Practical Application and Evaluation of Usace Sediment Models On The Flood-Prone Lower White River In Washington State

The channelized alluvial fan reach of the Lower White River in Washington State has been subjugated to over a century of large-scale river engineering “improvement” projects in response to a catastrophic avulsion of floodwaters into the Puyallup Basin from the Green River Basin in 1906. As documented by the USGS (Czuba et al. 2010, 2012), this reach located 68 miles downstream of Mt. Rainier, has been losing channel capacity from ongoing sediment deposition, impacting floodplain residents, emergency managers and reservoir operations at Mud Mountain Dam alike. This paper is focused on evaluation of the performance of several sediment models used to predict reach scale sedimentation in the lower White River, with an emphasis on recent improvements in the HEC-RAS software. The Lower White River was originally modeled by the USGS in the late 1980s, which was comprehensively updated in the late-2000’s (Czuba et al. 2010), which was followed by project specific updates by the Corps (USACE 2017) which led to a detailed 1D and 2D sediment model investigation of a large-scale levee setback project (Jones et al. 2018) that was implemented after completion of the study. Five years have elapsed since completion of the setback project allowing for post-project validation of the original models, recalibration of updated models to observed conditions, and forecasts of future conditions (channel capacity trends) relevant to real-time reservoir operations and emergency management. Primary findings are: 2D models are most informative for predicting geomorphic responses relevant to design of mainstem diversions into restored floodplains (capturing the spatial distribution of scour and deposition and bed material variation); 1D models are best for efficient analysis of long term sediment budgets and trends but are highly sensitive to small input variations; the Corps’ AdH 2D sediment model is robust and able to capture the physical response of the river to the setback with reasonable runtimes due to the HPC capability; RAS 2D sediment engine (beta) shows promise due to stability, ease of use/setup, and ability to capture spatial sedimentation patterns, but long compute times limit its practical application (at present). All post-project validation models were partially confounded by several factors which are presently outside the ability of numerical models to capture including vegetation recruitment and colonization, large wood recruitment and jam formation, and resultant influence on depositional and erosional processes and rates. On small and medium sized rivers these factors can exert significant influence on post project responses suggesting that numerical sediment transport models could benefit from inclusion of large wood and vegetation dynamics.