Sunteți pe pagina 1din 30

SIMULATION OF THE ASKERVEIN FLOW.

PART 1: REYNOLDS
AVERAGED NAVIERSTOKES EQUATIONS (k  TURBULENCE
MODEL)

F. A. CASTRO
Instituto Superior de Engenharia do Porto, Rua Dr. Antnio Bernardino de Almeida 431, 4200-702
Porto, Portugal
J. M. L. M. PALMA and A. SILVA LOPES
Faculdade de Engenharia da Universidade do Porto, Rua Dr. Roberto Frias s/n, 4200-465 Porto,
Portugal

(Received in final form 16 July 2002)

Abstract. The neutrally stratified flow over the Askervein Hill was simulated using a terrain-
following coordinate system and a two-equation (k ) turbulence model. Calculations were
performed on a wide range of numerical grids to assess, among other things, the importance of spatial
discretization and the limitations of the turbulence model. Our results showed that a relatively coarse
grid was enough to resolve the flow in the upstream region of the hill; at the hilltop, 10 m above the
ground, the speed-up was 10% less than the experimental value. The flows most prominent feature
was a recirculating region in the lee of the hill, which determined the main characteristics of the
whole downstream flow. This region had an intermittent nature and could be fully captured only in
the case of a time-dependent formulation and a third-order discretization of the advective terms. The
reduction of the characteristic roughness near the top of the hill was also taken into account, showing
the importance of this parameter, particularly in the flow close to the ground at the summit and in
the downstream side of the hill. Calculations involving an enlarged area around the Askervein Hill
showed that the presence of the nearby topography affected the flow neither at the top nor downstream
of the Askervein Hill.

Keywords: Askervein Hill, Atmospheric flow over topography, k  turbulence model.

1. Introduction

One of the drawbacks in the computer simulation of atmospheric flows over topo-
graphy is the lack of experimental data; this impairs the development and thorough
validation of numerical models. That difficulty is partly attenuated in the case of the
Askervein Hill, for which extensive field measurements have been made available
(cf. Taylor and Teunissen, 1983, 1985). However, despite the many studies after
this field measurement campaign in the early 1980s, to fully replicate the Ask-
ervein flow by computer simulations is still a major challenge (see, for instance,
Walmsley and Taylor (1996) and references therein). For instance, Kim and Patel
(2000), corroborating Teunissen et al. (1987), identify the downstream hilly region
 E-mail: fcastro@dem.isep.ipp.pt

Boundary-Layer Meteorology 107: 501530, 2003.


2003 Kluwer Academic Publishers. Printed in the Netherlands.
502 F. A. CASTRO ET AL.

as the cause for flow recirculation in the lee-side of the Askervein Hill. This shows
that, after so many years, there is still no agreement on how large and detailed the
integration domain must be to obtain a successful simulation of the Askervein flow,
not to mention the uncertainties due to turbulence modelling.
Predicting the flow in the lee-side of the Askervein Hill remains one of the
unsolved issues although the simpler, even linear, models have shown an ability
to resolve both the upwind flow, and the flow at the hill summit of isolated hills
of moderate slope (e.g., Mason and King, 1985). However, isolated hills are rarely
found in nature and the poor prediction of the flow in the hills lee-side may af-
fect simulations of the flow around a second hill or a set of hills located further
downstream.
Raithby et al. (1987) pioneered the use of three-dimensional computational fluid
dynamics (CFD), finite volume techniques, to simulate the Askervein flow, and thus
their study is used throughout this work as a reference with which our results were
compared. Those authors used what is currently considered a coarse grid (20 20
19), and the question arose as to whether the limitations identified in that study
were a limitation of the turbulence model or whether they could be removed simply
by using more accurate grids.
The present study is one single piece of a research programme whose ultimate
objective is the simulation of atmospheric flows over highly complex mountain-
ous regions (e.g., Castro, 1997; Maurizi et al., 1998; Silva Lopes, 2000). Here,
because of its well documented field campaign, the Askervein flow provided the
ideal framework for both development and appraisal of our computer model, based
on finite volume discretization of the flow equations in primitive variables and
a two-equation (k-) turbulence model. Particular emphasis has been placed on
the grids refinement because, besides reducing the spatial discretization error,
fine grids improve the topographys resolution, adding surface detail that would
be otherwise lost. We also assessed the importance of some basic issues, such as
variable roughness and time dependent formulation, not covered by the published
literature.

2. Mathematical Model and Numerical Techniques

2.1. F UNDAMENTAL EQUATIONS

The continuity (1), the momentum (2) and the turbulence model Equations (5) and
(6) given below were written in tensor notation for a generic coordinate system,
 
j
Uj k
= 0, (1)
j
(J Ui )  j
  j  j

+ j Uk Ui k = j Pi + j (ki + ki ) k , (2)
t
ASKERVEIN FLOW; k  TURBULENCE MODEL 503

where
 
Ui m Uj m
ij = + m i (3)
J m j
and
 
2 t Ui m Uj m
ij = kij + + m i . (4)
3 J m j

The eddy viscosity was given by t = C k 2 /, where k and  were obtained
from
   
(J k)  j
 1 t k m j
+ j Uk kk = j +
t J k m k k (5)
+J (Pk )

and
   
(J )  j
 1 t  m j
+ j Uk k = j +
t J  m k k
  (6)
C1  C2  2
+J Pk ,
k k
where
1 Ui m
Pk = ij . (7)
J m j
Coefficients are: C1 = 1.44, C2 = 1.92, C = 0.033, k = 1.0 and  = 1.85.
In the above equations, P and Ui are respectively the ensemble Reynolds av-
eraged pressure and velocity components in the x i Cartesian directions, and
are the airs density and molecular viscosity, and k and  are the turbulent kinetic
energy and its dissipation rate, whose transport Equations (5) and (6) and constants
C1 , C2 and k were set as in Launder and Spalding (1972), whereas the constants
C and  followed the recommendations by Beljaars et al. (1987).
The terrain-following coordinate system is defined by transforming a phys-
j
ical Cartesian coordinate system x i into a computational system i , where k =
J j /x k and J is the determinant of the Jacobian matrix of the coordinate trans-
formation (cf. Knupp and Steinberg, 1994). This transformation makes it relatively
simple to treat the boundary conditions and to use a structured mesh, where the
physical domains boundaries are the coordinate surfaces following the topography.
The transport Equations (1), (2), (5) and (6) were discretized by finite volume
techniques using a central differencing scheme for the diffusive terms. The advect-
ive terms were discretized by the hybrid scheme (cf. Patankar, 1980). In the case of
504 F. A. CASTRO ET AL.

