In July 1995, a heat wave killed more than 700 people in Chicago in less than a week. Investigators later mapped where the deaths had clustered. The pattern bore no relationship to the weather itself, which had been equally severe across the city. It followed the contours of neighborhoods that had been systematically divested for decades: predominantly Black, deeply poor, cut off from green space and social networks, housed in buildings that became ovens when the temperature climbed. The weather event was citywide. The dying was not. What separated the living from the dead was not luck. It was decades of policy choices made long before anyone checked the forecast.
New York City faces the same reckoning. Extreme heat is now the deadliest weather-related hazard in the city, killing more people each year than hurricanes, floods, and blizzards combined. Yet most of that danger is invisible in the data systems cities use to plan for it. Heat maps show where temperatures are high. They do not show who can actually do anything about it. A cooling desert is a neighborhood where the heat is severe and the capacity to escape it is structurally blocked: by a rent bill that leaves nothing for electricity, by a building too old to support air conditioning, by a language barrier that puts emergency information out of reach, by the absence of any shaded or cooled public space within walking distance. This project maps all 557 of them across New York City's five boroughs, and asks why they exist where they do.
The geography of heat risk in American cities is not accidental. It is the physical inheritance of decisions made at every level of government over a century. Federal housing policy from the 1930s through the 1960s explicitly directed public investment away from Black and immigrant neighborhoods, a process Richard Rothstein documents in meticulous detail in The Color of Law. Highways were routed through these communities. Trees were planted elsewhere. Buildings were left to deteriorate. The result was not a collection of unfortunate circumstances but a structured environment in which certain bodies, in certain places, would bear the full weight of heat without the tools to survive it. As sociologist Eric Klinenberg showed in his study of the 1995 Chicago heat wave, the city that killed hundreds of people that summer had been building toward that outcome for decades. New York City is no different.
What follows is an attempt to make that structure visible. Each section of this project traces a different layer of the same problem: where heat danger is highest, which households cannot afford to cool their homes, how race operates as an independent risk factor beyond income, where public cooling infrastructure fails to reach the people who need it most, and what targeted policy interventions could realistically accomplish. None of it is inevitable. All of it is a choice.
The same crisis. Two ways to understand it.
Where a Hot Day Becomes Dangerous
Every summer, the same weather passes over every borough. It does not land the same way.
Heat risk does not arrive fully formed. It accumulates. In some neighborhoods, decades of choices about where to plant trees, how to maintain buildings, and which households can absorb the cost of cooling have quietly compounded the odds against residents. The six panels below trace that accumulation layer by layer: from the official heat danger ratings, to the structural barriers that lock residents out of cooling, to the geographic concentration that leaves entire communities without any nearby refuge.
2,231 Neighborhoods. One City. Very Different Summers.
New York City is made up of 2,231 small neighborhoods, each home to roughly 1,000 to 4,000 people. That scale matters. Heat risk in New York is not uniform: it is hyper-local, shaped by tree cover, building age, population density, and the financial capacity of the people living there. A city-level average obscures the places where real danger concentrates. To understand where renters are most exposed to heat and least equipped to handle it, the city must be examined one neighborhood at a time.
Not Every Neighborhood Feels the Same Summer
New York City's Department of Health rates every neighborhood on a heat danger scale of 1 to 5. The score combines how hot temperatures get, how much shade and green space exists, how many households have air conditioning, and how much poverty there is. What that scale reveals is stark. The neighborhoods already bearing the weight of poverty and disinvestment are the same ones where a summer heat wave carries the greatest risk of death. In the most at-risk neighborhoods, residents are three times more likely to die from heat than those living in the safest ones.
557 Neighborhoods Without a Way Out of the Heat
When high heat danger and structural barriers to cooling overlap in the same neighborhood, the result is a cooling desert. In 557 neighborhoods across New York City, residents face elevated heat danger and one or more compounding constraints: extreme rent burden that makes electricity unaffordable, overcrowded housing that makes cooling impossible, language barriers that cut off access to emergency resources, or buildings so old their wiring cannot support air conditioning. These are not neighborhoods where people need more information. They are neighborhoods where the conditions that make cooling possible are simply absent.
Heat Does Not Fall Equally Across Race
When researchers control for income, green space, and poverty, one variable continues to predict higher heat danger independently: the racial composition of a neighborhood. For every 1% increase in Black residents, surface temperatures run 0.43 degrees Fahrenheit hotter. In Manhattan, that effect grows to 5.27 degrees. This is not a coincidence. It is the spatial legacy of a century of disinvestment, highway routing, urban renewal, and the systematic exclusion of Black communities from neighborhoods with parks, trees, and maintained housing stock.
