Abstract

The normal approach to classify urban sites for estimating stormwater characteristics is based on land use. This approach is generally accepted because it is related to the activity in the watershed, plus many site features are generally consistent within each land use. Two drainage areas with the same size, percentage of imperviousness, ground slope, sampling methods, and stormwater controls will produce different stormwater concentrations if the main activity in one watershed is an automobile manufacturing facility (industrial land use) while the other is a shopping center (commercial land use) for example. There will likely be higher concentrations of metals at the industrial site due to the manufacturing processes, while the commercial site may have higher concentrations of PAHs (polycyclic aromatic hydrocarbons) due to the frequency and numbers of customer automobiles entering and leaving the parking lots.

Previous studies indicated that there are significant differences in stormwater constituents for different land use categories (Pitt et al. 2004). This is supported for other databases like NURP (EPA 1983), CDM (Smullen and Cave, 2002) and USGS (Driver et al., 1985). The main question to be addressed in this chapter is if there is a different classification method that better describes stormwater quality, possibly by also considering such factors as geographical area (EPA Rain Zone), season, percentage of imperviousness, watershed area, type of conveyance, controls in the watershed, sampling method, and type of sample compositing, and possible interactions between these factors.

This chapter presents several approaches to explain the variability of stormwater quality using the National Stormwater Quality Database (NSQD). Several analyses had been performed using this database (Maestre et al. 2004, 2005a). Maestre et al. (2005b) for example, has shown that ignoring the non-detected observations can adversely affect the mean, median and standard deviations of the dataset, and the resulting statistical test results. The calculations presented in this chapter used the censored observations using the Cohen’s maximum likelihood method.

This paper is only available in PDF Format:

  View full text PDF