the momentum Equation (2), an alternative third-order truncation scheme was used
for the advective terms, identical to the QUICK scheme for non-uniform meshes
(Leonard,1979).
The resulting set of coupled algebraic equations was solved using the SIMPLE
algorithm of Patankar (1980) and the tri-diagonal matrix algorithm solver. The
pressure/velocity coupling in non-staggered meshes was treated following the
pressure-weighted interpolation as in Rhie and Chow (1983) and Miller and
Schmidt (1988).
The equations were solved either in time-dependent, ensemble-averaged RANS
(Reynolds averaged NavierStokes), using a second-order implicit scheme, or in
steady state formulation, time-averaged RANS, in which case the time derivatives
in Equations (2), (5) and (6) were dropped, i.e., a formulation appropriate for stat-
istically stationary turbulent flows. In either case the equations were solved until
mass and momentum budgets could be satisfied to a dimensionless residual below
5 104 .
The time dependent formulation, restricted to Section 4.5, accounted for the
possibility of low frequency unsteadiness of the mean flow, i.e., statistically non-
stationary turbulent flows, in a fashion identical to what has been used in the
computer simulation of flow over bluff bodies. The high frequency fluctuations
around the mean flow, which we conventionally accept as turbulence, is accounted
for in the turbulence model. This has been designated by URANS or T-RANS,
which stand respectively for unsteady or time-dependent RANS (cf. Nakayama and
Miyashita, 2001; Kenjere and Hanjalic, 2002). T-RANS differs from large-eddy
simulation (LES), which simulates the turbulent fluctuations, though restricted by
the filtering properties of the spatial mesh. The time averaging of results obtained
by ensemble Reynolds averaging may yield results different from a steady state
formulation (see, for instance, Durbin and Reif (2001), p. 189).
The development and validation of this computer code (named VENTOS, see
Castro and Palma (2002)) for a series of atmospheric flows was the subject of the
work by Castro (1997).

2.2. B OUNDARY CONDITIONS

The boundary close to the ground was modelled by a rough surface, using the wall
laws of Launder and Sharma (1974), based on a wall region of constant tangential
Reynolds stress t = u2 and a log-law tangential velocity ut ,
 
u z
ut = ln 1 + . (8)
z0
Here, = 0.4 is the von Krmn constant, u is the friction velocity, z is the
distance above the ground with characteristic roughness z0 . Following the work
by Beljaars et al. (1987), z0 = 0.03 m was used throughout this study, except for
Section 4.4 where z0 was specified as a linear function of the terrain elevation. The
ASKERVEIN FLOW; k  TURBULENCE MODEL 505

kinetic energy of turbulence and its dissipation in the control volumes adjacent to
the ground were given by

u2 u3
kwall = and wall = , (9)
C
1/2 (z + z0 )

and the tangential Reynolds stress and the mean production and dissipation of k
were as follows:
1/2
C1/4 kwall ut
t = , (10a)
ln (1 + z/z0)
ut
Pk = t  , (10b)
z
3/2
C3/4 kwall
= ln (1 + z/z0 ) . (10c)
z
The pressure at the boundaries was obtained by linear extrapolation from the in-
ner nodes. The inflow conditions were determined by the simulation of a turbulent
boundary layer over flat terrain, as in Section 3. The velocity was tangential at the
top and lateral boundaries. The velocity outflow condition was obtained by linear
extrapolation from the inner nodes, constrained by global mass conservation.

2.3. C OMPUTATIONAL DOMAIN AND MESHES

The Askervein Hill (Figure 1) is nearly elliptic, with minor and major axes of
roughly 1000 and 2000 m. The highest point (ht) is 116 m above the surrounding
terrain and the central point of the hill, along the major axis, is called point cp. Fig-
ure 1 also shows the reference site (RS), the 210o wind direction in our simulations
and the lines A, AA and B along which the measurements were made.
There are two maps to simulate the Askervein flow: Maps A and B (Walmsley
and Taylor, 1996). Map B, with higher resolution, focuses on the Askervein hill
only, whereas map A also includes the neighbouring hills, as shown in Figure 2.
There are also three other differences between maps A and B, most likely due to
the terrain digitisation procedure, which may affect the simulation: An elevated
surface in the lee-side of the hill in map A, a maximum height about 3 m lower in
map A, and a different profile of the hill along line B.
A total of 12 numerical meshes was used (Table I), comprising 7 levels of
resolution of the horizontal plane. Those meshes were expanded from the ht point
toward the edges of the domain, starting from a minimum control volume width
and length of 10 m for grids 1 to 11, and 20 m for grid 12. The five resolutions
along the vertical directions resulted from three different numbers of grid nodes
(NK = 15, 19 and 31) and five heights of the control volumes closer to the ground
("z = 0.3, 0.6, 1.2, 1.5 and 3.0 m).
506 F. A. CASTRO ET AL.

Figure 1. Askervein Hill topographic map.

Before the simulations, the maps were rotated to align the 210 incoming flow
with one computational direction. The bulk of the simulations (Table I) was based
on map B and a domain of 4000 4000 m2 , centred on ht. In all simulations the
top boundary was 700 m high. To assess the topographys effect downstream of
Askervein (Section 4.6), both maps A and B were based on an enlarged domain of
10000 8000 m2 (mesh 12 in Table I).
The domain dimensions were determined after a series of tests in Castro (1997),
which did not show any effect of the lateral, outlet and top boundary conditions in
the regions of interest.
ASKERVEIN FLOW; k  TURBULENCE MODEL 507

Figure 2. Perspective view of Askervein Hill. Map A (top figure) and map B (bottom figure). The
circle in map A identifies a feature not represented in map B. The vertical scale was magnified 22
with respect to the horizontal scale.

