Delivered by ICEVirtualLibrary.

com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
Bevan, M. A. et al. (2010). Geotechnique 60, No. 8, 635649 [doi: 10.1680/geot.9.P.010]
635
Inuence of large-scale inhomogeneities on a construction dewatering
system in chalk
M. A. BEVAN

, W. POWRI E and T. O. L. ROBERTS

Large-scale anisotropy and inhomogeneities resulting

from the presence of ssures, fractures and high-per-
meability zones can have a major impact on local
groundwater ows, and affect signicantly the perform-
ance of a construction dewatering system. A groundwater
model for the HS1 Channel tunnel rail link Thames
tunnel southern approach excavation is used to show that
the observed performance of the construction dewatering
system cannot adequately be explained unless such fea-
tures are incorporated. While experience suggested that
such features were likely to be present in this geological
setting, at the design stage there was signicant uncer-
tainty regarding their scale. Groundwater ow modelling
enabled the location and extent of the inhomogeneities
present to be deduced in some detail. Features such as
highly permeable zones might be identied by more
thorough site investigation, including the more wide-
spread use of pumping tests. However, the inuence of
anisotropy is harder to establish, primarily because its
effect is most signicant once a horizontal ow barrier
(for example, a retaining wall) is in place, which is
generally not the case during site investigation.
KEYWORDS: excavation; geology; groundwater; numerical
modelling; permeability
Lanisotropie et lheterogeneite resultant de la presence
de ssures, de fractures et de zones a` haute permeabilite
risque davoir un impact prononce sur les ecoulements de
la nappe phreatique locale, et daffecter de facon signi-
cative les performances dun syste`me dasse`chement pour
la construction. On utilise une maquette de leau phrea-
tique pour lexcavation de lacce`s sud du tunnel HS1 sous
la Tamise de la liaison ferroviaire du Tunnel sous la
Manche, an de demontrer que lon ne peut expliquer de
facon adequate les performances relevees du syste`me
dasse`chement pour la construction quen incorporant
ces caracteristiques. Bien que lexperience permettait de
penser que ces details etaient probablement presents dans
ce contexte geologique, il subsistait, au stade de letude,
une incertitude signicative quant a` leur envergure. La
modelisation de lecoulement de la nappe phreatique
permettait de deduire, de facon relativement detaillee,
lemplacement et lenvergure des heterogeneites pre-
sentes. Il pourrait etre possible didentier certaines
caracteristiques, comme des zones extremement perme-
ables, au moyen dun examen plus approfondi du site, y
compris lemploi generalise de tests de pompage. Toute-
fois, linuence de lanisotropie est plus difcile a` etablir,
principalement parce que son effet est particulie`rement
signicatif lorsquune barrie`re horizontale a` lecoulement
(par exemple un mur de soute`nement) est en place, ce
qui nest generalement pas le cas lors des reconnaissances
sur le site.
INTRODUCTION
Construction dewatering involves lowering the natural
groundwater level, usually temporarily by pumping from a
series of wells or sumps, to prevent excavations below the
water table from ooding and/or becoming unstable. Chalk
outcrops extensively in eastern and southern England and
provides a major aquifer, which is exploited for water supply
in many areas. Deep excavations for major construction
projects in these areas may extend to below standing ground-
water level and require dewatering measures to provide
stable conditions. Although the chalk is a heterogeneous
aquifer whose effective (bulk) hydraulic conductivity is
controlled by localised ssure development and connectivity,
it is often analysed for the design of a dewatering system as
a homogeneous unit with a bulk hydraulic conductivity
based on limited site investigation data (Roberts & Preene,
1990; Powrie & Roberts, 1995; Leiper et al., 2000). In many
cases a difference in horizontal and vertical hydraulic con-
ductivity, that is anisotropy, is not given full consideration.
Unforeseen large-scale inhomogeneities can have a signi-
cant impact on the performance of a dewatering system,
resulting in revisions to the design being required at short
notice. This paper demonstrates the effect of large-scale
inhomogeneities in the chalk on the performance of a major
construction dewatering system, with reference to the con-
struction of the HS1 Channel tunnel rail link (CTRL)
Thames tunnel in north Kent, England. A conceptual model
of the site hydrogeology is developed, tested using a nite
difference groundwater model and validated with reference
to drawdown and ow data from the site. The model is then
used to investigate the level of complexity required for
analysis and the effect of anisotropy.
The numerical model was originally prepared to develop
the design of the dewatering system. The modelling was
undertaken because it was recognised that variations in the
chalk permeability prole with depth would have a signi-
cant impact on ow rates and that these could not be
accurately determined by the site investigation information.
The model was used to review the design assumptions and
where appropriate to provide a basis for modifying the
scheme as ow and drawdown information was collected
following the start-up of the dewatering system.
The chalk outcrop boundary crossed the site and it was
known that it is common for a zone of higher permeability
chalk to be associated with this boundary. The possibility of
such a feature being present was discussed by the dewatering
design team but it was considered that the site investigation
Manuscript received 05 February 2009; revised manuscript accepted
24 August 2009. Published online ahead of print 26 March 2010.
Discussion on this paper closes on 1 January 2011, for further
details see p. ii.

