rdquo; \u003cem\u003eRisk Management and Insurance Review\u003c/em\u003e 21 (1): 11\u0026ndash;32.\u003c/li\u003e\n\u003cli\u003eKousky, Carolyn, and Roger Cooke. 2012. \u0026ldquo;Explaining the Failure to Insure Catastrophic Risks.\u0026rdquo; \u003cem\u003eThe Geneva Papers on Risk and Insurance Issues and Practice\u003c/em\u003e 37 (2): 206\u0026ndash;27.\u003c/li\u003e\n\u003cli\u003eKousky, Carolyn, and Howard Kunreuther. 2014. \u0026ldquo;Addressing Affordability in the National Flood Insurance Program.\u0026rdquo; \u003cem\u003eJournal of Extreme Events\u003c/em\u003e 01 (01): 1450001.\u003c/li\u003e\n\u003cli\u003eKousky, Carolyn, Erwann O. Michel-Kerjan, and Paul A. Raschky. 2018. \u0026ldquo;Does Federal Disaster Assistance Crowd out Flood Insurance?\u0026rdquo; \u003cem\u003eJournal of Environmental Economics and Management\u003c/em\u003e 87 (January):150\u0026ndash;64.\u003c/li\u003e\n\u003cli\u003eKousky, C., H. Kunreuther, B. Lingle, and L. Shabman. 2018. \u0026ldquo;The Emerging Private Residential Flood Insurance Market in the United States.\u0026rdquo; \u003cem\u003eWharton Risk Management and Decision Processes Center\u003c/em\u003e. https://riskcenter.wharton.upenn.edu/wp-content/uploads/2018/07/The-Emerging-Flood-Insurance-Market-Report.pdf.\u003c/li\u003e\n\u003cli\u003eKousky, C., and L. Shabman. 2017. \u0026ldquo;The Realities of Federal Disaster Aid.\u0026rdquo; https://media.rff.org/archive/files/sharepoint/WorkImages/Download/RFF-IB-12-02.pdf.\u003c/li\u003e\n\u003cli\u003eKunreuther, Howard. 2021. \u0026ldquo;Improving the National Flood Insurance Program.\u0026rdquo; \u003cem\u003eBehavioural Public Policy\u003c/em\u003e 5 (3): 318\u0026ndash;32.\u003c/li\u003e\n\u003cli\u003eLindsay, Bruce R., and Francis X. McCarthy. 2015. \u0026ldquo;Stafford Act Declarations 1953-2014: Trends, Analyses, and Implications for Congress.\u0026rdquo; \u003cem\u003eCongressional Research Service\u003c/em\u003e 42702.\u003c/li\u003e\n\u003cli\u003eMann, Henry B. 1945. \u0026ldquo;Nonparametric Tests Against Trend.\u0026rdquo; \u003cem\u003eEconometrica: Journal of the Econometric Society\u003c/em\u003e 13 (3): 245.\u003c/li\u003e\n\u003cli\u003eMichel-Kerjan, Erwann O. 2010. \u0026ldquo;Catastrophe Economics: The National Flood Insurance Program.\u0026rdquo; \u003cem\u003eThe Journal of Economic Perspectives: A Journal of the American Economic Association\u003c/em\u003e 24 (4): 165\u0026ndash;86.\u003c/li\u003e\n\u003cli\u003eMoss, Mitchell, Charles Schellhamer, and David A. Berman. 2009. \u0026ldquo;The Stafford Act and Priorities for Reform.\u0026rdquo; \u003cem\u003eJournal of Homeland Security and Emergency Management\u003c/em\u003e 6 (1). https://doi.org/10.2202/1547-7355.1538.\u003c/li\u003e\n\u003cli\u003eNayak, Adam, Pierre Gentine, and Upmanu Lall. 2024. \u0026ldquo;A Nonstationary Stochastic Simulator for Clustered Regional Hydroclimatic Extremes to Characterize Compound Flood Risk.\u0026rdquo; \u003cem\u003eJournal of Hydrology: X\u003c/em\u003e 25 (100189): 100189.\u003c/li\u003e\n\u003cli\u003eNCAR. 2025. \u0026ldquo;Precipitation Data Sets: Overview \u0026amp; Comparison Table.\u0026rdquo; 2025. https://climatedataguide.ucar.edu/climate-data/precipitation-data-sets-overview-comparison-table.\u003c/li\u003e\n\u003cli\u003eNCEI. 2024. \u0026ldquo;Billion-Dollar Weather and Climate Disasters.\u0026rdquo; 2024. https://www.ncei.noaa.gov/access/billions/state-summary/US.\u003c/li\u003e\n\u003cli\u003eNeumann, Barbara, Athanasios T. Vafeidis, Juliane Zimmermann, and Robert J. Nicholls. 2015. \u0026ldquo;Future Coastal Population Growth and Exposure to Sea-Level Rise and Coastal Flooding--a Global Assessment.\u0026rdquo; \u003cem\u003ePloS One\u003c/em\u003e 10 (3): e0118571.\u003c/li\u003e\n\u003cli\u003ePanagoulia, Dionysia, and George Dimou. 1997. \u0026ldquo;Sensitivity of Flood Events to Global Climate Change.\u0026rdquo; \u003cem\u003eJournal of Hydrology\u003c/em\u003e 191 (1): 208\u0026ndash;22.\u003c/li\u003e\n\u003cli\u003ePaprotny, Dominik, Antonia Sebastian, Oswaldo Morales-N\u0026aacute;poles, and Sebastiaan N. Jonkman. 2018. \u0026ldquo;Trends in Flood Losses in Europe over the Past 150 Years.\u0026rdquo; \u003cem\u003eNature Communications\u003c/em\u003e 9 (1): 1985.\u003c/li\u003e\n\u003cli\u003ePetroski, Henry. 