Comparing empirical sediment transport modeling approaches in Michigan rivers

Abstract

Excess or limited fluvial sediment transport can contribute to and exacerbate many environmental issues including nutrient loading, aquatic habitat degradation, flooding, channel navigation dredging, dam operation, and stream degradation or aggradation. However, fluvial sediment transport is difficult and expensive to comprehensively characterize because it can vary substantially both temporally and spatially. Having better estimates of fluvial sediment transport is important for understanding and solving these environmental issues when it is not possible to collect fluvial sediment samples. Different modeling approaches can be used to help estimate suspended sediment when sampling data are limited or unavailable. This study compared dimensionless sediment rating curves (DSRCs) developed in Pagosa Springs Colorado, Minnesota, and Michigan to determine if these DSRCs were suitable to make predictions of suspended sediment for Michigan rivers. Approximately 3,000 suspended sediment samples collected in or near Michigan from the mid-1960s through August 2022 were used to develop two DSRC models. The DSRCs developed in Michigan include a pooled DSRC model which uses nonlinear least squares regression, and a mixed-effects DSRC model which uses a mixed-effects modeling approach. In general, there was not a noticeable improvement in the performance of the Michigan mixed-effects DSRC model over the Michigan pooled DSRC model. The two Michigan DSRCs were evaluated against DSRCs developed for Pagosa Springs and Minnesota. The results showed DSRC models developed from Minnesota and Michigan were similar to each other. In contrast, the Pagosa Springs DSRC predicts higher suspended-sediment concentration (SSC) at low flows and increases at a higher rate due to having a greater exponent. The Pagosa Springs DSRC produces higher SSC predictions that do not approximate the observed data well at most of the Michigan sites in the study. The results suggest that the Pagosa Springs DSRC was not suitable to make predictions of suspended sediment for Michigan rivers. The similarity of the DSRC equations developed for Minnesota and Michigan compared to the Pagosa Springs DSRC equation suggest that there may be regional patterns of SSC in the upper Midwest rivers that differ from those in other areas of the country like Pagosa Springs. A regionally applicable model could be developed and strengthened by combining data from additional midwestern states. Since the Michigan DSRCs goodness-of-fit metrics were comparable to the site-specific simple linear regressions (SLRs) and outperformed them in the aggregate goodness-of-fit metrics, the Michigan DSRCs are suitable to make predictions of suspended sediment in Michigan rivers with limited data. However, the availability of the DSRCs from this study should not diminish the value of collecting physical samples and exploring alternative modeling approaches because of the uncertainty associated with using DSRCs. Approximately 3,000 suspended sediment samples collected in or near Michigan from the mid-1960s through August 2022 were used to develop two DSRC models. The DSRCs developed in Michigan include a pooled DSRC model which uses nonlinear least squares regression, and a mixed-effects DSRC model which uses a mixed-effects modeling approach. In general, there was not a noticeable improvement in the performance of the Michigan mixed-effects DSRC model over the Michigan pooled DSRC model. The two Michigan DSRCs were evaluated against DSRCs developed for Pagosa Springs and Minnesota. The results showed DSRC models developed from Minnesota and Michigan were similar to each other. In contrast, the Pagosa Springs DSRC predicts higher suspended-sediment concentration (SSC) at low flows and increases at a higher rate due to having a greater exponent. The Pagosa Springs DSRC produces higher SSC predictions that do not approximate the observed data well at most of the Michigan sites in the study. The results suggest that the Pagosa Springs DSRC was not suitable to make predictions of suspended sediment for Michigan rivers. The similarity of the DSRC equations developed for Minnesota and Michigan compared to the Pagosa Springs DSRC equation suggest that there may be regional patterns of SSC in the upper Midwest rivers that differ from those in other areas of the country like Pagosa Springs. A regionally applicable model could be developed and strengthened by combining data from additional midwestern states. Since the Michigan DSRCs goodness-of-fit metrics were comparable to the site-specific simple linear regressions (SLRs) and outperformed them in the aggregate goodness-of-fit metrics, the Michigan DSRCs are suitable to make predictions of suspended sediment in Michigan rivers with limited data. However, the availability of the DSRCs from this study should not diminish the value of collecting physical samples and exploring alternative modeling approaches because of the uncertainty associated with using DSRCs.