Historic Redlining, Tree Canopy, and Urban Heat in Chicago

Abstract

In 1938 the federal Home Owners' Loan Corporation drew lines around the neighborhoods of Chicago and shaded them into four grades, green for the ground it called best and red for the ground it called hazardous. This paper measures one thing about that map, which is its geometry, and borrows everything else. Working from 703 digitized HOLC polygons, of which 683 carry a letter grade, we find that the two lowest grades, C and D, cover 78.6 percent of about 626 square kilometers of graded residential land, while the best grade, A, covers 4.4 percent.^[1] The grading runs sharply north to south. The mean D-zone center sits about 27 kilometers below the mean A-zone center, and 78.2 percent of D zones lie south of the Loop against 22.4 percent of A zones.^[1] Joined to the city's 77 community areas, a D-grade zone touches 43 of them and an A-grade zone touches 4, with 41 community areas holding a D zone and no A zone.^[1]

That is the footprint, and it is the whole of what we measured. We computed no Chicago temperature and no Chicago tree-canopy value. What grew on top of the grades since 1938 is carried here entirely by published research, which reports, across dozens of American cities, that formerly D-graded neighborhoods now run several degrees hotter and hold roughly half the tree canopy of formerly A-graded ones.^[2][5][9] The contribution is the persisting shape, drawn to scale from real geometry, set beside a heat-and-canopy gradient that other researchers measured and we cite. We claim no cause. We describe a template and report what the literature finds resting inside it.

A map drawn in 1938

Begin with what the Home Owners' Loan Corporation actually made. In the late 1930s, appraisers, real-estate men, and lending officers working with the federal corporation went through American cities and sorted residential districts by their perceived security for mortgage lending. They arranged the sort as a ladder of four grades and keyed each grade to a color. Grade A, green, meant best. Grade B, blue, meant still desirable. Grade C, yellow, meant definitely declining. Grade D, red, meant hazardous.^[1] The red is where the word redlining comes from.

The grading was a risk classification, and it was explicitly hierarchical. The corporation that drew it, the Home Owners' Loan Corporation, was a New Deal agency set up earlier in the decade to rescue home mortgages during the Depression, and by the late 1930s it had a national interest in knowing which neighborhoods were safe to lend in. The four grades were its answer, a ladder from A down to D applied to residential districts city by city. A neighborhood's grade folded in the age and condition of its housing, but the surviving area descriptions that accompany the maps show it also folded in the race, national origin, and class of the people who lived there, with the presence of Black and immigrant residents recorded as a hazard to a neighborhood's rating.^[1] The point for this paper is structural rather than moral, and the moral case has in any event been made at length by others. The grades were a top-to-bottom ordering of neighborhoods, applied to the places people lived, keyed to race among other things, and committed to paper as colored polygons over a real city.^[1]

That last fact, that the grades were applied to neighborhoods rather than to buildings or parcels, is what makes them legible in present-day terms and is easy to pass over. The HOLC did not rate individual houses. It drew a boundary around a district, assigned the whole district a single grade, and shaded it accordingly, which is why the digitized product is a set of area polygons rather than a set of points. A neighborhood-scale judgment is exactly the kind of thing that can still be compared, decades later, against neighborhood-scale measurements of canopy and heat, because both are defined over areas of the same rough size. Had the HOLC graded parcels, the comparison the later literature runs would be far harder to make. It graded neighborhoods, and neighborhoods are what persist as units of a city.

What makes that 1938 act measurable today is the digitization. The Mapping Inequality project at the University of Richmond's Digital Scholarship Lab scanned the surviving HOLC maps and area descriptions city by city and traced each graded neighborhood into a precise polygon, published with its grade, its land-use flags, and its surveyor labels as open data.^[1] A scanned map is a picture. A polygon is a shape with an area, a center, and a boundary you can intersect with anything else drawn on the same city. The difference between the two is the difference between looking at the grading and counting it. The Chicago layer of that digitization is the object we worked on.

The distinction is the quiet enabling condition of this entire literature, and it is easy to take for granted. For most of the time the HOLC maps existed they were archival objects, sheets of paper in a federal records collection, reproducible as images but not as data. You could look at one and see that the South Side ran red. You could not, from the image alone, ask a computer how many acres were red, or whether the red sat north or south of a present-day boundary, or how the grades lined up against a satellite temperature reading, because the image carries no notion of area or coordinate. Tracing each zone into a georeferenced polygon converts the picture into something a machine can measure and overlay, and it is that conversion, performed once by the Mapping Inequality team, that makes the comparisons in this paper and in the studies it cites possible at all.^[1] Every finding here, ours and the literature's, sits downstream of that act of digitization. We did not perform it. We used its product, and the reliability of our counts is, in the end, the reliability of that tracing plus arithmetic.

We parsed the file. It holds 703 polygons for Chicago. Of those, 683 carry one of the four letter grades, and 20 do not, because the original surveyors set them aside by land use rather than by lending risk.^[1] Before counting we trimmed stray whitespace from the grade field, because a handful of entries arrived as "A " or "C " with a trailing space that would have split a single grade into two categories and thrown off the totals. After trimming, the grades resolve cleanly to 49 A, 160 B, 327 C, and 147 D.^[1] We confirmed that all 703 features carry usable polygon geometry, with none dropped for a missing or malformed shape. That last check matters, because a finding about how thoroughly the grading filled the city would weaken if a meaningful share of the shapes had to be discarded before counting. Nothing was discarded for want of a shape.

We recomputed the area of every polygon. The raw area field in the published file is expressed in squared degrees, and squared degrees distort with latitude, because a degree of longitude covers less ground in the north of the city than in the south. Rather than trust that field, we projected the geometry into a local equirectangular frame set at a reference latitude of 41.878 degrees north, near the latitude of the Loop, which converts the polygons to an approximately equal-area footprint in square kilometers around Chicago.^[1] This buys a sturdy comparison between grades at the cost of some precision in the absolute totals, a trade we return to in the section on limits. With the grades counted and the areas computed, we can describe the footprint.

