Grades of Heat

A Note to Mobile Users:

This visualization and data tool is designed for investigation and research and has not been optimized for mobile devices. Please try loading the application on a laptop or desktop device.

Grades of Heat

2023 marked the hottest summer on record. And this heat is not felt equally across the United States, cities, or even across neighborhoods. In a changing climate, it's crucial to understand how biophysical conditions impact different communities and urban settings. A pivotal figure in this narrative is the Home Owners' Loan Corporation (HOLC) of the 1930s, which fostered the redlining phenomenon: a racially biased grading system that informed lending risks and property investment potential for two decades, sowing deep-seated socio-economic rifts and spawning far-reaching environmental consequences.

Grades of Heat utilizes digital HOLC maps and a decade of satellite imagery analysis to unveil the thermal legacy of redlining at a city level. It offers users the chance to explore detailed temperature and vegetation data through interactive map and trend views, revealing the striking connection between historic HOLC grades and the experience of heat across the US.

Grades of Heat is produced and supported by the Brown Institute for Media Innovation.

Project Background

Project Overview

During the 1930s, the Home Owners' Loan Corporation (HOLC) emerged as a crucial entity with the primary aim to refinance home mortgages and prevent foreclosure. However, it initiated a controversial practice of grading neighborhoods from A to D, which not only determined lending risks but also sculpted lasting perceptions and influenced policies significantly based on racial composition. These grading criteria seeped deep into the urban fabric, affecting both public policy and private investment, consequently positioning lower graded neighborhoods at the periphery of developmental priorities, with implications echoing through the decades. This legacy of mapping bore not just segregation but epitomized systemic biases with lingering socio-economic repercussions.

Grades of Heat utilizes digital versions of HOLC maps along with advanced satellite image processing techniques to reveal the complex environmental patterns that have developed in areas categorized by these historical gradings. The initiative, focusing on metrics like temperature and vegetation, intends to highlight the environmental consequences of these gradings. It underscores the fact that these environmental conditions have been shaped over time by various factors, including the influence of the HOLCs, which were initially designed as socio-economic tools.

The tool is built on analysis of ten years of Landsat 8 imagery spanning June-August from 2013-2023. Its analysis includes two core metrics, land surface temperature (LST) and normalized difference vegetation index (NDVI). As a note, during summers in urban environments, LST will appear more extreme than air temperature, due to what is known as the heat island effect. In this tool, the LST provided represents the mean temperature captured across the 50+ readings for each location. For context, NDVI represents vegetation and greenspace, be it a park, a row of trees, or a forest. There is a strong connection between NDVI and lower temperatures for an environment.

How to use the Tool

This project is meant to provide journalists, researchers, and the public an opportunity to explore the relationship between HOLC grades and temperature and vegetation at a city level -- to interrogate the thermal legacy of HOLCs. The platform offers two main views for a nuanced analysis: the Map View and the Trend View.

Map View
The Map View serves as an interactive tool to investigate the selected region in detail. Users can click on a city directly on the map or utilize the search function to select a specific city or state. Once a city is identified, the legend provides an overview of data associated with the city, offering detailed information and trends about temperature and NDVI.

Upon selecting a city, a thermal layer for that city is loaded, representing the mean temperature readings derived from a decade-long analysis of Landsat imagery. This layer is specially designed to underscore the significant hot spots across a city, drawing attention to the substantial role both concrete and greenspaces play in affecting temperature dynamics. Upon selecting a city, users can further zoom in to explore a detailed view of the urban area under investigation, thereby gaining insights into intricate patterns and deviations.

Trend View
It is hard to ignore the direct connection between HOLC grades and higher temperatures. The Trend view is meant to provide a simple interface to highlight this. In this view, users have two modes of engaging with the data: a lollipop chart and a scatter plot. In both view, cities are arranged along their means, with HOLCs presented based on their deviation from the mean. This view is meant to provide a key takeaway from the data: Grades A and B are routinely cooler than Grades C and D during summers.

Support for Research and Journalism

At its core, Grades of Heat is an effort to illuminate the geophysical impacts of HOLCs on communities, particularly as our planet faces increasing temperatures. It's more than just numbers and analysis; it's about stories, histories, and lived experiences. Great reporting has been done about impacts of redlining on heat, but often at a single city with a single sample of temperature. This tool is meant to provide robust, processed data, built on thousands of satellite images over a ten-year span.

