Abstract
This study was conducted to determine if equations could be derived using multiple linear regression (MLR) to predict the removal of total and dissolved copper and zinc by vegetated biofilter strips treating highway runoff. The regression analysis was based on the data set collected during the Caltrans Roadside Vegetative Treatment Site (RVTS) Study. Eight RVTS study sites distributed across California and storms from several years were included. The predictors chosen for the MLR analysis were strip slope, strip width, vegetation coverage, percentage of clay content in the soil, rainfall duration, total event precipitation, and antecedent dry days. The predictands chosen were effluent concentration (Ce), concentration reduction (Ci-Ce), and fraction of concentration remaining (Ce/Ci). In a second analysis, a first order removal model was assumed to fit the field data, resulting in concentrations that should decline exponentially with strip width. MLR analysis was used to develop predictive equations for the exponential decay coefficients based on the same predictors (except width). Regression models were evaluated using criteria such as the value of the coefficient of determination (R2), whether or not the sign of the predictor coefficient matched expectations from physical processes and how well the equations conformed to MLR assumptions. Not all predictors proved to be statistically significant. Predictive equations were produced for effluent concentrations of total copper, total zinc, and dissolved zinc based on vegetation coverage, rainfall duration, total event precipitation, and antecedent dry days. The best equation for effluent dissolved copper concentrations was based on only vegetation coverage, rainfall duration, and antecedent dry days. The coefficient of determination (R2) values for these equations were 0.348 to 0.523. R2 values for the best equations predicting Ci-Ce and Ce/Ci were much lower. When the predictive Ce equations were graphed against vegetation coverage, dissolved copper and zinc concentrations are higher than those for total copper and zinc, thus these equations are unreliable. For the first order decay coefficient, the best fit equation had a low R2 value and the regression model could not meet all the assumptions of the MLR. Thus, these equations are also unreliable.