2016. \u003cem\u003eThe Road Taken: The History and Future of America\u0026rsquo;s Infrastructure\u003c/em\u003e. Bloomsbury Publishing.\u003c/li\u003e\n\u003cli\u003ePfahl, S., P. O\u0026rsquo;Gorman, and E. Fischer. 2017. \u0026ldquo;Understanding the Regional Pattern of Projected Future Changes in Extreme Precipitation.\u0026rdquo; \u003cem\u003eNature Climate Change\u003c/em\u003e 7 (June):423\u0026ndash;27.\u003c/li\u003e\n\u003cli\u003eRahmani, Vahid, and John Harrington Jr. 2019. \u0026ldquo;Assessment of Climate Change for Extreme Precipitation Indices: A Case Study from the Central United States.\u0026rdquo; \u003cem\u003eInternational Journal of Climatology: A Journal of the Royal Meteorological Society\u003c/em\u003e 39 (2): 1013\u0026ndash;25.\u003c/li\u003e\n\u003cli\u003eRentschler, Jun, Paolo Avner, Mattia Marconcini, Rui Su, Emanuele Strano, Michalis Vousdoukas, and St\u0026eacute;phane Hallegatte. 2023. \u0026ldquo;Global Evidence of Rapid Urban Growth in Flood Zones since 1985.\u0026rdquo; \u003cem\u003eNature\u003c/em\u003e 622 (7981): 87\u0026ndash;92.\u003c/li\u003e\n\u003cli\u003eRuig, Lars T. de, Toon Haer, Hans de Moel, Samuel D. Brody, W. J. Wouter Botzen, Jeffrey Czajkowski, and Jeroen C. J. H. Aerts. 2022. \u0026ldquo;How the USA Can Benefit from Risk-Based Premiums Combined with Flood Protection.\u0026rdquo; \u003cem\u003eNature Climate Change\u003c/em\u003e 12 (11): 995\u0026ndash;98.\u003c/li\u003e\n\u003cli\u003eSchubert, Erich, J\u0026ouml;rg Sander, Martin Ester, Hans Peter Kriegel, and Xiaowei Xu. 2017. \u0026ldquo;DBSCAN Revisited, Revisited: Why and How You Should (still) Use DBSCAN.\u0026rdquo; \u003cem\u003eACM Transactions on Database Systems\u003c/em\u003e 42 (3): 1\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eSen, Pranab Kumar. 1968. \u0026ldquo;Estimates of the Regression Coefficient Based on Kendall\u0026rsquo;s Tau.\u0026rdquo; \u003cem\u003eJournal of the American Statistical Association\u003c/em\u003e 63 (324): 1379\u0026ndash;89.\u003c/li\u003e\n\u003cli\u003eShi, M., W. Zhang, P. Fan, Q. Chen, Z. Liu, Q. Li, and X. Liu. 2022. \u0026ldquo;Modelling Deep Convective Activity Using Lightning Clusters and Machine Learning.\u0026rdquo; \u003cem\u003eInt. J. Climatol\u003c/em\u003e 42:952\u0026ndash;73.\u003c/li\u003e\n\u003cli\u003eTabari, Hossein. 2020. \u0026ldquo;Climate Change Impact on Flood and Extreme Precipitation Increases with Water Availability.\u0026rdquo; \u003cem\u003eScientific Reports\u003c/em\u003e 10 (1): 13768.\u003c/li\u003e\n\u003cli\u003eTeale, Natalie, and Jonathan M. Winter. 2024. \u0026ldquo;The Relationship between Extreme Precipitation and Damaging Floods in the Northeastern United States.\u0026rdquo; \u003cem\u003eJournal of Applied Meteorology and Climatology\u003c/em\u003e 63 (9): 1035\u0026ndash;47.\u003c/li\u003e\n\u003cli\u003eTellman, B., J. A. Sullivan, C. Kuhn, A. J. Kettner, C. S. Doyle, G. R. Brakenridge, T. A. Erickson, and D. A. Slayback. 2021. \u0026ldquo;Satellite Imaging Reveals Increased Proportion of Population Exposed to Floods.\u0026rdquo; \u003cem\u003eNature\u003c/em\u003e 596 (7870): 80\u0026ndash;86.\u003c/li\u003e\n\u003cli\u003eThe White House. 2021. \u0026ldquo;Fact Sheet: The Bipartisan Infrastructure Deal.\u0026rdquo; The White House. November 6, 2021. https://www.whitehouse.gov/briefing-room/statements-releases/2021/11/06/fact-sheet-the-bipartisan-infrastructure-deal/.\u003c/li\u003e\n\u003cli\u003eTobin, Graham A. 1995. \u0026ldquo;THE LEVEE LOVE AFFAIR: A STORMY RELATIONSHIP?\u003csup\u003e1\u003c/sup\u003e.\u0026rdquo; \u003cem\u003eJournal of the American Water Resources Association\u003c/em\u003e 31 (3): 359\u0026ndash;67.\u003c/li\u003e\n\u003cli\u003eVisser, Johan B., Conrad Wasko, Ashish Sharma, and Rory Nathan. 2023. \u0026ldquo;Changing Storm Temporal Patterns with Increasing Temperatures across Australia.\u0026rdquo; \u003cem\u003eJournal of Climate\u003c/em\u003e 36 (18): 6247\u0026ndash;59.\u003c/li\u003e\n\u003cli\u003eYin, Jiabo, Pierre Gentine, Sha Zhou, Sylvia C. Sullivan, Ren Wang, Yao Zhang, and Shenglian Guo. 2018. \u0026ldquo;Large Increase in Global Storm Runoff Extremes Driven by Climate and Anthropogenic Changes.\u0026rdquo; \u003cem\u003eNature Communications\u003c/em\u003e 9 (1): 4389.\u003c/li\u003e\n\u003cli\u003eZscheischler, Jakob, Olivia Martius, Seth Westra, Emanuele Bevacqua, Colin Raymond, Radley M. Horton, Bart van den Hurk, et al. 2020. \u0026ldquo;A Typology of Compound Weather and Climate Events.