Numerical modeling of a river site for in-stream energy

converters

E. Lalander1 and M. Leijon1

1 The Swedish Center for Renewable Electric Energy Conversion,
Division of electricity,
Department of Engineering Sciences,
Uppsala University,
Box 534, SE-751 21 Uppsala, Sweden

Abstract ηA = Measured water level (deviation from mean)[m]

ηL = Lake water level [m]
Current measurement in the river Dalälven, situated ηM = Simulated water level [m]
in Sweden, was compared with flow values. These were ηM∗ = Simulated water level extracted at ADCP location [m]
also used to validate a numerical simulation of the chan- ρ = Density of water [kg/m3 ]
nel. The results showed the numerical program was able Q = Flow from power station [m3 /s]
to assess the current speed variations, but the magnitude Si j = Deformation rate [m2 /s]
of the current speed was slightly lower than the measured τ~b = Bottom friction [N/m2 ]
values, a possible effect of bathymetry errors. The wa- ~uM = Simulated horizontal velocity[m/s]
ter level variations were not correctly assessed, which is UA = Measured current speed [m/s]
probably due to that wind data was excluded. UM = Simulated horizontal current speed [m/s]
The numerical program was used to evaluate how UM∗ = Simulated current speed extracted at ADCP location [m/s]
V = Velocity acted on a turbine [m/s]
large the effects on the surrounding water level and cur-
νt = Horizontal eddy viscosity [m2 /s]
rent speed can be for a set of ten turbines distributed
evenly along the channel. To investigate how the hy-
drography of the channel is affected by energy extrac-
tion, a set of ten turbines distributed evenly along the
1 Introduction
channel were put into the numerical model. It showed River flows are formed by atmospheric precipitation
that extracting 75 kW would increase the water level and the magnitude of the flow also varies with season.
at the power station (upstreams) from the level at the At high latitudes (>60◦ N), during winter, the precipita-
same place without any turbines with 5.5%, and 135 tion is locked as snow, and during spring time the water
kW would lead to an 8.8% increase, assuming a constant is released as the atmospheric temperature increases. In
water level at the downstream boundary. With increas- unregulated river courses the flow peaks at spring time as
ing cd -values, the velocity around the turbines increases, snow melts in the mountains and is reasonably constant
while it decreases at the turbine. the rest of the year. In regulated rivers which are con-
trolled by hydropower, the flow can be controlled by the
Keywords: numerical modeling, river, energy conversion
use of dams and varies with, among other factors, elec-
tricity consumption. In these rivers, the flow can vary
Nomenclature greatly within minutes, which can have a negative im-
Ae = Turbine area [m2 ] pact on downstream constructions.
cd = Turbine drag coefficient In Sweden, one of the main energy sources is hy-
cf = Bottom drag coefficient dropower, and most rivers are regulated. With the aim
cs = Smagorinsky coefficient of finding new sources for the production of electricity
d = Still water depth [m] in Sweden, together with the fact that river exploitation
Ft = Effective drag force by a turbine [N]
by conventional hydropower is reaching a maximum, in-
g = Acceleration due to gravity [m/s2 ]
h = Total water depth [m]
terest in other renewable energy sources has increased.
l = Grid length scale [m] Only between 2006 and 2007 the installed wind power
M = Manning number [m1/3 /s] increased by 40% [1]. Even in regulated rivers there are
watercourses with high kinetic energy density as will be

c Proceedings of the 8th European Wave and Tidal Energy discussed in this paper. With this background an assess-
Conference, Uppsala, Sweden, 2009 ment of the kinetic energy resource in Swedish rivers is

826
1
of interest.
Water in rivers flows downhill driven by potential en-
ergy. This potential energy has traditionally been con-
verted to electricity by hydropower turbines. Less com-
mon is to extract the kinetic energy in river flows using
in-stream mounted turbines, although this has gained in-
creased attention in the past years, and various methods
to do this are described in [2]. The concepts are similar
to tidal energy systems, although smaller in scale, but the
information can be mutually valuable for both research
areas. Stockholm
The extraction of kinetic energy from moving water is
highly dependent on the current speed and for a resource
assessment it is therefore important that the nature of the
flow is accurately described. Numerical methods have
been used in e.g. [3] where time series of the veloc-
ity around Portland Bill were produced, in [4] where (a) Outline map of Sweden.
effects of tidal energy extraction at Portland Bill was
estimated and in [5] where the extractable resource in Lake

Johnston Strait was calculated using a numerical model.

