Application of dimensionless sediment rating curves to predict suspended-sediment concentrations, bedload, and annual sediment loads for rivers in Minnesota

Abstract

Consistent and reliable sediment data are needed by Federal, State, and local government agencies responsible for monitoring water quality, planning river restoration, quantifying sediment budgets, and evaluating the effectiveness of sediment reduction strategies. Heightened concerns about excessive sediment in rivers and the challenge to reduce costs and eliminate data gaps has guided Federal and State interests in pursuing alternative methods for measuring suspended and bedload sediment. Simple and dependable data collection and estimation techniques are needed to generate hydraulic and water-quality information for areas where data are unavailable or difficult to collect. The U.S. Geological Survey, in cooperation with the Minnesota Pollution Control Agency and the Minnesota Department of Natural Resources, completed a study to evaluate the use of dimensionless sediment rating curves (DSRCs) to accurately predict suspended-sediment concentrations (SSCs), bedload, and annual sediment loads for selected rivers and streams in Minnesota based on data collected during 2007 through 2013. This study included the application of DSRC models developed for a small group of streams located in the San Juan River Basin near Pagosa Springs in southwestern Colorado to rivers in Minnesota. Regionally based DSRC models for Minnesota also were developed and compared to DSRC models from Pagosa Springs, Colorado, to evaluate which model provided more accurate predictions of SSCs and bedload in Minnesota. Multiple measures of goodness-of-fit were developed to assess the effectiveness of DSRC models in predicting SSC and bedload for rivers in Minnesota. More than 600 dimensionless ratio values of SSC, bedload, and streamflow were evaluated and delineated according to Pfankuch stream stability categories of “good/fair” and “poor” to develop four Minnesota-based DSRC models. The basis for Pagosa Springs and Minnesota DSRC model effectiveness was founded on measures of goodness-of-fit that included proximity of the model(s) fitted line to the 95-percent confidence intervals of the site-specific model, Nash-Sutcliffe Efficiency values, model biases, and deviation of annual sediment loads from each model to the annual sediment loads calculated from measured data. Composite plots comparing Pagosa Springs DSRCs, Minnesota DSRCs, site-specific regression models, and measured data indicated that regionally developed DSRCs (Minnesota DSRC models) more closely approximated measured data for nearly every site. Pagosa Springs DSRC models had markedly larger exponents (slopes) when compared to the Minnesota DSRC models and the site-specific regression models, and over-represent SSC and bedload at streamflows exceeding bankfull. The Nash-Sutcliffe Efficiency values for the Minnesota DSRC model for suspended-sediment concentrations closely matched Nash-Sutcliffe Efficiency values of the site-specific regression models for 12 out of 16 sites. Nash-Sutcliffe Efficiency values associated with Minnesota DSRCs were greater than those associated with Pagosa Springs DSRCs for every site except the Whitewater River near Beaver, Minnesota site. Pagosa Springs DSRC models were less accurate than the mean of the measured data at predicting SSC values for one-half of the sites for good/fair stability sites and one-half of the sites for poor stability sites. Relative model biases were calculated and determined to be substantial (greater than 5 percent) for Pagosa Springs and Minnesota models, with Minnesota models having a lower mean model bias. For predicted annual suspended-sediment loads (SSL), the Minnesota DSRC models for good/fair and poor stream stability sites more closely approximated the annual SSLs calculated from the measured data as compared to the Pagosa Springs DSRC model. Results of data analyses indicate that DSRC models developed using data collected in Minnesota were more effective at compensating for differences in individual stream characteristics across a variety of basin sizes and flow regimes than DSRC models developed using data collected for Pagosa Springs, Colorado. Minnesota DSRC models retained a substantial portion of the unique sediment signatures for most rivers, although deviations were observed for streams with limited sediment supply and for rivers in southeastern Minnesota, which had markedly larger regression exponents. Compared to Pagosa Springs DSRC models, Minnesota DSRC models had regression slopes that more closely matched the slopes of site-specific regression models, had greater Nash-Sutcliffe Efficiency values, had lower model biases, and approximated measured annual sediment loads more closely. The results presented in this report indicate that regionally based DSRCs can be used to estimate reasonably accurate values of SSC and bedload. Practitioners are cautioned that DSRC reliability is dependent on representative measures of bankfull streamflow, SSC, and bedload. It is, therefore, important that samples of SSC and bedload, which will be used for estimating SSC and bedload at the bankfull streamflow, are collected over a range of conditions that includes the ascending and descending limbs of the event hydrograph. The use of DSRC models may have substantial limitations for certain conditions. For example, DSRC models should not be used to predict SSC and sediment loads for extreme streamflows, such as those that exceed twice the bankfull streamflow value because this constitutes conditions beyond the realm of current (2016) empirical modeling capability. Also, if relations between SSC and streamflow and between bedload and streamflow are not statistically significant, DSRC models should not be used to predict SSC or bedload, as this could result in large errors. For streams that do not violate these conditions, DSRC estimates of SSC and bedload can be used for stream restoration planning and design, and for estimating annual sediment loads for streams where little or no sediment data are available.