What the geometry says

The graded zones cover about 626 square kilometers, and that ground is not split evenly across the four grades.^[1] Start with the counts, because they are the simplest cut. Among the 683 graded zones there are 49 A, 160 B, 327 C, and 147 D.^[1] The distribution leans hard toward the bottom of the ladder. The two lowest grades, C and D, together account for 474 of the 683 zones, which is 69.4 percent of the graded total.^[1] More than two of every three graded neighborhoods in 1938 Chicago were marked declining or hazardous. The best grade is the rarest, 49 A zones making up 7.2 percent of the count, and the single most common grade is C, the declining grade, at 327 zones, nearly half of everything graded.^[1]

Land sharpens the disparity, because the zones are not the same size and the grading did not spread large and small districts evenly across the ladder. By recomputed area, the C and D zones hold 78.6 percent of the graded land, with C alone at 51.5 percent and D at 27.1 percent, so the declining grade by itself covers more than half of all graded residential ground in the city.^[1] The A zones, the best grade, cover 4.4 percent.^[1] Set the two ends against each other and the ratio is stark. D-graded land outweighs A-graded land by roughly 6.2 to 1.^[1] For every square kilometer the surveyors marked best, they marked about six hazardous.

The scarcity of the top grade is the figure that stays with you. Just 4.4 percent of the graded land earned the green that told lenders a neighborhood was a safe bet, which means more than 95 percent of the graded residential ground in 1938 Chicago carried something less than the best grade.^[1] Green was not the baseline from which a few neighborhoods fell short. It was the exception, awarded to a sliver of the city, while the ordinary condition of a Chicago neighborhood, as the federal grading saw it, was some shade of doubt. That inversion, in which approval is rare and discouragement is the norm, is the single most important thing the area shares show, and it is the opposite of how the word redlining, with its focus on the red alone, tends to lead people to picture the map.

The gap between the count and the area is itself worth a moment, because the two measures disagree and the disagreement is informative. By count, D zones are 21.5 percent of the graded total; by area, they are 27.1 percent. By count, A zones are 7.2 percent; by area, 4.4 percent.^[1] The hazardous grade swells when you weigh it by ground and the best grade shrinks. A handful of large declining and hazardous districts carry more land than a larger number of small best-rated ones. That is why we lean on the area share as the more telling figure through the rest of the paper. A count tells you how many lines the surveyors drew. An area share tells you how much of the city fell inside the lines of each grade, and it is the city's ground, not the count of districts, on which present-day heat and canopy sit.

A caution belongs here, before the shares harden into something they are not. These are shares of land the 1930s surveyors graded, and nothing more. The 78.6 percent figure says that more than three-quarters of the graded residential ground carried a C or D grade. It does not say that 78.6 percent of Chicagoans lived on that ground, then or now, and it says nothing about who lives there today. The polygons record an act of classification performed by appraisers nearly ninety years ago. They are a faithful record of where the surveyors drew which grade, and they are silent on population, on tenure, on race in the present, and on every outcome that came after.

The middle grade carries its own weight in this accounting, and it is easy to lose. The intermediate grade B, still desirable, accounts for 160 zones and 17.0 percent of the graded land, a real share but a minority one, so the city did not pile up in the comfortable middle.^[1] It piled up at the bottom. Adding B's 17.0 percent to A's 4.4 percent leaves only about a fifth of the graded land in the two grades lenders were told to favor, against the 78.6 percent in the two grades they were told to avoid.^[1] However you cut the four categories, the encouraging end is thin and the discouraging end is thick.

The asymmetry is the first feature of the template, and it has a definite shape. The grading concentrated land in the declining and hazardous categories and reserved a thin sliver for the best, a wide base of C and D under a narrow cap of A. The binary version of the redlining story, a neighborhood either redlined or not, loses the middle of that shape. Grade C was not neutral ground. The HOLC called it definitely declining, and a yellow grade told lenders a neighborhood was on its way down, often because Black residents had begun to arrive or were expected to. Yellow was a warning about the future and red a verdict on the present, and together they covered the clear majority of the graded city. The optimism, the green and blue, was the exception.

It is fair to ask whether a four-grade share is the right way to read a lending map at all, since the HOLC did not allocate land by quota. The grades were judgments about particular neighborhoods, made one at a time, and the share that results is the sum of many separate verdicts rather than a plan. That is exactly why the share is worth reporting. No one set out to put 78.6 percent of Chicago's graded ground in the bottom two grades. The figure is what fell out when appraisers applied a rubric, neighborhood by neighborhood, to the city as they found it. A share that no one designed but that everyone's separate decisions produced is a fair description of where the lending system, in aggregate, came down. It came down hard against most of the city it bothered to grade.

A north-south line

The grading has a direction, and the direction is north to south. This is the spatial heart of what the geometry shows, and it is also where the temptation to overreach is highest, so the safest move is to let the polygon centers speak. A centroid is the geometric center of a polygon, the single point that balances its area. Average the centroids within each grade and you get a mean location for the grade as a whole, and those mean locations fall along a clean axis.

The mean A-zone centroid sits at 42.02 degrees north. The mean D-zone centroid sits at 41.78 degrees north.^[1] The gap is 0.24 degrees of latitude, about 27 kilometers at Chicago's latitude, so the best-graded land sits, on average, some 27 kilometers north of the hazardous-graded land.^[1] The middle grades fall in order between the two ends, which means the four grades line up down the map in the same sequence they line up down the risk ladder. A is highest and farthest north. D is lowest and farthest south.

The averages locate the center of each grade; the spread tells the sharper story, and the sharpest cut is to split the city at the Loop, near 41.88 degrees north, the line Chicagoans use to separate the North Side from the South. Among A zones, 22.4 percent sit south of that line. Among D zones, 78.2 percent do.^[1] The best grade is overwhelmingly northern and the hazardous grade overwhelmingly southern, with the two middle grades between them, 33.8 percent of B zones and 50.5 percent of C zones south of the Loop, so the southward share climbs steadily as the grade falls.^[1] The progression across the four grades runs 22.4, 33.8, 50.5, 78.2, a clean ascent from best to hazardous.

The far North Side completes the mirror. Take the zones whose centers reach above 41.95 degrees north, the latitude of the city's northern neighborhoods. Among A zones, 61.2 percent reach that far north; among D zones, 2.7 percent do.^[1] Read those two numbers together and the segregation of the grading is nearly total at the extremes. Almost two-thirds of the best-graded zones sit on the far North Side. Fewer than three in a hundred hazardous zones do. The middle grades again fall in between, 53.1 percent of B zones and 28.1 percent of C zones reaching the far north, so the northward pull weakens by exactly one step at each rung down the ladder.^[1] The A grade pools in the north and the D grade pools in the south, and the city the surveyors drew is split along a line that runs roughly through its middle.

There is an east-west component as well, fainter but present in the same numbers. The mean A centroid sits at about 87.78 degrees west and the mean D centroid at about 87.67 degrees west, which places the hazardous-graded land not only south of the best-graded land but somewhat east of it, toward the lake and the older industrial core.^[1] We do not lean on the longitude the way we lean on the latitude, because the east-west spread is smaller and a mean longitude is a blunt summary of a city that bulges and narrows along its lakefront. The dominant axis is north to south. The point of noting the longitude at all is honesty about what the centroids contain. They locate each grade in two dimensions, and the second dimension, while weaker, points the same way the history does, toward the South and West Sides and away from the northern lakefront.