3. Inflow Boundary Conditions

The inflow conditions were determined by simulating a turbulent boundary layer


over a flat rough terrain, with periodic longitudinal boundary conditions and the
hybrid discretization scheme, for each of the grids in Table I.
Figure 3 shows the vertical profiles of horizontal velocity, turbulence kinetic
energy and Reynolds shear stress at the reference site (RS) using a mesh of 97 97
nodes in a 4000 4000 m2 area. Our results closely match the horizontal velocity
measured by all anemometers for distances to the ground of less than 50 m. The
deviation of the TALA kite results for z > 50 m has been attributed to stratification
effects (e.g., Mickle et al., 1988). In the case of k , the largest deviations (about
65%) occurred at z  20 m when compared with experimental results obtained
with Gill anemometers. However, according to Taylor and Teunissen (1985), the
Gill results are likely to underestimate the actual turbulence levels. The sonic res-
ults are more reliable and showed a maximum difference to the numerical results

of 45% at z  50 m. In the case of xz , based on sonic anemometer data at only
two points, there is a maximum deviation of about 10% between our results and the
experimental data. The difference in the vertical distributions of k and xz

obtained
508 F. A. CASTRO ET AL.

TABLE I
Computational grids. NI and NJ are the number of nodes in each of the
two horizontal directions; NK is the number of nodes along the vertical
direction; "z is the height of the first control volume above the ground;
fxy and fz are the maximum expansion factors of the meshes in the
horizontal and vertical directions, respectively. The minimum horizontal
width and length of the control volume ("x and "y) was 10 m on grids 1
to 11, and 20 m on grid 12.

Grid NI NJ NK "z (m) fxy fz Map

1 27 27 15 1.2 1.46 1.61 B


2, 3 27 27 19 1.5, 3.0 1.46, 1.46 1.39, 1.31 B, B
4, 5 37 37 19 1.5, 3.0 1.26, 1.26 1.39, 1.31 B, B
6 57 57 31 0.6 1.13 1.22 B
7, 8 77 77 31 0.6, 0.3 1.08, 1.08 1.22, 1.26 B, B
9 97 97 31 0.6 1.05 1.22 B
10 117 117 31 0.6 1.04 1.22 B
11 155 155 31 0.6 1.02 1.22 B
12 155 155 31 0.6 1.05 1.22 A,B

Figure 3. Vertical profiles at reference site RS. V /Vref , horizontal velocity; k = k/(Vref
2 ), turbu-

lence kinetic energy and xz = u v (/(Vref ) Reynolds shear stress. Vref is the horizontal velocity
2

measured at 10 m above the ground.


ASKERVEIN FLOW; k  TURBULENCE MODEL 509

with two values of C (the standard value of 0.033 and 0.036, as suggested by the
experimental results of Taylor and Teunissen, 1985) was not enough to explain the
discrepancies found between measurements and simulations.

4. Presentation and Discussion of Results

This section includes simulations of the flow over the Askervein Hill for an incom-
ing flow at 210 (Figure 1). In Subsections 4.1, 4.2 and 4.3 we discuss the mean
and the turbulent flow fields. The results are compared with the field measurements
TU-03b, MF-03d and TK03 of Taylor and Teunissen (1985) wind-tunnel data of
Teunissen et al. (1987) and with the numerical results of Raithby et al. (1987),
Beljaars et al. (1987) and Zeman and Jensen (1987). Three separate studies, in
subsections 4.4, 4.5 and 4.6, assess the effects of variable roughness, time depend-
ent calculations and the inclusion of Askerveins nearby hills in the integration
domain.
Throughout the following sections, the nondimensional speed-up is defined by
Vh (z) Vref (z)
"S(z) = , (11)
Vref (z)

where Vh (z) is the modulus of the horizontal velocity and Vref (z) is that quantity
at reference point RS (Figure 1).

4.1. M EAN FLOW

4.1.1. Mean Flow (Lines AA, A and B)


The speed-up results along line AA (Figure 4) generally agree with the experi-
mental results. The prediction on coarser grids (Figure 4a) of the minimum of "S,
at about 400 m downstream of the hill, shows a dependence on "z that does not
occur in other regions of the flow. Simulations on finer grids (Figure 4b) showed
good agreement with experimental results at all locations along direction AA, for
all levels of grid refinement. Because of the linear treatment of the advective terms,
the MSFD method (Beljaars et al., 1987) cannot follow the steep decline of "S in
the hills lee side where non-linear effects are known to prevail.
Along line A (Figure 5), at x = 0 m the speed-up was 0.80, compared to
the experimental value of about 0.88. The angle between the local and inlet ve-
locity vectors along line A was always lower than the experimental value and
this difference could be as high as 10 degrees. The larger differences between
the numerical and the experimental speed-up occurred downstream of the hill at
about x = 400 m, where the simulations showed a larger scatter than in Figure 4.
The results of Raithby et al. (1987) are similar to our results on coarser meshes
(Figure 5a). Identically to line AA, downstream of the hilltop along line A the
results of Beljaars et al. (1987) were above the experimental results and, at the top
510 F. A. CASTRO ET AL.

