Wildfires and hurricane-related flooding are two events prioritized by STLT public health agencies, and important for CDC’s EHS program. The impact of drought on well water contamination was identified during a literature review and is detailed in Section 2.2. These are also priorities within CDC’s National Center for Environmental Health (NCEH)—a center that provides expertise and leadership in surveillance, epidemiology, technical assistance, training, and preparedness activities related to natural and human-induced disasters to domestic and international partners.

The PWWCII includes national, sub-county private well data for arsenic and nitrate concentrations. To briefly contextualize methodological decisions about contaminants, the potential for human exposure to and health impacts from nitrate and arsenic entering private wells may be exacerbated by the occurrence of flooding and drought respectively, and their presence or magnitude may serve as an indirect marker for the mobilization (or the inverse) of primary contaminants and/or other contaminants of concern for which no national, sub-county data is available. The justification for inclusion of these two contaminants was based on the literature and data search results detailed in Section 3.1.

We searched PubMed for articles on well water quality, natural hazards, and data modeling during the fall of 2021. Combinations of the following search terms were used: “private well water,” “domestic well water,” “natural disaster model,” “vulnerability model,” “catastrophe model,” “natural disaster,” “wildfire,” “hurricane,” “flooding,” and “drought”. Papers were identified from the search as relevant to the objectives of this work if they assessed risk or impact of contamination to private drinking water due to a natural hazard; additional papers that met the same criteria or provided nationwide private well water data were identified from the references of the relevant papers. As part of its mission, CDC’s Tracking Program provides nationally consistent, publicly available, sub-county spatially and temporally resolved public health and environmental data that drives data-informed decision making and action to improve community health throughout the country. Guided by this mission, governmental and academic data sources were identified from the literature search as well as from concurrent searches of governmental websites to obtain current, novel, and the highest spatially resolved data for the nation on private wells, natural disasters, and natural disaster relevant private well water contamination. Federal governmental website searches included CDC, Federal Emergency Management Agency (FEMA), Environmental Protection Agency (EPA), United States Geological Survey (USGS), and Drought.gov (a product of the National Oceanic and Atmospheric Administration’s [NOAA] National Integrated Drought Information System [NIDIS]). One author identified relevant papers as well as data and presented the extracted details for review by and discussion among all authors.

Prior to index creation, individual spatial variables identified from the FEMA National Risk Index (NRI), CDC Social Vulnerability Index (SVI), and USGS datasets were geospatially processed in ESRI ArcGIS Pro 2.6.7 (Redlands, CA) (Table 1). Briefly, Table 1 variables include drought: annualized frequency of drought occurrences (incident-days), arsenic in private well water: probability of arsenic concentrations exceeding 10 μg/L in private wells, coastal flooding: annualized frequency of coastal flooding incidents, hurricane: annualized frequency of hurricane incidents, riverine flooding: annualized frequency of riverine flooding occurrences (incident-days), nitrate in private well water: groundwater nitrate concentrations supplying private wells (mg/L), wildfire: area-weighted burn probability due to a large fire incident, nationally-normed and state-normed SVI: sum of 15 vulnerability variables’ percentile rankings, and private well water population: 2010 estimated persons dependent upon private well water.

The base spatial layer used for analysis was the CDC Tracking Program’s 2010 census tract shapefile restricted to the continental United States (CONUS). This resulted in 72,392 census tracts. We joined the 2017 FEMA NRI (Zuzak et al., 2021) and 2018 CDC SVI (Centers for Disease Control and Prevention/ Agency for Toxic Substances and Disease Registry/ Geospatial Research A, and Services Program, 2018) census tract datasets to the 2010 shapefile based on census tract FIPS codes. There were 25 census tracts from the 2018 CDC SVI file that required a manual spatial join because they did not join based on FIPS code. The USGS datasets were raster datasets per square kilometer (Ransom et al., 2021; Lombard, 2021; Johnson and Belitz, 2010). These data were converted to 2010 census tract measures (i.e., area-weighted persons dependent on private wells per 2010 census tract, area-weighted average nitrate concentration in mg/L per 2010 census tract, and area-weighted average probability of exposure to arsenic concentrations >10 μg/L per 2010 census tract) using the following geoprocessing tools prior to spatially joining to the CDC Tracking Program’s 2010 census tract shapefile: Raster to Point, Buffer, and Feature Envelope to Polygon. The final processed dataset was exported to SAS 9.4 (Cary, NC) for index creation and analysis, and bar charts were created in Microsoft Excel. Please see the Discussion section for data limitations.