A word on what a centroid can and cannot carry. The center of a polygon is a single point, and averaging the centers within a grade compresses dozens of irregular shapes into one coordinate. That compression is the source of the method's clarity and also its limits. It is why we can state the north-south gap in a single number, 27 kilometers, and also why that number should not be read as a precise distance between two places on the ground. It is the distance between two abstractions, the average position of the green zones and the average position of the red ones. The progression of those averages down the ladder, and the steady climb of the south-of-Loop share from 22.4 to 78.2 percent, are the trustworthy findings.^[1] They describe a real orientation in the data without pretending the data is tidier than it is.

This is the persisting template, stated flatly. A south-and-west concentration of the lower grades set against a northern concentration of the best grade, drawn to scale and locatable on any present-day map of the city. The centroids show that the grading was oriented north to south and that D zones cluster south while A zones cluster north. They do not, by themselves, explain why any neighborhood is hot or bare of trees today, and nothing in the geometry licenses that step. What the geometry gives us is the footprint. The inferential work, the case that this footprint has present-day consequences, belongs to research that took the temperature and canopy measurements we did not take.

The directional clarity is also what makes the footprint usable by that research. A pattern with a clear orientation, the lower grades pooling in a known part of the city, is something a present-day measurement can be laid against and checked. If the grades had been scattered at random across Chicago, with no north-south or east-west tendency, there would be little to compare a canopy map to; the historical layer would be noise. Instead the grades have a shape, and the shape points in a particular direction, which is precisely the property that lets a researcher ask whether present-day heat or canopy points the same way. The national studies do that across many cities at once. Our contribution is to draw the Chicago version of the shape exactly, so that anyone who later measures Chicago's canopy or heat has a precise historical layer to test against rather than a vague impression that the South and West Sides were graded down.

Lines around where people lived

One more feature of the geometry settles a question that matters for everything the literature will add. Each polygon carries land-use flags for residential, commercial, and industrial use, and the flags reveal what kind of districts the surveyors were grading. Every one of the 683 graded zones carries the residential flag.^[1] There is no graded zone in the Chicago file marked as anything other than residential. The grading was an assessment of residential neighborhoods, places where people lived, and the file says so without exception.

The commercial and industrial flags exist in the data, but they sit entirely off the graded ladder. There are 15 commercial flags and 5 industrial flags, and all 20 of them fall on the 20 ungraded zones.^[1] The surveyors set a small number of business and factory districts aside with a use label and declined to give them a letter grade, which is what you would expect from a system built to rate residential lending risk. Business land got a label. Residential land got a grade. The split is clean, and it carries one inference the data fully supports. The HOLC was grading the places people lived, which is exactly why later research can lay these grades over present-day residential heat and canopy. When a study finds that a formerly D-graded neighborhood runs hot today, it is comparing residential ground to residential ground, one neighborhood against another rather than a factory district against a leafy street.

This matters more than it might seem, because heat and canopy vary enormously by land use, and a comparison that mixed uses would be hard to read. Industrial land runs hot for obvious reasons that have nothing to do with lending grades, all roof and rail and tarmac, and it carries little canopy because it was never meant to. If the HOLC had graded factory districts and leafy residential ones on the same scale, a finding that low-graded land is hotter might simply reflect that more of it was industrial. The grading did not do that. It rated residential neighborhoods against one another, holding land use roughly constant across the comparison, so the heat and canopy differences the literature finds along the grades are differences between residential places, not artifacts of comparing a steel mill to a bungalow block. The residential-only structure of the grading is, in that sense, a quiet methodological gift to everyone who later used these maps, and it is worth stating because it is the kind of thing that makes the downstream comparisons cleaner than they would otherwise be.

A real limit comes with that finding, and it belongs in the open rather than in a footnote. Because every commercial and industrial flag sits on an ungraded zone, the file cannot distinguish residential subtypes within a grade. We cannot tell, from these flags, a single-family A district from a multi-family one, or a dense D neighborhood from a sparse one, because the only non-residential labels attach to zones that carry no grade at all. The land-use finding is therefore coarse, and we will not stretch it past what it says. It is the simple, fundamental fact that the graded universe is residential top to bottom, and that fact is what we need. It establishes that the template is a residential template, a sorting of neighborhoods, and that is the surface the heat-and-canopy literature works on.

The footprint on the present-day map

The last thing the geometry can do is locate the 1938 grades on the map Chicagoans actually use, the 77 official community areas, drawn by University of Chicago sociologists in the 1920s and held fixed since.^[1] We joined each graded HOLC zone to a community area through the zone's area-weighted centroid, asking which community area contains that single center point, and counted how many distinct community areas each grade reaches. This is a descriptive overlap count. We did not weight it by coverage, model it, or test it for significance.

A D-grade zone touches 43 of the 77 community areas. A C-or-D zone touches 68. An A-grade zone touches 4.^[1] The hazardous grade reaches into well over half the community areas of Chicago, and the best grade reaches into four. Widen the lower band to include the declining grade and the reach is nearly total. 68 of the 77 community areas, more than eight in ten, hold a C or a D zone, so the discouraging end of the ladder is not a feature of a few neighborhoods but a condition of most of the city.^[1] Put the asymmetry another way, 41 community areas hold a D zone but no A zone, and some graded zone of one grade or another reaches 72 of the 77 areas, so the 1938 map covers nearly the whole city while the part of it graded best is confined to a handful of places.^[1] The mirror figure, the number of community areas holding an A zone but no D zone, is necessarily small, because only 4 areas hold an A zone at all.

