Simulation and control of morphological changes due to dam removal in the Sandy River , Oregon , USA

A one-dimensional channel evolution simulation model (CCHE1D) is applied to assess morphological changes in a reach of the Sandy River, Oregon, USA, due to the Marmot Dam removal in 2007. Sediment transport model parameters (e.g. sediment transport capacity, bed roughness coefficient) were calibrated using observed bed changes after the dam removal. The validated model is then applied to assess long-term morphological changes in response to a 10-year hydrograph selected from historical storm water records. The long-term assessment of sedimentation gives a reasonable prediction of morphological changes, expanding erosion in reservoir and growing deposition immediately downstream of the dam site. This prediction result can be used for managing and planning river sedimentation after dam removal. A simulation-based optimization model is also applied to determine the optimal sediment release rates during dam-removal that will minimize the morphological changes in the downstream reaches.


INTRODUCTION
Dam removal projects in the United States are becoming increasingly common for river restoration purposes.However, the downstream impact of sediments released from the reservoir post dam removal is a major concern.For reservoirs with large sediment deposits, released sediments may significantly change the river morphology in both the reservoir and the downstream channels.If reservoir sediment deposits contain contaminants, sediment release may cause further environmental hazards downstream.Before decommissioning a dam, it is important to study sediment transport and morphological changes upstream and downstream of the dam site that result from dam removal, in order to find a better plan for managing the sediments to have a lower impact of the sediment release on the downstream.
Sediment transport processes following dam removal are highly complicated due to non-uniform spatial distribution of sediment in reservoirs and downstream and nonlinear dynamics of river flows and sediment transport in space and time.Quick release of reservoir sediments by a complete dam removal (i.e.without control of removal schedule and sediment release rate) may result in a supercritical flow in the river reach near the knickpoint, or the immediate downstream of the dam site.Eroded sediments from the reservoir usually deposit immediately in the downstream of the knickpoint.The rates of erosion inside the reservoir and the deposition downstream vary with time.A headcut may be triggered and migrate upstream through delta deposits if reservoir bed materials are cohesive.As a result, bank erosion and channel widening are important adjustment processes.Even if the dam removal process is gradual, river floodwaters during a storm may entrain a large portion of the deposits and pass it over the low-head dam.Furthermore, the bio-geomorphic impact of dam removal on the riverine environment is long-term and long-range both upstream and downstream.Thus, the assessment of potential impacts of dam removal on river hydrological and geomorphological environments requires integrated model capabilities to simulate multi-scale processes in river flow dynamics, morphodynamics, and ecosystem services.
Engineering planning and decision-making dealing with dam removal rely heavily on predictive simulation results to assess environmental impacts due to releases of sediment and water from reservoirs.A key challenge in the dam removal modelling is the simultaneous modelling of flow and sediment transport in reservoir and downstream reaches.This is because simultaneous modelling of reaches upstream and downstream of the dam needs to overcome the difficulties in simulating flow over very steep bed slopes, which is expected at the downstream portion of the reservoir sediment wedge immediately following a dam removal.In this transitional area around the knickpoint between reaches upstream and downstream of a dam, very steep slopes will probably generate transient/transcritical flow conditions, potentially resulting in numerical instabilities.
Most studies of riverine environmental impacts due to dam removal projects were assessed using one-dimensional (1-D) integrated hydrodynamic and morphodynamic models to predict sediment transport and morphological changes.Ding et al. (2013a) have applied a 1-D river flow and sediment transport model, CCHE1D, to assess morphological changes in a reach of the Sandy River, Oregon due to the Marmot Dam removal in October of 2007.This 1-D model was validated by simulating the sedimentation processes due to the release of the reservoir sediment release.Several sediment transport capacity formulations (e.g.SEDTRA formula, Ackers-White formula, Engelund and Hansen's formula, Wu-Wang-Jia's formula, etc.) were tested to find out their accuracy for simulating the sediment erosion and deposition due to the dam removal and sediment releases.CCHE1D is an integrated flow and morphodynamic model developed by the National Center for Computational Hydroscience and Engineering (NCCHE) at The University of Mississippi (Wu and Vieira, 2002;Wu et al., 2004).The model simulates non-uniform sediment transport, bed aggradation and degradation, bed material composition (hydraulic sorting and armouring), and bank erosion.CCHE1D is particularly effective in the estimation of long-term sediment loads and morphological changes in channel networks, evaluation of the effects of erosion control and channel remediation measures on sediment yield, and analysis of the influence of land use changes and agricultural management practices on sedimentation.The flow model in CCHE1D simulates flows that must be primarily subcritical in all reaches of the channel network.However, it can handle local supercritical and transcritical flows without hydraulic jumps in few cross-sections through a hybrid dynamic/diffusive wave model.The diffusive wave model is used for high Froude number (near critical and supercritical flow conditions).This hybrid dynamicdiffusive wave model can avoid numerical instability of the dynamic wave model in the transcritical flow regions and significantly reduce the computation time.Langendoen et al. (2005) applied the CONCEPTS (CONservational Channel Evolution and Pollutant Transport System) channel-evolution model to simulate post dam removal sediment dynamics along the Kalamazoo River between Otsego and Plainwell, Michigan.The erosion rate due to hydraulic forces in fine-grained, in-situ materials was determined from the field data using the submerged jet tests.This model is also capable of simulating unsteady river flows, nonuniform sediment transport, headcut advancement, and bank erosion processes.Cui et al. (2006a, b) developed the Dam Removal Express Assessment Models (DREAM): DREAM-1 for simulation of dam removal with the reservoir sediment composed of primarily fine sediment, and DREAM-2 for simulation of dam removal with the top layer of the reservoir sediment composed of primarily coarse sediment (gravel and coarser).These models are based on the separation of sand and gravel components in the modelling of sediment transport, rather than modelling of a mixed sand/gravel mixture as sand and gravel affect each other.Cui and Wilcox (2008) applied the DREAM models to assess the impacts on the river morphology due to the Marmot Dam removal project in the Sandy River, Oregon.Konrad (2009) used a 1-D movable boundary sediment-transport model to simulate morphological changes and sediment particle-size distribution variations after dam removal in the Elwha River, Washington.The Meyer-Peter and Muller formula was used to calculate sediment transport capacity in the river.
In the following of this paper, it presents the application of CCHE1D to assess morphological changes in a reach of the Sandy River, Oregon due to the Marmot Dam removal in the later summer of 2007.The model validation is reported by calibrating the model parameters such as bed roughness coefficients and selecting a better sediment transport capacity rate formulation from several commonly-used formulae for alluvial rivers.Then, the validated model is applied to assess long-term morphological changes in response to a synthetic 10-year-long hydrograph selected from historical storm water records.
To provide the most effective management plan to decommission the dam and release the reservoir deposits, it would be best if one could establish the optimal reservoir sediment release schedule (or the time-dependent controlled released process) so that the downstream impact of sedimentation can be minimized.Thus, this paper further presents a simulation-based optimization tool (Ding et al. 2013) to search for optimal control of sediment releases for minimization of morphological changes in the Sandy River after the Marmot dam was removed.This integrated optimization tool consists of a channel evolution simulation model (CCHE1D) and a variational adjoint sensitivity model for optimization of nonlinear flow and sediment transport processes.An adjoint sensitivity model for nonlinear and unsteady flow dynamics and morphodynamics was developed to quickly search for the best solution of the optimal control schedules of flow and sediment transport.The optimal result of controlled sediment release during the dam removal can provide engineering management guidance to plan a better scheduled dam decommissioning if releasing reservoir deposits can be operated.