However, input to these models has been data from sea- 1200 kHz
charts where the tidal current speed has been calculated
600 kHz
by tidal constituents. Data from water gauging stations ADCP measurements
have been used to validate these models, whereas real
time current speed data are rarely used. To measure ve-
locity variations in time over an entire water column, an
Acoustic Doppler Current Profiler (ADCP) can be used.
Power station
This was utilized in [6] to measure wave induced short (flow readings)

time fluctuations, and in [7] to compare with a numerical

(b) Map of site.
model of the tidal energy resource.
The aim of this work is to compare measured data to
predictions from a numerical model and to increase the Figure 1: Maps over the measurement site. In (b) the interpo-
understanding of the resource for in-stream current en- lated depths and the location for the deployment of the ADCPs
are shown.
ergy converters. In order to do that, the current velocity
was sampled with an ADCP in a Swedish river during lake. There are three bridges along the channel and the
one month. The velocity measurements were compared bridge furthest downstream was chosen as the measure-
with numerical simulations of the site and data of flow ment site, see Fig. 1(b).
and water level. By including a set of 10 turbines it was
possible to estimate how energy extraction could affect 2.1 Data
the surrounding water levels and velocity.
To measure the velocities two ADCPs from RD In-
struments, one 1200 kHz and one 600 kHz, were de-
2 Site description and measurement
ployed for one month in the channel 800 m downstream
There are approximately 15 rivers with a flow rate of of a hydropower station. During the measurement with
more than 100 m3 /s in Sweden. Most of these rivers are the 1200 kHz ADCP external forces interrupted the data
exploited by hydropower and the majority are situated series at two occasions, resulting in three series of data.
in the north of Sweden, having their discharge into the The measurement series for the 600 kHz ADCP was
Baltic sea above 61◦ N. Two of the largest rivers do how- good for a period of 24 days. Measurement parameters
ever discharge below 61◦ N and the measurement site lies are presented in Table 1. The ADCP uses an average of
in the northernmost one of these, Dalälven (Dal River). 50 pings to make one ensemble. Since the ensemble in-
The river originates in the west of Sweden in the moun- terval was 6 minutes, ten measurement readings were re-
tains, and flows eastward having its outflow at 60◦ N. ceived per hour. Due to the built-in pressure sensor, total
The current velocity was measured in a channel by depth is also measured at 6 minute interval. The time-
the city of Söderfors, located approximately 20 km up- varying total depth deviation from the mean depth is the
stream of the river’s outlet in the sea. The chosen chan- measured water level, denoted ηA . The 600 kHz instru-
nel is 1 km long and it discharges into a lake. The cross ment is meant to operate at depths greater than 20 m. At
section of the channel varies from a rectangular shape this occasion the instrument was deployed at a very shal-
with a depth of ten meters and width of 30–60 m, to a low depth, resulting in large error velocity, sometimes
parabolic shape 6 m deep and 100 m wide close to the exceeding 200 mm/s. The velocity data from this instru-

827
2
 600

400
m3 /s

200

2003 2004 2005 2006 2007

year

Figure 2: Weekly averaged flow in the channel by Söderfors

2003 - 2007.

Table 1: ADCP measurement parameters. Figure 4: Regression plot between hour averaged ADCP cur-
rent speed and flow.
Frequency 1200 kHz 600 kHz only basic properties such as bottom stress. The govern-
Ensemble interval 6 min 6 min ing equations are the well known depth averaged shal-
Pings/ensemble 50 50 low water model. The main equations are well described
Bin size 0.25 m 0.5 m in [8] and are therefore omitted from this work. The
Measured depth range 4.5 m 2.5 m equations used to describe the conditions at the site are
ADCP depth 6.5–7 m 5.5–6 m presented below.
Duration of deployment 33 days 29 days
Standard deviation 96 mm/s 119 mm/s 3.1 Theory
ment has thus been omitted, but the water level readings In Mike21 Flow model FM [8], bottom friction is as-
were used to compare with the 1200 kHz ADCP, Fig 3. sumed to vary quadratically with the velocity above the
bottom according to
0.2  
1200 kHz  
600 kHz τ~b = ρ c f u~M u~M  .
0.1

