The Center for Aging, Health, & Environment is pleased to announce seed funding for projects integrating social and environmental data to examine the intersections of aging, health & environment

Applications due October 30, 2026

We envision 2-3 awards of $25,000. Note that if your institution requires indirect costs on small grants, this amount must be included in the $25,000.

Eligibility: Investigators who are eligible to be Principal Investigators of NIA-funded projects. Doctoral students cannot be PIs for these awards.

Proposals will be evaluated in terms of merit and innovation; significance, both substantive and in terms of the potential for the resulting code and data to fuel further research; involvement of early-stage investigators; potential to lead to larger scale, NIA-funded research. The potential for external funding, primarily through NIA, is an important consideration for investment of CACHE funding. In addition, we aim to fund at least one scholar representing an NIA-funded Aging Centers.

The research must also involve the integration of social and environmental data.

Recipients of CACHE seed funding must agree to sharing data and code through the CACHE platform (or pointing to the relevant archive), thereby further fueling the development of Aging-Health-Environment research infrastructure.

The award period is one year, with a start date of January 1st, 2027, and the funds may cover faculty summer salary, graduate student salary, research travel, and/or data acquisition. The funds cannot be used for conference travel or for group meals.

Proposal (no more than 3 single‐spaced pages) to include: 

• Proposal title and investigator names and contact information; 
• Specific Aims (in NIH format) 
• Central research questions to be addressed;  
• Significance and potential impact of the research project including as related to fueling future research by sharing integration code and data; 
• Description of innovation 
• Short literature review; 
• Description of data and methods;  
• Progress to‐date (if applicable), timeline, and planned submission date and agency for external grant proposal. Submission should be within 16 months of award. 
• References (separate from 3-page limit) 
• Budget (separate from 3-page limit, 1‐page), with major budget items, including personnel, and a brief justification 
• CVs of all investigators and list of other current grant support, if applicable 

E‐mail applications to Julia Shipman, Program Manager, Center for Aging, Health & Environemnt: cache@colorado.edu  

NOTE: Both a progress and final report will be required. Funds cannot be expended until evidence is provided of all investigators’ completion of CITI human subjects training. The funding of seed projects is contingent on the continued availability of CACHE funding.

A PDF for this funding call can be found here. 

September 2026

Link to data

Click here

---

Prepared by: Kathryn Foster, Cornell University  

Date: August 2026

---

Authors: U.S Census Bureau, Department of Commerce  

About: The Local Air Conditioning Estimates (LACE) data is a part of the experimental data products produced by the Census Bureau and provides national, state, county, and tract-level estimates of the number of occupied housing units with or without air conditioning (AC). The data is derived and created from two sources: American Housing Survey (AHS) and the American Community Survey (ACS). 

Data are available on: The Local Air Conditioning Estimates (LACE) data is a response to comments from users of the Community Resilience Estimates (CRE) for Heat, who suggested that the lack of AC is an important measure of social vulnerability to extreme heat, yet it was not recorded in the data at granular geographic levels. LACE was created through cross-survey modeling by using the data from the AHS to train machine learning models to estimate how respondents in the ACS would have responded to questions from the AHS about AC. 

The LACE contains estimates (and margins of error) for the number and percentage of occupied households with and without any kind of AC at national, state, county, and tract-levels.  The data can be used for assessing vulnerability to extreme heat risks. Furthermore, the data can be combined with other dataset such as ACS age estimates or the Health and Retirement Study to estimate heat and health risks for older populations.  

Additional Resources about the LACE data:  

Local Air Conditioning Estimates Experimental Web Page https://www.census.gov/data/experimental-data-products/lace.htmlCross-Survey  

Modeling: Fusing Data from Multiple Data Sources to Enhance Multi-Dimensional Measures Working Paper 

https://www2.census.gov/library/working-papers/2025/demo/sehsd-wp2025-05.pdf

Link to data

Click here

---

Prepared by: Kathryn Foster, Cornell University  

Date: August 2026

---

Authors: Arizona State University’s Knowledge Exchange for Resilience and the U.S Census Bureau, Department of Commerce 

About: The Community Resilience Estimates (CRE) for Heat data builds from CRE measures of social vulnerability and community resilience to disasters to adjust CRE’s social vulnerability components to be relevanttoextreme heat exposure. The data is derived fromtheAmerican Community Survey (ACS), Census Bureau’s Population Estimates Program(PEP), theAmerican Housing Survey (AHS), and the North American Regional Reanalysis (NARR). The CRE for Heat measures heat exposure and social vulnerability at thenational,state, county, and tract-level. 