The latest framework endorsed by the Intergovernmental Panel on Climate Change’s (IPCC) 6th Assessment Report defines risk of climate change impacts as arising from the dynamic interactions of hazards (the potential occurrence of an event or trend that may impact health) with exposure (the presence of people in a place that could be affected) and vulnerability (the propensity to be adversely affected) (IPCC, 2022). While the framework is not prescriptive, risk and impact assessment models in the disaster and environmental health literature (Brody et al., 2012; Blaikie et al., 2014; Zuzak et al., 2022) and grey literatures (CalEPA, 2021; University of Washington Department of Environmental & Occupational Health Sciences and Washington State Department of Health, 2022; Gill et al., 2014) are based upon a common risk equation: risk=hazard×vulnerability. Of these models, CalEnviroScreen’s cumulative impact assessment tool (Version 4.0) characterizes pollution burdens and vulnerabilities affecting California communities at the census tract level with the goal of identifying communities that may benefit from policy change, investment, or intervention. Striving for a similar goal, we adapted the formula and percentile approach underlying CalEnviroScreen 4.0 to characterize climate-related hazards and vulnerabilities for persons dependent on private wells within communities at the census tract level. A summary and the details of our methods are presented in Fig. 1. The formula for this work is: Climate-relatedhazardscomponent*Populationcharacteristicscomponent=PWWCII

Variables (detailed in Table 1 and Section 2.3) were aggregated into components and then indexes (Fig. 1C). We created an Overall PWWCII and three climate incident-specific indexes—Drought, Flood, and Wildfire. Overall PWWCII included data on drought, flood, and wildfire (Fig. 1).

There is no published measure that currently captures the same construct that our novel index is intended to represent: the potential impact of climate on communities reliant on private well water. We assessed the nationally-normed indexes for construct validity, which is comprised of divergent and convergent validity. Divergent validity assesses to what extent two items measure different constructs, whereas convergent validity demonstrates whether two items measure the same construct. To do this, we examined correlations between our indexes and the 2008–2012 American Community Survey (ACS) variables on owner and renter occupied housing units with telephone service to assess divergent validity as we expected these constructs to be unrelated (Manson et al., 2022). For convergent validity, there is a new index, the climate vulnerability index (CVI), which characterizes community-level vulnerability to and risk from climate change and is intended to inform future research as well as prioritize resources and interventions (Tee Lewis et al., 2023). Therefore, this was considered the most comparable surrogate to our index. Lastly, according to index creation best practices, we assessed that the variables comprising each index were not highly correlated. In regards to construct validity, DeVon et al. (2007) suggest a correlation coefficient of 0.45 or greater to support convergent validity, and their literature review contained some studies that supported convergent validity at 0.30 or greater; below these correlation coefficient values was considered to be in support of divergent validity (DeVon et al., 2007).

To better understand climate-related impact burden to communities reliant on private well water across CONUS, the nationally-normed Overall and climate incident-specific indexes were examined by climate region and demographics. To examine these differences, conducting chi-square tests followed by post-hoc testing would result in all statistically significant differences due to the large populations being assessed (over 37 million estimated for the U.S. private well water population and over 306 million estimated for the U.S. population). Therefore, Cramer’s V was used to assess univariate associations between PWWCII categories (Very Low, Low, Moderate, High, and Very High) and climate region and demographics. If Cramer’s V indicated an association, simple ordinal regression was performed as post-hoc analysis. Utilizing the Score Test for Proportional Odds Assumption in SAS from PROC LOGISTIC would result in all statistically significant violations to the proportional odds assumption for ordinal regression models again due to the large populations being assessed. Therefore, plots of cumulative logits were generated for each model and examined to ensure cumulative logit curves were relatively parallel to meet the proportional odds assumption.