WJ Groundwater Limited, Bournehall House, Bushey, Hertford-

shire, UK.
School of Civil Engineering and the Environment, University of
Southampton, UK.
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
information did not specically identify such a zone and in
any case it was considered that, even if present, the high-
permeability zone would probably be cut off by the dia-
phragm wall.
PROJECT DESCRIPTION AND EXCAVATION
GEOMETRY
HS1 is the UKs rst high-speed railway line, connecting
London to the Channel tunnel with a journey time of
approximately 35 min. Construction of the Thames tunnel,
as part of Section 2 of the line, involved boring twin 8
.
15 m
diameter tunnels up to 25 m below the bed of the Thames
from Swanscombe, Kent to West Thurrock, Essex. Each
tunnel drive used a separate tunnel-boring machine (TBM)
attached to a 200 m long drive train, and commenced from a
southern launch chamber on the Swanscombe marshes. The
southern launch chamber was part of a longer excavation for
the tunnels southern approach. The southern approach struc-
ture is 445 m long and 26 to 29 m wide, and was con-
structed within diaphragm walls as either a cut-and-cover
tunnel or a retained cut. Construction dewatering was re-
quired to lower groundwater levels to 1 m below formation
level. As this varied along the length, the southern approach
structure was divided into four separate dewatering cells by
means of cement/bentonite slurry cross-wall cut-offs. The
excavation geometry and target drawdowns are summarised
in Table 1 and Figs 1 and 2. An aerial view of the southern
approach during construction is shown in Fig. 3.
GROUND CONDITIONS
The main water bearing strata at the site are the terrace
gravels and the underlying upper chalk, which are generally
in hydraulic connection. The chalk was eroded and weath-
ered prior to the deposition of the terrace gravels during
Quaternary interglacials as sea levels rose. Supercial allu-
vial deposits consisting of soft silty clay with peat horizons
act as a conning layer over much of the site (Fig. 4).
Detailed descriptions of the Lower Thames geological suc-
cession are given by Marsland (1986) and Gibbard (1994).
The depth prole shown in Fig. 2 is typical of much of
the site, with ground level at approximately +2 m OD. The
conning alluvial layer is generally 910 m thick, with
approximately 7 m of terrace gravels below. The thickness of
both the alluvium and terrace gravel decreases towards the
shallow end of the excavation, furthest away from the
Thames, as shown in Fig. 1.
The chalk outcrops to the south of the site, close to the
southern end of the excavation, as shown in Fig. 4. The
extent of the chalk outcrop was determined from borehole
logs and is consistent with the geological map (BGS, 1997).
An outline of the chalk geology associated with the CTRL
project is given by Warren & Mortimore (2003).
The upper chalk encountered during construction was of
the Seaford chalk formation. Typically there was 1 to 2 m of
structureless chalk at the interface with the terrace gravels.
Below this the chalk was generally classied as grade B2 or
B3 using the Construction Industry Research and Informa-
tion Association (CIRIA) grading scheme (Lord et al.,
2002), indicating that discontinuity apertures are less than
3 mm and the discontinuity spacing is between 60 and
Table 1. Design specications for the four dewatering cells
SLC + SCC1 SCC2 SCC3 + SRC1 SRC2 + SRC3
Description Launch chamber + cut and cover Cut and cover Cut and cover + 25 m retained cut Retained cut
Length: m 75 135 115 120
Width: m 2926 2628 28 29
Ground level: m OD +1
.
3 +1
.
3 +1
.
5 to +2
.
0 +2
.
0 to +3
.
2
Toe of d-wall: m OD 29 25
.
5 to 21 21 to 19
.
0 17 to 12
Formation level: m OD 17 to 14 14 to 11 11 to 7
.
25 7
.
25 to 4
.
5
Target drawdown: m OD 18 to 15 15 to 12 12 to 8
.
25 8
.
25 to 5
.
5
SL SCC1 C SCC2 SCC3 SRC1 SRC2 SRC3
120 m 115 m 135 m
75 m
m OD
5
0
5
10
15
20
25
30
Tunnel portal Toe of diaphragm
wall
Formation level Slurry bentonite
cross cut-off wall Made ground
Alluvial deposits
Terrace gravels
Upper chalk
m OD
5
0
5
10
15
20
25
30
Fig. 1. Longitudinal prole of the approach structure and geological strata derived from borehole logs
636 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
200 mm. This corresponds to Mundford grade III (Spink,
2002), which suggests a bulk hydraulic conductivity of 10
5
to 10
3
m/s (Roberts & Preene, 1990).
The dewatering system was designed using estimates of
hydraulic conductivity based on limited data from a pumping
test carried out some distance away from the actual excava-
tion. Analysis of the pumping test data from a single well
indicated an equivalent uniform horizontal hydraulic conduc-
tivity of 1
.
7 3 10
4
m/s (14
.
7 m/day) for the gravel and the
chalk above a level of 65 m OD. The level of 65 m OD
was taken as the aquifer base because the productive zone
of the chalk aquifer is generally assumed to be the top 50
60 m, see Price et al. (1993). Packer tests carried out in
boreholes at the southern approach site indicated chalk
permeabilities of 2
.
0 3 10
6
to 1
.
0 3 10
4
m/s (0
.
28
.
6 m/
day), although packer tests are likely to underestimate the
hydraulic conductivity because they tend not to intersect
major discontinuities.
After the dewatering system had been designed, more
detailed pumping test data, including tests in piezometers
with dened response zones and borehole packer tests,
became available from the site of the northern tunnel ap-
proach on the opposite bank of the River Thames. These
tests draw water from a limited and reasonably well-dened
horizon, and may therefore be used to provide an indication
of the variation in hydraulic conductivity with depth. The
stratigraphy and formative geological processes for the two
sites are almost identical, so it would seem reasonable to
assume that the data (shown in Fig. 5) are also representative
of conditions on the south side of the river. The data indicate
a clear decrease in the hydraulic conductivity of the chalk
with depth. Simple statistical analysis of the chalk perme-
abilities revealed that there was a signicant boundary at
approximately 25 m OD, marking the interface between the
more weathered surface chalk and the base chalk. Table 2
shows the mean hydraulic conductivity of the surface chalk
to be 3
.
06 3 10
4
m/s, compared with 4
.
71 3 10
6
m/s for
the base chalk. The hydraulic conductivity of the surface
chalk appears to be more spatially variable, which may
reect increased localised ssuring. The decrease in chalk
hydraulic conductivity with depth was conrmed by the
results of geophysical ow logging of screened wells at the
northern approach site.
Analysis of constant ow rate pumping tests carried out
C
L
Discharge
450 mm diameter
discharge main
Standing water
level
Dewatering
well
Roof
slab
Base
slab
Internal
wall
Formation
level
Target water level
Pump level
U
p
p
e
r