Data are available on: The CRE for Heat measures components of social vulnerability at the household and individual level. The original CRE provides a metric of social vulnerability and resilience to disasters such as hurricanes, floods, and pandemics. However, does not include measures specifically for extreme heat exposure. To account for this, the Census Bureau partnered with Arizona State University’s Knowledge Exchange for Resilience to develop measures of social vulnerability to extreme heat.  

The CRE for Heat provides more social vulnerability indicators such as whether a household’s housing costs are more the 50%, housing structure type, individual’s commute type (public transit, walking, biking, personal vehicle, or non-personal vehicle), and households that potentially lack air conditioning.  While the 2019 CRE for Heat only examined social vulnerability to an extreme heat event, the 2022 release provides heat exposure data based on air temperature and wet bulb temperature spatially joining temperature data to Census geographies.  

These data may be particularly helpful to researchers and practitioners who are interested in older populations’ risks to extreme heat. As past research has shown that older adults are more vulnerable to climate risks, the CRE for Heat, in combination with other datasets, may provide pathways for identifying social vulnerabilities at the intersection of heat exposure and aging.  

Additional Resources for the CRE for Heat data: 

https://www.census.gov/programs-surveys/community-resilience-estimates.html 

For additional questions about the data please contact: sehsd.cre@census.gov 

If you’re interested in contributing a short What Am I Reading post, we’d love to hear from you! Email us at cache@colorado.edu.

Written by Kathryn Foster, Cornell University 

Projections and forecasts are widely used within the demographic and climate communities (Balk et al., 2022). Projections are an especially useful tool in climate-epidemiology research, to estimate the future impacts of climate change on public health. A recent study, however, strongly cautions against projections that attribute future health risks to climate change without considering underlying demographic changes that may increase vulnerable populations’ susceptibility to climate hazards, such as global aging and population growth (Kephart and Bilal, 2026). The article delineates three drivers of future health risk that are often conflated in climate-health projections. First, there may be changes in the frequency or severity of climate hazards (e.g., longer heat waves, more intense hurricanes). Second, population growth (or decline) may contribute to the changes in the total number of people exposed. Third, demographic shifts in the population, such as increases in life expectancy and greater shares of older adults, may change the vulnerabilities of people exposed. Kephart and Bilal argue that if studies group together distinct drivers of future health risks, results will indicate only if and to what extent health risk may arise. But they will not clearly indicate why health risks arise, and  for whom they may disproportionately impact. Amid global aging and increasing environmental hazards, differentiating between these drivers allows researchers to accurately attribute risks to either climate or demographic changes.  

Kephart and Bilal (2026) provide four ways to incorporate population dynamics into climate-health projections to improve both transparency in research and attribution of factors that can serve as a basis for public health and climate adaptation (see Table 1): (1) Report climate and demographic counterfactuals on the same spatial or temporal scale; (2) Include normalized measures of health burdens alongside absolute measures in research; (3) Include age-standardized rates if available; (4) Avoid misattributions within titles, abstracts, or other products available to the public.  

The three examples below demonstrate innovations in current research that disentangle population aging, climate hazards, and health risks in projection modeling. In accordance with Kehart and Bilal’s recommendations, the following studies show how sociodemographic dynamics like aging, migration, and inequality interact with climate hazards to shape future climate-health risks. These examples each use a different spatial scale and extents from continental to cities to sub-city, respectively.  

Example 1: Population aging and heat exposure in the 21st century: which US regions are at greatest risk and why? 

This study by Carr et al. (2024) shows how population aging and rising temperatures may create regions in the U.S. where these two trends collide, increasing population exposure to heat. Drawing from two main data sources, county-level projections and gridded daily temperature projections, this research identifies changes from 2020 to 2100 in person-degree days, which is the cumulative population-level chronic heat exposure multiplied by the population over 69 years of age in the county at the given time period. The authors simulate future demographic dynamics under 5 different climate scenarios using a shared socioeconomic pathways (SSP) approach.  

Why use the SSPs to understand population futures?

The SSP approach uses expert opinion about future changes in demographic rates (fertility, life expectancy, migration between countries) and socioeconomic characteristics (education, urbanization, economic growth, and technological development) consistent with different scenarios of future carbon emissions (e.g., sprawling or concentrated spatial patterns of growth, economic growth consistent with fossil-fuel usage or sustainability futures). Using SSPs to guide narratives about the future, researchers and practitioners can quantify aspects of socioeconomic and environmental conditions and the pathways that lead to those outcomes (Climate Central (2021).