In the Bronx, There Is No Safer Street to Walk To
Cooling deserts do not scatter randomly across a borough. They cluster. In the Bronx, two out of three high-risk neighborhoods are completely surrounded by other high-risk neighborhoods. There is no adjacent street, no nearby park, no walkable location offering meaningful relief. This geographic concentration is not incidental: it reflects the same long-term disinvestment patterns that created heat risk in the first place, now locked into a spatial structure that compounds itself.
Public Cooling Infrastructure Misses the Neighborhoods That Need It Most
The city maintains 218 outdoor cooling features in parks: spray showers, misting stations, and fountains. But the distribution of those resources does not align with the distribution of heat danger. Many of the neighborhoods most at risk have no outdoor cooling feature within reasonable walking distance. And the indoor emergency cooling centers that do exist are closed on Sundays, which is the day heat-related deaths peak. A system designed for emergencies, operating on weekday hours, is not a system designed for the people who need it.
Navigate NYC's Heat Landscape
All 2,231 New York City neighborhoods, each scored across multiple dimensions of heat vulnerability. Choose a layer to shift the lens: composite risk score, cooling desert status, spatial clustering patterns, neighborhood type, official heat danger rating, or average summer surface temperature. Select a borough to zoom in, or click any neighborhood to view its full profile.
Five Ways a Neighborhood Gets Left Behind
Not every cooling desert faces the same barrier. Some neighborhoods are financially squeezed. Others carry decades of racial disinvestment. Some are cut off by language. Identifying the specific shape of each neighborhood's constraints is the precondition for targeted policy rather than broad gestures.
If 557 neighborhoods lack the capacity to cool their residents, what is actually holding each one back?
A neighborhood where rent consumes every dollar needs rent relief. One where residents cannot read emergency instructions in English needs multilingual outreach. One that runs measurably hotter because of decades of racial disinvestment needs environmental justice investment. The radar charts below show the structural profile of each neighborhood type: which risk factors are elevated and which are not. Hover each chart label to see what it measures.
Heat's Unequal Reach
Three patterns emerge from comparing heat risk across New York City's neighborhoods: race operates as an independent driver of heat exposure that income alone cannot explain, the Bronx carries a disproportionate share of the city's cooling deserts, and the racial composition of high-risk neighborhoods reflects a structure, not a coincidence.
Race is the strongest predictor of heat risk, independent of income
A statistical model tested which neighborhood characteristics most strongly predict heat danger, controlling for income, language access, overcrowding, and other variables. Race emerged as the dominant predictor. The share of Black residents has a standardized effect size more than twice that of household income, meaning that two neighborhoods with the same income level but different racial compositions will have measurably different heat risk. This is not explained by poverty. It reflects the spatial legacy of racially discriminatory housing policy: neighborhood disinvestment, lower tree canopy, aging building stock, and reduced access to green space, all of which accumulated in predominantly Black neighborhoods over decades of deliberate exclusion from the resources that make a neighborhood livable in summer.
Race is the single most powerful driver of heat risk, more influential than income, rent costs, or language barriers, even after all other factors are accounted for.
Statistical examination of 2,197 NYC neighborhoods · bars show relative influence as a percentage of the strongest predictor (race, set at 100%) · grey bars did not reach statistical significance
The Bronx bears a burden no other borough comes close to matching
More than half of the Bronx's neighborhoods are classified as cooling deserts. Its median household income is the lowest of any borough at roughly $50,000. These two facts are connected: lower incomes reduce every layer of adaptive capacity, from the ability to pay electricity bills to the ability to relocate to a cooler neighborhood or building. But the Bronx's position at the extreme end of this distribution cannot be explained by income alone. It also reflects the concentrated effects of urban renewal projects, highway construction, and decades of disinvestment that reshaped the borough's physical environment in ways that continue to produce heat today. The bar chart allows comparison across all five boroughs simultaneously on three dimensions: cooling desert share, mean risk score, and median income.
The Bronx concentrates the highest share of cooling deserts (53.8%) and the lowest median income ($50K), a borough-level gap that income alone cannot fully explain.
Heat risk data · U.S. Census 2022 · Hover any bar to highlight that borough across all three charts
As heat risk rises, the racial composition of neighborhoods shifts in one direction
When New York City's neighborhoods are sorted into five groups from lowest to highest heat risk, a clear demographic gradient emerges. In the safest neighborhoods, Black residents make up an average of 4.5% of the population. In the most at-risk neighborhoods, that share rises to 54.1%, a twelvefold increase across just five levels of risk. No other demographic group shows anything close to this pattern. Hispanic, white, and Asian population shares shift modestly or not at all along the same risk gradient. This is not a minor statistical observation. It means that heat danger in New York City is not distributed by chance or solely by poverty. It is distributed along racial lines in a way that reflects over a century of policy choices about where different communities were permitted to live, invest, and remain.
