Executive Summary

Redlining was predominately apparent in the everyday lives of black and minority residents of many major cities throughout the 20th century. While no longer a legal practice, the effects of such structural racism have persisted in more than just the racial and ethnic makeup of neighborhoods.

A rather new area of research looks at the relationship between historical redlining practices and current levels of green infrastructure. Using geospatial data, this body of research compares neighborhoods in Chicago based on each neighborhood’s percentage of area covered in redlining maps—broken down by grade—with the green infrastructure measures of tree canopy area and public park area. Tree Canopy area is by far the most common measure looked at for green infrastructure in the research already published on the topic. Because Chicago has been lauded by the Environmental Protection Agency (EPA) for their initiatives to bolster the urban tree canopy by a conglomerate of many community partners, this research provides a starting point for quantitative decision-making on strategic planning and funding for such initiatives.

We managed data collection through publicly available geospatial datasets that we manually cleaned and tabulated by neighborhood. Tree canopy information was collected using image processing techniques applied to manually generated images of each neighborhood. These manually generated images are archived and available for future research. We also conducted initial linear regression studies with publicly available datasets on energy and the housing market.

The main challenge in looking at the relationship between redlining practices and green infrastructure is the vast amount of time between which these two variables are collected. For instance, redlining information generally comes from around the 1940’s, whereas green infrastructure information was collected from as recently as possible. Using publicly available datasets, no strong linear relationships exist between the two categories of variables studied, however clustering neighborhoods into shared qualities from each category helps to bridge the vast time difference in variables studied. This allows us to see which neighborhoods have characteristics that line up with predictions on green infrastructure based on redlining, which may have vastly better green instructure measures as a result of neighborhood investment over time, and which, if any, may have worse green infrastructure measures as a result of declining investment in the neighborhood.

A major part of the research project was to also create visual representations at the scale of individual neighborhoods overlaying redlining information and green infrastructure measures. These are all archived in the Illinois Tech repository and available for use by the public. Each neighborhood has two maps: one comparing redlining information with tree canopy, and one comparing redlining information with public parks.

Our findings indicate that neighborhoods currently fit into four clusters that share characteristics related to the data collected. In order of lowest rates of green infrastructure to highest:

• Cluster 1 (N=28) were primarily redlined and have lower canopy coverage, but slighter higher rate of public parks as a percentage of land area compared to the cluster below

• Cluster 2 (N=33) were a mix of yellowlined and redlined with a canopy in line with the mean for all neighborhoods

• Cluster 3 (N=19) were primarily yellowlined and have a similar canopy to the cluster below, but a much lower concentration of public parks as a percentage of land area.

• Cluster 4 (N=18) were primarily not districted in redlining maps and have high canopy coverage and a high concentration of public parks as a percentage of land area.

Findings also indicate the potential for a relationship between redlining information and other factors in the built environment. While further studies should be conducted to strengthen the suggestion of correlated relationships, this broad overview of initial research suggests the potential for such predictive relationships to exist.

While we cannot go back in time to prevent redlining from happening, there is a societal need to help alleviate the disparities structurally erected from past practices that affected home ownership and segregated neighborhoods in such an outright way. As previous research on the subject indicates, neighborhoods with high instances of redlining are often hotter and have less clean air. Improving the urban tree canopy is a first step towards redressing those inequalities. There are already organizations dedicated to such a mission, however, they still face hurdles in funding and maintenance. They need to build a strong case for the allocation of funds into this important piece of green infrastructure. Perhaps the data collected, images produced, and findings outlined in this research could help these organizations get one step closer to securing the financial resources necessary to create lasting changes in the green infrastructure of neighborhoods currently suffering the most.

Abstract

This report looks at “redlining” maps produced by the Home Owners’ Loan Corporation (HOLC) to analyze the current green infrastructure levels in various neighborhoods of Chicago. Redlining maps, produced in the first half of the 20th century, essentially graded neighborhoods on their riskiness for mortgage lenders. These maps often followed strict racial lines marking neighborhoods with a majority of black and other minority occupants in red to indicate they were the riskiest, hence the term redlining.³ Gathering geospatial data from these HOLC maps and overlaying them with maps showcasing green infrastructure indicators provides a visual representation of the relationship between structural disinvestment in neighborhoods and their current green infrastructure levels. Additionally, some neighborhoods break the pattern, showcasing which areas have changed the most with investment, indicating changes that can be associated with things like gentrification. In the end, k-means clustering best showcases the patterns that have emerged when looking at two variables with such a large gap in time between when the data was collected: redlining information from the 1940’s, and green infrastructure indicators from to 2010’s.