The drag coefficient, c f , can be calculated using the

0
m

Manning number, M as
−0.1 g
cf =
2 . (1)
−0.2 Mh1/6
18/2,23.18 28/2,21.18 h is the total depth which is the sum of the still water
depth, d and water level deviation from this, ηM . The
Figure 3: Water level measured by the two ADCPs. Manning number represents the bottom surface rough-
ness and is the resistance to flow in channels. It depends
The data from the ADCP was compared with flow on factors such as the type and size of the material of the
and water level readings. The hourly averaged flow channel, and the shape of the channel [9].
data was provided by Vattenfall AB; the owners of the Lateral turbulence is modeled using a sub–grid scale
hydropower station upstream of the sampling location. horizontal eddy viscosity concept as proposed in [10].
Hourly water level measured in the downstream lake The horizontal eddy viscosity, νt , is related to a charac-
were received from Fortum AB and is denoted ηL . The teristic length scale (grid length scale), l, according to
data was taken from a water gauge station in the down-
p
stream lake (not on the map), but there was no informa- νt = c2s l 2 2Si j S ji , (2)
tion on the accuracy of the data.
The yearly flow variation, seen in Fig. 2, shows a where cs is the Smagorinsky coefficient and Si j is the
mean flow of 300 m3 /s and a maximum flow that exceeds deformation rate given by
600 m3 /s.
1  ∂ ui ∂ u j 
Si j = + (i, j = 1, 2) .
3 Numerical modeling 2 ∂ x j ∂ xi

To model the river, a 2D numerical program, Mike21, 3.2 Model setup

from the DHI Group in Sweden was used. The program
is capable of modeling various processes, e.g. sediment The studied area was modeled with a flexible mesh
transport, but the simulations were performed including system, allowing variable sized mesh where desired. The
meshed domain includes 2518 nodes and 4652 elements.

828
3
 Table 2: Mike 21 parameters. extracting energy. In this study, the numerical model
is two-dimensional and effects on the surrounding flow
Simulation Validation field can be expected to be somewhat different, since it
Time period 090204 – 090228 is possible to account for wake effects and a velocity in-
Time step 360 s crease around the turbines.
Nr of time steps 2500 A set of 10 turbines were included in the model and
South boundary Q (Fig 6(a)) distributed pair-wise along the channel (Fig. 5). The
North boundary ηL (Fig 6(c)) structure of the farm has not been optimized to give max-
Manning number 32 m1/3 s−1 imum power output, but to simulate effects on the sur-
cs 0.28 rounding flow. The distance was approximately ten di-
ameters in between each row.
Bathymetry data, Fig. 1(b), was received from mea-
Each turbine had a quadratic cross section area, Ae ,
surements, maps and drawings for the area. Due to the
see Table 3. The effects on the water level and velocity
bathymetry data being rather scarce, it is considered one
in the channel for the turbine setup could be calculated
of the greatest sources of errors for the simulation.
for several flow rates and cd -values.
As an input to the model both the Manning number
and a value for the Smagorinsky formulation is required.
Since it is a dredged channel and the surrounding area
is of sand and gravel, the Manning number was set to
32 m1/3 s−1 which corresponds to a rather smooth sur- T9
face [9, 11], and cs in Eq. 2 was set to 0.28.
T10
The simulations were executed as shown in Table 2 to T7
validate the numerical model. The two open boundaries T5 T8
were the power station (south) and the lake (north). Flow
T6
readings for February were used at the first boundary and T3
water level for February at the second. T4
Data from the numerical model was extracted at the T1
location of the two ADCPs, which were deployed in be- T2
tween and slightly downstream two pillars.
Further simulations were done to estimate current
speed for different flow rates with and without including
turbines. The same parameters for n and cs were used. Figure 5: Turbine configuration.