SSPs—and the use of scenarios—have a long history. Researchers and other users of scenarios around the world periodically reflect upon them. A Scenario Evolution Process is currently underway, and users (or potential users) are invited to participate in the scientific discourse.

In addition to SSPs, this study is consistent with Kephart and Bilal’s advice to avoid conflating demographic change with increased climate exposure. They use decomposition to calculate the proportion of the change in older adults’ chronic heat exposure attributable to climate change, population size, and population aging.  

The decomposition findings show that across the U.S., all three drivers contribute relatively equally to projected changes in older adults’ chronic heat exposure, with the contribution of climate change being slightly smaller than population size and aging. However, when examining the findings by U.S. Census Divisions, regional differences emerge. Climate change will account for a larger share of population heat exposure in historically colder regions, such as the Midwest and the northeastern regions of New England. However, population growth and aging are larger drivers of changes in older adults’ chronic heat exposure in the Southern and Mountain regions. These findings highlight regions in the U.S. that could be “hotspots” for both population aging and increased heat exposure. 

(Interested readers can find the same methodology applied to global regions in Falchetta et al. 2024.)

Example 2: Temperature-related mortality burden and projected change in 1368 European regions: a modelling study

This study by García-León et al. (2024) disentangles and projects the impacts of an aging population on temperature-related mortality risks by 2100 in 1368 regions in Europe. The authors use a three-stage modeling framework (found here) to estimate temperature-related mortality risks across age groups. The risks were then used to create temperature thresholds of mortality percentiles for both raw and age-standardized mortality rates for heat and cold. Similar to Kephart and Bilal’s suggestion, the authors isolated the effect of demographic change on future temperature mortality by calculating the difference in mortality between cases identifying mortality with (1) the present climate combined with the present population and (2) the present climate combined with future population exposure projections. 

While Europe varies by region, climate change is expected to widen disparities in mortality, more so than population aging, particularly in southern Europe, where heat-related deaths are projected to increase. The authors find that while the mean heat in Northern Europe is expected to increase, it will not cause additional deaths. Rather, the projected aging of the population will make the region more susceptible to extreme heat but will not cause additional deaths. Furthermore, they find that the effects of aging are much more pronounced for cold-related mortality in eastern regions of Europe, specifically for elderly populations, the authors do not pin-point mechanisms that may mediate this relationship such as whether they expect cold-weather mortality to increase due to increased exposure to cold and winter storms (e.g., falling on ice) or due to increased respiratory illnesses that are seasonally dominant and disproportionately affect older adults (e.g., influenza and pneumonia) For both heat- and cold-related mortality risks, the authors note they vary by region (and regional economic characteristics) and under projected demographic changes. 

Example 3: Equity dimensions of future vulnerabilities to climate hazards: Long-term spatial population projections for New York City consistent with Shared Socioeconomic Pathways (updated study is forthcoming, early results prepared for New York City’s Mayor’s Office were reported in  here). 

This study by Balk et al. (forthcoming 2026) adopts a multidimensional, multiregional demographic projection method to forecast long-term county-level population in order to understand the future health-related costs of morbidity and mortality associated with climate change in New York City (NYC) (McPherson et al. 2024). Most SSP-based projections are applied to large regional or national-level areas, whereas this study uses a fine-scale geography (counties) for population projections and downscales these projections to sub-city, fine-scale grid cells. The authors develop a novel set of local-level SSPs by notably including both migration within national borders and demographic rates for population subgroups that represent the diversity of a city (not previously done in SSPs). By applying these SSPs, they created multiple possible population scenarios for NYC, avoiding misattribution in population projections and providing a new pathway for local climate-health projections. The research team rescaled county projections to smaller units using historical parcel and Census block group data, allowing them to better assess local population change and future climate impacts with a detailed spatial scale.

The research finds that determinants of population growth change under different scenarios and across geographical locations. Results suggest that across most SSPs, population growth will decrease across NYC within the coming decades due to decreasing migration and fertility rates and increasing mortality rates and population aging. Through creating multiple population scenarios and spatial downscaling, the authors demonstrate a method for more accurate population-health projections and other impacts of future climate impacts on populations at risk at smaller geographical scales. Going beyond the recommendations of Kephart and Bilal, this work suggests a need to consider uncertainties in future demographic trends that may alter population size and age structure in a local area. 