Climate region data were obtained from NOAA (NOAA, 2004) and census-tract level demographic data were obtained from the 2010 U.S. Decennial Census (Manson et al., 2022). The 2010 estimated U.S. persons dependent on private wells was examined by climate region (Northwest, West, Southwest, West North Central, South, East North Central, Central, Southeast, Northeast). Because demographic information for the 2010 estimated U.S. persons dependent on private wells were not available, we used the 2010 U.S. Decennial Census to examine the total U.S. population by sex (male and female), age (under age 5, 5–24, 25–64, 65 and older), and race/ethnicity (Hispanic, non-Hispanic (NH) Black, NH White, NH Asian, NH Native Hawaiian and Other Pacific Islander, NH Other, NH American Indian and Alaskan Native, and NH Two or more races).

The methods in Section 2.4 produced nationally-normed climate-related impact indexes that compare census tracts across CONUS and highlight national differences. These methods were repeated to create state-normed impact indexes that compare census tracts within states and are relevant to public health equity work within states; state-normed impact indices can highlight the disproportionate burden of potential climate impacts to communities reliant on private well water within a state.

Four papers were identified from the literature search as relevant to the objectives of this work because they met the following criteria: assessed risk or impact of contamination to private drinking water due to a natural hazard (Pieper et al., 2021; Lombard et al., 2021a; Eisenhauer et al., 2016; Wing et al., 2002); we identified one additional paper providing nationwide private well water location data from the references of the relevant papers (Johnson et al., 2019).

Private well water testing for total coliform and E. coli are encouraged to screen and detect, respectively, for potential water-borne gastrointestinal illnesses from microbial contamination due to human or animal waste exacerbated by flooding events (Eccles et al., 2017; Pieper et al., 2021; EPA, 2024a). Testing for nitrate similarly indicates the presence of human and animal waste (from such sources as septic tanks or agriculture) as well as fertilizer contamination in private wells, and nitrates can affect everyone but can cause a serious condition called blue baby syndrome in infants and birth defects in developing fetuses (EPA, 2024b). Studies have demonstrated the impact of flooding or heavy rainfall on animal waste and the contamination of private drinking water (Eisenhauer et al., 2016; Wing et al., 2002; EPA, 2024b) and nitrate was predictive of fecal coliform/E. coli presence after precipitation in New Jersey (Procopio et al., 2017). While nationally consistent, sub-county total coliform and E. coli data are lacking, nitrate data were available and used as an indicator of potential contamination of private wells due to flooding (Ransom et al., 2022).

Lastly, chronic exposure to arsenic is associated with several cancers, such as lung, bladder, and prostate as well as adverse birth outcomes (Lombard et al., 2021a). Drought can contribute to groundwater wells running dry (Jasechko and Perrone, 2020), and a recent study indicated the potential adverse impact of drought on arsenic exposure via private drinking water wells; because national scale, sub-county data are available, arsenic data were included (Lombard et al., 2021b). Table 1 presents the results of the identification of nationwide, publicly available data sources at sub-county spatial resolution on private wells, private well water contamination relevant to natural hazards, natural disaster occurrence, as well as social vulnerability to natural disasters from governmental and academic data sources identified from the literature search as well as from searches of governmental websites.

Fig. 1C summarizes the four indexes created: Overall, Drought, Flood, and Wildfire PWWCIIs. Table S1 presents descriptive statistics on variables included in the PWWCIIs. While the results and analyses that follow are focused on the nationally-normed PWWCIIs, the state-normed PWWCII dataset will also be available on the CDC Tracking Network’s Interactive Data Explorer (CDC, n.d.) and available as a downloadable dataset (CDC, 2022b).

