Estimating solute loads from a small, urban watershed using a semiparametric regression approach

Abstract:

Accurate estimates of solute loads from small, urban watersheds are important, as these watersheds contribute large amounts of many pollutants relative to their size. However, many traditional regression methods for load estimation based on observed relationships with discharge, time, and season are known to be poorly suited to such watersheds. Using data from a tributary of a water-supply reservoir in central Massachusetts, we show that selection of appropriate hydrologic predictor variables can improve model fit and reduce load estimate uncertainties in urban watersheds. We found that the largest improvements were made by including 1-day change in stream discharge as a predictor variable and by using an exponential smoothing function to capture impacts of antecedent flow conditions. These variables better encapsulate storm conditions in flashy streams than discharge alone. Further, they are simple functions of discharge, facilitating their application to any stream for which a continuous record of discharge exists. In the case of nitrate-N and total organic carbon, model uncertainty was further reduced by introducing flexibility through the use of nonparametric smoothing of certain predictor variables. The results of this study demonstrate the need for careful selection of model structure when estimating solute loads in small, urban watersheds, as well as the potential for improved estimate accuracy through appropriate selection of model forms.