Professional Documents
Culture Documents
Three-Dimensional Numerical Modelling of The Flushing
Three-Dimensional Numerical Modelling of The Flushing
Abstract
Sediments filling reservoirs is a common problem in the world today, with an estimated 1% of the capacity of hydro-
power reservoirs being lost annually through sedimentation. One of the most used techniques for reducing this prob-
lem is reservoir flushing. During a flood, the water level is drawn down, causing increased velocities, therefore
facilitating erosion and sediment transport. During the flushing, water from the reservoir will be lost, resulting in signif-
icant economic implications for the reservoir owner. The success of reservoir flushing depends on several parameters,
including water discharge, sediment properties and reservoir geometry. This study describes the use of Computational
Fluid Dynamics (CFD) as a modern method to predict the reservoir flushing process. A three-dimensional numerical
model (SSIIM 2), with an adaptive, non-orthogonal and unstructured grid has been used. Through the application of
special modified algorithms (e.g., wetting ⁄ drying, free water surface), numerical modelling of sediment movement can
be an alternative for planning and optimizing the flushing process for complex reservoir geometries. The numerical
model was tested against data from a physical model study of the Kali Gandaki hydropower reservoir in Nepal. The
total quantity of flushed out sediments, and the bed deformation in six cross-sections, were compared, highlighting a
good correspondence between the results. These include the cross-sectional shape of a 90-degree bend, for which sec-
ondary currents influenced the results. The study indicates that numerical models might become a useful tool for reser-
voir flushing predictions.
Key words
hydropower, numerical modelling, reservoir flushing, reservoirs, sediments.
Table 1. Sedimentation rates for different regions in the world The availability of an accurate tool to estimate costs
as a percentage of the total reservoir volume and benefits of reservoir flushing would be very useful for
reservoir managers. Physical models are currently the
Location Sedimentation rate (%) Literature most applied method for predicting the flushing process.
World 1.0 Mahmood (1987) The water discharge is normally scaled using the Froude
Tunisia 2.3 Abdelhadi (1995) law in a physical model. Calibration of the roughness is an
China 2.3 Morris and Fan (1998) important task for correct modelling of the energy slope.
Turkey 1.2 Morris and Fan (1998) This is frequently done by using measurements from the
Morocco 0.7 Abdelhadi (1995) prototype, or data from historical events. Model-prototype
India 0.5 Morris (1995) comparisons often provide useful predictions of the mod-
USA 0.2 Crowder(1987) elled water flow and water level. Secondary currents
cannot be modelled correctly, if the model must be con-
structed in a distorted scale. This is because the velocity
increase when water levels are lowered. The increased profiles in the cross-sections of the physical model are not
shear stress at the channel bed causes erosion. The high comparable with the velocity profiles, as measured in nat-
turbulence during flushing will keep sediments in suspen- ure. Physical modelling of sediment transport is more diffi-
sion, allowing them to more easily negotiate the reservoir cult because of different scaling laws for suspended
outlet. An example of a successful application of reservoir sediments and erosion ⁄ bed load transport (Kobus 1978).
flushing is the Mangahao reservoir in New Zealand, where It also is difficult to scale the finer particles down because
59% of the original storage volume was lost after 45 years of the influence of cohesive forces occurring when the
of operation. With the first flushing, 75% of the deposited sand particles are very small. The magnitude of bed forms
sediments were removed within 1 month (Jowett 1984). is almost impossible to properly scale in a physical model.
As a result of the small ratio of water volume to annual This study utilizes a three-dimensional CFD model
water inflow, flushing could often be efficient and is very used to simulate reservoir flushing. The numerical model
common for run-off of the river reservoirs. The success of SSIIM 2 was chosen for the simulation (Olsen 2009).
