By Mary Becica, in support of Climate Justice Coalition for California Prisons
Executive summary
Extreme heat is a documented health hazard inside California’s prisons, but the legislature continues to avoid funding any interventions. To better address facilities planning and costs, a more detailed analysis of how heat is amplified and sustained per prison is necessary. This essay asks whether the public data on location, design and construction can explain how hot a facility gets inside vs outside, and finds that the currently available data is insufficient. A regression and clustering analysis confirms that older prisons tend to amplify heat more, and that the cooling system governs how much outdoor heat becomes indoor heat. Analyzed within each cooling group, though, temperature inconsistencies between otherwise similar prisons suggest that facility-level conditions the public data cannot capture are shaping indoor heat.
CDCR should implement continuous indoor and outdoor temperature monitoring from automated sensors and publish the readings. Monitoring is a safety tool, and an inexpensive investment compared to construction. Air conditioning added to an uninsulated, poorly maintained building costs more to run and maintain than the same equipment in an efficient one; measuring how each building actually performs should be the first step before committing to a cooling plan, and published indoor data would show where installing mechanical air conditioning alone would be insufficient.
Introduction
California sited and designed its prisons to make geographic isolation and extreme weather a tool for punishment. During the building boom of 1984–2000, the state built 23 prisons on idle rural land, concentrating its facilities in the Central Valley and inland deserts, the regions with the greatest heat hazard; CDCR has acknowledged that during this expansion, “the comfort level of the incarcerated population and staff was not a consideration or priority”. In environmental risk terms, exposure is the intersection of a hazard with human populations and infrastructure. For California’s prisons, heat exposure is defined by this history of location, design and construction.
Prison buildings and policies concentrate heat’s health harms on the people confined inside, and prison sites carry environmental hazards of their own. California’s prisons are uninsulated concrete that transmits radiant heat directly into cells and holds it overnight, with a repair budget that does not cover the deferred maintenance backlog. Housing unit buildings are cooled three ways. Refrigerated air conditioning cools the air mechanically. Evaporative (“swamp”) cooling passes air over water and cools it by evaporation; its efficiency decreases as humidity rises. Ventilation is not cooling at all; fans and vents move outside air through a building without lowering its temperature.
CDCR has admitted that its cooling infrastructure has degraded, but no analysis has measured what increases and sustains indoor heat at each of California’s 31 prisons. Without that measurement, the cost of a remedy cannot be calculated: air conditioning added to a poorly insulated concrete building costs more to run and maintain than the same equipment in an efficient building, and making cooling feasible there means first paying down years of deferred maintenance. California Penal Code Section 2067 already directs CDCR to weigh operational costs and facility condition, including the need for significant repairs and upgrades, when selecting prisons for closure; a full accounting of cooling costs may show that closing the most heat-burdened facilities is more financially rational than upgrading and cooling them.
To measure heat exposure in California’s prisons, this analysis compares outdoor temperature records against the one indoor record CDCR has published, covering May through October 2025 (an earlier essay documents what heat data CDCR does and does not publish).
Findings
Across all 31 prisons, building age is the only significant predictor of indoor heat
When all 31 prisons are analyzed together, only building age has a statistically significant relationship to how much a facility amplifies outdoor heat: older prisons have hotter indoor temperatures than expected compared to newer prisons. The data publicly available for this analysis only explained about a fifth of the temperature differential between facilities. To better measure the differences in heat exposure, the facilities are instead grouped by cooling type and examined one group at a time.
Figure 1. California’s 31 state prisons, by cooling type
Indoor heat by cooling group
The 31 facilities split into 5 majority-refrigerated, 15 majority-evaporative, and 11 majority-ventilation (herein the refrigerated, evaporative, and ventilation-only groups). The refrigerated group holds the newest prisons, built 1989 on average, in the hottest deserts; the evaporative group, built 1982 on average, sits in the hot Central Valley and desert; and the ventilation-only group holds the oldest, built 1959 on average, in mild coastal and inland climates. The refrigerated group logs the fewest indoor days above 78°F (63 on average, out of 184 days from May through October 2025), the evaporative group 99, and the ventilation-only group the most (101) (Table 1).