The share of Black residents rises twelvefold, from 4.5% in the safest neighborhoods to 54.1% in the most at-risk ones. No other group shows this pattern.
Average demographic shares across five heat risk levels · U.S. Census 2022 · ~440 neighborhoods per group
Shade as Infrastructure
Trees lower surface temperatures, reduce the urban heat island effect, and provide outdoor cooling that no budget line can replace. New York City gained canopy between 2010 and 2017 - but the distribution of that shade does not follow income. It follows heat risk. The neighborhoods most exposed to dangerous heat carry a persistent shade deficit that no amount of wealth correlation can explain away.
The most vulnerable neighborhoods have the least shade - and are now growing trees fastest
Canopy coverage falls as heat vulnerability rises, from 19.3% in the city's least vulnerable neighborhoods down to 16.7% in the most vulnerable. The gap is 2.6 percentage points - real, but modest. What the growth dots reveal is more significant: the most vulnerable quintile is adding canopy faster than any other group, at 2.31 percentage points since 2010, compared to 1.88 for the least vulnerable. This is evidence that urban greening programs are beginning to reach the right places. Two things remain true at once: the city is investing in shade where it is needed most, and the accumulated deficit is too large for tree planting alone to close.
The most vulnerable neighborhoods have 2.6 percentage points less canopy than the safest, yet are growing trees faster than any other group. The investment is real. The gap has not closed yet.
NYC Urban Tree Canopy Assessment 2010 & 2017 · CDI quintile groupings · Bars show 2017 canopy coverage · Dots show average growth rate since 2010 · Hover for detail
Shade deprivation tracks heat vulnerability - but income does not explain it
Each dot is one of roughly 600 sampled census tracts, plotted by its Cooling Desert Index score and its 2017 tree canopy coverage. Cooling deserts cluster toward the upper right: higher vulnerability scores, lower canopy. The relationship exists but it is noisy. Some high-CDI tracts have substantial shade. Some low-CDI tracts are almost treeless. The most striking finding from the full dataset is what is absent: the correlation between canopy coverage and median household income is essentially zero (r = 0.002). Shade inequality in New York City is not a function of neighborhood wealth. It reflects land use history, block density, and the persistence of environmental disinvestment in specific communities. Tracts with high proportions of limited-English-speaking residents show the strongest negative association with canopy of any demographic variable measured.
Cooling deserts have less shade, but the income-canopy correlation is near zero. Shade deprivation in NYC is not about wealth. It follows language access and heat risk more than any other factor.
~600 randomly sampled census tracts · CDI = Cooling Desert Index (higher = more vulnerable) · Canopy = 2017 coverage · Hover any dot for borough and values
Staten Island leads on canopy; the Bronx leads on cooling deserts
Tree coverage varies dramatically across boroughs. Staten Island, with the lowest density and most undeveloped land, carries 25.9% canopy and only 7.4% cooling deserts. Manhattan is the densest borough and has just 14.7% canopy - yet only 9.3% of its tracts are cooling deserts, because many high-income Manhattan neighborhoods have the financial and structural resources to compensate. The Bronx tells a different story: 18% canopy, near the city average, yet 53.8% of its tracts qualify as cooling deserts - the highest share of any borough. Queens had the most individual tracts lose canopy between 2010 and 2017, 149 in total, while Brooklyn saw the most severe losses by magnitude. Trees are necessary. They are not sufficient.
The Bronx has near-average canopy but the city's highest cooling desert rate, at 53.8%. Brooklyn and Queens account for the bulk of tracts that lost canopy. Shade and heat safety do not move together automatically.
NYC Urban Tree Canopy Assessment 2017 · CDI borough aggregates · Green bar = canopy coverage, orange bar = cooling desert share · Hover bars for detail
What If Policy Actually Intervened?
Identifying where cooling deserts exist is not itself a solution. What matters is whether targeted investments can move neighborhoods out of that category, and at what scale. This section tests four realistic policy interventions against the city's 557 cooling deserts, modeling how many neighborhoods would cross the threshold to safety under each scenario.
Each scenario is grounded in existing policy mechanisms. Adding public cooling sites addresses the outdoor cooling access gap in neighborhoods currently uncovered. AC retrofits and urban greening directly lower heat danger levels in the most exposed neighborhoods. Energy assistance for NYCHA residents removes the financial barrier that prevents one in three public housing tenants from using AC at all. Rent relief addresses the broadest financial constraint, the rent burden that leaves households with nothing for electricity.