reservoir flushing, however, depends on several parame- SSIIM 2 has already been used for sediment calculations
ters, making it difficult to predict the effects of an antici- for self-meandering channels (Ruether & Olsen 2007)
pated flushing. The location and size of the bottom outlets and for estimating local scour depths (Olsen & Kjellesvig
are among these parameters. An inadequate design of the 1998). It has also been used for a two-dimensional calcu-
bottom outlets would result in significant erosion down- lations of reservoir flushing (Olsen 1999). This study
stream. If the scour causes problem for dam safety, flush- clearly illustrates the weaknesses of two-dimensional
ing is not feasible (Morris & Fan 1998). models for modelling bends in the reservoir. One of the
One of the main questions facing reservoir owner is problems was the difficulty of replicating the cross-sec-
whether or not the benefits of flushing the reservoir are tional shape of the bed geometry in a bend, with experi-
in accord with the anticipated costs, as related to the ments indicating the deepest point of the bend was at the
quantity of water lost from the reservoir. Information on outside of the curve. The two-dimensional model, how-
whether the erosion takes place in the whole reservoir, ever, computed the deepest point to be on the opposite
or just in a small part of the reservoir, also is important, side, being attributed to its inability to replicate the sec-
an illustrative example being the Cachı̀ Reservoir in ondary currents in bends. In contrast, the model selected
Costa Rica. As a result of its significant width, a narrow for this study considers three-dimensional effects such as
erosion channel is formed during the flushing (Jansson secondary currents, a very important aspect for complex
& Rodriguez 1992). During subsequent flushings, erosion reservoir geometries. Through the application of an adap-
will take place in the same erosion channel, whereas the tive grid and implemented sediment algorithms, numeri-
sediments will continuously accumulate in the other cal modelling of sediment erosion can be an alternative
regions of the reservoir (Morris & Fan 1998). Accord- way to predict the success of a forthcoming reservoir
ingly, the volume of removed sediments will decrease flushing. Its application in an early stage of the design
over time, despite the flushings. A second important phase also is possible, and might help plan and optimize
aspect is the environmental implications of the flushing. the reservoir sediment management. Accordingly, the
A poorly conceived flushing can result in fairly high sedi- main purpose of the present study is to test the accuracy
ment concentrations, influencing downstream habitats of a numerical model in predicting the effects and suc-
and killing fish. cess for an upcoming reservoir flushing.
NUMERICAL SIMULATION
The fully three-dimensional numerical program SSIIM 2
was chosen for the computations in this study. SSIIM is
an abbreviation for ‘Sediment Simulation In Intakes
with Multiblock’ option. The numerical program was
chosen because of its large number of available algo-
rithms for sediment computations. A simulation of water
and bed elevation changes, and a computation of com-
plex bed geometries in the reservoirs, also is also possi-
ble (Olsen & Kjellesvig 1999). An introduction to the
code and the basic equations are provided in the follow-
Fig. 1. Location of the Kali Gandaki hydropower reservoir. ing section.
To calculate the sediment concentration, the formula ness was chosen at 0.005 m, and the active layer thick-
by Van Rijn (1984b) was used as a boundary condition ness at 0.01 m. No sediments were added to the model
for the cell closest to the bed, as follows: during the numerical calculation.
ssc;i 1:5
di sc;i
Other algorithms
cbed;susp:load;i ¼ 0:015 1=3 0:3 ð5Þ To calculate the free water surface, SSIIM 2 uses an algo-
a
di ðqs =ðqmw2 1Þ g rithm based on the computed pressure field. Thus, the
location of the water level is calculated with the Bernoulli
where cbed,susp.load,i = concentration of suspended load at equation, as follows:
the bed for the ith fraction; di = diameter of the ith frac-
tion; a = equal to the height of the bed cell; s = shear @p @z
¼ qg ð9Þ
stress; di,sc,i = critical shear stress for di, which was cal- @xi @xi
culated by an analytical formula from the Shields curve;
where p = pressure; q = density; g = acceleration of grav-
qs = sediment density; qw = water density; g = acceleration
ity; x = spatial geometrical scale; and z = water level eleva-
of gravity; and m = kinematic viscosity.
tion.
The bed load was calculated with an empirical formula
An iterative method, with a time step of 1.0 s, is used
from Van Rijn (1984a), as follows:
for producing a stable calculation of the free water sur-
face (Olsen & Haun 2010). A depth-averaged pressure
ssc;i 2:1
q sc;i
was used for the computations. The water level in the
qb;i
ffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 0:053 0:1 ð6Þ
ðqs qÞ g
di1:5 q di0:3 ðqs qÞ g
q m2
surface cells was smoothed by the following equation to
prevent instabilities through steep water levels in the cal-
where qb,i = transport rate of ith fraction of bed load per culation:
unit width; di = diameter of ith fraction; s = shear stress;
di,sc,i = critical shear stress for di, which was calculated z ¼ z0 ð1 rÞ þ r znb ð10Þ
by an analytical form from the Shields curve; qs = sedi-
where z = water level after the smoothing; z0 = water
ment density; qw = water density; g = acceleration of grav-
level before the smoothing; r = chosen relaxation factor
ity; and m = kinematic viscosity.
(a value of 0.3 is used); and znb = water level of the four
Modelling of multiple sediment sizes was also was a
neighbouring cells before the smoothing.
component of the numerical model. The roughness at the
As a calibration result, the value for the fraction of
bed corresponded to the grain-size distribution and bed-
compacted sediments was chosen to be 58%. In addition,
form height. Where the grain-size distribution was known
a sand slide algorithm was used, wherein the tangent of
from the physical model study, the bed-form height was
the angle of repose was set to 0.69.
calculated from the formula of Van Rijn (1984c), as follows:
The calculations were realized on an IBM cluster by
0:3 ! using 16 cores with 1.9 GHz, with a computational time
ssc;i
D d50 0:5 sc;i s sc;i of approximately 8.5 h.