DESCRIPTION OF SIMULATION MODEL: CCHE1D
CCHE1D, named for its research institute (Center for Computational Hydroscience and Engineering), is an integrated 1-D channel network model with the landscape analysis tool TOPAZ (Topographic PArametriZation) and the watershed models AGNPS (Agricultural Non-Point Source Pollution Model) (Young et al. 1987) and AnnAGNPS (Bingner and Theurer 2009), as well as SWAT (Soil and Water Assessment Tool) (Arnold et al. 1993).The models are managed through a graphical user interface based on ArcGIS and/or ArcView (Wu and Vieira 2002).It is capable of simulating unsteady flow and sediment transport processes in single channels or dendritic channel networks.The CCHE1D flow model simulates unsteady flow in channel networks using either the diffusive wave model or the dynamic wave model, taking into account the difference between the flows in main channel and floodplains of a compound channel, and the influence of hydraulic structures.For local supercritical and transcritical flows (without hydraulic jumps), the hybrid dynamic/diffusive wave model is applied.
In terms of the non-equilibrium transport modelling, the CCHE1D sediment transport model simulates non-uniform sediment transport, bed aggradation and degradation, bed material composition (hydraulic sorting and armoring), and bank erosion.In the model, multiple options for sediment-related parameters have been implemented.Four sediment transport capacity formulae are provided in the model, i.e.SEDTRA model (Garbrecht et al. 1995), Wu-Wang-Jia formula (2000), a modified Ackers and White's formula (Proffitt and Sutherland 1983), and a modified Engelund and Hansen's formula ( Engelund and Hansen 1967).The other parameters such as bedmaterial porosity, non-equilibrium adaptation length, and vertical mixing layer thickness can also be calculated by existing formula or given by users.The implementation of these options allows users to select the most appropriate formulae for practical river engineering problems, and allows the model to be widely utilized.For the details of CCHE1D, one may refer to the CCHE1D technical manual (Wu and Vieira 2002).

