Evaluating the Demographic and Geographic Characteristics of Green Stormwater Infrastructure

Over the past few decades, stormwater managers across the United States have increasingly turned to green stormwater infrastructure to mitigate flooding and improve the quality of stormwater runoff. Green infrastructure also offers a variety of co-benefits to the surrounding community compared with traditional gray infrastructure, including reduced urban heat island, improved water quality, and enhanced aesthetics.

Many cities have invested in green stormwater infrastructure directly or established incentive programs for property owners and developers to reduce stormwater runoff while providing these co-benefits to their communities. However, after several decades of green infrastructure investment and incentives, the question remains: Has green infrastructure been placed in areas where its residents stand to benefit the most?

A rain garden in Brooklyn, New York.

Photo by Alyson Youngblood/RAND Corporation

This study used an exploratory machine learning–based approach to evaluate green stormwater infrastructure investment across five cities in the United States—Boston, Detroit, New York City, Pittsburgh, and Washington, D.C. We compared the location of installed green stormwater infrastructure with demographic and land use characteristics of the surrounding area to understand whether green stormwater infrastructure is located in areas that, in addition to stormwater reduction, stand to gain from the co-benefits these investments provide.

A summary of the local characteristics that were frequently found in places with more green stormwater infrastructure assets (by number) for each city is shown in the graphic below.

Technical Approach

This graphic presents the findings of our machine–learning based analysis. To carry out this work, we collected city-level data on the types, sizes, and locations of existing investments to understand how much and where cities had installed green stormwater infrastructure. We also collected information to understand local characteristics—or the socioeconomic, geographic, and physical landscape—surrounding these investments in each city.

This information included data, for example, on income, race, climate, and infrastructure condition. These datasets were cleaned, combined, and used as inputs to a variety of popular machine learning methods. Using seven different machine learning methods,^[1] a model was trained to quantify the relationship between green stormwater infrastructure and local demographic and geographic characteristics. With this approach, we were also able to compare the goodness of fit for each model and examine the strength of the association between green stormwater infrastructure and each local characteristic.

The graphic above shows the results for the Random Forest model, which generally performed well across all cities. More information on the study, datasets, and our quantitative approach can be found in our full report.

Notes

1. The following approaches were used: Lasso Regression, ElasticNet, Gradient Boosted Decision Trees, Random Forest, Support Vector Regressor, and Shallow and Deep Neural Networks.Return to content ⤴

Credits

Alyson Youngblood (design) and Shawna Templeton (production)

About the Research

Topics

• Artificial Intelligence
• Environmental and Natural Resource Management
• Environmental Regulation
• Infrastructure