¼ 0:11 1e 25
y y sc;i
ð7Þ RESULTS AND DISCUSSION
The results from the computations in this study were
where D = bed-form height; y = water depth; d50 = charac- compared with the total quantity of eroded sediments,
teristic sediment size; s = shear stress; and sc,i = critical and with the measured cross-sections from the physical
shear stress for each fraction. model. Figure 3 illustrates the chosen grid at the begin-
The bed-form height was converted into the bed ning of the flushing simulation. The six cross-sections
roughness (ks) with the following formula: where bed level measurements were taken are also are
included in the figure.
kS ¼ 3:0 d90 þ 1:1D 1:0 e ð 7:3y Þ
25D
ð8Þ Figures 4–6 illustrate the movement of the adaptive
grid during the flushing simulation. After 3000 s (Fig. 4),
where d90 = characteristic sediment size; D = bed-form the first part dries up at the right side upstream of the
height; and y = water depth. weir. After another 3000 s (Fig. 5), a large area of the
A correction for sloping bed was used to calculate the reservoir is dried up, and after 9000 s (Fig. 6), a narrow
bed-sediment concentration (Brooks 1963). The rough- channel is formed at the outside of the bend.
Fig. 4. Grid after 3000 computation seconds. Fig. 7. Water levels from the numerical model after 9000 s with
the higher water level at the outside of the bend (all values are in
cm).
the flushing, and not over the whole reservoir width as it sections, where the physical model results were com-
was initially observed. pared with the calculated ones.
Figure 9 illustrates the bed and surface velocity vec- Lacking data on how the inflow area was incorporated
tors. This indicates that the secondary currents in the in the physical model, some assumptions were made. As
bend are taken into account by the numerical model. a result of this inaccuracy, cross-section A indicates a dif-
The bed levels after the flushing were taken at six ferent magnitude and location of the erosion between the
cross-sections (A–F). Figures 10–15 illustrate these cross- physical and numerical model. Cross-sections B and C, in
the bend of the reservoir, however, show that the loca-
Table 3. Erosion in the reservoir over time
tion and magnitude of the erosion correspond well. These
two cross-sections are the most important ones in regard
Eroded Eroded
to proving that the secondary currents in the bend were
sediments sediments Erosion
taken into account by the three-dimensional model. The
physical numerical rate per
results of the two-dimensional calculation are included in
Time (sec) model (m3) model (m3) 3000 s (m3)
cross-section B. The two-dimensional computation gives
0 0 0 0.673 incorrect results regarding the bend. Cross-sections D, E
3000 – 0.673 0.389 and F illustrate slight differences from the measured val-
6000 – 1.062 0.249 ues, indicating the numerical model also works well for
9000 1.363 1.311 – these areas. The difference in the eroded areas of the
cross-sections is between 13 and 27% for the cross sec-
tions B–F. As a result of the assumptions at the inflow
boundary, the difference in the eroded area of cross-sec-
tion A is much larger (141%).
The total quantity of eroded sediments in the numeri-
cal model was about 1.31 m3. This is 3.8% less than
observed for the physical model (1.363 m3).
CONCLUSIONS
An important aspect of reservoir management is to mini-
mize the loss of reservoir volume due to sedimentation
within the reservoir. A three-dimensional numerical
model was used in this study to successfully model the
flushing of sediments from a hydropower reservoir.
Fig. 9. Velocity vectors at the surface (black) and the bed (grey)
The results were compared with a physical model test on
of the reservoir at the end of the computations.
the Kali Gandaki hydropower reservoir in Nepal. The
numerical model used an algorithm to move the water was especially successful, wherein the secondary cur-
surface, which was based on the computed pressure rents computed in the three-dimensional model resulted
field. This enabled relatively long time steps, making it in maximum depths in the outer side of the bend. The
computationally feasible to compute processes taking total computed eroded volume also compared reasonably
place over a considerable time period. The numerical well with the observations from the physical model, after
model also used an adaptive grid wherein the number of calibration of the numerical model.
cells in the vertical direction changed, according to The study indicates the three-dimensional numerical
the varying water depth. The wetting ⁄ drying algorithm model can be an important tool in reservoir management
removed cells in areas of the reservoir that dried up. The planning procedures. Both the magnitude of the eroded
predicted areas of drying compared well with the obser- sediments and the location of erosion areas were pre-
vations in the physical model. A channel formed in the dicted in this study.
reservoir as the water surface was lowered, and the sides
of the channel dried up. The results from the numerical ACKNOWLEDGEMENTS
model also compared well with measurements of the The first author was funded by the Research Council of
cross-sectional shape of the channel formed after the Norway, through the RENERGI programme. Some of the
flushing. The prediction of the classical channel in a bend study computations were done on the IBM cluster of the
Norwegian University of Science and Technology. We Morris G. L. & Fan J. (1998) Reservoir Sedimentation
also thank NOTUR for providing assistance for computer Handbook. McGraw-Hill Book Company, New York.
use and computational time. Olsen N. R. B. (1999) Two-dimensional numerical model-
ling of flushing processes in water reservoirs.