c
h
a
l
k
T
e
r
r
a
c
e

g
r
a
v
e
l
s
A
l
l
u
m
i
a
l

d
e
p
o
s
i
t
s
M
G
2
0
5
10
15
20
25
30
L
e
v
e
l
:

m

O
D
Diaphragm wall
Fig. 2. Idealised cross-section through the cut and cover section of the approach
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 637
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
on both banks of the River Thames suggested that the
hydraulic conductivity of the terrace gravels was generally in
the range 5
.
8 3 10
4
to 4
.
6 3 10
3
m/s (50400 m/day).
This is broadly consistent with an estimate based on the D
10
particle size and Hazens formula
k 0
:
01(D
10
)
2
(1)
(where k is the hydraulic conductivity in m/s and D
10
is in
mm), which with 0
.
2 mm < D
10
< 0
.
6 mm suggests a hy-
draulic conductivity range of 4 3 10
4
to 3
.
6 3 10
3
m/s
(34306 m/day). Note that Hazens formula is not strictly
applicable because the uniformity coefcient of the gravels
is well above 3, which is the accepted upper limit of
application of Hazens formula (Preene et al., 2000). The
pumping test derived results are also comparable with the
estimate given by Marsland & Randolph (1978) based on
grading of bulk samples at Crayford marshes, 4 km up-
stream, of 2 3 10
4
to 1 3 10
2
m/s (17864 m/day), where
site pumping tests indicated a horizontal hydraulic conduc-
tivity of 5 3 10
3
m/s (432 m/day).
IN SITU GROUNDWATER CONDITIONS
Natural groundwater levels across the Swanscombe Penin-
sula are generally in the range 0 to +1 m OD. They are
affected by the tidal uctuations of the Thames to varying
degrees, depending on the geological conditions, as indicated
by the typical piezometer responses shown in Fig. 6 (piezo-
meter locations are given in Fig. 7). The magnitude of the
tidal inuence must be taken into account in assessing the
acceptable residual error of a groundwater model. The mean
Fig. 3. Aerial photograph of the construction of cells
SLC + SCC1 and SCC2
Thames
Tunnel portal
Confined
aquifer
Approach structure
Railway in open
cut
T
h
a
m
e
s
m
0 500
Unconfined
aquifer
N
Fig. 4. Schematic diagram of the Swanscombe Peninsula show-
ing the chalk outcrop
100 10
2
100 10
3
100 10
4
100 10
5
100 10
6
100 10
7
35
30
25
20
15
10
5
0
Permeability: m/s
D
e
p
t
h
:

m

O
D
Test in piezometers
Packer tests
Fig. 5. Hydraulic conductivity versus depth prole of the chalk
at the northern approach site
Table 2. Values of hydraulic conductivity (k) of the chalk in m/s (m/day) for the northern tunnel approach site
Surface chalk
(eight samples)
Base chalk
(ve samples)
Depth: m OD 12 to 25 25 to 35
Mean k 3
.
06 3 10
4
(26
.
47) 4
.
71 3 10
6
(0
.
41)
Standard deviation 4
.
08 3 10
4
(35
.
21) 2
.
88 3 10
6
(0
.
29)
Minimum k 3
.
30 3 10
5
(2
.
85) 1
.
96 3 10
6
(0
.
17)
Maximum k 1
.
28 3 10
3
(110
.
59) 6
.
21 3 10
6
(0
.
54)
638 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
tidal range of the Thames is up to approximately 5 m, with
uctuations generally between +2 m OD and 3 m OD.
SRA5945 was situated in the conned part of the aquifer
and comprised a dual installation with two separate piezo-
meters screened and sealed in the terrace gravels and in the
upper chalk. The monitoring data sets from the two piezo-
meters are virtually identical in Fig. 6; both show a tidal
peak-to-peak amplitude of approximately 0
.
5 m. The data
from each stratum were similar throughout the conned area,
suggesting good hydraulic connection between the gravels
and the chalk. The maximum recorded tidal amplitude was
2
.
50 m at the piezometer SR1232, 20 m from the shoreline.
Fig. 6 shows amplitudes of approximately 1
.
5 m for the
piezometer SA5981, approximately 300 m from the shore.
The methods described by White & Roberts (1993), based
on the theory set out by Ferris (1951), were used to interpret
the attenuation of tidal amplitudes across the conned area.
The attenuation is related to the aquifer transmissivity (T)
and storage coefcient (S) by
h
2
=h
1
exp

S=t
o
T
_
x
2
x
1
_ _
(2)
where t
o
is the tidal period (12
.
5 h), x
2
x
1
is the horizontal
distance between two standpipe piezometers and h
2
/h
1
is the
ratio of the mean tidal amplitudes. At this site, for piezo-
meters located along lines approximately perpendicular to
the centre-line of the Thames channel, this equation ade-
quately describes the attenuation with T 1
.
31 3 10
2
m
2
/s
(1134 m
2
/day) and S 0
.
005, for an effective aquifer base
depth of 65 m OD. According to White & Roberts (1993)
the resulting value of T/S 2
.
65 indicates partly conned
conditions. The analysis suggests that, in general, the effec-
tive source of the tidal uctuations is close to the centre of
the Thames channel, rather than at the bank; borehole
records conrm that the alluvial seal on the Thames bed is
only present close to the shore.
At the boundary between the conned and unconned
areas the tidal amplitude is reduced to ,0
.
2 m, as shown by
the data for AC1919 and NBH5 (Fig. 6). The uctuations
for the unconned standpipe piezometers further south, away
from the river, are negligible.
DEWATERING SYSTEM DESIGN
In view of the high ow rates anticipated, a deep-well
system was installed to lower the groundwater levels within
the tunnel approach structure. The proposed design incorpo-
rated 42 wells (W01 to W42), located as indicated in Table 3
and Fig. 8. The additional wells W43 to W54 will be
discussed later.
The design provided increased pumping capacity at the
deeper end of the excavation, where the required drawdown
was greatest. Either a 15 kW or 9
.
2 kW electrical submers-
ible pump, with ow capacities of 20 and 12 l/s respectively,
was installed in each well. Wells were installed to a depth of
between 29 and 24 m OD using the cable percussion
method, and screened from 1 m below formation level to the
well base. Well installation, pumping and excavation started
at the deep end of the tunnel approach (cell SLC SCC1)
to enable the TBM to be prepared at the earliest possible
stage. Dewatering and excavation then progressed towards
the shallow end.
05
00
05
10
15
20
0
7
-
N
o
v
0
8
-
N
o
v
0
9
-
N
o
v
1
0
-
N
o
v
1
1
-
N
o
v
1
2
-
N
o
v
1
3
-
N
o
v
1
4
-
N
o
v
1
5
-
N
o
v
1
6
-
N
o
v
1
7
-
N
o
v
1
8
-
N
o
v
Date
W
a
t
e
r

l
e
v
e
l
:

m

O
D
SA5945 chalk
SA5945 gravel
AC1919 chalk
NBH5 chalk
SA5981 gravel
G11 alluvium
Fig. 6. Example of tidal inuence on water levels in piezo-
meters. Data were logged at 15 min intervals. Piezometer
locations are shown in Fig. 7
SR5945
SR5958
AC1919
NBH1
NBH2
NBH3
NBH4
NBH5
G09
NBH7
NBH6
G10
G11
G12
G13
G14
G15
G16
G17
SA5981
Thames
Chalk piezometer
Gravel piezometer
Gravel and chalk piezometers
SR5957
m
T
h
a
m
e
s
NBH8
NBH6
Thames
0
C
o
n
fin
e
d
U
n
c
o
n
fin
e
d
500
SR1232
Fig. 7. Location of remote standpipe piezometers around the
Swanscombe Peninsula (not all piezometers are labelled)
Table 3. Distribution of pump capacity within the excavation
Dewatering cell Length: m Number of wells Pump size: l/s Total pump capacity: l/s
20 12
SLC + SCC1 75 16 16 320
SCC2 135 14 4 10 200
SCC3 + SRC1 115 6 6 72
SRC2 + SRC2 120 6 6 72
Totals 445 42 20 22 664
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 639
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
INSTRUMENTATION
The dewatering system was monitored with standpipe
piezometers installed both inside the excavation (the PI
series), and outside but close to the diaphragm walls (the PE
series). The PE series included dual installations into the
gravel and the chalk strata. Pore pressures in the alluvial
clay layer were monitored using pressure cells (VE series).
All of these piezometers (PI, PE and VE) incorporated a
vibrating wire transducer wired to a datalogger to record
water level readings hourly. The instrument locations are
shown in Fig. 8. The instruments were calibrated against
manual dip readings and corrected for the effects of changes
in atmospheric pressure. Piezometers and cables that became
damaged during construction were repaired where possible;
cases of instrument breakdown and repair are indicated by a
gap in the data record.
Remote standpipe piezometers in the chalk and gravel, at
the locations shown in Fig. 7, were monitored manually
using a dip meter. In addition to a selection of the site
investigation piezometers, the monitoring programme re-
quired the installation of standpipe piezometers in the chalk
(NBH series) and in the gravel (G series). Pore pressures in
the alluvial clay were monitored by pressure cells installed
at the G series locations. In general, the remote standpipe
piezometers were monitored once every two to ve days
during commissioning of the dewatering system, reducing to
once every seven to 14 days once the system was fully
operational and ows had stabilised. Baseline data of in situ
groundwater levels were obtained by monitoring each remote
piezometer for at least 24 h, using a stand-alone pressure
transducer and datalogger, to establish the tidal range (as
shown in Fig. 6).
Groundwater abstraction ow rates from the dewatering
system were monitored using in-line owmeters installed in
each of the two discharge mains. Each discharge main was
capable of delivering 300 l/s of water from the excavation to
the discharge outfall into the Thames. The owmeters were
logged by the datalogger at hourly intervals. Initially, air in
the discharge main, which was drawn in when the pumps
drew down to the pump intake level, resulted in erratic and
incorrect measurement. This was resolved by trimming back
individual wells using the valve at the well-head, and by
installing air escape valves on the discharge main.
DEWATERING SYSTEM PERFORMANCE
The dewatering system was commissioned in mid-Novem-
ber 2001 with all 16 wells in SLC+SCC1 pumped briey, to
give an initial ow in excess of 200 l/s (Fig. 9). It was
immediately evident that this cell could be dewatered to the
target level (approximately 18 m OD) by pumping from
just two of the 16 wells. Dewatering in SCC2 began on
1 January 2002, with the pumps being switched on progres-
sively. By the end of January the discharge ow was
approximately 200 l/s, but two additional wells, W43 and
W44 (Fig. 8) were required to achieve the target drawdown.
Dewatering of SCC3+SRC1 began in March 2002, in-
creasing the total extraction ow rate (from all cells) to
about 370 l/s. The required drawdown in the retained cut
sections (SRC2+SRC3) could not be achieved with the
designed pumped capacity, and nine additional wells (W45
to W49 and W51 to W54) had to be installed in this area.
The required drawdowns were achieved across the entire
southern approach structure with a total abstraction rate of
just under 600 l/s.
EVIDENCE FOR LARGE-SCALE INHOMOGENEITIES
During the commissioning of wells in the retained cut
section, it became clear that locally the ground was far more
permeable than expected. This is indicated by the variation
in the specic capacities of the wells (dened as the yield
per unit drawdown) across the excavation as a whole (Fig.
10). The specic capacity depends both on well depth and
diameter and on the hydraulic conductivity of the ground. In
this case the well depth and diameter did not vary greatly
across the site, so the distribution of specic capacity
provides a good indication of zones of increased hydraulic
conductivity. It is apparent from Fig. 10 that there is a zone
of elevated hydraulic conductivity present diagonally across
the excavation between 300 and 400 m from the tunnel
portal.
As discussed previously, the upper chalk is predominantly
of the CIRIA grade B2 or B3. In the highly permeable zone,
the chalk was more difcult to classify but is described as
possibly grade C4C5 for the top 5 m. This implies a
discontinuity aperture greater than 3 mm and a discontinuity
W
0
3
W
0
7
W
0
1
W
0
2
W
1
0
W
1
4
W
1
1
W
1
3
W
1
6
W
1
5
W
1
7
W
1
8
W
1
9
W
2
0
W
2
1
W
2
2
W
2
3
W
2
4
W
2
5
W
2
6
W
4
3
W
4
4
W
2
7
W
2
8
W
2
9
W
3
0
W
3
1
W
3
2
W
3
3
W
3
4
W
3
5
W
3
6
W
4
5
W
5
3
W
5
4
W
4
6
W
4
7
W
4
9
W
3
8
W
3
7
W
4
8
W
5
1
W
3
9
A
W
5
2
W
4
0
W
4
1
W
4
2
P
I0
9
P
I0
8
P
I2
3
P
I2
2
P
E
1
3
V
E
1
3
P
E
1
4
V
E
1
2
P
E
1
2
P
I2
1
P
I0
7
P
I2
0
P
E
1
1
P
I0
6
P
E
2
1
P
I1
9 P
I0
5
P
I1
8
P
E
2
0
V
E
0
9
P
E
0
9
P
I1
7
P
E
1
9
P
I1
5
P
I1
6
P
E
0
7
P
I0
4
P
I0
3
P
I1
4
P
E
1
8
P
I1
3
V
E
0
6
P
E
0
6
V
E
0
5 P
E
0
5
P
E
1
6
P
I1
2
P
I1
0
P
E
1
7 P
E
1
5
P
I1
1
P
E
0
2
V
E
0
2
P
E
0
3 V
E
0
3
Designed dewatering well
Additional dewatering well
Piezometer
N
SLC SCC1 SCC2 SCC3 SRC1 SRC2 SRC3
0 50
m
W
0
6
Fig. 8. Layout of wells and piezometers
Nov-01 Feb-02 May-02 Aug-02 Nov-02 Feb-03 Aug-03
q o n a c d e h
0
100
200
300
400
500
600
700
May-03
Date
D
i
s
c
h
a
r
g
e