The key insight from this exercise is not which single intervention works best. It is what happens when all four are combined. Even with every intervention applied simultaneously, 257 neighborhoods remain cooling deserts. These are the most structurally entrenched places in the city: neighborhoods where heat danger is severe, incomes are extremely low, and rent burden is so high that even meaningful policy improvements are insufficient to cross the threshold alone. They represent the irreducible core of the problem, the neighborhoods that require sustained, multi-year, multi-agency investment rather than single policy fixes.
Impact on 557 cooling deserts
Even with all four interventions combined, 257 neighborhoods remain as cooling deserts. These are the places where heat danger is highest, poverty is most severe, and rent burden is most extreme. They are beyond the reach of any single policy lever. They require comprehensive, sustained investment.
Risk That Clusters and Doesn't Let Go
Cooling deserts are not scattered randomly across the five boroughs. High-risk neighborhoods tend to be surrounded by other high-risk neighborhoods, forming continuous zones where there is no adjacent lower-risk area to escape to. Two charts measure the scale of that clustering.
A neighborhood's risk score closely predicts the risk score of every neighborhood around it
Each dot in this chart represents one neighborhood. Its position on the horizontal axis shows its own heat risk score. Its position on the vertical axis shows the average heat risk score of all surrounding neighborhoods. If risk were distributed randomly, the dots would form a flat cloud. Instead, they form a strong upward slope: neighborhoods at high risk are almost universally surrounded by other high-risk neighborhoods. Neighborhoods at low risk are almost universally surrounded by other low-risk neighborhoods. That clustering can be measured. On a scale where 0 means risk is distributed randomly and 1 means every neighborhood perfectly predicts its neighbors, heat risk in New York City scores 0.867. That is one of the highest scores recorded for any urban variable in the city's data. Higher than how income clusters across the city. Higher than how race clusters. Once a neighborhood is in a danger zone, the neighborhoods around it almost certainly are too. Heat risk in New York City is spatially locked in.
The strong upward slope confirms that high-risk neighborhoods are almost always surrounded by other high-risk neighborhoods. This is one of the strongest geographic clustering patterns in the city's data.
Neighborhood clustering chart · x-axis: neighborhood heat risk score · y-axis: average risk score of surrounding neighborhoods · 2,122 neighborhoods plotted
The Bronx is nearly one continuous high-risk zone
This map shows which neighborhoods form statistically significant clusters of high risk (red) and low risk (blue). A high-risk cluster means a neighborhood is not just at risk itself but is surrounded by other at-risk neighborhoods, amplifying the problem: residents cannot find relief nearby, emergency responders are stretched across an entire zone rather than a single block, and the resources that might reduce heat danger, parks, trees, well-maintained buildings, are absent across an entire area rather than just one street. In the Bronx, 66% of all neighborhoods sit inside such a zone. Brooklyn shows a significant cluster in Brownsville and East New York. Staten Island and parts of eastern Queens form the city's largest low-risk cluster, where proximity to cool, well-resourced neighborhoods is protective in itself.
66% of Bronx neighborhoods are embedded in a continuous high-risk zone, not isolated pockets. Brooklyn follows, with 29% of its neighborhoods in the same situation.
Neighborhood risk clustering map · only statistically significant clusters shown · High-risk clusters: 582 neighborhoods · Low-risk clusters: 516 neighborhoods
Behind the Cooling Desert Index
The Cooling Desert Index combines seven census-tract-level variables into a single score, each selected through principal component analysis and validated against the binary cooling desert classification. This section explains what went into the index, how it was tested, where the data came from, and what it cannot yet measure.
01 How the Cooling Desert Index Was Built
The Cooling Desert Index (CDI) is a composite score combining seven census-tract-level variables into a single continuous measure of structural heat vulnerability. Each variable is min-max normalized to 0–100 and multiplied by its assigned weight. The CDI ranges from 2.9 (lowest risk) to 72.0 (highest risk), with a citywide mean of 39.4.
Variables were selected using principal component analysis (PCA). Three components were retained explaining 64.4% of total variance. Variables with redundant axes (poverty and income) or non-significant regression coefficients (Hispanic ethnicity, elderly share) were excluded.
| Variable | Weight | Direction | Role in index |
|---|---|---|---|
| HVI Rank (Heat Vulnerability Index, NYC Dept. of Health, 1 to 5 scale) | 25% | ↑ Higher = more risk | Primary heat exposure |
| Rent burden ≥50% | 20% | ↑ | Financial cooling constraint |
| Median income | 20% | ↓ Inverted | Economic adaptive capacity |
| % Black residents | 15% | ↑ | Race as independent risk factor |
| % Limited English | 8% | ↑ | Communication access barrier |
| % Disability | 7% | ↑ | Mobility and physiological risk |
| % Overcrowded renter | 5% | ↑ | Physical cooling constraint |
02 Statistical Methods
- Linear Regression on Heat Vulnerability Index (OLS, Ordinary Least Squares)
- Three nested models tested whether race predicts heat vulnerability independently of income. Model 3 (full specification with borough fixed effects) achieved R² = 0.567, a +27.7 percentage-point gain over income alone. The standardized β for % Black (0.726) was the strongest predictor in the model. n = 2,197 tracts.