Future Research and Policy: In climate-health projections, it is important to disaggregate the drivers of future health risks. Determining what process is impacting future health risks, whether it is climate or demographic change, will inform subsequent adaptation and policy intervention.

Carr et al. (2024) and García-León et al. (2024) highlight regions where population aging may pose a greater health risk or where climate change may be the greater contributor to future risk. In ‘hotspot’ areas, or areas where aging populations and rising temperatures equally contribute to future health risks, they suggest that public initiatives should focus on educating older adults about heat risks by sending heat alerts, informing them about and offering transportation to cooling centers, and offering home-based cooling solutions (Carr et al., 2024). Based on the varying size and age structure in the future NYC example (Balk et al., 2026), policies in cities, which may be particularly vulnerable to heat due to the urban heat island effect and could have more (or less) diverse populations, may also need to embrace increased uncertainties in future population.

Accurately attributing demographic change and climate change in future health risk and exposure research yields projections with more usable information, subsequently, better evidence for policy and planning. These practices from researchers will support effective action in protecting aging populations and mitigating the hazards associated with future temperature extremes.

References 

• Anon. 2021. “IPCC 6th Assessment Report – the Physical Science Basis | Climate Central.” Climate Central. Retrieved July 9, 2026 (https://www.climatecentral.org/climate-matters/ipcc-6th-assessment-report-the-physical-science-basis).

• Balk, D., Tipaldo, J., Leiwen, J., & Zoraghein, H. (2026). Equity of future vulnerabilities to  climate hazards: Long-term spatial population projections for New York City consistent  with Shared Socioeconomic Pathways [Forthcoming]. Proceedings of the National  Academy of Sciences of the United States of America. 
• Carr, D., Falchetta, G., & Sue Wing, I. (2024). Population aging and heat exposure in the 21st century: which US regions are at greatest risk and why?. The Gerontologist, 64(3), gnad050.  

• Climate Matters. 2021. “IPCC 6th Assessment Report – the Physical Science Basis | Climate Central.” Climate Central. Retrieved July 9, 2026 (https://www.climatecentral.org/climate-matters/ipcc-6th-assessment-report-the-physical-science-basis). 

• Falchetta, G., De Cian, E., Sue Wing, I., & Carr, D. (2024). Global projections of heat exposure of older adults. Nature Communications, 15(1), 3678. 

• García-León, D., Masselot, P., Mistry, M. N., Gasparrini, A., Motta, C., Feyen, L., & Ciscar, J. C. (2024). Temperature-related mortality burden and projected change in 1368 European regions: a modelling study. The Lancet Public Health, 9(9), e644-e653. 

• Kephart, J. L., & Bilal, U. (2026). Disentangling climate hazards from demographic change in climate–health projections. Nature medicine, 1-2. 

• McPhearson, T., Towers, J., Balk, D., Horton, R., Madajewicz, M., Montalto, F., Neidell, M., Orton, P., Rosenzweig, B., Bader, D., Chen, Z., DeGaetano, A., Evans, C., Golkhandan, M. R., Gurian, P., Kaatz, J., Herreros-Cantis, P., Lo, R., Ortiz, L., Braneon, C., Branco, B., Campbell, L., Dubay, F., Graziano, K., Jiang, L., Johnson, M., Kennedy, C., Kioumourtzoglou, M.-A., Kleyman, J., Kleyman, J., Knowlton, K., Limaye, V., Matte, T., Munoz Perez, S., Reed, D., Shakya, M., Svendsen, E., Tipaldo, J., Zoraghein, H. (2024). New York City Town+Gown Climate Vulnerability, Impact, and Adaptation Analysis Final Report. New York City Mayor’s Office of Climate and Environmental Justice. 

Link to data

USC/UCLA Center on Biodemography and Population Health (CBPH)

Click here

---

Prepared by: Alex Mikulas, PhD, CACHE postdoctoral associate 

Date: July 2026

---

Principal Investigator: Dr. Jennifer Ailshire, University of Southern California 

About the data:

(Paraphrased from CDR website) The Contextual Data Resource (CDR) is a collection of user-friendly datasets that integrates contextual data with several extensive datasets and surveys on health and aging. The CDR enables researchers to study the impact of place on health and well-being among older adults within the structure of existing and well-used data sets. Available contextual data within the CDR include measures on socioeconomic and demographic structures, economic conditions, social stressors, health care, physical hazards, amenities, and the built environment. Depending on the underlying data source, measures are available at multiple spatial scales over several decades. 