IMPACT ASSESSMENT OF THE MARMOT DAM REMOVAL IN SANDY RIVER
Marmot Dam was the only dam on the main-stem of the Sandy River, Oregon (Figure 1).The Sandy River extends approximately 89 km from its headwater to its confluence with the Columbia River.The dam was located near the middle of the basin at RM 30, downstream of the Salmon River confluence but upstream of the Bull Run River confluence and 2 km upstream of the entrance to the Sandy River gorge.The Bull Run River is the largest tributary of the Sandy River.Marmot Dam, a 47-foot-high, 345-foot-long roller-compacted concrete dam, was built in 1989 to replace an earlier timber structure, which originally constructed in 1913.Portland General Electric (PGE), the owner and operator of the dam, decided to build a temporary cofferdam upstream of Marmot Dam to facilitate the removal of the dam body and prevent the disturbance of the deposited sediment behind the dam.After the dam was removed, the cofferdam was breached to allow the 80-km-long Sandy River to flow freely from Mount Hood, Oregon, to its confluence with the Columbia River.The breaching of the cofferdam was scheduled for October 19, 2007, between seasonal fish runs, to minimize adverse impacts to resident and migratory fish.Associates, 2000).The deposited sediments were mainly composed of a surface gravel layer (Unit 1), a sand layer underneath (Unit 2), and the pre-dam channel bed representing the third distinct unit.Figure 2 provides the size composition of the reservoir sediments for the two prescribed layers.The uppermost unit (Unit 1) ranged from approximately 2-5.5 m in thickness and was composed of sandy gravel with a small amount of cobbles and boulders, becoming thicker toward the dam.The next unit (Unit 2) was predominantly fine sediment (silty sand to sand with a small amount of gravel, ranging from 4 to 11 m in thickness).The pre-dam channel, below Unit 2, consisted primarily of coarse sediment and its thickness ranged from 0.8 to 3 m.At the time of decommissioning, the 3-km-long reservoir behind the dam was filled with nearly 750,000 m 3 sand and gravel, a volume equivalent to about 5 to 10 years of average annual sediment load.Most sediment transported by the river had been passing the dam for decades.The sediment transport rate was about 250,000 ton/year at the Marmot Dam, of which the majority was fine sediment (Stillwater Sciences, 2000).To establish a 1-D model of the Sandy River including the dam and the reservoir, the surveyed cross-section data provided by PGE (Stillwater Sciences, 2000) have been used to generate crosssections for the numerical simulation of hydrodynamic and morphodynamic processes in the river.The study reach (i.e.computational domain) extends from 2.5 km upstream of the dam site to 18.0 km of its downstream.To improve the simulation results, a number of cross-sections have been To Columbia River linearly interpolated or extrapolated in several subreaches.The spacing between cross-sections varies from 12 m to 325 m and a total of 126 cross-sections are used in this model.
A discharge series spanning the period of model runs is required as input hydrograph to the model.It is found that the USGS operates several stream flow gauges on the Sandy River.The daily discharge data from the Sandy River is available at the Marmot gauge (USGS station number 14137000), which is located 0.5 km upstream of the Marmot Dam site, and has been in operation since 1911.This station's hydrograph is used as a boundary condition at the inlet of the study reach.The maximum average daily discharge in the river in the first year after the dam removal from Oct. 19, 2007to Sept. 30, 2008 reached 180 m 3 /s.Another USGS hydrologic station (number 14142500) is located 0.2 km downstream of the Bull Run River confluence.For specifying the water depth boundary condition at the downstream end for the numerical model, the water depths at the station are calculated by assuming a state of uniform flow at a cross-section.A base flow of 5.0 m 3 /s has been assumed as the initial flow condition.
For non-uniform sediment transport modelling, the sediment size distributions, Unit 1 and Unit 2 in Figure 2, are used for creating sediment compositions in the reach behind the dam.Unit 1 is assumed to be the bed compositions (i.e.sandy gravel with some cobbles and boulders) in the downstream cross sections from the dam site.The graded sediments in the model are divided into 12 size classes.The representative sizes of the sediments vary from 0.09 mm to 188.34mm.
The calibration of those model parameters including the selection of sediment transport capacity formulae in CCHE1D with the multiple size classes was carried out by simulating the flow and sediment transport starting at the moment of the dam removal, Oct. 19, 2007to Sept. 30, 2008.To investigate the performance of the sediment transport capacity formulae, all the four available formulae in CCHE1D were examined by computing the first year morphodynamic processes.The intercomparisons of average bed elevation changes after the one year computed by the four formulae are presented in Figure 3.All the capacity formulae give reasonable bed change patterns in comparison with the observed bed changes after the first one year natural flush of reservoir sediments: the erosion before the knickpoint and the deposition downstream are well simulated.By evaluating the bed changes in both upstream and downstream, it is found that the Engelund and Hansen's formula produced a slower erosion rate in the reservoir, but an overestimated change in the downstream reaches; Wu-Wang-Jia's formula provided better morphological changes in the whole study reach.Therefore, Wu-Wang-Jia's formula was used for the following studies of modal validation and the long-term assessment of morphological changes in the Sandy River.Then, the distribution of the bed roughness coefficients along the river reach was also fine-tuned.As shown in Figure 4, with the range of the roughness value from 0.03 to 0.06 varying from upstream to downstream, it looks like that the bed changes downstream are more sensitive to the roughness coefficient.It indicates that the large roughness value is appropriate to the downstream reach because the bed surface pre dam removal was armoured and consisted mainly of cobbles and boulders, and the morphology downstream is more complex.By using the calibrated model parameters and the selected sediment transport capacity formula (Wu-Wang-Jia's formula), a 10-year-long impact assessment of morphological changes in the Sandy River after the Marmot Dam Removal was performed.Similar to the approach implemented by Stillwater Sciences (2000) and Cui and Wilcox (2008), a 10-year long period containing ten yearly hydrographs was assumed to be the synthetical hydrological forcing in the river.The recorded first-year (i.e., 2008) hydrograph after the dam removal is used as the input for the first year of simulation.The other nine years following the first year were selected randomly by Stillwater Science (2000) and Cui and Wilcox (2008) from the historic yearly records in the Sandy River to represent the normal and extreme flow events.The peak discharge in the 10-year hydrograph varies from 180.4m 3 /s in the first year to 390.8 m 3 /s in the 10 th year.
For the 10-year average bed changes, Figure 5 shows that there is an increasing erosion upstream (Figure 5a), and a continuing aggradation in the immediate downstream of the dam (Figure 5b).It is obvious that the travelling wave of the eroded sediments from the reservoir causes these morphological changes.The rate of upstream bed erosion seems slightly slower than that of deposition downstream; One of reasons is that the downstream cross-sections are narrower than those upstream inside the reservoir.Meanwhile, erosion is also highly related to the initial bed material size.It is expected that following dam removal, the slope of the riverbed would gradually return to the pre dam condition.The results show that the bed changes from year 5 to year 10 are much more than those from year 3 to year 5.This is because the peak discharges in the second five years are significantly greater than those in the first five years.And more river waters in the second five years create more bed changes at a few cross sections around 4km and 7km (didn't plot in the figure).It is also found that there is an increasing erosion at a reach from 1.4 -1.8km, due to accelerated river flows by the narrow cross sections in the reach.The computed trend of the bed changes depends on the hydrological conditions of the selected water years.If more cycles of simulations of the ten water years can be repeated, an equilibrium state of morphology may be found, which remains as a future topic of this study.