Interested in using data on a city? After selecting a city on the map, download links for that city will appear providing access to a dataset containing geographic and geophysical HOLC data specific to your selected city. This data encompasses mean NDVI and LST values extracted from a complete collection of Landsat 8 raster images taken across a 10-year period (2013-2023) for HOLCs across the US. Users also have access to the raw temperature values (in Celsius) in tif format. To access all data used for this tool, in its creation (raw data) and in its output, click on the Data + Methods button in the bottom right of most views.

If at any point you find yourself with questions, or you seek guidance on how to leverage this data to its fullest potential, please don't hesitate to contact us. Collaboration and knowledge sharing are essential to crafting narratives rooted in science but focused on people. If you would like additional data for a city, let us know.

This project was created by Michael Krisch and was supported by The Brown Institute for Media Innovation at Columbia Journalism School and Stanford Engineering School.

The Brown Institute logo

GRADES OF HEAT

The thermal legacy of HOLCs

Select a city on the map or use the search bar in the lower right to find a state or HOLC to investigate.

Of the HOLCs represented here, 173 out of 202 Grade C and Grade D neighborhoods experienced hotter summer temperatures across 2013-2023 than bordering Grade A and Grade B neighborhoods.

Land Surface Temperature and Normalized Difference Vegetation Index data driving this tool are derived from ten summers of USGS satellite imagery captured June-August from 2013-2023.

Temperature (℉) across HOLCs

NDVI across HOLCs

Nearby Cities

Data Downloads

Deviation from the Mean

In this chart, cities are arranged alphabetically and aligned along their mean temperatures. For each city, grades are situated according to their variance from the mean. Cooler temperatures are to the left of the mean and hotter temperatures are to the right.

In scrolling through the list, a core trend emerges: it is cooler in the majority of grades A and B relative to their counterpart grades C and D.

Ascending Temperatures

In this chart, cities are arranged left to right based on mean temperatures of each city. For each city, grades are situated according to their variance from the mean. Cooler temperatures are to the bottom of the mean and hotter temperatures are to the top.

In this view, a core trend emerges: it is cooler in the majority of grades A and B relative to their counterpart grades C and D. A trend mirrored in the other chart view.

Cooler Hotter
Mean

Data + Methodology

Definitions

Home Owners' Loan Corporation (HOLC)
The HOLC was a U.S. government agency established in 1933 as part of the New Deal response to the Great Depression. Its primary mission was to assist homeowners facing foreclosure by refinancing their mortgages. However, it is infamously known for its role in institutionalizing redlining, a discriminatory practice wherein neighborhoods, often with large minority populations, were designated as high-risk for loan purposes.

The Home Owners' Loan Corporation utilized a grading system for neighborhoods which consisted of four designations: A, B, C, and D. Each grade reflected HOLC's assessment of the neighborhood's desirability and creditworthiness:

  • Grade A (Green): These were deemed the "Best" areas. They were typically newly built, affluent neighborhoods, often located in the suburbs, with residents predominantly being white.
  • Grade B (Blue): Deemed "Still Desirable," these neighborhoods were a mix of older and newer homes. The areas might have been somewhat dated but were still considered stable and without significant "objectionable presence."
  • Grade C (Yellow): Termed "Definitely Declining," these neighborhoods often had older homes and were more racially mixed. Lenders considered these areas as less desirable, making loans more difficult to obtain or having less favorable terms.
  • Grade D (Red): These were the "Hazardous" areas. They often had the oldest homes, were proximate to industrial areas, and had a considerable population of ethnic minorities, particularly Black residents. The term "redlining" is derived from this grade, as these neighborhoods were often outlined in red on HOLC maps. It was challenging, if not impossible, for residents of these areas to secure mortgages or loans at reasonable rates.