The potential to link this contextual data to aging surveys will increase opportunities to analyze prospective effects of environmental conditions on health and aging, the effects of residential mobility on aging-related outcomes, and the ways environments change around older adults as they age in place. 

Data are available on: 

The CDR datasets can be accessed for use with several restricted-level datasets via the MiCDA Enclave Geographic Linkages Repository, including the Health and Retirement Study, Panel Study of Income Dynamics, and National Health and Aging Trends Study. The data is also available for use with the Understanding America Study via a Tier 3 Data User agreement and with the Hispanic EPESE (Established Population for the Epidemiological Study of the Elderly) via the Hispanic EPESE study team. Contact CDR administration for data access for use with other datasets, cdradmin@usc.edu. 

The Contextual Data Resource gathers and processes data from multiple sources into user-friendly data that can be easily integrated into aging surveys. These datasets include the following: 

• USDA Food Environment Atlas and Food Access Atlas 

• Uniform Crime Report from the FBI and NACJD 

• US Census Bureau decennial counts and American Community Survey 

• Dartmouth Atlas of Healthcare 

• Air Pollution (0.3 and PMD2.5) from the Fused Air Quality Surface Using Downscaling Files (FAQSD) 

• Air Pollution (ATMOS) from the Atmospheric Composition Analysis Group (ACAG) 

• Street Connectivity from the US Census Bureau TIGEWR/Line shapefiles 

• Noise data from the National Parks Service Geospatial Sound Monitoring Files  

• Area Health Resource Files from the US Health Resources and Services Administration (HRSA) 

• Weather and environmental data from gridMET 

 

If you’re interested in contributing a short What Am I Reading post, we’d love to hear from you! Email us at cache@colorado.edu.

Written by Sophia Arabadjis, MA, MSc, PhD. Institute for Implementation Science in Population Health, City University of New York Graduate School of Public Health and Health Policy

---

Have you ever considered linking population data to information from a thematic map on, for example, coastal flooding, social mobility, wildfire hazard or social vulnerability? Maps are powerful communication tools, especially when interactive, allowing users to explore and visualize finely resolved geographies like ZIP code tabulation areas or census tracts or block groups. However, these maps and measures — and the data underlying them — are not always precisely measured, which can misconstrue spatial patterning of local risk and vulnerability.  

These issues of measurement have very real implications for researchers and decision-makers who often use maps like these to provide environmental context or guide policy and resource allocation, respectively. Ignoring measurement issues can yield inaccurate insights.   

This “What am I Reading?” entry provides an example of the implications of measurement error. Using material from our recent Annals of the American Association of Geographers paper, we explore how the sampling uncertainty in different data sources impacts wildfire vulnerability index construction. We describe various methods to construct vulnerability indices; discuss the sources of data used to create measures; and show how ignoring the sampling uncertainty can lead to false conclusions about vulnerability and a misallocation of resources[1]. 

Creating Composite Measures 

In both academic literature and policy applications, it is common to summarize an area’s vulnerability using a singular score or number based on a scale or rubric, called a composite index. Composite indices are functions of several different measures recorded at each location or geographic unit across a landscape. Measures generally reflect population (social vulnerability components) and environmental context (locational vulnerability components) depending on the application. For example, the Center for Disease Control’s Social Vulnerability Index (SVI) combines national survey measures of socioeconomic status, household composition, racial and ethnic minority status, and local housing types/transportation to estimate overall vulnerability across a landscape[2]. For each location, or areal unit, the measures are summed to create a singular SVI score; scores between 0.75 and 1 indicate high vulnerability. The SVI and other similar composite indices (e.g. SoVI, SEDAC [3-6]) are increasingly used to prepare and direct resources towards areas in the top ~10% of scores, a.k.a. “highly vulnerable” areas[7-11].  

Considering Measurement Error 

While environmental measures may come from a variety of sources, the population measures are often from the same source: the American Community Survey (ACS). The ACS is a monthly survey of roughly 3.5 million people in the United States. The survey asks detailed questions about economic, social, and demographic characteristics, which are then aggregated (across areas and years) to different census geographies, from census block groups (600 – 3,000 people) all the way up to national estimates. Though 3.5 million people may seem like a large survey, this equates to roughly 14 people per census block group annually[12]. These small sample sizes mean that the errors within a given measurement may be quite large[12], which in turn means that the true value of the measurement could feasibly be any value within a large range of values — simply because of the sampling design. 

