Stormwater Management Using Bayesian Networks with Satellite Imagery

Mi-Hyun Park

Department of Civil and Environmental Engineering

University of California, Los Angeles

Stormwater runoff management is important to protect receiving water quality since most

wastewater sources in the Santa Monica Bay Watershed have been treated to secondary

standards or beyond. Urban stormwater runoff has become the primary source of many

pollutants, which is caused by runoff from highly developed, impervious land use, and

managing stormwater has become the primary objective of new regulatory efforts.

However, monitoring and modeling is inherently difficult due to the lack of accumulated

data that are often site and event specific. Empirical methods to estimate stormwater

pollution have been developed using land use data. This study explored a new approach

using satellite imagery since satellite imagery provides information with high spatial and

temporal resolution at low cost. For satellite imagery classification, we used Bayesian

networks that are robust artificial intelligence tools whose structure explicitly shows the

relationships between variables, which is not provided by conventional classifiers such as

neural networks and maximum likelihood estimation. A subset of Landsat Extended

Thematic Mapper+ imagery was extracted focusing on Marina del Rey and its vicinity in

Los Angeles, CA. We used the image for classification of urban land use and for

mapping stormwater pollutant loading of selected water quality parameters (TSS, COD,

nutrients and heavy metals). The resulting maps spatially estimated the stormwater

pollutant loadings, which identified areas generating high stormwater pollutant

emissions. The results showed that stormwater pollutants were highly correlated to

impervious areas because of their high runoff coefficients, even when they had low event

mean concentrations. The maps were also used for estimating the annual mass pollutant

loading of each water quality parameter to Santa Monica Bay. The results suggest that

treatment of a small region of high pollutant loading area would achieve significant

reduction of stormwater pollution. These results are useful in developing best

management strategies for stormwater pollution and in establishing total maximum daily

loads in the watershed.