The list of community areas a D zone touches reads as a tour of the South and West Sides with the expected northern exceptions. On the South Side it runs through Englewood and West Englewood, Auburn Gresham, Greater Grand Crossing, Woodlawn, Grand Boulevard, Douglas, Kenwood, Hyde Park, Washington Heights, Roseland, Pullman and West Pullman, Burnside, Riverdale, South Chicago, South Deering, the East Side, and Hegewisch, a near-continuous band from the historic Black Belt down to the old steel district at the city's southern edge.^[1] On the West and Southwest Sides it includes North Lawndale, East Garfield Park, the Lower West Side, the Near West Side, New City, McKinley Park, Brighton Park, Archer Heights, Garfield Ridge, Chicago Lawn, and Ashburn.^[1] It also reaches the central area and the North Side, the Loop and the Near North and Near South Sides, along with Bridgeport, West Town, Logan Square, Avondale, Belmont Cragin, Lincoln Park, and Rogers Park, which shows the hazardous grade was not exclusively southern even as it concentrated south.^[1] Two of the names, Beverly and Mount Greenwood on the far Southwest Side, are places more often associated today with stability than with redlining, a reminder that a touch records only that some D polygon's center fell inside the area, not that the whole area was graded down.^[1]

The geography of the list matches the centroid finding from the other direction. Where the centroids gave an average position, the community-area list gives the actual named places, and the two agree. The hazardous grade is spread widely, anchored on the South and West Sides, and it lands in neighborhoods a Chicagoan can picture. The presence of a handful of northern and lakefront names, and of two far-Southwest areas usually thought of as comfortable, is the kind of detail that keeps the finding honest. The pattern is strong and directional, not absolute, and the overlay reports it as such.

The caveats on this join are the loudest in the paper, and they belong in plain view. The join uses each zone's single area-weighted centroid, so a zone straddling two community areas is credited only to the one holding its center, which means the counts slightly understate how many community areas each grade physically overlaps. A community area counts as touched if at least one zone of a grade has its center inside it, so these are presence counts, not measures of how much of an area a grade covers. The larger limit is the reach of the 1938 map itself. Only 329 of the 703 zone centroids fall inside Chicago proper. The other 374 fall outside the city, because the HOLC graded well into suburban Cook County, and the original map spans a wider latitude range than the city does.^[1] We report those 374 out-of-city centroids rather than dropping them silently, because dropping them without saying so would misrepresent how much of the original map lies beyond the community-area grid. The counts above describe the 329 centroids that land inside the city. The 374 outside are real polygons in a real map, and they are accounted for here.

The split between the 329 in-city and the 374 suburban centroids is itself a fact worth pausing on, because it says something about what the HOLC was doing. More than half of the graded polygons in this Chicago map sit beyond the city line, out in the towns and unincorporated stretches of Cook County. The agency was not grading a city in isolation; it was grading a metropolitan region, the city and the suburbanizing ring around it, in a single sweep. That is consistent with what the lending system was for, sorting the whole field in which mortgages might be written, and it foreshadows the postwar pattern in which federally backed credit flowed heavily to new suburban construction. For this paper the practical consequence is narrow. The community-area cross-walk can only account for the 329 centroids inside Chicago, because the community areas are a city geography, and the suburban half of the map falls outside the frame the cross-walk uses. We do not draw conclusions about the suburbs from a city overlay. We note only that the original document was metropolitan in scope, that we kept the count honest by reporting the part that fell outside our frame, and that a fuller study of the suburban grades would need a different administrative geography than the 77 community areas.

Two cautions close the section. The overlap count is not a causal estimate. Saying a D zone touches 43 community areas says nothing about why those neighborhoods look the way they do today. It is also not a population figure. A community area touched by a D zone need not hold any particular number of people living under any particular condition, and we make no such claim. What the cross-walk delivers is reach. The 1938 template maps onto named neighborhoods a present-day reader can locate, which is what lets the national literature bear on Chicago at all. The studies report patterns across formerly redlined neighborhoods nationally. The cross-walk shows which Chicago neighborhoods carried the grades those studies key on.

What we did not measure, and who did

Here the paper changes register, and the change is deliberate. Everything to this point came from our own measurement of the HOLC geometry. Everything that follows comes from other people's published research, and we cite it as such. We computed no value for Chicago tree canopy and no value for Chicago land surface temperature. We have no satellite imagery, no temperature readings, no canopy survey of our own. To pretend otherwise would be to invent data, and the value of the exercise depends on not doing that. The reason to bring heat and canopy in at all is that a clear and convergent body of recent research has already measured them against the same HOLC grades we mapped, and the findings are consistent enough to report plainly.

This division of labor is the honest genre of the paper, and it is worth naming plainly so a reader knows what kind of document this is. It is a synthesis with one original component. The original component is the geometry, measured by us from a public file. The synthesis is the surrounding literature, measured by others and cited. We did not run an experiment, model an outcome, scrape a website, interview anyone, or take a reading of the present-day city. The strongest claim the paper is entitled to make is descriptive and comparative, that a footprint we can measure precisely sits beside a present-day pattern that others have measured, and that the two describe the same neighborhoods. Everything below is held to that standard. Where a number is ours it carries the analysis citation; where a number is borrowed it carries the citation of the study that produced it; and the seam between the two is kept visible on purpose.

Start with heat, because it is the most direct reading of what a thin canopy and a wide field of pavement do to a summer afternoon. Hoffman, Shandas, and Pendleton studied 108 United States urban areas and compared present-day summer land surface temperatures against the historical HOLC grades. They found that the formerly D-graded neighborhoods run hotter than the non-redlined ones in 94 percent of those cities, with the formerly redlined areas averaging about 2.6 degrees Celsius, roughly 5 degrees Fahrenheit, warmer than the formerly A-graded ones, and Chicago is among the cities in their sample.^[2] The figure is theirs, computed from satellite readings across the country, not a Chicago value we generated. We report it because the city whose grading we mapped sits inside the set of cities where that pattern was found. The HOLC surveyors were not measuring temperature. Something about what their grades tracked, and what those grades helped set in motion, shows up in the thermal imagery eight decades later, in nine of every ten cities examined.

It helps to be clear about what land surface temperature is, because it is not the same as the air temperature on a thermometer. It is the temperature of the ground and the surfaces, derived from satellite sensors, and it responds strongly to what covers the land. Asphalt, rooftops, and parking lots drive it up. Vegetation and shade pull it down. A finding that redlined zones carry higher land surface temperature is, in large part, a finding about what is on the ground in those zones, which leads straight to the trees. The thermal map and the land-cover map are reading the same neighborhoods, and they agree.

The scale of the Hoffman finding is what makes it hard to dismiss. The figure is not a single dramatic city held up as an example. It is an average drawn across 108 urban areas, and the direction of the effect holds in 94 of every hundred of them, which is to say nearly everywhere the comparison was run.^[2] A gap of about 2.6 degrees Celsius between the formerly best-graded and formerly worst-graded neighborhoods is, in the abstract, a couple of degrees, the sort of difference a person might not notice walking from one block to the next on a mild day.^[2] On the worst days it is not mild. A few degrees added to an already dangerous afternoon, and held through the night, is the margin that turns a heat wave from an ordeal into a lethal event, and it is added precisely to the neighborhoods with the least shade and the least means to cope. The average is small. The tail is where people die.