- Principal Component Analysis
- PCA on 10 candidate variables identified 3 components explaining 64.4% of total variance. Component 1 captures economic deprivation (income, poverty, rent burden). Component 2 captures racial heat exposure. Component 3 captures language and overcrowding barriers. Variables with eigenvalues < 1 or theoretical redundancy were excluded from the CDI.
- K-Means Clustering (k = 5)
- Tracts were clustered on 7 normalized CDI variables. The optimal k was determined by elbow method on within-cluster sum of squares and silhouette scores. Five typologies emerged with distinct risk profiles: Low Risk, Financially Strained, Immigrant Heat Burden, Racial Heat Burden, and Compound Deprivation.
- Spatial Autocorrelation (Moran's I)
- Global Moran's I = 0.867 (z = 67.9, p < 0.001, 999 permutations) using PySAL queen contiguity weights. Local Indicators of Spatial Association (LISA) identified 582 HH hot-spots and 516 LL cold-spots at p < 0.05. This is the highest Moran's I of any variable tested.
All scripts are reproducible Python (pandas, geopandas, scikit-learn, PySAL, statsmodels). Source code is available in the analysis/ directory of this repository.
03 Data Sources
| Dataset | Source | Vintage | Known limitation |
|---|---|---|---|
| Heat Vulnerability Index | NYC DOHMH | 2022 | NTA-level only; counts AC ownership, not usage |
| American Community Survey | U.S. Census Bureau | 5-Year 2018–2022 | 5-year pooled estimates; margins of error in small tracts |
| NYC Housing Vacancy Survey | NYC Dept of City Planning | 2021 | Borough-level microdata only; not tract-disaggregatable |
| Cool It! NYC Sites | NYC Parks Dept | 2024 | Outdoor cooling features only; no indoor emergency centers |
| NYCHA Developments | NYC Housing Authority / NYC Open Data | 2024 | No building-level AC infrastructure data available |
| Census Tract Boundaries | U.S. Census Bureau TIGER/Line | 2020 | 2020 tract definitions; minor boundary changes from 2010 |
04 Where the Index Has Limits
Three data gaps constrain what this project can currently measure. Future work with these datasets would materially improve the accuracy of the Cooling Desert Index and the policy scenario models.
- Energy burden at the tract level
- The U.S. Department of Energy's LEAD (Low-Income Energy Affordability Data) tool provides energy burden estimates by census tract. This project uses rent burden as a proxy for unaffordable cooling costs, but energy burden data would directly measure whether households can afford to run air conditioning, a more precise measure of the constraint identified in the NYC Comptroller's energy insecurity research.
- Emergency indoor cooling center accessibility
- The current scope includes outdoor public cooling features but lacks geocoded data for the city's indoor emergency cooling centers, including their hours of operation, ADA accessibility status, capacity, and actual utilization during heat events. Without this, the access gap measure captures only outdoor cooling proximity, not the full emergency cooling infrastructure.
- Building-level heat and hot water failures
- NYC's Housing Preservation and Development (HPD) tracks heat and hot water violations at the building level. These violations are concentrated in the same neighborhoods identified as cooling deserts, but the building-level linkage would allow a direct connection between structural landlord failures and individual cooling desert neighborhoods, strengthening the evidence base for enforcement-based interventions.
A Problem Built by Policy. A Solution That Requires It.
The structure of heat risk in New York City is not a product of geography or bad weather. It is the spatial record of a century of decisions: where to invest, where to plant trees, where to allow housing to deteriorate, and whose neighborhoods to protect from the physical consequences of urban poverty. Those decisions are visible in the data. They are also reversible, but only if the policy response matches the scale of the problem that created them.
Every summer in New York City, 1.47 million renters enter the heat without the structural capacity to protect themselves. They do not lack information about the dangers of heat. They lack working air conditioning they can afford to run, buildings that can support it, access to public cooling spaces within walking distance, or emergency communications they can read in their own language. These are not individual failures. They are the accumulated effects of disinvestment, rent extraction, and exclusion operating over decades, concentrated with extraordinary precision in the same neighborhoods that have borne the weight of every other form of structural disadvantage the city has produced.