Fortunately, along with the population measure point estimates (for example, the population of older adults in a given area), the ACS also provides an estimate of the uncertainty due to the sample design[12-13]. The ratio of the sampling uncertainty value to the point estimate, called the coefficient of variation, can give us a quick insight into the estimate’s precision. For example, suppose the ACS suggests that 15% of the population in a given census tract is above the age of 65 with a standard deviation of 1.8%; the ratio would be 0.018/0.15 = 0.12, which is a relatively low value, so we have relatively high confidence that ~15% of the population in that area is actually an older adult. However, if we have the same estimate with a standard deviation closer to 8%, then the ratio is over 0.5 and suddenly we have a lot less confidence that 15% of the population is accurate. Figure 1 maps the ACS coefficient of variation for a subset of urban census tracts and block groups in Santa Barbara County. A coefficient of variation (CV) greater than 0.12 is relatively good value; a CV of 0.5 or greater suggests much less confidence and more uncertainty. The table at right gives the percent of Census block groups that fall in each category for this urban subset. The maps show higher coefficient of variation values at smaller census block group geographies, which indicates more error within smaller geographic units.  

[Caption: Figure 1 maps the coefficient of variation (standard deviation/estimate) for a subset of urban Census tracts and Census block groups.] 

Environmental measures also have uncertainties and measurement issues, though these generally take a different form. The error or noise in physical measurements may come mechanistic constraints of instruments (for example specificity and tolerance in thermometers), from spatial interpolation (e.g. creating a rain surface from point measurements across a landscape), or model-based variation (e.g. gridded outputs of complex physical or statistical models.) Some of these uncertainties may be knowable, such as the tolerance of an instrument, but others are not so easy to quantify.  

Simulation is one way to propagate measurement error and assess its effects on the stability of findings. In a simulation framework, samples of each variable are repeatedly drawn from a range of acceptable values with an assumed shape (e.g. a normal distribution), and then summarized across those values. For example, one could take the pixel values of an environmental variable and the tolerance of that instrument and treat those as the mean and variance of normal distributions across a landscape (see Figure 2). Then, over the course of several samples (perhaps 1000), trends in values can be summarized. In our case, we are interested in how many times a specific pixel appears in the top 10% of values across each simulation. This is the recurrence rate and provides us insight into the stability of the top 10% of the distribution (“highly vulnerable”) areas. 

[Caption: Figure 2 Visible Atmospherically Resistant Index (VARI) is a satellite-derived measure that is closely correlated with live fuel moisture. Displayed are the VARI values per pixel in an urban area of Santa Barbara County in the period just before the Thomas Fire of 2017.[18]] 

An example of the implications of sampling error using a wildfire risk index 

To ground our investigation of wildfire risk, we chose the 2017 Thomas Fire as a case study. The Thomas Fire began December 4th, 2017 in Ventura County California and quickly spread into neighboring Santa Barbara County. Wildfires in this region of California are characterized by chaparral vegetation (burn-adapted shrublands), steep canyons, mesas, and unique wind patterns that make wildfires in this area both fast-moving and highly destructive[14-16]. By December 6th, 2024, the Thomas Fire had engulfed more than 100,000 acres; it would go on to consume 281,000 acres, destroy 1000 structures, and force more than 100,000 people to evacuate (see Figure 3)[17]. 

[Caption: Figure 3 The Thomas Fire final fire perimeter (red, hexagonal pattern) burned into densely populated coastal regions of Santa Barbara and Ventura counties (a-d). The estimated population aged 65 years or more per US Census block group (left column) and tract (right column) shows the proportion of older adults varies spatially across counties. Source: Arabadjis et al., 2025[1]] 

With the Thomas Fire as our backdrop, we construct a simplified wildfire risk index with two measures common to the literature: a measure of the share of the population of older adults from the ACS (aged 65 years or older) and a satellite-based measure of vegetation moisture called the Visible Atmospherically Resistant Index (VARI). The VARI is derived from the green-to-red band signals and is strongly correlated with live fuel moisture which in turn, is strongly associated with wildfire ignition, spread and intensity in chaparral environments[18]. We take several steps.  

First, we use a simulation framework within which samples are repeatedly drawn from the range of values (proportion of older adults) dictated by the sampling uncertainty. In mathematical terms, we take a draw from a normal distribution with mean as the logit-transformed point estimate (μˆ) and variance (σ2)$${\mathrm{}}^{}$$specified from the variance estimate replicate tables for each Census geography (s). (See equation 1.) 