Two degrees also compounds with what the same literature finds in the same places. The formerly redlined neighborhoods that run hotter are, by the canopy and land-cover studies, the neighborhoods with less tree cover and more impervious surface, the two conditions that drive the heat in the first place.^[5][9] The land surface temperature gap does not arrive alone. It arrives in neighborhoods already short on the shade that would relieve it, which is why the same two-degree difference on a satellite map can mean something sharper on the ground in one neighborhood than in another the weather report would treat identically.

There is a fair question about how far a national average travels to a particular city, and Chicago's place in this evidence answers part of it. Chicago is not outside the Hoffman sample looking in; it is one of the 108 urban areas the study examined, so the city whose grades we mapped is among the cities where the redlined-runs-hotter pattern was tested and, in the large majority, found.^[2] That does not tell us Chicago's exact gap, which the study reports as part of an aggregate rather than as a city-by-city table we have transcribed, and we will not invent a Chicago figure from a national mean. What it tells us is that Chicago is a member of the population the finding describes, not a city to which a foreign result is being stretched. The canopy and greenspace studies cover overlapping national samples on the same HOLC grades, and the present-day heat studies cover hundreds of cities and counties including the country's largest, so the weight of evidence bearing on Chicago is the weight of a pattern Chicago belongs to, reported by others, set beside a Chicago footprint we measured ourselves.^{[2][4][5][7][8][9]}

The same signal holds when researchers narrow the lens to individual metros. Wilson examined Baltimore, Dallas, and Kansas City and found higher mean land surface temperature in the districts redlining had targeted decades earlier, and argued that equity planning has to reckon with that inheritance rather than treat present-day heat as a blank-slate condition.^[3] The argument matters as much as the measurement. Heat in a formerly redlined neighborhood tracks a built environment that earlier policy helped shape rather than a fresh accident of weather, and a planning response that ignores the history will tend to reproduce it.

Li and colleagues followed the chain one link further, from grade to heat to the body. Across 11 Texas cities they modeled the relationships running from HOLC redlining to higher urban heat and onward to higher rates of heat-related emergency department visits.^[10] That is the part a temperature map alone cannot show. A few degrees of additional warmth on a July afternoon stays abstract until it arrives as a person in an emergency room, and their work connects the grade a surveyor wrote in the 1930s to a hospital admission in our own decade. The cities are in Texas, not Illinois, and the finding belongs to them, but the mechanism it describes, heat meeting the body in places with the least relief from it, is not regional.

The Li result is worth pausing on because it completes a chain that the other studies leave at the edge of the body. Hoffman gives the temperature, the canopy studies give the surface that produces it, and Li gives the clinical consequence, the visit to the emergency department that is the measurable form of heat doing harm to a person.^[2][5][10] Each study covers a different link, and read together they describe a pathway from a 1930s lending grade to a present-day medical event, by way of the built surface in between. No single paper proves the whole pathway end to end, and Li's own contribution is a model of associations across 11 cities rather than a tracing of any individual case. But the links line up, each measured by someone, and the pathway they sketch is coherent. The lending map sorted neighborhoods, the sorting shaped the surface, the surface holds heat, and the heat sends people to the hospital. Where any link is missing for Chicago specifically, it is missing because we did not measure it, not because the chain breaks.

The canopy underneath the heat

Turn to the trees and greenness beneath the heat, because the canopy is the lever the heat literature keeps pointing back toward. Nardone and colleagues measured greenspace across 102 urban areas and found it thinning step by step as the grades worsened, with the mean normalized difference vegetation index falling from 0.47 in A-graded areas to 0.43 in B, 0.39 in C, and 0.36 in D, a gradient that held after they adjusted for the historical characteristics of the neighborhoods.^[4] The ordering is the thing to notice. It is not a single gap between best and worst. It is a staircase that descends in the same A over B over C over D sequence the surveyors drew, which is the same sequence the Chicago land-share and centroid figures trace across the South and West Sides.^[1]

The measure Nardone uses is worth a sentence, because it is broader than tree canopy alone. The normalized difference vegetation index is a satellite reading of how much living, green, growing vegetation covers a patch of ground, lawns and shrubs and parks as well as trees. So the Nardone finding is not only about the street canopy that the Locke and Nowak studies count; it is about greenness of every kind, and it thins in the same order.^[4][5][9] A formerly redlined neighborhood tends to have less of all of it, fewer trees and less green underfoot, which matters because vegetation cools through the same two routes whatever its height, by shading the ground and by releasing water into the air. Three studies, three slightly different instruments, one staircase.^[4][5][9]

The detail in Nardone that bears most on this paper is the adjustment. The authors did not simply observe that worse-graded areas are greener or less green; they checked whether the gradient survived once they accounted for the historical characteristics of the neighborhoods, and it did.^[4] That is the move that separates a finding from a coincidence. If the greenspace gap vanished the moment you controlled for what the neighborhoods were like to begin with, the grade would be carrying no independent information. It did not vanish. The grade still tracked the present-day greenness after the adjustment, which is the kind of result that makes the historical map look like more than a curiosity, though, as the section on evidence will insist, still not a proven cause.

Locke and colleagues put a number on the canopy itself. Across 37 metropolitan areas they found that formerly A-graded neighborhoods average about 43 percent tree canopy today while formerly D-graded ones average about 23 percent, the present-day canopy ranking mirroring the 1930s risk ranking almost exactly.^[5] Roughly twice the shade in the best-rated neighborhoods as in the worst-rated, eight decades after the rating. Nowak, Ellis, and Greenfield reached the same shape at national scale, reporting tree cover of 40.1 percent in HOLC class A against 20.8 percent in class D, with impervious surface moving the opposite way, climbing from 30.6 percent in A to 53.0 percent in D, and the dollar value of the services the urban forest provides falling with each worse grade.^[9] Read those two facts together and the physical picture closes. The best-graded neighborhoods carry the most canopy and the least pavement. The worst-graded carry the least canopy and the most pavement. Shade and asphalt arranged themselves along the grade lines, and the heat readings follow from that arrangement.

The Nowak result is the one to dwell on, because it treats the canopy as infrastructure rather than amenity. Tree cover and impervious cover are the two halves of a single surface, and they trade off against each other, so a neighborhood heavy on roads and roofs and parking and light on trees is, by construction, a place built to absorb and hold heat.^[9] If the canopy were ornamental, its uneven spread would be an aesthetic complaint. Because it does measurable cooling work, and because that work has a dollar value Nowak and colleagues put on the ledger, its uneven spread is an infrastructure gap, and infrastructure gaps track investment.^[9] A mature street tree is not a quick fix a neighborhood can install the year it finally gets attention. It is the visible result of decades of someone having tended the parkway, the budget, and the soil. Where that tending was withheld, the canopy never grew, and a canopy that never grew cannot be conjured back in a season.

