Development and application of demonstration MIKE 21C morphological model for a bend in Mekong River September 2015 0
Table of Contents 1. Introduction... 2 2. Data collection... 3 2.1 Additional data... 3 2.2 Bathymetry data... 4 2.3 Hydrometric data... 5 2.4 Sediment grain size data... 6 2.5 Sediment transport data... 6 3. Model development... 9 3.1 Curvilinear grid... 9 3.2 Model bathymetry... 9 3.3 Hydraulic calibration... 10 3.4 Sediment transport calibration... 11 4. Model applications... 13 5. Conclusions... 18 1
1. Introduction A demonstration model, which is documented in a PowerPoint presentation, has been developed. This document provides a more formal documentation of the development. The purpose of the exercise is to develop and apply a morphological demonstration model, which can be used for evaluating the impacts of human interventions in the bend, in this case focused on sand mining. Figure 1: The selected study site. The selected study site is the very characteristic sharp bend in the Mekong River just upstream of Phnom Penh, see Figure 1. The study site is a sharp bend with a substantial side channel and good bathymetry data, which makes it suited for sand mining and for modelling. 2
2. Data collection The following data is required for morphological models: Planform data Bathymetry data Discharge time-series observed over longer time Water level time-series at two stations in the study area corresponding to the observed discharge timeseries Sediment particle size data Observed sediment transport rates for a range of discharges 2.1 Additional data In addition the following additional data can be nice to have available to make models more reliable: ADCP data. Historical planform data can be useful for evaluating the stability of the bend. Historical bathymetry data can be used for calibration of morphological models by performing hind-cast simulations of the bathymetric development. Observed Simulated Figure 2: Comparison between ADCP data and MIKE 21C in Chaktomuk Junction during the 2000 flood. 3
ADCP data is very useful if velocity profiles covering the whole river width area available in combination with bathymetry data reasonably representing the time of the ADCP collection. If such ADCP data is available, it can be used for determining the flow resistance distribution in the river cross-section, which can in many cases deviate from the trivial constant Manning M or Chezy C that usually has to be assumed when making models. Figure 2 shows an example of ADCP data collected in Chaktomuk Junction during the 2000 flood. For the present application the flow resistance distribution between the outer and inner channel in the bend is critically important for the long-term behavior of the system. Since the flow distribution has to be estimated, the model behavior over longer time may not be reliable. 2.2 Bathymetry data Two bathymetry data sources were available: Cross-section surveys, only river levels (bathymetry) DEM (50 m grid), only ground levels (topography) Figure 3: DEM transformed to 10 m grid. 4
Figure 4: Bathymetry data, left: Transects, right: Interpolated bathymetry (10 m grid) from transects. 2.3 Hydrometric data The discharges are observed in the period 1985-1995, while there are no water level stations in the study reach. This situation is quite common, and is often remedied by using a calibrated hydraulic model covering a longer stretch of the river to generate water levels for the local model. This is indeed also done here, and we used a MIKE 11 model developed for the Mekong River to generate a water level time-series downstream (boundary condition) and upstream (calibration). Figure 5: Observed hydrograph along with water level extracted from MIKE 11 at the downstream boundary. 5
Figure 6: Observed hydrograph along with water level extracted from MIKE 11 at the downstream boundary 1985-1987. Scenario simulations were conducted in the period 1 Jan 1985 to 1 May 1987, the time-series shown in Figure 6. 2.4 Sediment grain size data The sediment grain was set to 0.35 mm, which was also used by DHI in the Chaktomuk Junction model. 2.5 Sediment transport data Sediment transport data was available from two stations, namely Kratie and Chroy Chang Var, here we only used for data from Kratie, which is located upstream of the study reach. The data was processed by using the observed 80% cohesive sediment content to arrive at the non-cohesive sediment load. DHI is currently leading a major study of the Mekong River from which the Kratie data has been obtained. Sediment transport measurements are always associated with uncertainty, and the adopted Kratie data shows a total load of 70 mill tonnes/year of which 80% is cohesive, which means the non-cohesive suspended load is around 14 mill tonnes/year. The measurements are obviously suspended load only, while bed-load measurements are not available. For the Mekong bed-load would probably be, say, 25% of the total noncohesive load, which brings the total sediment load a bit higher. Additional data for the Mekong suggests that the Kratie data provide a low estimate on the total non-cohesive load, which DHI estimates to 15-30 mill tonnes/year. The sediment transport magnitude in this model is hence in the low end of the estimated range for the Mekong River. The sediment transport magnitude is not as important, as one would think for a model like the present. It is quite easy to show, using the model, that the sediment transport magnitude only influences the time-scale over which the system approaches dynamic equilibrium. So by having a slightly low sediment transport, the system will have a slower response than if we had used a higher magnitude. 6