Even with every policy intervention this project models, including new cooling sites, AC retrofits, NYCHA energy relief, and rent burden reduction, 257 neighborhoods remain cooling deserts. This is not a ceiling on ambition. It is a map of where the structural damage is deepest: neighborhoods where heat danger is most severe, poverty is most extreme, and the path to safety requires sustained multi-agency investment over years, not a single policy fix. The 300 neighborhoods that targeted intervention can reach are evidence that the approach works. The 257 that remain are evidence of how much work those policies still leave undone.
Climate projections are unambiguous: extreme heat will intensify. The neighborhood structures that produce cooling deserts today will not correct themselves as temperatures climb. Each summer without intervention compounds the risk that is already structural. The 557 cooling deserts identified in this analysis are not a prediction. They are the present condition, and the question is not whether to act, but whether the scale of the response will match the scale of the harm.
Make Cooling Infrastructure Universal
Ensure every cooling desert neighborhood has an accessible public cooling site within walking distance. Extend cooling center hours to Sundays and evenings, when heat deaths peak and most facilities are closed. No neighborhood should face a heat emergency without a reachable refuge.
Establish a Legal Right to Cooling for Renters
Create enforceable building-level standards for cooling capacity. Give tenants in non-compliant buildings the right to rent reduction. Remove the NYCHA air conditioning surcharge entirely and expand energy assistance to low-income renters citywide. Owning an air conditioner you cannot afford to run is not cooling access.
Invest in Neighborhoods That Have Run Hotter by Design
Direct urban greening, tree canopy expansion, and building retrofit programs to the neighborhoods where racial disinvestment has produced measurably higher temperatures, not as charity but as repair. The physical environment of these neighborhoods did not deteriorate by accident. It will not recover without deliberate, sustained, and spatially targeted public investment.
Every NYC neighborhood, colored by cooling desert status
What comprehensive policy intervention changes
It is July 15th. Maria pays $1,650 a month for her apartment in the South Bronx - 58 cents of every dollar she earns. She has lived here for eleven years. Tonight it is 94°F and rising.
She reaches for the air conditioner in the window. The apartment has held the day's heat since morning. But there is a problem she has faced every summer for years.
Heat is not just weather. Anything that changes what is possible for people is a force - and this heat acts differently on different bodies in different neighborhoods.
Running the AC costs $80 more on the electricity bill. After rent, she has $237 left for everything. There is no $80 here. The machine sits in the window. The switch stays off.
1 in 5 NYC renters who own an air conditioner say they cannot afford to run it. The appliance exists. The bill prevents its use.
Even if she could afford it, her building was constructed in 1938. The electrical panels were never designed for air conditioning. Running the unit overnight risks tripping the circuits. The landlord has not upgraded the wiring in thirty years.
47% of New York City's housing was built before 1980. The building does not choose to block cooling - but it does.
The city scores her neighborhood 3 out of 5 on its official heat danger index - moderate risk. Cooling resources go to the 4s and 5s. Maria's street does not make the priority list.
The index counts whether households own an AC. It does not count whether they can afford to use it. That single omission shapes where the city sends help.
This is not a string of bad luck. The electricity bill, the old wiring, the city's measurement tool - these forces are connected. Each reinforces the others. Together they form a structure that makes cooling unreachable.
This is a cooling desert.
The framework behind the map
Actor-Network Theory
Developed in the 1980s by sociologists Bruno Latour, Michel Callon, and John Law, Actor-Network Theory makes one foundational claim: outcomes in the world - inequality, risk, exclusion - are not produced by people alone. They are produced by networks of both human and non-human forces working together.
An air conditioner sitting unused in a window is not just a household decision. It is the end result of a chain: a rent burden that consumes most of the household's income, an electricity tariff that makes cooling unaffordable, a building whose wiring cannot safely carry the load, and a government index that marks the neighborhood as moderate risk and routes resources elsewhere. Each of these - the bill, the building, the index - is an actor. Together they form a network. ANT is the framework that makes that network visible.
The 1.47 million renters living in NYC's 557 cooling deserts are not there because of bad choices or bad luck. They are held there by a network assembled over decades through decisions made at every level - by landlords, city agencies, utility companies, federal housing policy, and the data systems that define what risk even looks like. Understanding that network is the first step toward breaking it.
Why does the city's own heat map make some neighborhoods invisible?
ANT concept: Black-boxing + Inscription
The city's Heat Vulnerability Index is not just a number - it is the number that determines where cooling resources go. When the index scores a neighborhood at 3 out of 5, emergency cooling centers are sent elsewhere. The index's design is not neutral: what it counts shapes what gets funded.
