Identification of road salt contamination using multiple regression and GIS

A multiple regression model of atmospheric deposition of salt, combined with geographic information systems (GIS) data on four classes of roads, is used to predict sodium concentrations in 162 randomly chosen streams in Massachusetts. All four classes of roads, as well as atmospheric deposition, were found to be highly significant in a model that explains 68% of the observed variation in sodium concentration. The highest salt loading rates are associated with interstate and major state roads with an estimated 22,500 and 17,700 kg of salt per kilometer, respectively. Our mass balance calculations indicate road salt is the major source of salt to the streams in Massachusetts. We examined some of the common statistical problems associated with the use of multiple regression for this type of analysis. Our confidence in the accuracy of the loading rates estimated above are limited by the collinear nature of environmental data and uncertainties related to model specification. Our results suggest multiple regression techniques can lead to overconfidence in the accuracy of the estimated loading rates and thus should not be used as the basis for policy unless the model is validated.