Although considerable effort has been expended during the past two decades to control nonpoint-source contamination of streams and lakes in urban and rural watersheds, little has been published on the effectiveness of various management practices at the watershed scale. This report presents a discussion of several parametric and nonparametric statistical techniques for detecting changes in water-chemistry data. The need for reducing the influence of natural variability was recognized and accomplished through the use of regression equations. Traditional analyses have focused on fixed-frequency instantaneous concentration data; this report describes the use of storm load data as an alternative.
Selected statistical techniques were applied to three urban watersheds in Texas and Minnesota and three rural watersheds in Illinois. For the urban watersheds, single- and paired-site data-collection strategies were considered. The paired-site strategy was much more effective than the singlesite strategy for detecting changes. Analysis of storm load regression residuals demonstrated the potential utility of regressions for variability reduction. For the rural watersheds, none of the selected techniques were effective at identifying changes, primarily due to a small degree of management-practice implementation, potential errors introduced through the estimation of storm load, and small sample sizes. A Monte Carlo sensitivity analysis was used to determine the percent change in water chemistry that could be detected for each watershed. In most instances, the use of regressions improved the ability to detect changes.