Figure 4. "S at z = 10 m above the ground along line AA on coarser (a) and finer grids (b). x = 0 m
represents the cp point.
ASKERVEIN FLOW; k  TURBULENCE MODEL 511

of the hill, in agreement with the experimental results because the linear treatment
used in the MSFD method tends to overestimate the speed-up at the hilltop, as
found in Xu et al. (1994) and Walmsley and Taylor (1996).
For "z = 0.6 m there was a reduction of "S, downstream of ht, with the grid
refinement (Figure 5b). The minimum "S varied from 0.65 to 0.8, for grids
6 and 10, respectively with 57 57 and 117 117 in horizontal planes. The
minimum "S on grid number 6, the coarser grid, agreed with the experimental
result, whereas that on grid 10 differed by 20%. Grid number 8 (77 77 31,
"z = 0.3 m) also agrees with the field measurements. However, this is simply the
consequence of two opposing effects: the results showed that if the reduction of "z
tends to increase the minimum value of "S, the grid refinement in the horizontal
plane tends to do the opposite.
Figure 5b includes the results with a third-order discretization scheme, and the
scatter of the numerical data is larger when compared with Figure 5a. The third-
order scheme reduced the minimum speed-up ratio even further compared with the
results in Figure 5b: "S was close to 1 and, above all, displayed an inflection
point at x = 300 m found to be associated with the occurrence of a region of
reversed flow, as discussed in Section 4.2.
Figure 5 shows that 400 m downstream of the hill the results were very sensitive
to numerical parameters, namely the distance of the first grid node to the wall, the
order of the finite difference approximation and the discretization of the horizontal
plane. In this region there was a strong interplay between numerical and physical
modelling, and the higher numerical accuracy increased the discrepancy between
the numerical and experimental results. If we had been interested in other regions
of the flow, the coarser grids with the larger distance from the ground would have
been appropriate (grid 2 in Table I). A possible explanation for these findings will
emerge from the analysis in the following sections.
The speed-up predictions along line B (the direction along the hills crest per-
pendicular to both lines A and AA, Figure 1) were not very sensitive to the grid
size and the results were in good agreement with the experiments, except for the
region around ht, x = 0.

4.1.2. Mean Flow (Vertical Profiles at cp and ht)


Accurate prediction of the speed-up at heights between 10 and 60 m is of great
importance for wind energy applications. At this height range, a deviation of nearly
10% from the measurements was found over ht and cp (Figure 6). However our
results compare well with the numerical results of Zeman and Jensen (1987) and
Raithby et al. (1987) and also with the wind-tunnel results of Teunissen et al. (1987)
(AES rough model). The largest deviations from the experimental results occurred
near the ground, below z = 5 m, where the experimental "S increases for lower z
and our results were nearly constant, in agreement with all previous studies except
for the numerical results of Raithby et al. (1987) (Figure 6a) that followed the
512 F. A. CASTRO ET AL.

Figure 5. "S at z = 10 m above the ground along line A and (a) finer grids, as in Figure 4b, and (b)
finer grids and discretization schemes of first- and third-order accuracy. x = 0 m represents the ht
point.
ASKERVEIN FLOW; k  TURBULENCE MODEL 513

Figure 6. Vertical profile of "S at (a) ht and (b) cp.

experimental results. Either over ht or cp (Figure 6) the results were marginally


affected by the grid resolution or the order of the spatial discretization scheme.
The agreement between our calculations and the experimental results in Figure
6 was inferior to that achieved by Raithby et al. (1987), where the turbulence
boundary conditions were implemented by simply excluding the production and
dissipation terms in the k transport Equation (5) at the first control volume, 1.5 m
above the ground. By doing so, as reported in Castro (1997), we could replicate
both their results and the measurements. With the grid refinement both in the
horizontal and vertical directions, the results obtained with Raithbys formulation
converged to those obtained with our original boundary condition, which deviated
from the trend in the experimental data. This showed how important the wall
boundary condition is in the prediction of "S for z < 5 m, and that the good
agreement between the experimental results and those of Raithby et al. (1987) was
fortuitous.

4.1.3. Momentum Budget


To identify the transport mechanisms in different regions of the flow, Figure 7
shows the magnitude of the terms in the equation for longitudinal momentum
transport, at 2 and 10 m above the ground along line A.
514 F. A. CASTRO ET AL.

Figure 7. Longitudinal momentum budget along line A at z = 10 m. Advective (), diffusion ( )


and pressure ( . ) term. The vertical axis (0 = Tp + Td Tconv ) was made dimensionless using
the maximum of the diffusion term. Grid 11 in Table I.

In the case of z = 10 m, the region between x = 200 and 800 m, downstream


of the hilltop, was the only one where diffusion was a dominant term. This was the
region where interaction between the mean and turbulent fields was stronger and
weakness in the turbulence modelling and numerical discretization schemes were
expected to show up (cf. Figure 5). This is a recurring characteristic in our results.
The diffusion term was null or negligible in other regions of the flow, dominated
by pressure gradient and advective terms. This was consistent with the fact that
z = 10 m roughly corresponded to the top of the inner layer, where the mean is
hardly affected by the turbulent field (Jackson and Hunt, 1975). For z = 2 m, inside
the inner layer, diffusion plays an important role on both sides of the hill.

4.2. F LOW PATTERN IN THE LEE - SIDE OF THE HILL

It was somehow intriguing that the good agreement between our results and the
experimental speed-up along line AA (Figure 4) could not be replicated along line
A and was so sensitive to grid size (Figure 5). This was caused by a shallow recir-
culation zone formed downstream of ht for grid 7. This region became thicker as
the numerical grid was refined (Figures 8a and b) or the order of the discretization
scheme increased (cf. Figures 9a and b) and was responsible for the increasingly
ASKERVEIN FLOW; k  TURBULENCE MODEL 515

Figure 8. Streaklines of particles regularly spaced at the outlet section, 1 m above the ground. left)
grid 57 57 31; right) grid 117 117 31. The vertical scale was magnified 5 with respect to
the horizontal scale.

Figure 9. Streamwise velocity contours 2 m above the ground. (a) hybrid scheme on grid
155 155 31 (Grid 11); (b) third-order scheme on 77 77 31 (Grid 7). Points dcp and dht,
located downstream of cp and ht, are located close to positions where the mimimum values of "S
occur, along lines AA and A.

lower "S values predicted at 10 m above the ground (Figure 5). Figure 8 shows
the flow pattern of two steady state calculations that could satisfy the stopping
criterion. The ability to capture the reversed flow, up the hill, in the downstream
side was due to nothing but an increase of the spatial resolution in the horizontal
plane.
For grid sizes above 155 155 and 77 77, for hybrid and third-order
schemes respectively, more difficulties were found in meeting the stopping cri-
terion. In those cases the flow field downstream of ht showed large variations from
one iteration to another, when trying to resolve the aforementioned recirculation
zone, finally preventing the convergence of calculations based on a steady state
formulation.
516 F. A. CASTRO ET AL.