The impervious-cover half of the Nowak finding deserves its own moment, because it is where the heat actually comes from. The figures move in mirror image, tree cover falling from 40.1 percent in the best-graded class to 20.8 percent in the worst while impervious surface climbs from 30.6 percent to 53.0 percent.^[9] The two are not independent. Ground is either covered by something living or by something hard, and a neighborhood that lost its trees gained pavement and roofs in their place. The worst-graded neighborhoods, on Nowak's national figures, are more than half hard surface, which is the physical recipe for the heat the thermal studies measure, asphalt and rooftop soaking up the sun by day and giving it back slowly after dark.^[9] So the canopy gap and the heat gap are not two problems that happen to coincide. They are one surface seen two ways, the missing trees and the extra pavement that replaced them, and the temperature reading is what that surface does in the sun.

Why the canopy settled where it did is its own question, and Schwarz and colleagues supply one piece of the answer. Across seven cities they found urban tree canopy strongly and positively correlated with median household income, the relationship dependable enough that they titled the paper for it.^[6] Trees, in their phrasing, grow on money. That correlation is the hinge between a 1930s lending map and a present-day canopy map. The grades shaped where mortgage capital flowed and where it was withheld. Withheld capital meant disinvestment, and disinvestment shaped neighborhood income over decades. Income, the Schwarz work shows, predicts canopy.^[6] None of those steps is something the Chicago polygons can prove, and we do not claim them. We point to the chain because it is the mechanism the rest of the literature keeps assuming, drawn out and measured by the people who studied it.

The income link also explains why canopy is so slow to change, which matters for any neighborhood hoping to close the gap. Money buys trees in more ways than one. A wealthier neighborhood can afford to plant on private lots, can press a city for public planting and maintenance, and, crucially, has had the stability over decades for trees to reach the size at which they actually shade a street. A neighborhood that lost its mortgage credit in the 1940s did not just miss a round of planting. It missed the long, quiet accumulation of canopy that wealthier areas banked year after year, and that accumulation cannot be repeated quickly, because a sapling planted today is not shade for a generation. The Schwarz correlation is a snapshot of that long divergence, the present-day distance between neighborhoods that could afford to grow trees and neighborhoods that could not.^[6] It is the economic reason the 1938 footprint is still legible in the 2020 canopy.

Wilson carries the same point into planning practice. The canopy is not a decoration that happens to be unevenly distributed, she argues; it is environmental infrastructure whose distribution tracks the historical and racial inequities written into maps like the HOLC's, and treating urban heat seriously means treating that uneven infrastructure as the problem to be solved rather than the backdrop to it.^[3] The places that most need cooling are, on this account, the places least equipped to demand or fund it, which is why a heat plan blind to the legacy will tend to widen the gap it set out to close.

The practical edge of Wilson's argument is in how a city decides where to act. A heat-mitigation program that distributes its trees and its cooling investments evenly, the same effort to every neighborhood, leaves the existing gap exactly where it was, because the neighborhoods that start with less canopy stay proportionally behind. A program that distributes by present need does better, but only if it can see where the need is, which requires measuring the present-day canopy and heat directly. And a program that wants to undo a historical pattern, rather than merely soften it, has to know the pattern's shape, which is where a measured footprint like the one in this paper earns its keep. Wilson's contribution is to insist that the choice among those approaches is a real one with real distributional consequences, and that pretending heat is a blank-slate condition is itself a choice, the one that quietly preserves the inherited map.^[3] The geometry cannot make the decision. It can keep the decision from being made in ignorance of where the lines fell.

The same lines, measured a different way

A skeptic could grant every study above and still ask whether the historical grade is doing any work, or whether it is just a stand-in for present-day poverty and race that would show the same pattern with no HOLC map at all. Two of the studies in this literature answer that question from the present side, and they are worth holding up precisely because they do not depend on the 1938 grades.

Hsu and colleagues examined surface urban heat island intensity across the 175 largest United States urbanized areas and found people of color living in census tracts with higher heat-island intensity than non-Hispanic white residents in all but 6 of them, with the same disparity holding for households below the poverty line.^[7] The near-universality of that result is the part to hold onto. Not a majority of cities, not most, but all but six of the 175 largest urbanized areas in the country show people of color living in the hotter ground.^[7] A disparity that consistent across a continent of cities with different climates, different histories, and different governments is not a local accident. It is a structural feature of how American cities are built and who lives in which part of them.

Benz and Burney covered wider ground, 1,056 counties, and found the poorest census tracts significantly hotter than the richest in 76 percent of them, with racial heat disparities persisting after adjustment for income in 71 percent, driven by density, built-up cover, and a shortage of vegetation.^[8] Two features of that result matter here. The first is the mechanism they name, less vegetation and more built surface, which is the same mechanism Locke and Nowak measured along the HOLC grades, so the present-day disparity and the historical one run through the identical physical channel.^[5][9] The second is what survives the income adjustment. When Benz and Burney held income constant and the racial heat gap remained in 71 percent of counties, they showed that the disparity is not merely a story about rich and poor that happens to track race.^[8] Race carries its own weight, on top of income, in determining who lives on the hotter ground. That is the finding most resistant to the easy dismissal that this is all just about money, and it is the finding that most clearly echoes a lending map whose grades, in the surviving area descriptions, were drawn around race directly.

That last clause is the one that ties the present-day findings back to the canopy studies and forward to the lending history. Heat in American cities falls along lines of both class and race, it falls there through the physical surface of the neighborhood, the trees that are present or absent and the pavement that is there instead, and it falls on the same kind of neighborhood the HOLC marked down eight decades ago.^[7][8] None of these three studies cites our Chicago geometry, and our Chicago geometry measures none of their heat. They meet only in the neighborhoods they both describe.

Put the two bodies of evidence together. The historical studies show that a 1938 grade predicts present-day heat and canopy across dozens of cities.^[2][4][5][9] The present-day studies show that heat falls on poor and non-white neighborhoods across hundreds of cities and counties, through the same vegetation-and-surface mechanism, whether or not anyone consults an old map.^[7][8] The grades and the present-day disadvantage are not two separate findings. They are the same population seen at two moments, because the lending map helped sort who would live where and with what investment, and the sort held.

Reading the evidence honestly

Before leaning on those studies, be exact about what kind of evidence they are, because the strength of this paper's argument depends on not overstating them. Every study cited here is observational. No one assigned grades to neighborhoods and then watched what happened. The grades were assigned by the HOLC eighty years ago and the outcomes were measured in the present, so what these studies establish is association, that the historical grade and the present-day outcome move together, and they are clear about that.