The variable percentiles comprising each of the PWWCIIs were not highly correlated — Spearman correlations ranged from −0.38 to 0.46 (Table S2A); similarly, component percentiles for each PWWCII and PWWCII percentiles were not highly correlated — Spearman correlations ranged from −0.13 to 0.27 (Table S2B). Lastly, it was interesting to note that population characteristics percentiles were marginally more correlated with the Flood, Wildfire, and Overall PWWCII score percentiles than their respective climate-related hazard percentiles, except for Drought in which case they were equally correlated (Table S2B).

The convergent validity Spearman correlation coefficients between our PWWCIIs and CVI ranged from 0.38 to 0.55, and divergent validity Spearman correlation coefficients between our PWWCIIs and owner and renter occupied housing units with telephone service ranged from −0.10 to 0.00 (Tables S2C–D). Taken together, this initial evaluation supported the validity of the PWWCII construct.

The distribution of the estimated private well water population living in PWWCII census tracts by climate region are visualized in Fig. 3 and presented in Table S3. For the 2010 estimated U.S. persons dependent on private wells, the univariate associations between the Overall, Flood, and Wildfire PWWCIIs and climate regions were strong (Cramer’s V = 0.32–0.34), and the univariate association between the Drought PWWCII and climate regions was moderate (Cramer’s V = 0.23) (Table 2).

Compared to the Southeast where the majority of the 2010 estimated U.S. private well water population lived, the estimated persons dependent upon private well water living in the West had an increased odds of living in higher impact census tracts for the Overall PWWCII (OR = 2.12, 95 % CI: 2.11,2.13) (Table 3 A). Whereas, for the Drought PWWCII, the estimated persons dependent on private wells living in the West (OR = 14.43, 95%CI: 14.36,14.50), Southwest (OR = 5.78, 95%CI: 5.75,5.81), West North Central (OR = 4.69, 95%CI: 4.67,4.71), South (OR = 2.09, 95%CI: 2.09,2.10), and Northwest (OR = 1.28, 95%CI: 1.27,1.28) had an increased odds of living in higher impact census tracts compared to the Southeast. For the Flood PWWCII, no climate regions in the U.S. had an increased odds of their estimated private well water population living in higher impact census tracts compared to the Southeast (Table 3 A). However, for the Wildfire PWWCII, the estimated persons dependent on private wells living in the West (OR = 1.50, 95%CI: 1.49,1.51) and South (OR = 1.31, 95%CI: 1.30,1.31) had an increased odds of living in higher impact census tracts compared to the Southeast.

For the 2010 census total U.S. population, the univariate associations between the Overall, Drought, Flood, and Wildfire PWWCIIs and race/ethnicity were weak (Cramer’s V = 0.10–0.12); whereas, the univariate association between the four PWWCIIs and age as well as sex were negligible (Cramer’s V = 0.00–0.02) and therefore not analyzed further (Table 2). The percent of the 2010 census total U.S. population living in PWWCII census tracts and no private well water population census tracts are visualized in Fig. 4 by race/ethnicity with corresponding n’s presented in Table S4 and presented in Table S5 by sex and age.

Compared to non-Hispanic White persons, non-Hispanic American Indian and Alaska Native (AI/AN) persons (OR = 2.72, 95 % CI: 2.71,2.72) and Hispanic persons (OR = 1.06, 95 % CI: 1.06,1.06) had an increased odds of living in higher impact census tracts for the Overall PWWCII (Table 3B). Similarly non-Hispanic AI/AN persons (OR = 3.14, 95 % CI: 3.13,3.15) and Hispanic persons (OR = 1.22, 95 % CI: 1.22,1.22) had an increased odds of living in higher impact census tracts for the Drought PWWCII compared to non-Hispanic White persons. Non-Hispanic AI/AN persons (OR = 1.23, 95 % CI: 1.23,1.23) had an increased odds of living in higher impact census tracts for the Flood PWWCII (OR = 1.23, 95 % CI: 1.23,1.23) and the Wildfire PWWCII (OR = 3.12, 95 % CI: 3.11,3.13) compared to non-Hispanic White persons (Table 3B). Plots for simple ordinal regression models in sections 3.4 and 3.5 are presented in Fig. S1. With the exception of the plot for Wildfire PWWCII and Climate Region, the plots exhibit parallelism.