This behaviour reveals the type of numerical phenomena well illustrated in the
work by Gresho and Lee (1981), and we interpreted it as an indication of the
collapse of the steady state formulation, which may drive the iterative procedure
in the search of a solution that might not exist. There have been references to this
incipient recirculation zone since the early reports on the Askervein campaign (e.g.,
Taylor and Teunissen, 1985), though it has proven elusive to modelling approaches
used in most of the numerical studies. In recent calculations, Kim and Patel (2000)
also predicted this recirculation region, which they attributed to the inclusion, in
the integration domain, of the topography north-east of the Askervein Hill. We
may recall the study by Taylor (1977) for two-dimensional topography, accord-
ing to which the formation of recirculating regions is possible when the h0 /Lh
(height/width) ratio is higher than 0.52. In the present case, the topography is
nearly two-dimensional, with h0 /Lh  0.5 along line A, and the flow is on the
verge of forming a recirculation zone. We will try to uncover some of this regions
properties by time dependent calculations in Section 4.5.

4.3. T URBULENT FLOW

4.3.1. Turbulent Flow (Lines AA, A and Vertical Profiles)


The nondimensional turbulence kinetic energy k was greater than the experimental
data along lines A and AA, 10 m above the ground (Figure 10) and upstream of the
hill. The trend of the experimental values upstream of ht was only partly matched
by our numerical results. The use of both hybrid and third-order schemes did not al-
ter the results much, since the third-order scheme was not used in the discretization
of the turbulence model equations for numerical stability. It is likely that rapid-
distortion and streamline curvature effects, if included, as in the case of Zeman and
Jensen (1987), would reduce the magnitude of k , improving the agreement with
the experimental values of Walmsley and Taylor (1996). Moreover, it is also known
that the measurements obtained with the Gill anemometers may be about 20% low
compared with the sonic anemometer results (Taylor and Teunissen, 1985).
With respect to the vertical profiles (Figure 11), in both locations ht and cp
the turbulent kinetic energy increased as the distance to the ground was reduced.
Although the trend was identical, the numerical results could be as low as half
of the field measurements. Changes of the vertical profiles of turbulence kinetic
energy at ht and cp caused either by grid refinement, order of the discretization
scheme or terrain roughness (also in Figure 11 and to be adressed in Section 4.4)
were restricted to the first 10 m above the ground.

4.3.2. Turbulence Kinetic Energy Budget


In this section we analyse and justify some of the trends of the turbulent field along
line A and at the top of the hill, based on the kinetic energy budget (Figure 12) and
strain rates (Figure 13) at z = 2, 10 and 70 m. These are locations representative
of the interior, and tops of the inner layer and the outer layer, respectively.
ASKERVEIN FLOW; k  TURBULENCE MODEL 517

Figure 10. k at 10 m above the ground along (a) line A and (b) line AA. x = 0 m corresponds to
the cp and ht points, respectively.
518 F. A. CASTRO ET AL.

Figure 11. Vertical profiles of k in points a) ht and b) cp. Numerical results obtained using a
77 77 31 mesh and "z = 0.6 m.

In the upstream region, x 1000 m, and for all three heights, the turbulent
transport and advective terms were nearly zero. The production and dissipation
terms were balanced and the strain rate U/z dominated this region, indicating
a state of nearly local equilibrium of the turbulent field. This was consistent with
observations at the reference site (RS) where the incoming flow was well described
by a log-law. In this region the turbulence model yielded vertical profiles of k and
xz in agreement with the experimental results (Figure 3).
At z = 2 m, Figure 12a, the budget was almost entirely dominated by the pro-
duction and dissipation terms, i.e., a local equilibrium state. Exceptions occurred
near the hilltop, where the advection and turbulent transport terms showed up but
with small magnitude, and downstream of ht, in the region 200 x 600 m,
where all terms were of similar magnitude. Here, equilibrium does not hold for
z = 2 m, and this agrees with Zeman and Jensen (1987), who found that this
assumption may be restricted to locations below 1 m above the hilltop. In this
region, U/z was always the largest strain rate (Figure 13a), followed by U/x
and W/z which, related by continuity, respond to the presence of the topography.
For locations at z = 10 m, Figure 12b, the production and dissipation terms
dominated regions x 300 m and x 700 m; in between, the advection and tur-
bulent transport terms were active in establishing the budget. For 300 x 0 m,
ASKERVEIN FLOW; k  TURBULENCE MODEL 519

Figure 12. Terms of the turbulence kinetic energy transport equation along line A: Advective term
(), turbulent transport (- - -), production ( . ) and dissipation (. . . . . . ). Terms are of the form
0 = Tp + Td + T Tconv .

the production decreased and was almost null near the top of the hill. Advection
and turbulent transport balanced out this budget but with advection in a more in-
tervening role. In the 0 x 300 m region, production and dissipation increase
in magnitude, reaching their maxima about 250 m downstream of the hilltop, also
the location for the maxima of k (Figure 10). Near the top of the hill, U/z was
marginally the largest strain rate and its maximum was also found at x = 250 m.
In the upstream slope all the other strain rates had identical magnitudes.
At z = 70 m and upstream of ht, advection was the second largest term, almost
balancing production, Figure 12c. The assumptions behind the rapid distortion
520 F. A. CASTRO ET AL.

