Identifying main areas generating stormwater pollution from satellite imagery and ancillary data

Abstract

Urban stormwater runoff has been recognized as a major source of pollutants in Santa Monica Bay watershed because of the watershed's highly developed, impervious landuses. Estimating stormwater pollutant mass emissions is inherently difficult and often requires land use information. Many approaches have been developed to estimate land use from satellite imagery. This research uses an alternative approach, which estimates stormwater pollutant loadings from satellite imagery. We classified a Landsat ETM/sup +/ image of the areas in the Santa Monica Bay watershed using Bayesian networks. Identified water quality parameters were heavy metals such as copper, lead and zinc, and oil and grease. The parameters were classified into low, medium and high loads and the most polluting areas are the target area to be identified. We examined the effect of incorporating ancillary data: locational data, such as X and Y coordinate values of each pixel, and elevation data such as SRTM. The results show that incorporating coordinate values or elevation data improves overall accuracy and lowers the omission error compared with using Landsat ETM/sup +/ spectral data only. In fact, incorporating coordinate values was more useful than incorporating elevational data to provide better accuracies and less omission errors. These results will be useful in developing best management strategies for stormwater pollution.