\u0026rdquo; \u003cem\u003eNature Reviews Earth \u0026amp; Environment\u003c/em\u003e 1 (7): 333\u0026ndash;47.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Footnotes","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003e We note that FEMA also provides pre-disaster mitigation grants that are not contingent upon presidential disaster declaration through two programs: the Building Resilient Infrastructure and Communities (BRIC) grants and Flood Mitigation Assistance (FMA) grants for NFIP policyholders (Congressional Research Service \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; FEMA 2021). However, since these funds are non-event contingent and can be requested at any time, not all grants have associated extreme events with a return period, so we do not analyze these records here.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Since daily duration maximum precipitation returns were most consistently the highest across all three modes of loss events, we report daily durations for the remainder of the results, but sensitivity to varying durations are presented in the SI.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e Note that we simplify these categories into four major groupings: hurricane/tropical cyclone, flood, storm, and dam/levee break for clarity. We do not consider snowfall-related declarations here due to the inconsistent time lags associated with snowmelt-related flooding, and neglect tsunamis due to their seismologic forcing.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6025742/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6025742/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eFlooding in the U.S. leads to billions of dollars in financial losses annually, with projected increases due to escalating extreme precipitation, population growth, and deteriorating flood infrastructure. While federal regulation mandates flood insurance purchase within 100-year floodplains, analysis of millions of federal insurance claims reveals that most flood losses arise from frequent, low-intensity precipitation events relative to regional climatology, with average regional precipitation return periods of under five years. Similarly, precipitation linked to disaster aid and property buyouts has return periods averaging less than 20 years. Using unsupervised learning, we identify that space-time precipitation clusters associated with major storms dominate losses, emphasizing the need for flood risk assessments and mitigation strategies that account for recurrent spatiotemporal compound events. The findings bring the putative 100-year flood protection strategy into question and provide a focal point for the ongoing national discussions that underscore systemic challenges in U.S. flood preparedness.\u003c/p\u003e","manuscriptTitle":"Financial losses associated with US floods occur with surprisingly frequent, low return period precipitation","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-26 06:22:31","doi":"10.21203/rs.3.rs-6025742/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-water","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"natwater","sideBox":"Learn more about [Nature Water](https://www.nature.com/natwater/)","snPcode":"44221","submissionUrl":"https://mts-natwater.nature.com/cgi-bin/main.plex","title":"Nature Water","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"1c5d178d-3ca2-4379-b1e1-e8a91b72f294","owner":[],"postedDate":"March 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":44576367,"name":"Earth and environmental sciences/Hydrology"},{"id":44576368,"name":"Earth and environmental sciences/Natural hazards"}],"tags":[],"updatedAt":"2025-09-20T07:06:24+00:00","versionOfRecord":{"articleIdentity":"rs-6025742","link":"https://doi.org/10.1038/s44221-025-00506-8","journal":{"identity":"nature-water","isVorOnly":false,"title":"Nature Water"},"publishedOn":"2025-09-19 04:00:00","publishedOnDateReadable":"September 19th, 2025"},"versionCreatedAt":"2025-03-26 06:22:31","video":"","vorDoi":"10.1038/s44221-025-00506-8","vorDoiUrl":"https://doi.org/10.1038/s44221-025-00506-8","workflowStages":[]},"version":"v1","identity":"rs-6025742","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6025742","identity":"rs-6025742","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}