Use of Regression Models for Analyzing Highway Storm-Water Loads

Abstract

Storm-water data collected from an expressway in the Austin, Tex. area were used to develop regression models for predicting loads for a number of constituents commonly found in highway runoff. The goal of the model development was to identify the processes that affect the quality of highway runoff. Linear regression was selected as the most appropriate technique for analyzing the data because of its ability to identify constituent specific causal variables. The regression equations indicate that the majority of variations observed in highway storm-water loading can be explained by causal variables measured during the rainstorm event, the antecedent dry period, and the previous rainstorm event. Loads for each of the constituents were dependent upon a unique subset of the identified variables, indicating that processes responsible for the generation, accumulation, and washoff of storm-water pollutants are constituent specific. Loads of some constituents, such as total suspended solids, were dependent on the characteristics of the current storm, antecedent dry period, and the preceding storm indicating the importance of buildup and washoff processes. Other constituents, such as oil and grease, were dependent only on conditions during the current storm, such as runoff volume and number of vehicles during the event. The identification of constituent-specific explanatory variables suggests the type of mitigation that would be appropriate for specific constituents in non-point-source pollution control.