Quality Sampling Strategy

Water quality monitoring programs have expanded in an effort to quantify loadings to streams and lakes from various watershed activities and managements. At the core of monitoring programs are strategies or schemes that determine when and how samples are taken for estimating stream loadings. Quantification of the differences between these schemes has not been adequately documented. An analytic approach was used to evaluate 45 commonly used sampling strategies that included time–based (5, 10, 15, 30, 60, 120, 180, 300, and 360 min) and flow–stratified (2.5, 5.0, 7.5, 10.0, 12.5, and 15.0 mm) schemes using discrete and composite sampling procedures. A total of 300 storm hydrographs from 87 different watersheds in the U.S. were coupled with two concentration graphs (a 100% positive correlation of concentration to flow, and a 100% negative correlation to flow) to estimate average bias values for each sampling strategy. The mean bias and absolute error for time–based sampling, as determined by the standardized root mean square error (SRMSE), always increased with a greater sampling time interval. For time–based sampling, a positive correlated concentration graph generally resulted in under–prediction (positive bias) from the true load, while a negative correlated concentration always resulted in over–prediction (negative bias). For flow–stratified sampling, the direction of bias was generally reversed from the time–based case, but the SRMSE increased with a greater flow interval. Even at the lowest flow interval used in this study (2.5 mm), the median residual values were significantly different from zero ( = 0.05). Time discrete sampling schemes <15–min provided the only bias and mean residual values not significantly different from zero ( = 0.05). When an equal number of samples was obtained, the flow–stratified approach had less absolute error than did the time–based approach. The results indicate that, prior to water quality monitoring, careful consideration should be given to the sampling strategy and its overall impact on load estimates.