The size of the samples is what gives the association its force. A single hot, treeless, formerly redlined neighborhood in one city proves nothing, because any city has accidents of topography, industry, and local politics that could produce such a place for reasons unrelated to the HOLC. A pattern that holds across 108 metropolitan areas, governed by different councils and shaped by different histories, is far harder to wave away.^[2] When 37 cities independently show less canopy in their formerly redlined zones, and 102 show the same staircase in greenspace, the local accidents begin to cancel and a common structural cause becomes the most economical reading.^[4][5] That is the logic the field rests on, and it is sound as far as it goes.

There is one feature of these studies that gives them more weight than an ordinary correlation, and it is worth stating because it is also easy to oversell. The grades were assigned in the 1930s, fixed in the historical record, and not revisable in light of what neighborhoods later became. That ordering in time rules out one common way a correlation goes wrong, the possibility that the present-day outcome caused the predictor rather than the other way around. A neighborhood's 2020 canopy cannot have reached back and changed its 1938 grade. So when a 1938 grade predicts 2020 canopy, the arrow of time at least points in the direction the argument needs. That is more than many observational studies can say, and it is part of why this literature is taken seriously.

It is not, however, a natural experiment, and the careful papers do not pretend it is. What the time ordering cannot rule out is that some third thing present in 1938, the racial and economic composition the grades were partly drawn from, caused both the grade and, through its own long persistence, the present-day outcome. In that account the grade is a marker rather than a cause, a label the system stamped on conditions it then helped perpetuate by other means. Isolating redlining as the single cause is therefore not something these studies do. A 1938 grade is correlated with the present-day pattern in part because it is correlated with everything else that ran along the same lines, the restrictive covenants, the parallel federal underwriting, the disinvestment, the contract selling, and the later rounds of urban renewal and highway construction that so often cut through the graded-down neighborhoods. A grade is, in effect, a marker for a century of treatment. The studies show that the marker predicts heat and canopy. They cannot, and do not try to, divide the credit among the many forces the marker stands in for. We hold to the same discipline. The association is real and well replicated. The clean causal story is not available, and we will not pretend it is. This is also why the present-day evidence from Hsu, Benz, and Burney matters so much. It shows the disparity is live and mechanical now, not merely a fossil of where lines once fell.^[7][8]

A Chicago note, and a moving target

The literature this paper leans on is mostly national, which is both its strength and a gap. The strength is the breadth that lets a pattern across 108 cities stand for something structural. The gap is that none of those studies is about Chicago in particular, and our own work supplies only the historical geometry, not a present-day Chicago measurement. One recent study closes part of that gap, and it is worth treating on its own because it is the single source here that follows Chicago specifically and follows it over time.

Yin and colleagues traced greenness, tree canopy, and new parks across the city's formerly redlined C and D areas from 2010 to 2020.^[11] What they found was not a simple story of recovery or of further decline. It was divergence. Some formerly redlined Chicago neighborhoods gained green over the decade while others did not, and the greening that did occur often slowed after about 2015.^[11] The trajectories split. That finding is more useful than a single number would be, because it answers a question the historical studies leave open. A reader who learns that formerly redlined neighborhoods hold less canopy today might reasonably ask whether the gap is closing on its own, whether time and ordinary investment are quietly undoing the old pattern. The Chicago evidence says not reliably, and not everywhere. Where greening happened it was uneven, and where it had begun it tended to lose momentum.^[11]

We cite the Yin work as a present-day Chicago counterpoint rather than as a measurement of our own, and we hold it at the same arm's length as the rest of the literature. But its lesson sharpens the point of the whole paper. The footprint we measured from the 1938 map is a starting condition, not a verdict on the present. What has happened on top of that footprint since is its own question, measured by its own data, and in Chicago specifically the answer as of the last full decade was a set of diverging paths rather than a uniform trend.^[11] The shape persists, but it is also being worked on, unevenly, in real time. That is the most precise thing that can be said about the present, and it is the strongest argument for measuring Chicago's canopy and heat directly rather than inferring them from a historical map, our own included.

It also reframes what a reader should take the historical footprint to mean. The 1938 geometry does not predict 2026. It marks where the city started the long divergence that the Yin study catches in its most recent decade. A neighborhood the HOLC graded D is not condemned by that grade to be hot and treeless forever; some formerly redlined areas have gained green, after all.^[11] What the grade marks is the starting line of a race that most such neighborhoods began far behind, with the canopy and the investment already pooled elsewhere, and that some have been closing and others losing. The map tells you where the gap began. Only present-day Chicago data, tracked forward, can tell you where it stands.

On method, and its limits

Set down in one place, exactly, what was computed and what was borrowed, because the credibility of a synthesis like this rests entirely on keeping the two apart.

The computation was small and clean by design. We took the Mapping Inequality file for Chicago, parsed its 703 polygons, confirmed that all 703 carried usable geometry with none dropped, trimmed stray whitespace from the grade labels, and from the geometry alone calculated three things, the count of zones at each grade, the area of each polygon and the share of graded land each grade holds, and the latitude of each polygon's center.^[1] From those we derived the comparisons reported above, the 78.6 percent C-and-D land share, the 6.2-to-1 ratio of D land to A land, and the north-south split of the centroids. We then performed one spatial join, intersecting the polygon centers with the 77 community areas to count how many areas each grade reaches.^[1] That is the entire extent of our original analysis. Counts, areas, shares, centroids, and one descriptive overlay. The grade definitions, the dates, the digitization, and the coordinate system all come from the Mapping Inequality project, and the numbers follow arithmetically from the geometry that project released.^[1] We redrew no boundary and reinterpreted no grade.

Two of those steps deserve a fuller word, because they are the places where a careless choice could quietly bend a result. One is the whitespace trim. A handful of grade labels in the file arrived with a trailing space, as "A " or "C " rather than "A" or "C," and to a literal-minded counter those are different categories. Left as they came, the spaces would have split single grades into pairs and undercounted the totals. Catching it is unglamorous, the kind of thing that does not appear in a finding but determines whether the finding is right, and we mention it because a count is only as trustworthy as the cleaning behind it. After the trim the grades resolve to 49, 160, 327, and 147, and they sum to the 683 graded zones the integrity check reports.^[1]

The other is the projection. The published geometry arrives in latitude and longitude, an angular coordinate system in which a degree of longitude covers less ground the farther north you go. Computing area straight from those coordinates would treat a degree as the same width everywhere and would distort the result, and the raw area field in the file carries exactly that distortion, which is why we did not use it.^[1] Instead we projected the polygons into a local equirectangular frame anchored at 41.878 degrees north, near the latitude of the Loop, which rescales the geometry to an approximately equal-area footprint across the small north-south span of one city.^[1] The choice is a standard one for a single metropolitan area and it is the reason we trust the grade shares while treating the absolute square-kilometer totals as estimates. The whole analysis was done in code against the published file, so any of these steps can be rerun and checked rather than taken on faith.^[1]

Several limits follow directly, and we would rather state them than have a reader find them.