OPTIMIZATION OF SEDIMENT RELEASE
Most of dam removal practices resulted in a natural (freely) flush of the reservoir sediments to downstream.A concern for the natural flush sediment routing is the significant impact of sedimentation on downstream in a relative short period after dam removal.For a better sedimentation management purpose, it should be the best to find a better way to minimize this kind of geomorphic impact, which could be the controlled sediment routing at the dam site.Thus, we (Ding et al. 2013b) have developed a simulation-based optimization model to find the optimal sediment release schedule so that the morphological changes in the downstream reaches become minimal (as less as possible).The simulation model is based on the sediment transport model and the flow model of CCHE1D.The optimization technique was developed based on the so-called variational adjoint sensitivity approach which is an efficient and effective nonlinear optimization method to seek the optimal control action such as the time-varying sediment release rate (i.e. the release schedule).
To apply this innovative optimization model to control the downstream deposition due to the flush of reservoir sediments, it is assumed that by a certain engineering way (e.g.dredging or pumping) the sediment release can be regulated by diverting extra sediments at the dam site out of the river system.The model is able to find out the best diverting schedule of the reservoir sediments so that the morphological changes are minimized at the downstream.It has to note that to divert all the reservoir sediments out of the river won't make a less impact on the morphology downstream because the clear water release after the dam removal should generate significant erosions at the downstream reaches.
The optimal control duration for the Marmot Dam removal was started immediately after the dam removal and lasted one year.Figure 6 presents the optimal sediment control results and comparisons with the bed changes without control (natural flush): the left figure shows two release sediment rate curves from reservoir based on natural flush and the optimal control; the right one plotted the morphological changes through the natural flush (without sediment release control) and the control sediment routing.The controlled release peaks of the sediments in the blue curve are corresponding roughly to the storm peaks of the river flows.The optimization results provide the best performance of the controlled sediment release which can reduce significantly the sedimentation at the downstream, if a certain amount of reservoir sediments can be diverted out of the river (the blue curve is the optimal sediment diversion schedule).However, a question on how to control the sediment release so precisely still remains for further discussions.

