Watershed Typologies

Full Title

Identifying socio-environmental watershed typologies based on stormwater pollution using machine learning


This project represents the first socio-environmental-technological system (SETS) study of the generation of stormwater pollution in urban watersheds across the United States. Urban stormwater pollution poses a major and growing threat to local water bodies, yet its study and management has consistently ignored the human activities and behaviors that release pollution. In this project, we will combine data on population (social) and urban form (technological) along with landscape and climate factors (environmental) to model human activities. These four factors interact to drive stormwater pollution and thus underlie our SETS conceptual framework. 

We will analyze the data using machine learning clustering and classification algorithms to identify typologies of urban watersheds based on the stormwater pollution they produce. Then, we will build a regression model to predict stormwater quality based on any given SETS characteristics. We will first conduct this analysis at a broad national level, using data from 10 American metropolitan areas from 1992 to 1996, followed by conducting a detailed analysis of 3 metropolitan areas from 1992 to 2013. 

By identifying watershed typologies, we will expose relationships between SETS characteristics and stormwater quality. These results will suggest where public outreach may be needed to influence human behavior and will have implications for local urban planning policy. Overall, this project advances socio-environmental research, especially for urban areas, with its conceptual framework and methods that capture interactions between social and physical factors— with a high level of spatial detail— that together are determinants of environmental outcomes.  

Project Type
Team Synthesis Project (Graduate Student Led)
Principal Investigators
Celina Balderas Guzman, University of California, Berkeley
Matthew Smith, Harvard University
Oliver Muellerklein, University of California, Berkeley
Runzi Wang, Michigan State University
Caitlin Eger, Syracuse University