Figure 13. Mean strain rates along line A.

theory may be valid in many locations along line A at z = 70 m. Near the hilltop,
U/z was no longer the largest strain rate, Figure 13c. Downstream of ht, the flow
evolved to a state dominated by production and dissipation but with non-negligible
advection and turbulent transport terms.
According to either Zeman and Jensen (1987) or Walmsley and Taylor (1996),
the streamline curvature is important in establishing the turbulent field near z =
10 m. Here, this was suggested by Figure 13b showing that for z = 10 m and
around x = 0 all strain rates had similar magnitudes. The reduction of U/z
observed near the hilltop, together with W/x < 0, convex curvature, with a
similar magnitude, should have led to a larger reduction of k around x = 0 m,
than observed in Figure 10. We tend to think that this was not only linked with
ASKERVEIN FLOW; k  TURBULENCE MODEL 521

the inability of the present method to deal with the streamline curvature effects,
but also with the Boussinesq formulation of the Reynolds stresses, where there
are no negative contributions in the production term of k. It can be seen for
instance in the upstream side of the hill, where the strain V /y is reduced
that large errors may occur in the account of the production term. For instance,
because U/x  W/z, u 2 U/x and w 2 W/z should give two op-
posite sign contributions to the k production term, which will not occur in the
present case because those terms are modelled by t (U/x)2 and t (W/z)2.
Another problem arises because of the turbulence anisotropy, where u 2 is larger
than w 2 . Because U/x  W/z in the present case, this should give a
larger contribution to the u 2 U/x production term, instead of the t (U/x)2
 t (W/z)2 contributions.
A well-known limitation of the k  model, when applied to the Askervein
flow, is the negative shear stress xz above the hilltop, as found in Beljaars et al.
(1987). The increased magnitude of the advection term upstream of the hilltop with
increasing z, accompanied with a reduction in U/z, shown in Figures 12 and 13,
indicated the collapse of the present shear stress model, based on the local nature
of the flow.
Other important aspects of the Askervein case study are the effects of the con-
stant roughness length, steady state flow assumptions and also the possible impact
of neighbouring hills downstream of Askervein. The following sections deal with
these aspects.

4.4. T HE EFFECT OF THE ROUGHNESS LENGTH

The speed-up ratio "S, both at the top and on the lee-side of the hill, is strongly
dependent on the roughness (see, for instance, the wind-tunnel study of Teunissen
et al., 1987 or Walmsley and Taylor, 1996). The most common value for charac-
teristic roughness in Askervein studies has been 0.03 m. However, it is known,
for instance Zeman and Jensen (1987), that the characteristic roughness near the
hilltop could be 3 times lower, 0.01 m, and the question arose whether this could
be responsible for the lower "S values obtained by the present numerical method
in the lee-side of ht (Figures 5 and above points ht and cp (Figure 6).
In this section the characteristic roughness was varied between 0.03 m and
0.01 m as a linear function of the height, starting at two values of height, zc0 = 60
and 90 m, to determine to what extent the prediction of the mean and turbulent
fields could be improved. This was the simplest way to influence the near-wall
region and enabled us to assess the level of uncertainty associated with the apparent
roughness length in the Askervein Hill (Walmsley and Taylor, 1996).
The vertical profiles of "S above points ht and cp are shown in Figure 14,
where the constant roughness results are also displayed. As expected, the velocity
momentum close to the ground increased, improving the agreement between field
measurements and predictions, particularly in the case of zc0 = 60 m, at both
522 F. A. CASTRO ET AL.

Figure 14. Vertical profiles of "S in points (a) ht and (b) cp. Numerical results obtained using a
77 77 31 mesh and "z = 0.6 m.

ht and cp locations (Figure 14). The reduction in roughness length produced a


great improvement, better than that obtained by Zeman and Jensen (1987). The
best results were obtained in the case of zc0 = 60 m, which followed very closely
the experimental results between 1 and 4 m above ht. In both locations, cp and ht,
the differences associated with the order of the spatial discretization scheme were
minor.
In the case of turbulence kinetic energy (Figure 11), the changes in roughness
also affected the first 6 m, resulting in a reduction of about 30% of k 1 m above
the ground. Once again, the better results were found using zc0 = 60 m and no
measurable differences could be attributed to the two numerical schemes.
In the horizontal evolution of "S, 10 m above the ground along line A (Figure
15a), the main differences caused by the reduction in roughness length occurred
in the flows downstream region, where the lower z0 yielded higher "S values
(Figure 15a), as shown mainly by the hybrid scheme. The third-order results with
lower characteristic roughness led to higher "S values in the 100 m x 400 m
range, but still below the experimental results. In the case of k (Figure 15b), only
the third-order results with a reduction in roughness length improved the results in
the downstream region, with a maximum close to that found by the experimental
results. The results of the hybrid scheme showed an inverse behaviour with the
ASKERVEIN FLOW; k  TURBULENCE MODEL 523

Figure 15. "S(z) and k at 10 m above the ground along line A. Numerical results obtained using a
77 77 31 mesh and "z = 0.6 m.
524 F. A. CASTRO ET AL.

Figure 16. Time evolution of the x-component of the velocity field at 4 locations, 10 m above the
ground: (a) ht; (b) cp; (c) downstream of ht, x  390 m along line A; and (d) downstream of cp,
x  390 m along line AA. For a clearer identification of these four locations, refer to Figure 9.
Results for "z = 0.6 and meshes 155 155 31 and 77 77 31 (Grids 11 and 7 in Table I),
respectively in the hybrid and third-order discretisation.

constant roughness results presenting higher values in this region of the flow. The
results obtained along line AA, both in the case of "S and k , were similar to those
found with constant roughness length and were not included.

4.5. T RANSIENT SIMULATIONS

Following the discussion in Subsection 4.2 on the difficulties of convergence on


fine meshes, simulations with time-dependent formulation were also performed.
The equations were integrated during 3600 s ("t = 1 s) on grids 7 and 11, starting
from converged steady state solutions.
The velocitys x-component was monitored at four locations (Figure 16), all of
them 10 m above the ground: At ht, cp and x = 390 m, along both lines A and AA,
ASKERVEIN FLOW; k  TURBULENCE MODEL 525

Figure 17. "S(z) at 10 m above the ground along line A. Numerical results obtained with
"z = 0.6 m. Both third-order results represent time averaged values over two periods.