CONCLUDING REMARKS
In this paper, a brief review of numerical models and their applications for impact assessment of dam removal on sediment transport and morphological changes in alluvial rivers is given.As an example, a one-dimensional river flow and sediment transport model, CCHE1D, was applied to assess morphological changes in a reach of the Sandy River, Oregon due to the Marmot Dam removal in October of 2007.The sediment transport model parameters (e.g., sediment transport capacity, bed roughness coefficient) are calibrated by comparing the simulated bed changes with the observed changes after one-year of the dam removal in the Sandy River.After the successful model calibration, the model was applied to assess long-term morphological changes in response to a synthetic hydrograph consisting of 10 years selected from historical storm water records.The long-term assessment results on morphological changes are reasonable and can be used for managing and planning river sedimentation after the dam removal.This study has also applied a simulation-based optimization model to find the optimal control of the sediment release rate during the dam removal to minimize the morphological changes in the downstream river reaches.The optimal result of controlled sediment release can provide engineering management guidance to plan a better scheduled dam decommissioning if releasing reservoir deposits can be operated.

Figure 1
Figure1Sandy River Basin and Marmot Dam before removal(Keller 2010) Due to long-term sediment accumulation in the reservoir, different sources and sizes of sediments deposited in the reservoir formed a layer structure.(Squier Associates, 2000).The deposited sediments were mainly composed of a surface gravel layer (Unit 1), a sand layer underneath (Unit 2), and the pre-dam channel bed representing the third distinct unit.Figure2provides the size composition of the reservoir sediments for the two prescribed layers.The uppermost unit (Unit 1) ranged from approximately 2-5.5 m in thickness and was composed of sandy gravel with a small amount of cobbles and boulders, becoming thicker toward the dam.The next unit (Unit 2) was predominantly fine sediment (silty sand to sand with a small amount of gravel, ranging from 4 to 11 m in thickness).The pre-dam channel, below Unit 2, consisted primarily of coarse sediment and its thickness ranged from 0.8 to 3 m.At the time of decommissioning, the 3-km-long reservoir behind the dam was filled with nearly 750,000 m 3 sand and gravel, a volume equivalent to about 5 to 10 years of average annual sediment load.Most sediment transported by the river had been passing the dam for decades.The sediment transport rate was about 250,000 ton/year at the Marmot Dam, of which the majority was fine sediment(Stillwater Sciences, 2000).

Figure 2
Figure 2 Reservoir sediment size distributions in the upper (Unit 1) and lower (Unit 2) layers

Figure 3
Figure 3 Comparisons of 1-year bed changes by using four sediment transport rate formulae

Figure 4
Figure 4 Comparisons of 1-year bed changes by using different roughness coefficients

Figure 5
Figure5Long-term bed change evolutions computed by using calibrated model parameters

Figure 6
Figure 6 Optimal sediment control results: Left: release rate from reservoir based on natural flush (green line) and the optimal control (blue line); Right: Morphological changes through the natural flush (without sediment release control) and the controlled sediment routing (blue line)