We then summarize across the simulations to make sense of impacts of the sampling uncertainty of the ACS. We show that the selection of the top 10% most vulnerable areas — the areas that would likely be identified to receive resources — is sensitive to sampling uncertainty. We find that at smaller analytic scales (i.e. census block groups), the top 10% most vulnerable areas selected are not necessarily the same as if we took the proportion or share of older adults at face value. 

Figure 4 provides a visual explanation of this finding, comparing the top 10% most vulnerable Census block groups (left column) and Census tracts (right column). Panels (a) and (b) show the top 10% areas according to the raw point estimates of the share of older adults in each area. Panels (c) and (d) show the recurrence rates of each Census block group or tract over the course of 1000 simulations. Areas shaded in brown-to-gold consistently appear in the top 10% most vulnerable (in 90+% of simulations); areas shaded in green-to-tan appear with less frequency in the top 10%, but any draw from the distribution is equally likely under the distributional assumptions. Panels (e) and (f) show the difference in the top 10% using the raw point estimates versus the simulation. Areas shaded in blue were ranked in the top 10% using the raw point estimates and the simulation. Areas in yellow were only ranked in the top 10% using the raw estimates, and areas in pink were only selected using the simulation method. Given the different sizes of Census tracts and block groups in Ventura and Santa Barbara county, these are difficult to see. For Census block group map (e), 2 areas are shaded pink; in the Census tract map (f) 1 area is shaded pink. These differences suggest that resources may be misallocated (or more resources may be needed). Areas in red are excluded from analysis.

[Caption: Figure 4 displays the Census tracts (right) and Census block groups (left) of Santa Barbara and Ventura counties. In the top panel, geographies with a top 10% share of the older population are highlighted in yellow. In the middle panel (c-d) show the recurrence rates for the top 10% from the simulations. The lowest panel (e-f) maps differences in which geographies are in the top 10% of older adult population share by simulation versus raw point estimate. Source: Arabadjis et al., 2025[1]] 

Second, we propose a statistical model and simulation procedure that combines the older adult population measure and the vegetation measure to create an example composite wildfire risk index. (See equation 2.) Importantly, our statistical model and procedure accounts for the sampling uncertainty of both the population and environmental measures and has a closed form expression of the variance[1]. (See paper for details.)  

Similarly to Figure 4, each point in panels (c) and (d) in Figure 5 are colored to represent the proportion of simulations for which the index value at that point appeared in the top 10% of each sample (i.e. vulnerable areas). Areas in green almost never appeared in the top 10%; areas in tan appeared in the top 10% in roughly ⅔ or more of samples; and areas in brown-to-gold appeared in the top 10% in ninety percent or more of the simulations. These brown-to-gold areas are consistently highly vulnerable per our index.   

Ultimately, our simulations show that accounting for the sampling design makes a difference in which areas are designated as “highly vulnerable to wildfire,” and that the uncertainty in the older adult measure likely dwarfs any uncertainty in the vegetation measure. 

In this way, wildfire risk, in particular, would benefit from more finely resolved person-level data, such as parcel-level indicators from county tax assessor data and specialized survey data measuring risk, mitigation, and specific demographic indicators. This is especially important for older adults who may live in wildfire prone areas, but have different risk profiles in terms of knowledge, capacity, and underlying health[19]. 

Our results extend beyond our simplified wildfire example. We show that ignoring the sampling uncertainty in any ACS population estimates used in an index may misidentify risk across a landscape. We also note that the more complex the index (more measures), the more complicated the statistical model needs to be to incorporate the uncertainty, and the more trouble the subsequent risk distribution may be (multimodal, for instance).  

However, simulation is another powerful tool that can help practitioners overcome sampling design constraints. Generating maps of simulation summaries (such as Figures 4 and 5) are potentially just as interpretable, but more true to the underlying unknowns in the data.

[Caption: Figure 5 maps the recurrence rates for each point (s) in the top 10% of wildfire risk index values. Subfigure (c) shows a Census block group-based-index and subfigure (d) shows a Census tract-based index. Source: Arabadjis et al., 2025[1]] 

The article contributes to a robust and growing literature on vulnerability to environmental hazards. We show that uncertainty in the underlying data can distort vulnerability indices, especially as areal units get smaller.  However, simulation is a pragmatic tool to help practitioners rigorously identify vulnerable areas and improve resources targeting across a landscape.  

The real take-home message is that the next time you click on a map that highlights certain geographies as ‘highly vulnerable’, interpret with care!  