close to the location of the minimum experimental values of "S (see also Figure
9). One can see in Figure 16 that the hybrid scheme does not yield any transient
behaviour, contrary to the results with the third-order scheme, which, after an initial
time of about 500 and 1500 s displayed a steady periodic behaviour with periods of
about 360 s and 200 s, respectively, in the case of constant and variable roughness
(Figure 16c). After approximately 2000 s, the variable roughness case does not
show negative velocities downstream of ht, indicating that the recirculation zone
is now restricted to lower distances to the ground. The vertical profiles of "S
and k at ht (not included here), averaged over two periods, were identical to the
corresponding profiles in a steady state formulation, Figure 6.
Based on the time averaged "S along the line A, we may conclude that the time
dependent formulation affects only the flow downstream of ht (Figure 17). The flow
at 210 degrees over the Askervein Hill possesses a time dependent nature restricted
to the region downstream of the hill. Compared with the steady state formulation,
one can see that the best agreement with the experimental data was obtained using
the transient formulation and variable roughness, in which case our results at the
position of the experimental minimum were about 15% lower.
The experimental studies report on this behaviour, but there is no quantification
that gives us a more precise assessment of our results. There is a high degree of
resemblance with what has been observed in the real flow and we are inclined to
conclude that this is in fact the phenomenon described in the field measurement
526 F. A. CASTRO ET AL.

reports (cf. Taylor and Teunissen, 1983, 1985), which is not much different from
what can occur in pressure-induced boundary-layer separation (cf. Schlichting and
Gersten, 2000).
Based on our experience of computer simulation of atmospheric flows other
than the Askervein flow, it is our current belief that, provided the numerical accur-
acy is sufficiently high and the terrain topography is moderately steep, atmospheric
flow calculations will likely display time dependent features. The qualitative state-
ments sufficiently high and moderately steep reflect our current understanding,
because until now we found no way, if one exists (c.f. Wood, 1995), of knowing
beforehand whether the calculation will display a time dependent behaviour. Under
such circumstances, the steady state formulation may impose unwanted constraints
on the physics of the flow.

4.6. T HE INFLUENCE OF THE HILLS DOWNSTREAM OF THE A SKERVEIN H ILL

To asses the influence of the hills downstream of the Askervein Hill, simulations
were also performed, using both maps A and B (Figure 2) and a hybrid scheme.
Kim and Patel (2000) attribute the improved agreement between their results and
the experimental data, compared with Raithby et al. (1987), to the blockage caused
by the downstream hills.
The "S over cp was identical in either map A or B. In the case of ht, the "S on
map A was slightly lower than on map B; however, this might be a consequence of
the lower peak height of map A. As for the results along lines AA and A (Figure
18), map B showed a better agreement with experimental results in nearly all loc-
ations. Based on the smoother trends of the results, map A seems to include less
topographic detail. The elevated surface in Figure 2 caused a small plateau, at x
between 200 and 300 m, and a shift in the downstream direction of the "S curve
along line A.
Despite the differences between the two maps, in the region where they should
be strictly identical, either along lines A or AA, we see no differences that can be
attributed to the hills downstream of the Askervein Hill because the results in maps
A and B tend to overlap for x > 600.

5. Conclusions

The neutrally stratified flow over the Askervein Hill was simulated using a terrain-
following coordinate system and a two-equation (k ) turbulence model. Calcula-
tions were performed on a wide range of numerical grids and flow conditions from
which the following conclusions could be drawn.
1. The mean flow upstream of the Askervein Hill was well predicted, even with a
relatively coarse grid (27 27 19). The main deficiencies found in predic-
tions, based on steady state formulation and a constant roughness, concerned
ASKERVEIN FLOW; k  TURBULENCE MODEL 527

Figure 18. "S at 10 m above the ground along lines (a) A and (b) AA. Results from simulations
using 155 155 31 grid nodes and the hybrid scheme.
528 F. A. CASTRO ET AL.

the flow in the lee-side of the hill and the speed-up values over points ht and
cp; for instance, at the hilltop and 10 m above the ground the speed-up was
underpredicted by less than 10% of the experimental value.
2. At the hilltop, the field-measured speed-up increased for lower distances to
the ground, whereas our predictions showed a constant value for distances of
less than z = 4 m and a lower speed-up. This was found to be related to the
thickness of the first control volumes above the surface and the wall boundary
laws, based on local equilibrium of turbulence. The vertical speed-up profiles
at ht and cp, however, agreed well with the wind-tunnel results of Teunissen et
al. (1987).
3. At 10 m above the ground, along line AA and on the lee-side of the hill, the
speed-up values were well predicted, whereas along line A the predictions
depended on the grid size. The grid refinement revealed a recirculation zone
with low speed-up and high turbulence intensity.
This was a prominent feature, which set the main characteristics of the
whole downstream flow. This region, of an intermittent nature, could be
captured fully only in the case of a time-dependent formulation and a third-
order spatial discretization scheme for the advection terms of the momentum
equations.
4. Upstream of the hilltop, 10 m above the ground and along line A, k was over-
predicted. That was attributed to modelling limitations of the production term
and the inability of the k model to accommodate the effects of the streamline
curvature.
5. For locations closer to the hilltop, the reduction in the characteristic roughness
improved the agreement between the numerical and the experimental data in
the vertical profiles, both of k and speed-up at ht and cp.
6. Based on calculations involving an enlarged area around the Askervein Hill, we
concluded that the nearby hills did not affect the flow at the top or downstream
of the Askervein Hill.

Acknowledgements

This work was partly sponsored by the Portuguese Foundation for Science and
Technology (FCT), under contracts PRAXIS XXI N. 3/3.2/EMG/1967/95 and Bu-
let N. 33980/99. We are grateful to Prof. P. A. Taylor for providing copies of the
reports and terrain data on the Askervein Hill.

