What this research found
Given a fixed budget of 1,000 cooling and weatherization program slots, which Richmond, Virginia neighbourhoods should receive them? A 2021 street-level heat-mapping campaign was combined with Department of Energy modelled energy-burden estimates across 126 census tracts to build three competing allocation strategies — heat exposure only, energy burden only, and an equal-weight blend — with measurement uncertainty propagated through 1,000 Monte Carlo trials. Targeting on heat and targeting on energy burden agree on fewer than half their selections, and only 2 of the 27 blended selections survive every trial.
- Heat and energy burden point to different neighbourhoods. Selections driven by heat exposure alone overlap those driven by energy burden alone with a Jaccard index of just 0.47 (Pearson correlation 0.61 on slot counts), while heat-only and the blended index agree closely at Jaccard 0.83 and Pearson 0.89.
- Selection is fragile under input uncertainty. Across 1,000 trials only 2 tracts are chosen with probability of at least 0.999, while 64 are marginal, meaning most of the 27 blended selections hinge on values within their error bars.
- The three strategies spread money differently: the heat index funds 28 tracts, the energy index 22, and the blend 27, with per-tract awards ranging from 25 to 53, 25 to 93, and 26 to 69 slots respectively.
- Robustness falls off with rank. Among the top 30 tracts by blended allocation the leading ones have tight confidence intervals sitting above the 25-slot floor, whereas mid-ranked tracts have intervals reaching zero and are selected only conditionally.
- Ambient heat varies measurably but modestly across the mapped footprint: tract-mean afternoon air temperature spans 89.5 to 92.5 °F, afternoon heat index spans 91.9 to 98.5 °F, and evening air temperature spans 85.7 to 90.0 °F.
How it was done
Six 10-metre heat rasters from the 2021 NIHHIS-CAPA community heat-mapping campaign in Richmond were reprojected to UTM zone 18N and reduced to per-tract means and standard deviations for afternoon air temperature, afternoon heat index, and evening air temperature, leaving 127 tracts with valid coverage and a 126-tract allocation pool. Department of Energy Low-Income Energy Affordability Data for Virginia was aggregated per tract with a ratio-of-sums burden estimator restricted to households below 80% of area median income. Three min-max normalized indices were built — a heat mean, a multiplicative energy index combining burden intensity with household count, and an equal-weight blend — and 1,000 slots were allocated in proportion to index times low-income households under a floor of 25 and a ceiling of 200 per tract. Uncertainty was propagated through 1,000 trials perturbing temperatures by their standard error and the energy variables log-normally, and the whole framework was written up as a 20-page paper with 6 figures and 39 references.
Data sources
- NIHHIS-CAPA Richmond community heat-mapping campaign, 2021 — six 10 m rasters of air temperature and heat index (OSF project 3xvmg)
- US Department of Energy Low-Income Energy Affordability Data, Virginia 2022 — 612,705 rows across 2,163 census tracts
- US Census Bureau 2022 TIGER/Line census tracts for Virginia — 2,198 tracts, 224 intersecting the Richmond footprint
- 39 verified references on heat mortality, urban heat islands, energy burden, and environmental justice indices
Limitations
The heat surface is a single campaign-day snapshot of ambient conditions rather than a measure of household heat risk, which also depends on building envelope, air-conditioning access, and occupant health. Energy burden is modelled rather than metered, census tract boundaries impose a modifiable areal unit problem, and the allocation bounds and index weights are policy choices rather than empirical quantities.
Figures from this analysis
How this research was produced
K-Dense Web planned and ran this environment 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.