This research was taken a step further to begin descriptive analytical assessments of the ways in which redlining information relates to a variety of other datasets on the built environment. In this way, trends started to emerge that could suggest redlining as a predictor for other variables. However, the neighborhood scale at which the data was grouped does not produce strong enough indications. Nonetheless, continuing this line of study at a smaller scale, perhaps at the census tract level, could lead to more conclusive findings with a stronger correlation between redlining information and other datasets on the built environment that highlight the need for, and lack thereof, of green infrastructure in disenfranchised areas.

Background

The term “redlining” comes from maps drawn by the HOLC, and later taken up by both the Federal Housing Administration (FHA) and the Veterans Administration. These maps indicated where it was deemed safe, and where it was deemed risky to insure mortgages, which were needed in order to buy property anywhere in a major city.⁴ Areas colored red indicated that a neighborhood was deemed too risky to insure mortgages, and these areas were almost always black neighborhoods or neighborhoods adjacent to them. This made home ownership nearly impossible for many black and minority residents in major cities following World War II. The rest of the maps were broken down into yellow districts for declining, blue for still desirable, and green for best. Odds of getting an insured mortgage increased respectively moving through the list from red to green.

Green Infrastructure was defined in the Water Infrastructure Improvement Act, passed by Congress in 2019, as “the range of measures that use plant or soil systems, permeable pavement or other permeable surfaces or substrates, stormwater harvest and reuse, or landscaping to store, infiltrate, or evapotranspirate stormwater and reduce flows to sewer systems or to surface waters.”⁵ Green infrastructure comes in a variety of scales and when implemented can also help provide cleaner air, provide flood protection, diversify habitats, and beautify green spaces.⁶

Other research has been done on the relationship between redlining and green infrastructure in other cities in the United States. One study claims the disparities in air pollution measurements were more pronounced when looking at HOLC grade rather than race or ethnicity.⁷ Another analyzed data from many U.S. cities and found that in 94% of regions studied, formerly redlined areas were hotter than formerly non-redlined areas by an average of 4.7°F.⁸ It’s not hard to imagine why—less shade from trees means the sun can warm larger areas on the street, composed of cementitious elements that have a high heat retention. Relating this information into visual maps has also been conducted for many cities to showcase the inequalities present in green infrastructure. Clearly, inequalities in green infrastructure affect many areas of life from public health to energy consumption and housing.

Chicago offers a great starting point for a detailed look at this relationship because of its initiatives towards improving the urban tree canopy—a major component of green infrastructure. The EPA lists the Chicago Region Trees Initiative (CRTI) under its examples for improving urban tree canopy as a way to strengthen green infrastructure. CRTI is composed of several community organizations, business partnerships, and government organizations aiming to expand and diversify the urban tree canopy. The largest initiative of its kind in the United States, CRTI cites urban development and a lack of funding for planting and proper care as major hurdles to their mission.⁹

This research focuses on tree canopy as a major indicator of green infrastructure in Chicago neighborhoods for this reason. Because there are publicly available datasets produced by community stakeholders invested in the Chicago area urban tree canopy, we imagine this area of research would be useful for helping these organizations with strategic planning as well as having something to use when seeking resources such as grants in order to improve the urban tree canopy in Chicago. This research can also be potentially utilized when making decisions on the approval of developments within the city.

Technical details

Geographic Imaging Software (GIS) was primarily used for data collection. By isolating layers of publicly available geospatial datasets, we were able to collect information on the area of neighborhoods and redlining districts, as well as park district areas and tree canopy coverage. The table below outlines the layers, sources, and file types.

The code we wrote for the report can be found on Github, along with the datasets. A manual for following along is located with Github, as well as listed out in the Appendix

Raster Information

Because the tree canopy was a raster layer, image processing techniques were utilized to calculate the percentage of each neighborhood covered by the tree canopy. For producing images, the raster information for the tree canopy was overlaid with the vector information for the neighborhood boundary manually, which leads to some deviation in accuracy. However, this error is minimal given the resolution achievable within the software utilized, namely QGIS and Adobe Illustrator. A .png image of the tree canopy was eventually exported, which was masked using the neighborhood boundary vectors. Each of these masks was separated to different layers and exported as a new .png for each neighborhood to calculate the percentage of each neighborhood covered by the tree canopy.

Image processing techniques were used to calculate the percentage of each neighborhood covered by the tree canopy. With the .png images exported as outlined above, a count of black pixels (which represent tree canopy) over the count of alpha pixels (opaque pixels as compared to transparent pixels to define the border) gave the percentage. The image was then recreated in Python given the count to check for accuracy and resolution. Had the Python-generated image come back with less granularity and sharp contrast, the count would have been rejected. An example of an original manual .png output and a python generated image of the count is presented below.