REFERENCES J. Hydraul. Res. 37(1), 3–16.
Abdelhadi M. L. (1995) Environmental and Socio-eco- Olsen N. R. B. (2009) A Three-Dimensional Numerical
nomic impacts of erosion and sedimentation in North Model For Simulate Of Sediment Movements. In:
African Countries. Proceedingsof the Sixth Interna- Water Intakes with Multiblock Option. Users‘s Manual
tional Symposium, River Sedimentation; New Delhi, (ed. N. R. B. Olsen) Department of Hydraulic and Envi-
India, A.A.Balkema, Rotterdam. ronmental Engineering, The Norwegian University of
Brooks H. N. (1963) Discussion of ‘‘Boundary Shear Science and Technology, Trondheim, pp. 1–196.
Stresses in Curved Trapezoidal Channels’’ by Ippen, Olsen N. R. B. & Haun S. (2010) Free surface algorithms
A.T. and Drinker, P.A. J. Hydraul. Div. 89(2), 189– for 3D numerical modelling of reservoir flushing. Dit-
91. trich et al. Preprints of the River Flow Conference
Crowder B. M. (1987) Economic Costs of Reservoir 2010. Bundesanstalt für Wasserbau, 1105–10.
Sedimentation: A Regional Approach to Estimating Olsen N. R. B. & Kjellesvig H. M. (1998) Three-dimen-
Cropland Erosion Damages. J. Soil Water Conserv. sional numerical flow modelling for estimation of maxi-
42(3), 194–7. mum local scour depth. J. Hydraul. Res. 36(4), 579–90.
Jansson M. B. & Rodriguez A. (1992) Sedimentological Olsen N. R. B. & Kjellesvig H. M. (1999) Three-dimen-
Studies in the Cachı̀ Reservoir, Uppsala University, sional numerical modelling of bed changes in a sand
Department of Physical Geology, Costa Rica. UNGI trap. J. Hydraul. Res. 37(2), 189–98.
Rapport 81; ISSN 0375-8109, ISBN 91-506-0913-0. Patankar S. V. (1980) Numerical Heat Transfer and Fluid
Jowett I. (1984) Sedimentation in New Zealand hydroelec- Flow. McGraw-Hill Book Company, New York.
tric schemes. Water Int. 9, 172–6. Ruether N. & Olsen N. R. B. (2007) Modelling free-form-
Kobus H. (1978) Wasserbauliches Versuchswesen, Mittei- ing meander evolution in a laboratory channel using
lungsheft 4, Schriftenreihe des deutschen Verbandes three dimensional computational fluid dynamics. Geo-
für Wasserwirtschaft und Kulturbau, Parey Verlag, morphology 89(3–4), 308–19.
Hamburg. Schlichting H. (1979) Boundary Layer Theory. McGraw-
Launder B. E. & Spalding D. B. (1972) Lectures in Mathe- Hill Book Company, New York.
matical Models of Turbulence. Academic Press, London. Shen H. W. (1999) Flushing sediment through reservoirs.
Mahmood K. (1987) Reservoir Sedimentation: Impact, J. Hydraul. Res. 37(6), 743–57.
Extent and Mitigation. World Bank Technical Paper, Van Rijn L. C. (1984a) Sediment transport. Part I: bed
World Bank, Washington, DC. 71; ISSN 0253-7494, load transport. J. Hydraul. Eng. 110(10), 1431–56.
ISBN 0-8213-0952-8. Van Rijn L. C. (1984b) Sediment transport. Part II: sus-
Morris G. L. (1995) Reservoir Sedimentation and Sustain- pended load transport. J. Hydraul. Eng. 110(11), 1613–
able Development in India: Problem Scope and Reme- 41.
dial Strategies. Proceedings of the Sixth International Van Rijn L. C. (1984c) Sediment transport. Part III: bed
Symposium, River Sedimentation, New Delhi, India, forms and alluvial roughness. J. Hydraul. Eng.
A.A.Balkema, Rotterdam. 110(12), 1733–54.