Figure 7: Observed total (cohesive and non-cohesive) sediment transport rates at Kratie and Chroy Chang Var. Cohesive sediment transport is also known as wash-load, which is very descriptive for the behavior: It is washed through the river because the shear stresses are too high to deposit the cohesive sediment (except on the floodplain), and hence the cohesive sediment is morphologically neutral. 80% cohesive content in river sediment samples is quite common, i.e. 80% of the sediment transport is morphologically inactive. Figure 8: Observed non-cohesive sediment transport as function of the discharge in Mekong River at Kratie. 7
Figure 8 shows the processed sediment transport rating curve at Kratie, which is used for model calibration. Such sediment rating curves can usually be represented in the form: Q s (t) = f(q(t)) This form is convenient because it implies that the sediment transport is purely a function of the discharge, which is usually a good assumption for rivers with well-defined water level rating curves. A power-law was applied in this study in the form: Q s (t) = a Q(t) b When using this sediment rating curve in combination with the observed discharges 1985-1995 the annual sediment transport becomes around 5 mill m 3 /year, corresponding to 13 mill tonnes/year. As stated earlier the estimated Mekong River non-cohesive sediment transport from the Mekong River study is 15-30 mill tonnes/year, which means we are using the low end of the estimated range in this demonstration model. However, the adopted sediment transport is within the accepted range and it only influences the time-scale of the system; with a sediment transport in the low end of the estimated range the model gets a slightly slower morphological response. 8
3. Model development The model development can be divided into the steps: Generate curvilinear grid Interpolate bathymetry Calibrate hydrodynamic model (flow resistance) Calibrate sediment transport model (formula and calibration factor) 3.1 Curvilinear grid MIKE 21C uses curvilinear grids, which are good for resolving the flow paths in bends. Figure 9: Curvilinear grid. The bank lines were based on the available GIS data. We chose a grid size of 400x50 grid cells, see Figure 9. 3.2 Model bathymetry The model bathymetry was based on the two bathymetry sources, such that the river bathymetry was first used and then the missing elevations in the large island were based on the DEM. 9
Figure 10: Curvilinear bathymetry based on the DEM and river transects. Figure 10 shows the bathymetry based on these two data sources. It is noted that the elevations on the island in the bend are quite high, up to 13 m. Considering that the flood levels in the area are up to 14 m, these elevations seem reasonable. We do not know whether the island is vegetation covered, but it has been assumed in the morphological model by making the island initially non-erodible (can still deposit sediment). 3.3 Hydraulic calibration Hydraulic calibration is the process of adjusting the flow resistance to make the model match observed water levels. A Manning resistance formulation was selected. 10
Figure 11: Calibration of the hydraulic resistance. The calibration process was then conducted iteratively by running the model in hydrodynamic only mode (no morphological activity), while adjusting the Manning M value so the model matched the upstream water level. The result was a Manning M=25 m 1/3 /s, which is a bit high resistance for a large river, but we did not investigate any further whether the number was realistic. It is not impossible, but the resistance is a bit on the high side. It can be noted that the calibration is best for high discharges, while the MIKE 21C model over-predicts the dry season water levels compared to MIKE 11. Discrepancies between high and low flow calibrations are not uncommon, and for morphological models residuals are often accepted for dry season water levels because the sediment transport due to its non-linearity is much higher for flood conditions. 3.4 Sediment transport calibration Having adjusted the flow resistance we can continue and adjust the sediment transport. This can be done by just activating the sediment transport without updating the bathymetry, but here it was done with bathymetry updating activated as well. 11
Figure 12: Calibrated sediment transport model. The Engelund-Hansen sediment transport formula was selected, and a good match to the observed sediment rating curve was obtained by using a factor 0.4 on the Engelund-Hansen formula. Such a modification to a generally accepted sediment transport formula is considered reasonable, as long as the modification factor is not too far from unity; e.g. using 0.01 on the formula is too much modification. The Engelund-Hansen formula is a total load formula, so the bed-load and suspended load were assumed to be distributed as 75% suspended load, which is reasonable for high discharges. 12
4. Model applications The model is applied in the following for the three simulations covering 1 January 1985 to 1 May 1987: Existing conditions, also known as Do nothing or Baseline Scenario 1: 5 mill m 3 sediment removed from the bar Scenario 2: 5 mill m 3 sediment removed from the side channel Baseline simulations are very useful for morphological models because they allow some reduction of uncertainties by essentially having the uncertainties at play in both the baseline and scenario simulations. The 5 mill m 3 sand volume is strongly exaggerated in order induce large morphological changes in the initial bathymetries. In reality such large volumes cannot be expected realistically mined from a single bend. Realistically the annual sand removal from a single site would be 100-200,000 m 3. It is not a problem to consider more gradual mining of sand over many years, but the simulations will take longer to conduct. If wanting to induce large morphological changes in the considered bend, a capital dredging operation would be required. Figure 13: The two sand mining scenarios Results from the simulations can be evaluated at many levels: Bed levels Bed level changes over time (subtract initial bed level) 13
