Climate and Punishment: Methodology and Data

The U.S. carceral system is at a critical juncture. With reporting, data analysis, and an interactive map, The Intercept’s Climate and Punishment investigation captures the scale of peril faced by America’s detention system amid a deepening climate crisis. It is our hope that risk factors for individual facilities serve as a starting point for families, community members, organizers, and journalists to ask questions of officials with power over those inside. Here’s how we collected the data, what it can tell us, and its limitations. The full dataset is available at GitHub.

About the Facility Data

Our investigation started with the Department of Homeland Security’s Homeland Infrastructure Foundation-Level Data, which describes the U.S. prison system as of June 19, 2020. The database includes a range of details about jails, prisons, juvenile detention centers, and immigration detention centers across the U.S., including the locations of the facilities’ boundaries. We focused on 6,637 institutions in the contiguous U.S.

Some of the facilities included in the DHS database are not currently operating. The agency included a “status” column that lists 613 facilities as “closed.” We found that some number of the “closed” sites appear to actually be open and some of the “open” sites are now closed. Other facilities have changed their name or location since the database’s creation.

DHS declined to answer questions about how it compiled its data. Generally, we left facility details unchanged. Given the ongoing shifts in prison openings and closures, our Climate and Punishment map includes institutions from the DHS database regardless of their status. The map’s individual facility profiles feature a label indicating whether DHS marked the prison as open or closed. The data points in our stories do not include facilities labeled “closed” by DHS. (Updated numbers were used in article texts when they became available through our inquiries to specific authorities, such as the California Department of Corrections and Rehabilitation, about systems and facilities we focused on.)

Our dataset also includes information about whether each facility is used for U.S. Immigrations and Customs Enforcement detention, based on ICE’s online list of holding facilities as of April 13, 2020. The Carceral Ecologies team at UCLA, led by Nick Shapiro, collected the ICE data and reconciled the information with the DHS prison data, then shared their analysis with us.

Some of the ICE detention facilities identified by Shapiro’s team were not part of the DHS database. Those were left out of our investigation because our flood and wildfire analyses required geographic data about institutions’ boundaries. Since two additional facilities in the DHS database also lacked that geographic information, we removed them from our analysis as well. Finally, some privately run detention facilities identified by the Carceral Ecologies team are not listed in the DHS database.

About the Heat Data

Our heat data is based on the Union of Concerned Scientists’s Killer Heat analysis, which calculated daily maximum heat indexes county by county across the U.S. The organization not only calculated the historical average number of days per year with heat indexes over various thresholds, but also offered projections of future heat indexes under various greenhouse gas emissions scenarios.

Our interactive map is based on historic heat indexes from 1971 to 2000. Building off the Union of Concerned Scientists’ map, we broke down risk severity based on a county’s average number of days in a year over 90 degrees.

• 0: minimal
• 1 to 10: 1 or low
• 11 to 25: 2 or moderate
• 26 to 50: 3 or major
• 51 to 100: 4 or severe
• 101 and up: 5 or extreme

The number of days over a particular heat index is an imperfect proxy for risk. In areas where people are not acclimated to heat or where facilities widely lack infrastructure to relieve heat, a brief spike in heat can also have serious health impacts.

The data from the Union of Concerned Scientists, which published its methodology in its full report, does not include Alaska, Hawaii, Puerto Rico, Guam, the Virgin Islands, or the Mariana Islands, because the climate models the organization used only included the contiguous U.S. The analysis is also limited because the data are averaged by county, within which heat indexes can vary significantly. Individual facilities’ building materials, cooling infrastructure, and surrounding landscape also significantly impact heat risk.

In our article and in our larger dataset, we also consider facilities’ average number of days per year with heat indexes over 105 degrees. We include projections describing how the averages for days over 90 and 105 degrees are likely to change by the end of the century under three different greenhouse gas emissions scenarios. The naming conventions in our spreadsheet are:

• Hist: Historical heat index data, covering 1971 to 2000
• Late: Late-century heat index projections, covering 2070 to 2099
• 45: A “slow-action” emissions reduction scenario, in which greenhouse gas emissions start to decline at mid-century and temperatures rise approximately 4.3 degrees Fahrenheit across the globe by the year 2100
• 85: A “no-action” emissions reduction scenario, in which greenhouse gas emissions increase throughout the century and temperatures rise nearly 8 degrees Fahrenheit by the year 2100
• Paris: A “rapid-action” emissions reduction scenario, in which temperature rise is limited to 3.6 degrees Fahrenheit (2 degrees Celsius) above pre-industrial temperatures, as prescribed by the Paris Agreement of 2015

About the Wildfire Data

To understand the wildfire risk that incarcerated people face, we turned to the U.S. Forest Service’s Wildfire Risk to Communities project. The “Risk to Potential Structures” data describes the probability that a fire will burn in a given area, as well as the potential intensity of a fire. It takes into account the type of vegetation in the area and historical climate data, including wind speed, wind direction, temperature, and humidity. Historical wildfire records covering 1992 to 2015 were used in the analysis. Climatology data was drawn from a similar historical period, from 1992 to 2016.

