Multiple Regression Approach in Predicting Highway Runoff Pollutants Concentration

Abstract

Multiple linear regression (MLR) analysis was performed to develop mathematical relationships to predict highway runoff pollutant concentrations from various site and storm event predictor variables. Partial correlations were used to select the independent variables for the MLR models. The final “optimized” MLR model was used to generate a new fitted variable calculated as the cumulative effects of the significant predictor variables for each constituent. This fitted variable was then included as the single covariate and final MLR models were developed for each constituent. Predictor variables found to have significant impacts on highway runoff pollutant concentration include: total event rainfall (TER), cumulative seasonal rainfall (CSR), antecedent dry peri od (ADP), contributing drainage area (DA), and annual average daily traffic (AADT). The MLR models then were validated using actual highway runoff measurements that were not included in model development.