What this research found
Where should a heat-resilience research programme spend its money? All 3,109 contiguous-US counties were scored on compound heat exposure, chronic-disease prevalence, and adult population, then ranked to recommend 25 counties for an illustrative $10 million portfolio. A 1,000-iteration Monte Carlo simulation that re-ran the entire ranking under climate and prevalence uncertainty found the shortlist stable, and comparison against single-axis rankings showed the joint index picks out counties that neither heat nor health alone would surface.
- The 25 recommended counties are led by Harris (Texas), Maricopa (Arizona), Dallas (Texas), Los Angeles (California), and Bexar (Texas). Twenty-one sit in Sun Belt states, with 6 in Texas and 5 each in California and Florida, and together they contain 43,621,450 adults, roughly one-sixth of the contiguous-US adult population.
- The shortlist holds up under uncertainty. Across 1,000 iterations that bootstrap the 30-year climate record and redraw each disease prevalence from its 95% confidence interval, 24 of 25 counties retain better than a 0.7 probability of staying in the top 25, mean 0.95. The exception is Tulsa, Oklahoma, ranked 25th with a 90% interval of 22 to 29 and a probability of 0.41.
- No single dimension reproduces the joint ranking. Jaccard overlap of the top-25 set is 0.00 against a health-only intensity ranking, 0.09 against heat-only intensity, 0.39 against population-weighted heat, and 0.61 against population-weighted health.
- Compound heat days, defined as a calendar day reaching at least 35 °C by day and staying at or above 20 °C at night, average 8.13 per county-year nationally and range from 0 to 148. Imperial County, California leads at 120.6 per year, followed by Maricopa, Arizona at 114.5, while counties across northern Montana, Michigan, and New York record none.
- Splitting the budget in proportion to composite risk hands Harris County $1.47M and Tulsa $0.21M, with the top five counties absorbing 39% of the $10 million total.
How it was done
Daily maximum and minimum temperatures for 1991 to 2020 were pulled from the Daymet single-pixel service at each county centroid, and every county-year was reduced to a count of compound heat days. Chronic-disease burden came from CDC PLACES 2024 as the mean national percentile across five age-adjusted conditions: coronary heart disease, stroke, chronic obstructive pulmonary disease, diabetes, and disability. Multiplying the heat percentile by the health percentile by adult population gives a composite score that reads as hazard times sensitivity times scale. The whole ranking was then recomputed 1,000 times, bootstrapping the 30 years of climate and drawing prevalence from the published confidence intervals, to produce rank intervals and top-25 probabilities. Funding was allocated in proportion to composite score, and the shortlist was benchmarked against four single-axis rankings in a 19-page manuscript with 7 figures and 33 references.
Data sources
- Daymet V4 R1 daily temperature via the ORNL single-pixel API, 1991-2020 (93,270 county-year records, DOI 10.3334/ORNLDAAC/2129)
- CDC PLACES 2024 county data for 3,109 contiguous-US counties across 48 states and DC, with 95% confidence intervals on five conditions
- US Census cartographic boundary file cb_2024_us_county_20m, the vintage needed to match Connecticut's nine planning regions
Limitations
A single Daymet pixel at the county centroid stands in for the whole county, whether that is a dense metro or a large rural expanse, and three coastal centroids fell on water and had to be shifted to the nearest land pixel. Because the composite multiplies by adult population by design, large metropolitan counties dominate the list, which is a prioritisation exercise rather than an evaluated funding decision.
Figures from this analysis
How this research was produced
K-Dense Web planned and ran this public health investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