References

Beljaars, A. C. M., Walmsley, J. L., and Taylor, P. A.: 1987, A Mixed Spectral Finite-Difference
Model for Neutrally Stratified Boundary-Layer Flow over Roughness Changes and Topography,
Boundary-Layer Meteorol. 38, 273303.
ASKERVEIN FLOW; k  TURBULENCE MODEL 529

Castro, F. A.: 1997, Numerical Methods for the Simulation of Atmospheric Flows over Complex
Terrain, Ph.D. Thesis, University of Porto, Portugal, 267 pp. (in Portuguese).
Castro, F. A. and Palma, J. M. L. M.: 2002, VENTOSTM: A Computer Code for Simulation of
Atmospheric Flows over Complex Terrain, Technical Report, Available from the authors, 21 pp.
Durbin, P. A. and Reif, B. A. P.: 2001, Statistical Theory and Modeling for Turbulent Flows, John
Wiley & Sons, Chichester, New York, 285 pp.
Gresho, P. and Lee, R.: 1981, Dont Suppress the Wiggles Theyre Telling you Something!,
Comput. Fluids 9, 223253.
Jackson, P. S. and Runt, J. C. R.: 1975, Turbulent Wind over a Low Hill, Quart. J. Roy. Meteorol.
Soc. 101, 929955.
Kenjere, S. and Hanjalic, K.: 2002, Combined Effects of Terrain Topography and Thermal Strat-
ification on Pollutant Dispersion in a Town Valley: A T-RANS Simulation, J. Turbul. 3(26),
(http:/jot.iop.org).
Kim, H. and Patel, V.: 2000, Test of Turbulence Models for Wind Flow over Terrain with Separation
and Recirculation, Boundary-Layer Meteorol. 94, 521.
Knupp, P. and Steinberg, S.: 1994, Fundamentals of Grid Generation, CRC Press, Boca Raton, 286
pp.
Launder, B. E. and Sharma, B. I.: 1974, Application of the Energy-Dissipation Model of Turbulence
to the Calculation of Flow near a Spinning Disc, Lett. Heat Mass Trans. 1, 131138.
Launder, B. F. and Spalding, D. B.: 1972, Mathematical Model of Turbulence, Academic Press,
London, 169 pp.
Leonard, B. P.: 1979, A Stable and Accurate Convective Modelling Procedure Based on Quadratic
Upstream Interpolation, Comput. Meth. Appl. Mech. Eng. 19, 5998.
Mason, P. J. and King, J. C.: 1985, Measurements and Predictions of Flow and Turbulence over
Isolated Hill of Moderate Slop, Quart. J. Roy. Meteorol. Soc. 111, 617640.
Maurizi, A., Palma J. M. L. M., and Castro, F. A.: 1998, Numerical Simulation of the Atmospheric
Flow in a Mountainous Region of the North of Portugal, J. Wind Eng. Ind. Aerodyn. 7476,
219228.
Mickle, R. F., Cook, N. J., Hoff, A. M., Jensen, N. O., Salmon, J. R., Taylor, P. A., Tetzlaff, G., and
Teunissen, H. W.: 1988, The Askervein Hill Project: Vertical Profiles of Wind and Turbulence,
Boundary-Layer Meteorol. 43, 143169.
Miller, T. F. and Schmidt, F. W.: 1988, Use of a Pressure-Weighted Interpolation Method for
the Solution of the Incompressible NavierStokes Equations on a Nonstaggered Grid System,
Numer. Heat Transfer 14, 212233.
Nakayama, A. and Miyashita, K.: 2001, URANS Simulation of Flow over Smooth Topography,
Int. J. Numer. Meth. H. 11, 723743.
Patankar, S. V.: 1980, Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing Corporation,
Washington, DC, 197 pp.
Raithby, G. D., Stubley, G. D., and Taylor, P. A.: 1987, The Askervein Hill Project: A Finite Control
Volume Prediction on Three-Dimensional Flows over the Hill, Boundary-Layer Meteorol. 39,
107132.
Rhie, C. M. and Chow, W. L.: 1983, Numerical Study of the Turbulent Flow Past an Airfoil with
Trailing Edge Separation, AIAA J. 21, 15251532.
Schlichting, H. and Gersten, K.: 2000, Boundary Layer Theory, Springer, Berlin, 8th revised and
enlarged edition, 799 pp.
Silva Lopes, A.: 2000, Flow Simulation in Complex Geometries by Large Eddy Simulation, Ph.D.
Thesis, University of Porto, Portugal, 169 pp. (in Portuguese).
Taylor, P. A.: 1977, Some Numerical Studies of Surface Boundary-Layer Flow over Gentle
Topography, Boundary-Layer Meteorol. 11, 439465.
Taylor, P. A. and Teunissen, H. W.: 1983, ASKERVEIN 82: Report on the September/October 1982
Experiment to Study Boundary Layer Flow over Askervein, South Uist, Report: MSRS-83-8,
530 F. A. CASTRO ET AL.

Technical Report, Meteorological Services Research Branch Atmospheric Environment Service


4905 Dufferin Street, Downsview, Ontario, Canada M3H 5T4.
Taylor, P. A. and Teunissen, H. W.: 1985, The Askervein Hill Project: Report on the Sept./Oct.
1983, Main Field Experiment, Research Report MSRB-84-6, Technical Report, Meteorological
Services Research Branch Atmospheric Environment Service 4905 Dufferin Street, Downsview,
Ontario, Canada M3H 5T4.
Teunissen, H. W., Shokr, M. E., Bowen, A. J., Wood, C. J., and Green, D. W. R.: 1987, The Askervein
Hill Project: Wind Tunnel Simulations at Three Length Scales, Boundary-Layer Meteorol. 40,
129.
Walmsley, J. L. and Taylor, P. A.: 1996, Boundary-Layer Flow over Topography: Impacts of the
Askervein Study, Boundary-Layer Meteorol. 78, 291320.
Wood, N.: 1995, The Onset of Separation in Neutral, Turbulent Flow over Hills, Boundary-Layer
Meteorol. 76, 137164.
Xu, D., Ayotte, K. W., and Taylor, P. A.: 1994, Development of a Non-Linear Mixed Spec-
tral Finite Difference Model for Turbulent Boundary-Iayer Flow over Topography, Boundary
Layer-Meteorol. 70, 341367.
Zeman, O. and Jensen, N. O.: 1987, Modification of Turbulence Characteristics in Flow over Hills,
Quart. J. Roy. Meteorol. Soc. 113, 5580.

S-ar putea să vă placă și