3.3 Including turbines

4 Results and discussion
In this study the interest is in estimating how the wa-
ter level and velocity along the channel are affected by There is a good correlation between measured hourly
energy extraction. Turbines can be added in the model averaged ADCP data and flow readings seen from the
and are then assumed to induce a drag force, Ft , on the correlation coefficient, R2 , of 0.94 as seen in Fig. 4.
flow. This is calculated according to This could be expected due to the proximity of the mea-
surement site to the power station. The correlation value
1 indicates the flow data is good to use to describe the cur-
Ft = ρ cd AeV 2 .
2 rent speed in the channel.
Ae is the effective area of the turbine on which V acts. The level data for the lake (ηL ) was received sepa-
Note that V is the velocity experienced by the turbine, rately from the flow values, and no accuracy is available.
and not the undisturbed velocity, UM . The drag coeffi- It can be seen in Fig. 6(c) that ηA and ηL vary inde-
cient, cd , increases the blockage effect on the flow, which pendently from each other, which is a result of spatial
in turn decreases the velocity, V . The power extraction variations of the lake water level.
by each turbine can be calculated using
4.1 Validation of the numerical model
P = Ft V. (3)
The above mentioned results indicate that the veloc-
As the value of cd increases, the extracted power in- ity in the channel depends mostly on the flow. Thus,
creases, but this effect is diminished at high values of cd , the received flow readings were applied at the power sta-
since the velocity goes toward zero (Fig. 7). However, tion boundary. ηL was applied at the second boundary.
at these values the devices could be regarded as pillars The simulation was performed to compare with mea-
since no flow is passing through, and it is thus not real- sured data, and due to the interruption during measure-
istic to apply very high values of cd . ments, the simulation was only run for existing sets of
Turbines were modeled as a retarding force on the data. This explains the gap in the Mike21 series in Fig.
flow in e.g. [12]. The one-dimensional model indicated 6.
the velocity in the channel would decrease in total when

829
4
 UA
1.3 UM∗ 380
Q
50
330
T1
1.1 280

flow / m3/s
40 T2
m/s

T3
0.9 30
T4

kW
20 T5
0.7
T6
11/2 18/2 28/2 10 T7
T8
(a) Current speed. 0 T9
0 10 20 30 40 T10 50
Mike data ADCP data
T8
90 5000 90 4000
120 60 120 60
2500 30 2000 30 90
150 150

180 0 180 0 60 0.8

m/s
kN
210 330 210 330 0.6
30
240 300 240 300 0.4
270 270

0 0.2
(b) Rose plot. 0 10 20 30 40 50
c
d
ηA
0.2 ηM∗
ηL Figure 7: Top: Effects of cd on power output modeled in
Mike21 at a flow of 500 m3 /s. Maximum power for each tur-
bine is denoted by *. Bottom: Force and velocity at turbine T8
m

with increasing cd .

m while for ηM∗ the maximum variation was 0.13 m.

−0.2

11/2 18/2 28/2 The model assesses the current speed reasonably
(c) Water level.
well, but not the water level deviations. Using more ac-
curate bathymetry data and including other parameters
such as wind data over the lake could have improved
Figure 6: Mike data extracted from the measurement location
by the bridge and ADCP data from the 1200 kHz instrument. these results, but no such data was available.
(a) Current speed at the bridge (left axis, UM∗ and UA ) com-
pared with flow at power station, Q. (b) Rose plot between 4.2 Turbine influence
measured and modeled current direction data (c) Water level at The power output from each turbine was calculated
the bridge (ηL and ηA ) compared with water level at the lake for different cd - and flow values using Eq. 3. In Table 3,
boundary.
the power output for three flow rates, 300 m3 /s (mean),
In the same figure it is seen that UM∗ follows UA 500 m3 /s and 700 m3 /s ( maximum), is presented, show-
closely, but the magnitude of the simulated current speed ing the total power output is more than 10 times higher
is slightly lower. This is probably due to bathymetry er- in the case for maximum flow as compared to the mean
rors. As mentioned previously, there were some difficul- flow.
ties in re-creating the bathymetry due to scarce data. The The power is proportional to the force of the turbine,
bathymetry could be improved by more measurements. which in turn depends on Qt2 (through the turbine) and
At the shallow depths as in the present channel, small cd . Maintaining a constant flow in the channel, an in-
changes in the bathymetry greatly alter the flow field. creased cd -value will decrease the flow through the tur-
The rose plot in Fig. 6 shows there is a slight de- bine, a result of an enhanced flow at the sides of the tur-
viation in the modeled current direction compared with bine. At a certain cd , the power output starts decreasing.
the measured one. The ADCP was located downstream a Maximum power for each turbine in a flow of 500 m3 /s
few bridge pillars where the flow is turbulent. Therefore, is plotted in Fig. 7. Comparing this plot with simple cal-
depending on the cross-channel position, the flow direc- culation using P = 12 ρ AeU 3 shows that for a velocity of
tion can vary. The deviation can be a result of that the 1 m/s and a cross sectional area of 100 m2 , as is the case
data from the simulation was extracted from a location for turbine T1, the maximum kinetic energy is 50 kW.
slightly different than from the measured one. For turbine T1 and T2 this value is reached already at a
Fig. 6(c) shows the simulated (ηM∗ ) and measured cd -value in between 1 and 2. It is thus not recommended
(ηA ) water level deviations from mean. Although ηM∗ at to apply a cd -value above 1 for realistic simulations.
times shows the same variation as ηA , the model is not In [12] the model was forced by a varying water level,
able to simulate the magnitude of the water level. The and increased turbine drag resulted in a decreased veloc-
maximum variation (max(η ) – min(η )) of ηA was 0.47 ity in the whole channel. Here, the driving force is a