References: 

[1] Arabadjis, S. D., Zheng, Z., Strange, L. P., Murray, A. T., & Sweeney, S. H. (2026). Social Vulnerability, Locational Vulnerability, and Uncertainty in Wildfire Risk Index Construction. Annals of the American Association of Geographers, 116(5), 1211–1234. https://doi.org/10.1080/24694452.2025.2604851 

[2] Flanagan, Barry E. et al. (2011). A Social Vulnerability Index for Disaster Management. 8(1).  

[3] Cutter, S.L., B.J. Boruff, and W.L. Shirley. (2003). Social Vulnerability to Environmental Hazards. Social Science Quarterly 84(2). DOI: https://doi.org/10.1111/1540-6237.8402002  

[4] Cutter, S.L. (1996) Vulnerability to Environmental Hazards. Progress in Human Geography 20(4). https://doi.org/10.1177/030913259602000407  

[5] Cutter, S.L. (2024) The Origin and Diffusion of the Social Vulnerability Index (SoVI). International Journal of Disaster Risk Reduction. 109(104567). https://doi.org/10.1016/j.ijdrr.2024.104576   

[6] NASA SEDAC. 2023. Center for International Earth Science Information Network (CIESIN) Documentation for the U.S. Social Vulnerability Index Grids: NASA Socioeconomic Data and Applications Center (SEDAC). New York. Columbia University. 

[7] South Carolina Office of Resilience. (2026). Retrieved June 11, 2026, from https://scor.sc.gov/ 

[8] State of California Governor’s Office of Land Use and Climate Innovation (2026). Retrieved June 11, 2026, from https://vcp.lci.ca.gov/  

[9] Federal Emergency Management Agency, FEMA (2026). Retrieved June 11, 2026, from https://www.fema.gov/emergency-managers/practitioners/recovery-resource-library/social-vulnerability-environmental  

[10] U.S. Department of Agriculture and Rural Development, USDA (pre-2024). Retrieved 2024 from https://www.rd.usda.gov/priority-points/equity-search (No longer available.) 

[11] Maine Infrastructure and Adaptation Fund MIAF (2026). Retrieved June 11, 2026 from https://www.maine.gov/future/climate/community-resilience-partnership and https://www.maine.gov/future/sites/maine.gov.future/files/inline-files/CAG2026-7-ProgramStatement.pdf  

[12] Spielman, S. E., Folch, D., & Nagle, N. (2014). Patterns and causes of uncertainty in the American Community Survey. Applied Geography, 46, 147–157. https://doi.org/10.1016/j.apgeog.2013.11.002 

[13] U.S. Census Bureau. (2017). Documentation for the 2013-2017 variance replicate estimates tables. https://www2.census.gov/programs-surveys/acs/replicate_estimates/2017/documentation/5-year/2013-2017_Variance_Replicate_Tables_Documentation.pdf 

[14] Murray, A. T., Carvalho, L., Church, R. L., Jones, C., Roberts, D., Xu, J., Zigner, K., & Nash, D. (2021). Coastal Vulnerability under Extreme Weather. Appl. Spat. Anal. Policy, 14(3), 497–523. https://doi.org/10.1007/s12061-020-09357-0  

[15] Park, I., Fauss, K., & Moritz, M. A. (2022). Forecasting Live Fuel Moisture of Adenostema fasciculatum and Its Relationship to Regional Wildfire Dynamics across Southern California Shrublands. Fire, 5(4), Article 4. https://doi.org/10.3390/fire5040110 

[16] Storey, E. A., Stow, D. A., Roberts, D. A., O’Leary, J. F., & Davis, F. W. (2021). Evaluating Drought Impact on Postfire Recovery of Chaparral Across Southern California. Ecosystems, 24(4), 806–824. https://doi.org/10.1007/s10021-020-00551-2 

[17] CAL FIRE. 2025. Statistics–CAL FIRE. https://www.fire.ca.gov/our-impact/statistics  

[18] Peterson, S., D. Roberts, and P. Dennison. 2008. Mapping live fuel moisture with MODIS data: A multiple regression approach. Remote Sensing of Environment112 (12):4272–84. doi: 10.1016/j.rse.2008.07.012. 

[19] De Fries, C., C. Melton, R. Smith, L. Reyes Mason. (2022). The Impacts of Wildfires on Older Adults: A Scoping Review. Innovation in Aging, 6:Supplement 1. https://doi.org/10.1093/geroni/igac059.2307 

[20] National Interagency Fire Center. 2025. National Interagency Fire Center. https://data-nifc.opendata.arcgis.com