Induced bed level changes (subtract the baseline bed level at the same point in time) Flow distributions Water levels 1 Jan 1985 Baseline (z) Sand 1 (z) Sand 2 (z) 1 May 1987 Baseline (z) Sand 1 (z) Sand 2 (z) Baseline (Δz) Sand 1 (Δz i ) Sand 2 (Δz i ) Figure 14: Simulated bathymetries (1 May 1987), bed level changes for baseline and induced bed level changes for scenarios. General observations are made from the bed levels and bed level changes, see Error! Reference source not found. and Figure 14. Induced morphological changes upstream are very small, which is typical due to the hyperbolic nature of the morphological problem. However, it is possible to induce changes upstream via lowered water level. Both sand mining scenarios lead to weakening of the main channel and growth of the side channel. For the bar mining scenario ( Sand mining 1 ) the side channel grows due to flow crossing over the lowered bar and into the side channel, while in the second sand mining scenario ( Sand mining 2 ) the side channel grows 14
directly via mining. Sand mining 2 has a much bigger impact on the main channel than Sand mining 1, but it is not consistently increasing the outer bend bed levels, as seen in the figure. Growth of the side channel and weakening of the main channel are both predictable consequences, which should also be considered over a longer time-scale, as the balance may be different over longer time. The most likely development is that the side channel will weaken over time, so maintenance dredging will have to be conducted if wanting to establish a permanently enhanced side channel flow. Figure 15: Simulated flow distributions for the baseline and scenarios. Simulated flow distributions between the inner and outer channel are shown in Figure 15. It is seen that the side channel discharge increases in both scenarios. 15
Figure 16: The relative flow in the inner channel as function of the total Mekong discharge for the three scenarios. Another way to illustrate this is to calculate the relative flow in the side channel and plot it as function of the total Mekong discharge, see Figure 16. This figure shows that the side channel picks up relatively more flow as the Mekong discharge increases, and that the character changes much more when sand is mined from the side channel compared to mining sand from a bar. Mining the side channel increases the side channel flows much more and also for all discharges. Baseline Sand mining 1 Sand mining 2 Figure 17: Simulated flow speeds [m/s] for baseline on 1 September 1986 and changes to flow speed compared to baseline. Another way to illustrate the changes to the flow fields is to look at the flow speeds and changes to flow speeds, see Figure 17. The figure illustrates the increase in flow speed in the side channel for the two scenarios compared to baseline, and the associated decrease in flow speed in the other channel. 16
Figure 18: Simulated water level upstream for baseline along with water level changes for the two scenarios. Finally a bit about water levels, as seen in Figure 18. Typical consequences of sand mining is that the removal of sediment will lead to flood level reductions upstream, which is clearly induced by sand mining in the side channel (sand mining 2), while the impact is more irregular when mining a bar. There are other aspects, which can be investigated with a morphological model. Especially the downstream impact has not been addressed here for the basic reason that the model does not extend far enough downstream. If extending the model downstream, the sediment deficit associated with mining will evolve into general scour, which can also lead to increased bank erosion, but the downstream extension requires bathymetry data. 17
5. Conclusions This document describes the development of a demonstration MIKE 21C morphological model for the sharp bend upstream of Phnom Penh in the Mekong River. The model was developed based on available data, which included: Planform Bathymetry Discharges Water levels Sediment grain size data Sediment transport rates A model like this is useful for evaluating scenarios, but one should be cautious using the model for long-term simulations because the simulated morphological development over long time-scales is much more sensitive to uncertainties in the model calibration. Of particular importance is the distribution of flow and sediment between the inner and outer channels in the bend. Uncertainties can be reduced by using e.g. ADCP data to accurately determine the flow distribution between the channels. Model development essentially consists of generating a curvilinear grid, interpolating the bathymetry data, calibrating the hydraulic resistance and calibrating the sediment transport. Scenario simulations were done at a deliberately exaggerated level in which the annual sediment transport was mined from a bar or the inner channel in the bend. The mining volumes adopted in the demonstration model are not realistic, and should be considered capital dredging volume rather than mining volumes. Typical examples of how to analyze and understand the developments associated with sand mining were shown. 18