BAYESIAN PREDICTION OF FLUVIAL TRANSPORT PARAMETERS (Invited Presentation)

Fluvial sediment transport and flow conditions can be predicted from water discharge, slope, and grain size, but physically-based predictive models cannot be applied when one or more variables are unknown. Practical problems rarely conform to physical theory, and it is often necessary to estimate unknown variables from limited data. For example, suspended sediment concentration and grain size are regularly measured using acoustic or physical sampling techniques, but bed grain size and bedload flux are difficult to measure directly. Empirical models provide a useful substitute for estimating unknown parameters in the absence of physically-based relations.

We utilize Bayesian inference to constrain power-law expressions for bedload flux and bed material grain size from routinely measured variables in sand bed rivers. Grouped and hierarchical models allow for prediction from both global datasets and continuous records at a single site. This approach is demonstrated in two scenarios. First, we simulate bedload flux and bed material grain size at the Diamond Creek Sediment monitoring station on the Colorado River. This station has an eight-year record of high temporal-resolution (fifteen minute) suspended sediment data. Next, we estimate bedload flux and bed material grain size at all USGS sediment monitoring stations that report suspended sediment concentration. We find that conventional methods of estimating bedload flux fail to capture fluctuations driven by the interaction between flow strength and sediment supply and can introduce large persistent biases to estimates of total load.