Authors: Celina Balderas Guzman*, University of California, Berkeley, Runzi Wang, Michigan State University, Oliver Muellerklein, University of California, Berkeley, Caitlin Eger, Syracuse University, Matthew Smith, Florida International University
Topics: Water Resources and Hydrology, Quantitative Methods, Urban and Regional Planning
Keywords: stormwater, watersheds, machine learning, urban development
Session Type: Paper
Presentation File: No File Uploaded
Urban stormwater jeopardizes the health of water bodies worldwide. Despite local efforts to mitigate water quality, there remains a widespread disconnect between the types of contaminants in stormwater, the landscape through which they are transported, and the human activities that influence them. The large spatio-temporal heterogeneity of stormwater pollution stems from differences in urban development and small-scale activities that make source identification of contaminants difficult. In this environmental data science study combining urban planning, biogeochemistry, and hydrology, we apply machine learning techniques to a national-level analysis examining the relationship between watersheds and the stormwater pollution they produce.
We analyzed contaminant concentration data from stormwater samples taken from over 1,000 storm events across 25 American cities from 1992 to 2003. First, using a clustering algorithm, we identified 6 stormwater “signatures”— distinct combinations of 9 common contaminants found in stormwater samples. Next, we combined the stormwater quality data with data from multiple sources detailing landscape, climate, and weather characteristics of the watershed. Using a classification algorithm, we analyzed the combined data to determine how location, storm event, and watershed characteristics are associated with each stormwater signature. Our data, Python workflow, and supplementary Python libraries will be publicly available.
The results yielded a set of watershed typologies that correspond to unique stormwater signatures that are reproducible from regional landscape and climate characteristics. These typologies show the importance of place and time in the generation of stormwater pollution and inform best management practices to mitigate water quality at local and regional scales.
To access contact information login