The HVI counts whether households own an air conditioner. It does not count whether they can afford to run it. That single omission means that the 21% of renter AC owners who cannot afford to use their units are invisible to the measurement system that is supposed to protect them. And because the HVI is the obligatory passage point - the one tool every resource flows through - its blind spot becomes everyone's blind spot.
What does a building's age have to do with who survives a heat wave?
ANT concept: Non-human actants
47% of New York City's housing was built before 1980. Buildings from that era were not designed to support air conditioning. A window AC unit draws 1,000–1,500 watts. Many pre-1940 buildings still run on electrical systems that cannot safely carry that load without tripping circuits or creating fire hazards.
The building is not making a choice - but its physical structure enforces a decision made decades ago into every summer heat emergency. Statistical analysis confirms this directly: building age is a significant negative predictor of AC access, independent of household income. The building itself is an actor holding the cooling desert in place.
How does a rent bill become a heat weapon?
ANT concept: Translation + Enrollment
When rent consumes more than half a household's income, the arithmetic of survival becomes extremely tight. Severely rent-burdened renters face energy costs that consume more than 10% of income - more than four times the rate of unburdened renters, and well above the DOE's 6% threshold for "high energy burden."
The result is a chain: the rent check translates directly into an electric bill that cannot be paid in full. NYCHA's monthly AC surcharge - around $25 - deters one in three public housing seniors from using cooling at all. What looks like a personal decision not to use the AC is actually the endpoint of a financial chain that the measurement system never traces back to its source.
Why can't a cooling center cool the people who need it most?
ANT concept: Scripts + Network misalignment
New York City maintains emergency indoor cooling centers throughout the five boroughs. But the system was designed with assumptions built in - about who would use it, when they would come, and how they would get there. Those assumptions do not match the actual constraints of the people most at risk.
83% of indoor emergency cooling centers are closed on Sundays - the day heat-related deaths peak in New York City. 47% are age-restricted to seniors. 38% of high-risk, high-rent-burden neighborhoods have no outdoor cooling feature within walking distance. The infrastructure has scripts encoded into it. Those scripts exclude the households the system is supposed to serve.
What would it actually take to break a cooling desert?
ANT concept: Obligatory Passage Points + Network disruption
Scenario modeling tests four interventions: adding cooling sites, AC retrofits and urban greening, NYCHA energy assistance, and rent burden reduction. Applied separately, each removes 12–23% of cooling deserts. Applied all at once, they remove 67% - but 257 neighborhoods remain.
This is the strongest evidence for the ANT framework's core claim. A cooling desert is not caused by one actor. It is held in place by a network of constraints that are mutually reinforcing. Removing one actor leaves the others intact. Only multi-point disruption - interventions that simultaneously address financial constraints, building infrastructure, measurement systems, and access barriers - can destabilize the assemblage that produces the cooling desert.
Mapping the Network
The visualization to the right maps every actor involved in producing a cooling desert: people, objects, and systems, and the connections between them. As you scroll through the panels, the network highlights the specific actors and relationships each concept describes. Hover any shape to see its role in the data.
Not Just People
ANT asks a specific question: which actors, human and non-human, produce this situation? In ANT, anything that changes the outcome for someone counts as an actant. That means heat itself is an actor in this network. So is the electricity bill that stops a household from running AC. So is the pre-1940 building whose wiring cannot carry the load. So is the HVI dataset, which defines risk in a way that leaves affordability out.
Hover any shape to see exactly what role it plays. Drag nodes to rearrange. Each shape connects to others through the lines on screen.
Five People. Five Constrained Positions.
Renters absorb the full heat exposure without owning the building. Landlords control infrastructure and decide whether cooling upgrades happen. City agencies define heat risk through the data tools they choose. NYCHA runs 178,000 units under a separate financial structure. Policymakers direct resources based on what the data shows them.
Each group makes decisions. Each group is also shaped by the network around it. No single actor holds full control, and no single actor bears full blame.
Objects That Do Things
The objects in this network act. The electricity bill stops AC use in 21% of households where owners say they cannot afford to run it. The pre-1940 building blocks installation because its wiring fails under load. The NYCHA AC surcharge rations cooling access in public housing by existing as a monthly line item.
The English-only emergency alert system shuts out 10.5% of NYC renters who are limited English proficient. These are not background conditions. Each one produces a measurable effect you can find in the data.
How Heat Becomes Policy
ANT calls it translation: each actor reshapes another's situation to fit its own goals. Heat becomes a spatial risk score at DOHMH. That score becomes cooling center locations at city agencies. Follow the highlighted path on the left. This is the official chain that turns a heat emergency into a government response.
Rent burden, the financial constraint that stops households from using cooling, never enters this chain. It stays background context. The system responds to what it measures, and it does not measure this.