Table 1. California’s prisons by cooling type, 2025
Refrigerated
The five desert facilities with majority-refrigerated cooling averaged 139 outdoor days above 90°F in 2025 but only 63 indoor days above 78°F, and the indoor–outdoor correlation is indistinguishable from zero (r = −0.31, p = 0.61). Refrigerant cooling has largely decoupled indoor conditions from outdoor heat; this is the expected result when air conditioning works as it should. Calipatria is the exception: it reported 137 indoor days above 78°F despite near-total AC coverage. The published data offers no explanation: it could be equipment failure, systems overloaded during heat waves, a few hot buildings, or indoor temperatures set out of regulation.
Evaporative
Outdoor temperature is a significant but only partial predictor (r = +0.55, p = 0.036; r = +0.52, p = 0.047 against outdoor days above 90°F) in facilities with majority-evaporative cooling; it accounts for about a third of the variation in indoor days, and the rest does not follow outdoor conditions. Facilities with nearly identical climates diverge sharply: Mule Creek logged 11 indoor days above 78°F against 138 outdoor, while Avenal logged 132 against 162; both opened in 1987, so building age does not account for the gap. High Desert has the coldest nights in the group and still logged 89 indoor days above 78°F against 128 outdoor, though cool nights should help a building release the day’s heat. All use evaporative cooling in comparable outdoor heat, so the gap between them comes down to their buildings or their equipment; the available data is insufficient to explain which, or whether these facilities need repaired cooling, a different system, or work on the building itself.
Ventilation
Indoor days above 78°F rise with outdoor days in the majority-ventilation facilities (r = +0.75, p = 0.008). With only airflow to work against the heat, these buildings follow the weather. But because they sit in the mildest climates, outdoor temperature rarely reaches the 90°F Stage I threshold that triggers CDCR’s heat response; Pelican Bay, San Quentin, the California Men’s Colony, and the Correctional Training Facility almost never cross it, with an average of about one outdoor day above 90°F from May through October 2025.
Conclusion
The public data on location, design and construction is not enough to explain what makes each prison hot inside. A count of days above a single threshold, reported once a year, cannot show how long a housing unit stayed hot, whether it cooled overnight, or whether its cooling equipment held through a heat wave. Continuous indoor and outdoor temperature monitoring from automated sensors, published as it is collected, would show each of those, and would make clear which facilities need repaired equipment, different cooling, or work on the buildings themselves.
Those distinctions carry costs. Air conditioning added to an uninsulated, poorly maintained building costs more to run and maintain than the same equipment in an efficient one. Any accounting of what it would cost to cool California’s prisons, or of which facilities are too costly to cool at all, has to start from continuous indoor data.
Methodology
Determinants. To identify which characteristics drive indoor heat, each facility’s indoor/outdoor ratio (indoor days above 78°F divided by outdoor days above 78°F) was regressed on each candidate building and site characteristic individually, and combined models were validated with leave-one-out cross-validation.
Cooling-type groups. Facilities were grouped with k-means clustering (k = 3). Three clusters was the most stable choice: the grouping held across resampling (adjusted Rand index ≈ 0.85) and reproduced under an alternative distance-based method, and it matches the three distinct cooling systems. The inputs were each facility’s dominant cooling system (refrigerated, evaporative, or ventilation-only) as a categorical variable, plus seven climate and site measures: building age, outdoor days above 78°F, overnight low temperature, census-tract hot-nights percentage, normalized urban heat island intensity, elevation, and latitude. The continuous measures were Yeo-Johnson transformed to correct skew and then standardized before clustering. Indoor temperature was not used as a clustering input.
Correlations. Pearson r was computed within each cooling group and system-wide, between indoor days above 78°F and outdoor days above each threshold.
Data
| Data | Source |
|---|---|
| Indoor temperature (days each housing unit type exceeded 78°F, May–October 2025) | CDCR Air Cooling Pilot Program Supplemental Report, January 2026, Table 1 |
| Outdoor temperature (daily maximum and minimum, extracted at each facility’s coordinates for the same window) | gridMET |
| Cooling-system type | CDCR facility records and the Air Cooling report’s Table 2 |
| Urban heat island intensity | Benz & Burney (2021) |
| Hot-nights percentage | CalEnviroScreen 4.0 |
| Elevation | USGS National Elevation Dataset |
| Roof and envelope project records | CDCR Master Plan Annual Reports, LAO budget analyses, and CDCR’s roof replacement program records |
Analysis code and the full facilities data table are in the California prison climate justice repository.