Since that time, wildfires have become more severe. In turn, risk to some facilities, especially in the Pacific Northwest and Northern California, may be understated in our data.

We loaded the “Risk to Potential Structures” data for each state into a spatial database alongside DHS’s geographic footprint data for each detention facility. We then ran a spatial query that buffered, or expanded, each facility’s footprint by 100 meters to better match the risk data. Next, a query was run against the “Risk to Potential Structures” data for all the wildfire risk values inside each area. The mean of those values was calculated for each facility to determine a risk value. We then compared the risk values against one another to come up with a risk percentile for each site.

• 0 to 75th percentile: minimal
• Above 75th to 80th: low
• Above 80th to 85th: moderate
• Above 85th to 90th: major
• Above 90th to 95th: severe
• Above 95th: extreme

Wildfire risk exists on a rapidly increasing curve. Three-quarters of prison structures across the U.S. have a similarly minimal chance of being hit by a wildfire. After the 80th percentile, risk increases exponentially among the most endangered facilities. On our scale, the difference between “severe” and “extreme” will be greater than the difference between “low” and “moderate.”

To stay consistent with our other datasets, we did not include Alaska or Hawaii, though wildfire risk data exists for those states. The federal wildfire data does not include U.S. territories, such as Puerto Rico, Guam, the Virgin Islands, or the Mariana Islands.

About the Flood Data

To understand the flood risks that incarcerated people face, we turned to the First Street Foundation’s flood model, which provides flood risk ratings for individual properties. The “Flood Factors,” as the group calls the ratings, account for flooding from rivers, high tides, precipitation, and storm surges and reflect both the probability of flooding and the potential depth of that flood. Some locations that have relatively low flood risk now will see higher risks within the next 30 years due to the atmospheric warming effect of greenhouse gas emissions; First Street’s flood risk scores account for that.

The model used by First Street relies on property boundaries and building footprints within those properties. We sent First Street a file with the center points of all the buildings associated with each facility listed in the DHS database, and in most cases, the group sent us a corresponding Flood Factor for each institution.

First Street designed its flood model with residential properties in mind, so prison campuses present some challenges. Since some jail, prison, or detention center campuses are expansive, the Flood Factors do not perfectly capture the range of risks to a single facility. The building to which a Flood Factor is tied, often the largest building on a property, may not be a building that houses people or that would be most impacted by flooding.

For some detention facilities, we lacked a precise Flood Factor. That’s in part because some counties, especially in rural areas where prisons are often located, do not make available the land parcel data necessary for First Street’s flood model. At the same time, some institutions are located on parcels of land that have multiple uses. In turn, the largest building on the tract may not be part of that prison. Rikers Island, for example, is a single parcel of land that holds 10 jails. First Street’s model assigned one Flood Factor to the land parcel, based on water reaching just one building footprint at one of the jails.

In cases in which First Street did not have a Flood Factor for building footprints associated with a facility, the organization’s researchers instead provided us with a Flood Factor for the nearest building with data available.

We gave each facility a “confidence level” to describe the distance between the detention facility and the location for which the risk rating was calculated.

• High: exact match
• Medium-High: less than 0.1 kilometers away
• Medium: 0.1 to 0.5 kilometers away
• Low: more than 0.5 kilometers away

Most facilities in our database are high confidence. For our analysis in the article, we included only high and medium-high confidence data. In our map, we displayed low-confidence data as “not available.”

In our full dataset, risk scores correspond to the following scale:

• 1: minimal
• 2: low
• 3 to 4: moderate
• 5 to 6: major
• 7 to 8: severe
• 9 to 10: extreme

In our map sidebar, we ranked facilities first by flood risk ratings, then by the estimated depth of a 100-year flood.

At the time of publication, the First Street data did not include Alaska, Hawaii, Puerto Rico, Guam, the Virgin Islands, or the Mariana Islands, since separate flood and climate models would apply to noncontiguous parts of the nation. Much more information about First Street’s flood modeling can be found here.