830
5
 Table 3: Power [kW] for three different flow rates. cd -value is 0.8.

Turbine T1+T2 T3+T4 T5+T6 T7+T8 T9+T10 Total

Cross section m2 100 49 49 49 25 –
Flow 300 m3 /s 11 4.6 5.6 7.6 3.1 32 kW
Flow 500 m3 /s 48 19 23 32 12 135 kW
Flow 700 m3 /s 123 48 60 81 32 344 kW

0.5
No turb
0.4 cd=0.4
0.3 cd=0.8
m

cd=1
0.2
cd=2
0.1

0
T1,2 T3,4 T5,6 T7,8 T9,10

0.08

0.06
m

0.04

0.02
(a) No turbines.
0
T1,2 T3,4 T5,6 T7,8 T9,10

Figure 8: Effects of turbines on the water level along the chan-

nel for different cd – values and a flow is 500 m3 /s. Top: In-
crease from the value at the lake. Bottom: Change in water
level calculated as the difference from the case of no turbines.

constant flow, and increasing the turbine drag leads to a

higher water level upstream. For continuity to be valid,
the same flow must pass through all sections in channel,
leading to an enhanced pressure force. This is the prob-
able reason for the high power values achieved when in-
creasing the drag force.
Modeling of turbine effects was performed both with (b) 10 turbines.
and without turbines included; hence the difference in
water level could be extracted for different cd -values. Figure 9: Velocity field in the channel for the case of 10 tur-
The effects of this particular turbine configuration on bines, cd = 10.
surrounding water level are shown in Fig. 8 for flow rate
of 500 m3 /s. For a case with no flow, the water level is where strongest currents can be found. The flow field
zero everywhere. At a flow rate of 500 m3 /s and without around the turbines is modified locally, with enhanced
any turbines, the water level at the power station is 36 current speed around the turbines, and decreased flow
cm higher than at the lake boundary. When including 10 speed through the turbine and a few diameters down-
turbines the water level at the power station is increased stream.
by an additional 1.6 cm using a cd of 0.4 (75 kW) and 3 The modeled turbines have not been limited to a max-
cm for a cd of 0.8 (135 kW). This corresponds to a 5.5% imum power extraction, which is a questionable assump-
and a 8.2% increase of the water level respectively, com- tion. However, this report was made to investigate how
pared to the case without any turbines (Fig. 8). Note, the hydrography is altered when extracting kinetic en-
however, that this result regards a constant water level at ergy, and not to study the theoretical potential of the
the downstream boundary. channel. The results showed that power extraction and
In Fig. 9 the velocity field around all turbines is effects on the upstream water level is highly dependent
shown. The kinetic energy is proportional to the cube of on the chosen cd -value.
the velocity, and the largest extraction rate can be found It should be noted that for the parameters used in the
at the location of strongest currents. In Table 3 the pre- simulation, the model was not able to correctly assess
sented values are per turbine, indicating T1 and T2 ex- the water level deviation at the measurement site down-
tracts most power. However, power per cross sectional stream the bridge. When simulating the turbine influ-
area is highest for T7 and T8, which is coincident with ence on the hydrography of the channel, the whole chan-