Land Surface Temperature (LST)
Land Surface Temperature (LST) represents the temperature of the Earth's surface and is a crucial metric obtained from remote sensing data, especially in the context of understanding urban heat and microclimates. Unlike air temperature which is commonly measured by weather stations at a standard height above the ground, LST provides a direct measure of the heat emitted from the ground, buildings, and other surfaces. In urban areas, LST can vary significantly across short distances due to variations in land cover, with built-up regions typically registering higher temperatures than vegetated areas. Monitoring LST is vital for urban planning and environmental health, as high LST values can exacerbate heat stress on populations, affect local air quality, and influence energy consumption patterns.

Normalized Difference Vegetation Index (NDVI)
The Normalized Difference Vegetation Index is a remote sensing metric used to quantify and monitor vegetation health, density, and greenness from multispectral imagery, and it's crucial for studies spanning both natural environments and urban areas. It is calculated by taking the difference between the near-infrared (NIR) and red reflectances and dividing by their sum: NDVI = (NIR - Red) / (NIR + Red). NDVI values range between -1 and 1. In urban contexts, where vegetation is intermixed with built structures, NDVI serves as a key indicator of urban greenery and heat island effects. Values close to 1 suggest dense, healthy vegetation, such as parks or green roofs; values around 0 might signify barren spaces or highly urbanized areas with minimal vegetation.

Data Processing Methods

gif of methodology

  • Step 1: Data Cleaning - A complete set of Home Owners' Loan Corporation (HOLC) geographies was obtained from the Mapping Inequality project. The dataset was cleaned to remove fields not used for analysis, and modified to fix broken geometries.
  • Step 2: Hydrography Clipping - HOLCs were clipped using a US Hydrography dataset to ensure water would not influence LST/NDVI readings.
  • Step 3: Dissolve Geography - For each Home Owners' Loan Corporation (HOLC) city/neighborhood, geographies were dissolved by grade and city for grade-level analysis.
  • Step 4: Parameter Definition - A Region of Interest was declared for individual HOLCs as well as dissolved HOLC grades by city, state. A date range for imagery was set, incorporating June, July, August (days 152-243) across 2013-2023.
  • Step 5: Curate the Landsat Collection - Landsat Collection 2, Tier 1, Level 2 image collections were selected, and reduced to ST and QA_PIXEL bands. Images were spatially filtered to the parameters defined in step 4. Additionally, Landsat 8 Collection 1 Tier 1 Raw Scenes were selected for NDVI calculations.
  • Step 6: Cloud Removal - Images were filtered to mask clouds and cloud shadows based on the QA_PIXEL band, for LST calculations. Images were then filtered to exclude those with cloud cover > 20%.
  • Step 7: Scale Factors - Scale factors were applied for deriving LST in Kelvin, Celsius, and Farenheit.
  • Step 8: Generate Analysis - From the working collection, statistics were generated for each image, including HOLC- and grade-level Land Surface Temperature
  • Step 9: Calculate NDVI - NDVI was calculated across the same filtered time and region of interest.
  • Step 10: Generate Composites - For LST composite images that represent min, mean, and max values for each HOLC and grade were generated. For NDVI, mean values for these geographies were calculated. LST was resampled from the thermal band resolution of 100m to 30m.

Notes on Deriving LST from Landsat

Processing for steps 1-3 were completed using GeoPandas in Python. Processing for steps 4-10 were completed using Google Earth Engine. There are varied approaches taken to derive land surface temperature readings from raw Landsat products. While this method uses steps defined by the NASA Applied Remote Sensing Training (ARSET) program and leverages pre-processed products for surface temperature and quality assurance, other methods were explored. For comparison, methods to derive LST using top of atmospheric (TOA) spectral radiance (Avdan & Jovanovska - 2016) were followed. LST values were on average 18.56 degrees hotter for each HOLC grade using steps defined by ARSET. Importantly, trends pertaining to LST, NDVI, and HOLCs did not change, regardless of method. Mean LST derived using TOA is found in the statistics available for download, for those interested.

Data / Tools

Data Downloads

Credits

This project was created by Michael Krisch and was supported by The Brown Institute for Media Innovation at Columbia Journalism School and Stanford Engineering School.

A special thanks to Grga Bašić at the University of Chicago, for inspiration and concept design for exploring the the use of heat as a proxy for vulnerability. And thanks to Shivani Agarwal at Columbia University, for support and guidance on scientific presentation and accuracy.