What the HVI Hides
ANT calls it a black box: something so accepted that nobody looks inside it anymore. The HVI is a black box. It produces a number and that number drives policy. But someone made choices when building it: which variables to include, which scale to use, what counts as protective.
One of those choices: the HVI counts AC ownership, not AC use. When we opened that box using NYCHVS data, we found that 21% of renter AC owners report not being able to afford to run it. The box counted the machine. It missed the gap between having a unit and being able to turn it on.
557 Cooling Deserts. One Network.
ANT calls this an assemblage: a situation produced by multiple actors locking into a pattern that reinforces itself. No single actor creates a cooling desert. It forms when heat exposure, rent burden, aging buildings, absent cooling infrastructure, and language exclusion all hold each other in place at once.
This is why single fixes fall short. All four scenario interventions combined, cooling sites, AC retrofits, NYCHA energy relief, and rent reduction, still leave 257 neighborhoods as cooling deserts. You need to break the pattern at multiple points simultaneously. Scroll down to see what that looks like.
Three Objects That Shape Who Gets to Stay Cool
These are not passive background conditions. Click each one to see exactly how it acts on people's ability to cool their homes.
The Electricity Bill
An air conditioner sitting in the window is not the same as a cool apartment. The bill decides whether the switch gets flipped. For households spending more than half their income on rent, energy costs can consume another 10% - leaving almost nothing left for summer electricity.
This is how a physical appliance that exists in the home becomes effectively inaccessible - not because of ignorance or choice, but because of arithmetic.
The Pre-1940 Building
Buildings constructed before modern electrical codes were not wired for air conditioning. A window AC unit draws roughly 1,000–1,500 watts. Many older NYC apartments cannot safely support that load without tripping circuits or creating fire hazards.
The building is not making a choice - but its physical structure enforces a 1930s decision into today's heat emergency. No amount of financial assistance helps if the wiring cannot carry the load.
The Heat Vulnerability Index
The HVI is not just a number - it is the number that determines where the city sends resources during a heat emergency. When the index says a neighborhood scores 3 out of 5, cooling centers go elsewhere. The index's design is not neutral. What it counts shapes what gets funded. What it leaves out shapes who gets ignored.
The HVI counts whether households own an AC unit. It does not count whether they can afford to run it. That single omission means 21% of AC owners are invisible to the policy system that should be protecting them.
How Heat Becomes Policy - And Who Gets Left Out
Each dot is one renter. As the city responds to heat, watch how many slip through the cracks at each stage - before a single cooling center opens its doors.
Heat Event
A 94°F July night in the South Bronx. Every renter in the city feels the same heat - but not the same risk. No one has been counted yet. No resources have been deployed.
Data Collection
Agencies begin measuring: surface temperature, green space, AC ownership. But rent burden, building age, and language access are never collected. Forty renters vanish from the data before any response begins.
HVI Score
The Heat Vulnerability Index scores the 60 remaining neighborhoods. But 21% of renter AC owners cannot afford to run their units. The HVI counts ownership - not use - scoring 12 neighborhoods as "moderate" risk when they are not.
City Response
Cooling centers open in HVI-priority neighborhoods. But 83% close on Sundays - the peak mortality day. 47% are age-restricted. Emergency alerts go out in English only. The system was built for a different renter.
A Network That Holds the Desert in Place
Each of the five questions revealed one layer. Assembled together, they show why a cooling desert stays stable and why fixing one piece changes little. The network below is not a metaphor. It is the actual structure of constraints that produce 557 cooling desert neighborhoods in New York City. Hover each node to read what it does.
What the Network Tells Us About Policy
ANT does not offer a single cause or a single fix. It offers a way to see how separate systems, a building's wiring, a city's measurement tool, a monthly surcharge, a rent check, connect into one outcome. The cooling desert is that outcome. Fixing it means disrupting several connections at once.
The scenario modeling in the data section tests this directly. Four interventions applied separately each remove 12 to 23 percent of cooling deserts. Applied together, they remove 67 percent. The remaining 257 neighborhoods signal that deeper structural change, in rent regulation, building codes, and energy assistance eligibility, is part of the picture.
Three specific policy directions follow from this analysis. First, the HVI needs to count AC use alongside AC ownership. The current measurement gap sends resources to the wrong places. Second, the NYCHA AC surcharge should be suspended for income-qualified residents. One in three NYCHA seniors cite it as the reason they do not cool their apartment. Third, cooling center hours need to extend through Sundays, when heat mortality peaks and 83 percent of centers are closed.
None of these changes are technically difficult. Each one addresses a specific actor in the network. The challenge is political, not logistical. The data here makes the case for why the network needs to be addressed as a whole.
See the spatial and demographic data behind these findings.