831
6
nel was considered, not only a single point. For this the References
model has not been validated. However, we are regard- [1] Vindkraftsstatistik 2007. The Swedish En-
ing an increase of water level due to energy extraction, ergy Agency, ES 2008:02, publikationsser-
and not the absolute value of the water level, and for this [email protected], 2007.
the results are still valid.
[2] M. J. Khan, M. T. Iqbal, and J. E. Quaicoe. River cur-
rent energy conversion system: Progress, prospects and
challenges. Renewable & Sustainable Energy Reviews,
5 Conclusion 12:2177–2193, 2008.

Measurements of current speed, current direction and [3] L. S. Blunden and A. S. Bahaj. Initial evaluation of tidal
water level have been compared with 2D-simulations stream energy resources at Portland Bill, UK. Renewable
done with the program Mike21. The simulation program Energy, 31(2):121–132, February 2006.
was able to assess the current speed variations, but the
[4] L. S. Blunden and A. S. Bahaj. Effects of tidal energy
modeled magnitude was slightly low, which is thought extraction at Portland Bill, southern UK, predicted from
to be a result of bathymetry errors. The program could a numerical model. In Proceedings of the 7th European
not assess the water level deviations well. This could be wave and tidal energy conference, EWTEC07, Porto, Por-
improved by adding wind data to the model, but no such tugal, pages 1–10, September 2007.
data was available.
[5] G. Sutherland, M. Foreman, and C. Garrett. Tidal cur-
Variation of the hydrography due to turbines was rent energy assessment for Johnstone Strait, Vancouver
studied. The set of turbines were chosen to experiment Island. Proc. IMechE Part A: Journal of Power and En-
how the upstream water level is affected input flow ergy, 221:147–157, 2007.
and turbine drag is varied, and not to optimize energy
extraction or to calculate the potential of the site. It [6] J. V. Norris and E. Droniou. Update on EMEC activi-
showed that extracting 75 kW would increase the water ties, resource description, and characterisation of wave-
induced velocities in a tidal flow. In Proceedings of
level at the power station from the level without any
the 7th European wave and tidal energy conference,
turbines at the same location with 5.5%, and 135 kW
EWTEC07, Porto, Portugal, September 2007.
would lead to an 8.8% increase, using the specified
level at the downstream lake. Choosing the appropriate [7] R. Carballo, G. Iglesias, and A. Castro. Numerical model
cd -value for each turbine is of major importance since evaluation of tidal stream energy resources in the Ría
the effects on the hydrography is large. de Muros (NW Spain). Renewable Energy, 34(6):1517–
1524, 2009.

[8] Mike 21 and Mike 3 flow model FM, hydrodynamic and

transport module. scientific documentation, 2008. DHI.
Acknowledgments
[9] G. J. Arcement and V. R. Schneider. Guide on select-
The numerical program Mike21 was used with per- ing Manning’s roughness coefficients for natural chan-
nels and flood plains. Water-supply paper WSP2339,
mission from Stefan Ahlman at the DHI Group in Swe-
Unites States Geological Survey, 1984.
den (www.dhi.se). The flow data was provided by Vat-
tenfall AB (www.vattenfall.se) and water level readings [10] J. Smagorinsky. General circulation experiments with the
are from FORTUM AB (www.fortum.se). This work primitive equations. Monthly Weather Review, 91(3):99–
was financed by Ångpanneföreningen’s Foundation for 164, March 1963.
Research and Development, Östkraft’s Environmental
[11] Pierre Y. Julien. River mechanics. Cambridge University
Fund and Vattenfall AB. I would also like to thank my
Press, Cambridge, 2002.
colleagues at The Swedish Centre for Renewable Elec-
tric Energy Conversion for helping out with the measure- [12] I. G. Bryden, S. J. Couch, A. Owen, and G. Melville.
ments and for their valuable comments on this article, Tidal current resource assessment. Proc. IMechE Part A:
and to Antoine Baudoin for his initiation of the grid. Journal of Power and Energy, 221:125–135, 2007.

832
7