Cooling-system shares. Each facility’s cooling-system shares reflect the fraction of its housing buildings with each system type; the dominant system is the one covering the most buildings. Group medians near 100% show that most facilities run almost entirely on one system, while a few mixed facilities (as low as 53%) pull the means down, so no group is uniform.
| Group | Facilities | Share range | Median | Mean |
|---|---|---|---|---|
| Refrigerated | 5 | 53–100% | 100% | 91% |
| Evaporative | 15 | 64–100% | 96% | 93% |
| Ventilation | 11 | 57–100% | 90% | 84% |
Outlier. Pelican Bay (indoor/outdoor ratio 15.8, from just 4 outdoor days above 78°F against 63 indoor) was excluded from the ratio models as an outlier, leaving n = 30.
Results
Determinants. Building age was the strongest predictor (r = −0.53, R² = 0.28, p = 0.003); refrigerated-AC share was second and weaker (R² = 0.11, p = 0.08); all other characteristics explained ≤ 0.07 individually and none was significant. With n = 30, only large effects can reach significance; non-significance here does not rule out real physical effects. The best combined model reached a cross-validated R² of about 0.17, and adding all candidate variables lowered out-of-sample performance.
Table 2. Building characteristics vs. how much a facility amplifies outdoor heat
| Characteristic | R² | p |
|---|---|---|
| Building age (year opened) | 0.28 | 0.003 |
| Refrigerated-AC share | 0.11 | 0.08 |
| Evaporative-cooling share | 0.07 | 0.16 |
| Overnight low temperature | 0.07 | 0.16 |
| Hot nights · UHI · elevation · latitude · roof/envelope work | ≤ 0.03 | n.s. |
Roof and envelope work. Completed roof and building-envelope projects show no association with indoor outcomes (R² ≈ 0.02, not significant). Ten of the 31 facilities completed such work between 2017 and early 2026 and six more have active projects, but they span nearly the full range of indoor days; Mule Creek (11 indoor days) and Avenal (132) had no envelope work in this window, so roofing cannot account for the gap between them. Records before 2017 barely exist: earlier roof work was funded from a pooled repair appropriation with no facility-level itemization, and much of it was performed by incarcerated day labor with no public procurement record. Smaller envelope measures such as shade structures, insulation, or reflective coatings have no public record.
Correlations. Indoor heat tracks outdoor heat significantly in the ventilation-only and evaporative groups and not in the refrigerated group. The system-wide correlation mixes all three groups and describes none of them. Group sizes of 5–15 facilities are small; the refrigerated-group values (n = 5) are descriptive only.
Table 3. Indoor–outdoor correlations by cooling type, 2025
| Group | Indoor 78°F days vs. outdoor 78°F days | vs. outdoor 90°F days |
|---|---|---|
| All 31 facilities | r = +0.32 (p = 0.08) | r = +0.08 (p = 0.68) |
| Ventilation | r = +0.75 (p = 0.008) | r = +0.49 (p = 0.12) |
| Refrigerated (AC) | r = −0.31 (p = 0.61) | r = −0.19 (p = 0.76) |
| Evaporative | r = +0.55 (p = 0.036) | r = +0.52 (p = 0.047) |
Figure 2. Indoor vs. outdoor days above 78°F, by cooling type
indoor_outdoor_heat_2025.csv, California prison climate justice repository.Residual variation in the evaporative group. Outdoor temperature accounts for about a third of the variation in indoor days within the evaporative group. The residual (what outdoor temperature leaves unexplained) shows no relationship to overnight low temperature (r = 0.08) or to urban heat island intensity (r = 0.34), so neither outdoor overnight warming nor heat-island exposure accounts for the remaining spread.
Limitations
The indoor figures are facility self-reports of a single threshold, not sensor records; they cannot distinguish a unit that briefly passed 78°F from one that held high temperatures for hours, and they carry the same reporting inconsistencies the OIG’s 2025 audit documented in CDCR’s manual heat logs. All temperature variables come from 2025 alone; an unusually hot or mild summer at a given facility could shift its indoor count and its group. Urban heat island and hot-nights values are census-tract proxies that may poorly represent a specific facility parcel. Cooling shares count buildings, not capacity or condition; a facility coded predominantly evaporative may be running undersized or failing equipment. This analysis covers the built environment only; deciding where monitoring or cooling matters most also requires facility-level counts of heat-vulnerable people, which CDCR does not publish.