Using experimental tsunami deposits to evaluate and reduce uncertainty in hydraulic reconstructions

A goal of paleotsunami research is to quantitatively reconstruct wave hydraulics from sediment deposits. Inversion models tend to be based on assumptions about sediment transport and open channel flow that may poorly describe tsunamis. Published hydraulic reconstructions are reasonably consistent with some recent and historical tsunamis, but little flow data exist with which to independently evaluate model accuracies. We conducted controlled flume experiments to measure flow hydraulics, source grain size distributions and deposit characteristics. Scaled tsunamis were created in a 32 m flume by raising a computer-controlled liftgate and releasing ~6 m3 of impounded water. The resulting bore entrained sediment from an upstream "source dune" and deposited it downstream. Bores produced thicker bedload waves in the upstream ~1/3 of the flume, followed by thin suspension-dominated deposits that fined downstream. First, we compare experimental velocities to predictions of TsuSedMod, a previously published inversion model. Assuming a hydraulic roughness for flow over ponded water of Manning's n=0.025 (consistent with a previously published inundation model), TsuSedMod tended to overpredict flow velocity by a factor of 2 to 3. Model results are fairly sensitive to hydraulic roughness over the experimental parameter space. Second, a previously proposed advection-settling model based on suspended D95 grain sizes is also applied to the experimental data to quantify model uncertainty. At 95% confidence, the model predicted time-averaged flow depths to within nearly a factor of two, and time-averaged flow velocities to within a factor of 1.5. Finally, we find that the advection-settling model tends to more accurately predict depths and velocities for our data if the median grain size (D50) is used, rather than D95. We propose a mechanistic interpretation for why D50 may give more accurate results than D95 when using advection-settling models to infer flow hydraulics from suspension-dominated deposits.