The absolute areas are approximate. We recomputed them under a local equirectangular projection rather than trusting the raw squared-degree field, which gives a graded footprint of about 626 square kilometers, accurate enough to compare grades against one another but not a survey-grade measurement of ground.^[1] The shares are the sturdy quantity, because whatever distortion the projection introduces affects the numerator and the denominator alike, so the ratio of D land to A land survives the approximation intact while the bare square-kilometer total should be read as a close estimate.

The community-area overlay is descriptive and nothing more. It is a yes-or-no record of whether a polygon center sits inside an area, not a measure of how thoroughly any area was graded, and a zone that straddles a boundary is credited only to the area holding its center.^[1] Only 329 of the 703 centroids fall inside the city, the other 374 lying in suburban Cook County, and those 374 are reported here rather than quietly discarded.^[1] We chose this conservative form on purpose, because anything more would imply a precision the simple intersection does not have, and because the 1920s community-area boundaries and the 1938 HOLC polygons were never drawn to align.

The land-use reading is coarse. Every commercial and industrial flag in the file sits on an ungraded zone, so the data shows every graded zone as residential and cannot distinguish a single-family district from a dense multi-family one.^[1] We report the fact that the graded universe is residential, which is all the flags support, and nothing finer.

The largest limit is the one the analysis names most sharply. We measured no heat and no canopy. Every temperature and canopy figure in this paper comes from cited literature, not from our own computation, and the reader should hold those figures to the standard of the studies that produced them, which are peer-reviewed and national in scope, rather than to the standard of the geometry we measured directly.^[2][4][5][9] We have tried never to phrase a borrowed finding as one of our results, and where one of our numbers and a borrowed one appear in the same sentence we have marked the seam with a citation. We could have tried to manufacture a Chicago canopy or temperature figure of our own, and we declined on purpose, because doing it honestly would require a real, machine-readable Chicago canopy or land-surface-temperature table joined to these same polygons and recomputed in the open, and transcribing numbers by hand out of published reports would invite exactly the kind of fabrication a civic record cannot afford.

And the deepest limit is causal, stated once more because it is the line a reader is most likely to cross on the paper's behalf. The HOLC grade is a 1938 assessment. The heat and canopy outcomes are present-day, and the relationship between them is correlational in every source we cite. Redlining did not plant or fail to plant a single tree by itself. It was one instrument in a long apparatus of disinvestment, segregation, and neglect, and disentangling its specific contribution from everything that followed is beyond what any of the cited studies, or this synthesis, can do. We describe a persistent shape and a pattern other researchers find resting inside it. We do not, and cannot from this evidence, prove that the one caused the other.

One last limit is conceptual, and it cuts the other way. The HOLC map is not the origin of Chicago's segregation. It is a snapshot of a color line that already existed and a tool that helped harden it. Treating the 1938 grades as the starting point risks giving the map too much credit, as if inequality began when the surveyors drew it. The map's significance is that it took an existing arrangement, gave it the precision of geometry and the authority of the federal government, and helped carry it forward. That is a strong claim about durability and a weak one about origin, and we mean both. The geometry we measured is best read not as a first cause but as a uniquely exact record of a line that many forces drew and many forces kept.

What persists

The clean finding of this paper is geometric. A federal agency in 1938 sorted about 626 square kilometers of Chicago's residential ground into four grades, reserved 4.4 percent of that land for its highest grade, and marked 78.6 percent of it as declining or hazardous, with the hazardous grade pooling on the South and West Sides some 27 kilometers, on average, below the best grade in the north.^[1] Those polygons still exist, precisely, in the digitized record, and they still reach across most of the city, into 43 of 77 community areas at the lowest grade and just 4 at the highest.^[1]

The harder finding belongs to the literature we read rather than the data we computed, and we have tried to keep that line bright. Across many American cities, the neighborhoods those red lines enclosed run hotter today and hold roughly half the tree canopy of the best-graded ones, the heat and the missing shade arranged along the same surface of pavement and vegetation, and present-day heat falls hardest on poor and non-white neighborhoods whether or not anyone consults the old map.^{[2][5][9][7][8]} In Chicago itself, recent greening of the formerly redlined areas has been real but uneven and often slowing, which is to say the shape is still being worked on.^[11]

The honest relationship between those two findings is the whole point, and it is worth stating once more in the plainest terms. We proved the first. We borrowed the second. We did not connect them with a measurement of our own, because we have none, and we did not connect them with a causal claim, because the evidence will not carry one. What we did was lay a precisely measured 1938 footprint next to a precisely measured present-day disparity and observe that they fall on the same neighborhoods. That is a smaller claim than the one a reader in a hurry might want, which is that redlining made these neighborhoods hot. It is also a claim we can fully stand behind, where the larger one is not, and on a civic record the difference between a claim you can defend and a claim you would like to make is the difference that matters.

A map is meant to describe the world it is drawn from. The strange durability of this one is that it kept describing a world that came after it, a world its makers did not live to see and could not have rendered in thermal infrared. The 1938 surveyors were rating mortgage risk, not forecasting summer temperatures or counting trees that had not yet been planted or cut. They were not trying to describe the Chicago of the 2020s. Yet the literature keeps finding the Chicago of the 2020s, and dozens of other cities, arranged along the lines they drew, which is less a tribute to their foresight than a measure of how durable a federal judgment becomes once capital, maintenance, and neglect have spent decades flowing along it.

We can measure the lines to the square kilometer. The canopy surveys, the temperature studies, and the health analyses were done by others, and they point back, again and again, at the shape we measured. The canopy that cools a summer street is being planted, or not planted, right now, neighborhood by neighborhood, on top of a footprint that is eighty-five years old and still legible.^[11][3] A city that has the old grades drawn to scale, and the present-day disparities measured beside them, can choose to grow the next generation of trees where the map left the ground barest, or it can let the inherited distribution stand. The geometry does not make that choice. It only shows, with some precision, where the choice was made once before, and against whom. The map is finished. What it described is not, and the part that is still being written is the only part anyone can still change.