Original .png image for Hegewisch created in Adobe Illustrator.	Python generated image for Hegewisch used to check accuracy and resolution.

The .png images of all of the neighborhoods are archived and can be accessed in the Illinois Tech repository for further research.

These three sets of information–redlining boundary percentages, public park percentages, and tree canopy percentages–are combined in Table 1 in the appendix. Note the names of the neighborhoods from the Chicago Data Portal are referenced as the neighborhoods for which data was collected.

Visual Representation

Maps of each neighborhood comparing redlining boundaries to public parks, and redlining boundaries to the tree canopy were created in a similar manner, combining layers in QGIS and exporting in Adobe Illustrator to change visual characteristics. All of these maps can be found in the Illinois Tech repository. An example for Albany Park has been provided below. For this neighborhood, you can see the yellow portions that indicate yellowlining and the blue portions that indicate bluelining.

Map of Albany Park comparing redlining and tree canopy	Map of Albany Park comparing redlining and public parks

Findings

Linear Regression

Another method for extracting findings from the data set would be to combine all of the redlining percentages into a single quantifiable score for each neighborhood. This would allow for comparison between redlining and the green infrastructure variables to find trends with a regression line. This was tested by creating a score for each neighborhood on a scale of 0-1 in which neighborhoods closer to 0 would have been more affected by redlining, and neighborhoods closer to 1 would have been less affected. To calculate the score, the following equation was used:

(%RL*1 + %YL*1.25 + %BL*1.5 + %GL*1.75 + %UL*2) - 1.

This equation weights redlining to skew a neighborhood score closer to zero, and weights areas not districted in redlining maps to skew closer to one. In the equation above, %RL is the % redlined column of Table 1, with other variables named respectively following the table. This method is far from soundproof, but was conducted as an introductory look at the data. For instance, areas not districted in redlining maps were either undeveloped at the time, or slated for industrial use and thus had no housing in the area. These two use-cases shape neighborhoods in vastly different ways. On one hand, undeveloped areas often provided a haven for new development to house the white flight fleeing from their neighborhoods due to blockbusting, which would reflect an influx of wealth into a neighborhood. On the other hand, industrial zones were often concentrated near redlined areas because housing located in areas with high noise and air pollution weren’t ideal.

Clearly no strong relationships can be drawn from these rudimentary tests. This is likely due to the large gap in time between the two variables being compared. The redlining percentages are taken from maps produced around the 1940’s, whereas the green infrastructure data was taken from information collected within the past decade. With over half a century between datasets, it’s safe to assume that neighborhoods have changed in terms of investment and development between the 1940’s and now.

Future Work

The groupings of clusterings indicate that most neighborhoods follow a trend in which redlining relates to lower levels of green infrastructure. Visualizing these clusters allowed us to gain insights into how redlining information and green infrastructure might interact in reference to other factors on energy and the housing market. An initial look at clusters indicates that clusters are not bound geographically in concentric bands radiating out from the city center. This means that certain neighborhoods on the outskirts of the city have the same concentration of green infrastructure in relation to the city center which is much more dense. Taking into account factors on energy and the housing market can help explain the impact redlining may have had in these areas. Regression tests for these new datasets compared to redlining and green infrastructure can be found in Table 6.

Better understanding the relationships between our initial datasets on redlining and green infrastructure with the new datasets on energy and the housing market would likely need to be pursued through non-linear relationship tests. In order for these tests to have potential for prediction power, a smaller unit of scale in comparison to the neighborhood level, would likely need to be utilized to have a higher sample size. This could be done at the census tract level for instance, which has almost 900 occurrences in Chicago, as compared to the just under 100 neighborhoods within the city.

License

The author of this technical report, which was written as a deliverable for a SoReMo project, retains the copyright of the written material herein upon publication of this document in SoReMo Reports.

Acknowledgements

The authors would like to thank the SoReMo Fellowship Program at the Illinois Institute of Technology, and in particular the participating faculty from the Applied Mathematics department.

The authors would like to thank the Camras Scholars Program at the Illinois Institute of Technology who provided feedback on an initial draft of the findings.

Lastly, the authors would like to thank Gemma Brizzolara, a student in the College of Architecture at the Illinois Institute of Technology, who provided assistance with the map productions for each neighborhood.

Appendix

Table 2: Distribution of Features

Table 3: QQ-Plots

Table 4: Scatterplots Comparing HOLC Map Identification with Green Infrastructure Measures

Map 1: Clusters Map

Clustering Map