Use of Artificial Neural Networks in Predicting Highway Runoff Constituent Event Mean Concentration

In this paper, the large amount of highway runoff characterization data that were collected in california, during a 3-year monitoring season (2000-2003), were assessed in order to develop an artificial neural network (ann) model for predicting the event mean concentration (emc) of the constituent. the initial data analysis performed by a multiple linear regression (mlr) model revealed that the total event rainfall (ter), the cumulative seasonal rainfall (csr), the antecedent dry period (adp), the contributing drainage area (da) and the annual average daily traffic (aadt) were among the variables having a significant impact on the highway runoff constituent emc. these parameters were used as the basis for developing an artificial neural network (ann) model. the ann model was also used to evaluate the impact of various site and storm event variables on highway runoff constituents' emcs. the ann model has proven to be superior to the previously developed mlr model, with an improved r2 for most constituents. through the ann model, one was able to see some non-linear effects of multi variables on pollutant concentration that, otherwise, would not have been possible with a typical mlr model. for example, the results showed that copper emc is more sensitive at higher annual average daily traffic (aadt), with respect to adp, compared with lower range aadt.