Comparison of two regression techniques for the estimation of contaminant loadings to streams

Abstract

The study describes two simple methods to estimate contaminant loads from point and diffuse sources and to separate groundwater and surface contributions to streams and compares the results of two regression techniques applied to these methods. The methods are composed of expressions whose coefficients are estimated by regression techniques using monitoring data from stations situated along a stream. The Kendall-Theil Robust Line is shown to give comparable results to its parametric counterpart without the need to eliminate records from the data. Parametric regression is burdened by outliers, leverage and influential points which also, besides influencing slope and intercept, give rise to non-normal residuals, thus making hypothesis testing difficult. The methods are applied to a number of conventional contaminants. Monitoring data from a heavily polluted stream and from streams draining natural salt deposit areas in Turkey are utilized.