f
l
o
w
:

l
/
s
Recorded flow
Representative flow for period
b f i j k l m p r s
Dewatering periods
Fig. 9. Total abstraction of groundwater by the dewatering
system. A description of the dewatering schedule is provided in
Table 4
640 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
spacing less than 60 mm, suggesting a more permeable
material.
Further evidence of an increased hydraulic conductivity in
the chalk at the retained cut end of the approach excavation
was provided by the core samples taken during the pre-
construction site investigation. Fig. 11 compares the quality
of the core samples for two rotary drilled boreholes:
SR5957, located within the supposed highly permeable zone;
and SR5958, located on the exposed chalk outcrop to the
south of the site. The locations of the boreholes are shown
in Fig. 7. Caution must be exercised in assessing the state of
rock cores, because they are largely a function of the drilling
method and the care taken by the driller during boring and
extraction (Clayton et al., 1995). However, the total core
recovery (TCR) gives an indication of the presence of
natural voids and is expressed as a percentage of the rock
recovered during a single coring run. The solid core recov-
ery (SCR), which is the percentage of full diameter core
recovered during a single coring run, gives an indication of
the fracture state; a low SCR suggests a high degree of
fracturing.
Figure 11 suggests a high degree of fracturing in the top
5 m of the chalk at SR5957. At this location the SCR
generally appeared to increase, suggesting a decrease in the
degree of fracturing, with depth. Overall, the proles suggest
a greater degree of fracturing at the highly permeable zone
(SR5957), particularly above 22 m OD, than at the chalk
outcrop (SR5958). This evidence was only assembled retro-
spectively once the precise location of the high-permeability
zone in the excavation had been identied and following a
detailed review of the borehole logs. Borehole SR5957
proved to be the only borehole which was known to pass
through the high-permeability zone.
The highly permeable chalk lies at the extremity of the
Thames oodplain in the old river valley. An increase in
transmissivity of chalk in valleys has been widely observed,
most notably by Ineson (1962). Price (1987) (after Rhoades
and Sinacori (1941)) identied the inuence of converging
owlines towards the area of groundwater discharge and the
development of secondary ssures due to dissolution by
water containing carbon dioxide as explanations for elevated
chalk hydraulic conductivity in valleys.
DEVELOPMENT OF THE NUMERICAL MODEL
A numerical model had been used in the dewatering
system design and was further used to develop the scheme
as information became available following start-up. It was
recognised in the course of this process that the inhomo-
geneities and extent of the variation in permeability with
depth identied were rather greater than anticipated at de-
sign stage. The programme constraints for the works did not
allow time for these conditions to be fully explored during
the construction phase. However, the data set collected was
considered to be sufciently comprehensive that there was
value in subsequently further developing the numerical mod-
el to help understand the hydrogeology at the site and to
investigate in more general terms the level of complexity
required in analysis for the design of a large-scale dewater-
ing system. In particular, the inuence and effect of three
large-scale features were investigated. These were
(a) a high degree of anisotropy of the surface chalk,
thought to be responsible for unexpectedly low ow
rates in the SLC+SCC1 cell
(b) a generally highly permeable zone, leading to unfore-
seen high ow rates in the retained cut sections
(c) a transition zone between the highly permeable zone
and the anisotropic surface chalk, which led to the need
to install the additional wells (W43 and W44) in SCC2.
Table 5 shows the four steady-state models investigated.
Model 1 includes all three features and models 2 to 4 each
have one feature omitted. For each model the hydraulic
conductivities of the hydrostratigraphic units were varied,
within the limits of the expected hydraulic conductivity
ranges, to nd the best calibration (i.e. the best t between
calculated and observed piezometer levels).
The models were developed using Groundwater Vistas, a
Windows interface for processing input and output les for
the three-dimensional nite difference code Modow
(McDonald and Harbaugh, 1988). The steady-state model
represented the dewatering period l (Fig. 9 and Table 4),
when the dewatering system was yielding its maximum ow.
This was a relatively stable period in terms of both ows
and drawdowns, as shown in Fig. 9.
Each model represented an area 2
.
5 km 3 2
.
5 km on plan
with the excavation situated close to the centre. The grid
was oriented so that the diaphragm walls around the excava-
tion were aligned with the x and y axes. Cells of varying
size were used to allow more closely spaced nodal points
inside the excavation (4 m by 4 m) than at the model
boundaries (33
.
33 m by 33
.
33 m) (Fig. 12). This gave an
increased sensitivity in the zone where the hydraulic gradi-
ents were greatest without using an impractically large num-
ber of nodal points. The widths of adjacent cells did not
differ by a factor of more than 1
.
5.
The vertical spacing of the layers of the three-dimensional
400 350 300 250 200 150 100 50
SRC2 SRC3 SCC3 SRC1
0
30
Distance from tunnel portal: m
25
2550
50 100
100 200
200 400
400 800
800
SCC2
Specific capacity: m /day
2
0
W
i
d
t
h
:

m
SLC SCC1
Fig. 10. Distribution of specic capacity inside the southern approach excavation, where data
are available (not to scale). Shaded area indicates the estimated extent of the highly permeable
zone
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 641
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
model was dictated by the need to be able to simulate
approximately the variable depth of the diaphragm walls and
the levels of the interfaces between each of the hydrostrati-
graphic units. The base of the model, which was assumed to
be impermeable, was set at 65 m OD. Fig. 13 shows the
discretisation into horizontal layers, together with the
stepped diaphragm wall prole. The diaphragm walls and
cut-off cross walls were incorporated into the model using
the Modow wall facility, as elements of thickness 1
.
2 m
and hydraulic conductivity 1
.
1 3 10
7
m/s. Zones of poten-
tially different hydraulic conductivity were incorporated into
the grid as indicated in Fig. 14. The chalk from 17 down
to 26 m OD was split into an anisotropic surface chalk
zone and a transition zone, as shown. It was assumed that
the highly permeable zone followed the line of the chalk
outcrop and the old river valley. The effect of varying the
depth of the highly permeable zone was investigated during
the modelling.
The model boundary positions, shown in Fig. 12, should
be sufciently far away from the excavation to exceed the
estimated distance of inuence (L
o
), calculated using
Sichardts empirical formula
L
o
Ch

k
_
(3)
where h is the drawdown at the excavation and C is a
factor of 3000 for radial ow. Assuming a drawdown of
18 m and a Darcy hydraulic conductivity (k) of 3 3 10
4
m/s, the estimated distance of inuence (L
o
) should be in
the order of 1000 m. However, measured drawdowns at
piezometers within this range suggested that the distance of
inuence was rather greater than this. Therefore, the bound-
ary heads in the gravel and chalk were set to 2 m OD or
3 m OD, as shown in Fig. 12. As drawdown was not
observed, or expected, in the conning alluvium, the bound-
aries surrounding this material (in model layers 1 and 2)
were set at a head of 0
.
5 m OD. The Thames shoreline was
used to dene the boundary position around the upper and
right sides of the model.
Each boundary cell was assigned a hydraulic conductivity
equal to k
h
of the surrounding aquifer material, although
parametric analysis suggested that the model was insensitive
to changes in the boundary conductance.
Each dewatering well was represented by an analytical
well in the model, with the well coordinates and the depth
of well screen specied in accordance with reality. Fig. 15
shows the recorded abstraction ow rates which were applied
to the wells and remained unchanged for each model run.
The total system abstraction rate was 51 149 m
3
/day, or
592 l/s for the model. The model was used to calculate the
drawdowns at 59 target locations, corresponding to the
response zones of 11 internal, 16 external and 32 remote
piezometers installed in the eld, and were compared with
those measured on site. Fig. 16 shows the mean measured
water levels during the steady-state period for three example
piezometers. The modelled and measured drawdown data
sets were compared using the statistical methods shown in
Table 6.
RESULTS
Model 1, with all the large-scale features, gives a close t
between the measured and modelled internal drawdowns
(Fig. 17), and this model represented the overall hydrogeol-
ogy convincingly (Fig. 18, and see Table 8 later in this
section). The modelling suggested that the hydraulic conduc-
tivity of the terrace gravels (150 m/day) (Table 7) was
slightly lower than expected based on pumping tests (190
400 m/day), but within the estimated range based on particle
size distribution (34306 m/day). Pumping tests are gener-
ally considered to provide a more accurate estimate. In this
case analysis of the pumping test data was complicated by
the fact that the gravel was in good hydraulic connection
with chalk below.
To achieve the measured drawdown in the cell
SLC + SCC1, an anisotropy ratio k
h
/k
v
350 had to be
100 80 60 40 20
36
32
28
24
20
16
12
0
Percentage
(a)
D
e
p
t
h
:

m

O
D
TCR SCR
30
26
22
18
14
10
6
2
0
Percentage
(b)
D
e
p
t
h
:

m

O
D
TCR SCR
100 80 60 40 20
Fig. 11. The quality of chalk cores from boreholes (a) SR5957,
in the highly permeable zone, and (b) SR5958, on the outcrop
chalk. Total core recovery (TCR) and solid core recovery (SCR)
are shown
642 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
applied to the surface chalk. Assuming a low anisotropy
ratio of just 1
.
5 for the surface chalk, a k
h
in the order of
1
.
25 m/day was required to achieve the same drawdown in
SLC + SCC1, as shown for model 4. Such a low horizontal
hydraulic conductivity in the top 9 m of the upper chalk
could be explained by the presence of putty chalk (dened
as structureless chalk that is devoid of factures) but little if
any thickness of this material was observed at this site. To
avoid a signicant overestimation of drawdown along SCC2
in model 4, the hydraulic conductivity of the transition zone
was increased.
In model 1, the degree of anisotropy in the transition zone
is low (1
.
5), and the horizontal hydraulic conductivity
slightly greater than the surface chalk. Omitting the transi-
tion zone, as in model 2, overestimates the internal draw-
downs in the cell SCC2 by up to 12
.
5 m at P104 (Fig. 17),
and overestimates some external drawdowns. However, the
t to the remote drawdowns remains largely unaffected (Fig.
18, Table 8), suggesting that anisotropy only has a signi-
cant inuence on drawdowns in and immediately around the
excavation where the cut-off walls restrict horizontal ow.
In addition to the four steady-state models, a study of the
dewatering of section SLC + SCC1 was used to conrm the
anisotropy of the surface chalk unit. A transient model, with
the same mesh and conceptual hydostratigraphy as model 1,
was constructed to simulate a period of dewatering in
December 2001 when only this section was being dewatered.
Records showed that a pumping rate of 20 l/s, evenly
distributed between wells W04 and W12, lowered the water
level in the standpipe piezometer PI01 to 9
.
2 m OD.
Doubling the pumping rate to 40 l/s reduced the water level
in the piezometer to 18
.
5 m OD. A series of transient
model runs was carried out with different degrees of aniso-
tropy, where k
h
was constant at 35 m/day. All other hydro-
stratigraphic units were assigned the hydraulic conductivities
of model 1, as listed in Table 7. Fig. 19 shows that the
drawdown inside the cell is highly sensitive to the degree of
anisotropy at ratios k
h
/k
v
between 10 and 1000. Interpolation
of the data suggests that a ratio k
h
/k
v
of 350 to 750 is
Table 4. Description of the dewatering schedule and total system ow rates
Period Start date Description Approximate
total ow: l/s
a 26-Nov-01 Wells in SLC + SCC1 commissioned 220
b 27-Nov-01 Only wells W04 + W12 operational 20
c 01-Jan-02 First pumping in SCC2 using wells W19 + W20 40
d 11-Jan-02 Wells W21 to W30 started 200
e 18-Feb-02 Additional wells W43 + W44 started 235
f 08-Mar-02 First pumping in SCC3 + SRC1 using wells W32 to W34 290
g 19-Mar-02 All wells in SCC3 + SRC1 operational 370
h 24-Mar-02 All wells in SCC3 + SRC1 stopped 235
i 01-Apr-02 All wells in SCC3 + SRC1 operational 370
j 01-May-02 First pumping in SRC2 + SRC3 using wells W37 + W38 430
k 21-May-02 Wells W39 to W42 started. Wells W41 + W42 then permanently decommissioned 470
l 20-Jul-02 Additional wells W45 to W54 started 590
m 13-Sep-02 Wells W47 to W54 stopped, except W52 500
n 18-Oct-02 Wells W45 and W46 stopped 430
o 07-Nov-02 Additional wells W55 to W58 commissioned 500
p 17-Nov-02 All wells in SRC2 + SRC3 operational 580
q 12-Mar-03 Gradual decline in ow rate from SRC2 + SRC3 as dewatering operations are scaled down to allow
drawdown recovery
580 to 490
r 10-May-03 Increase in pumping in SRC2 + SRC3, to prevent temporary leakage of base slab 550
s 22-May-03 Decline in pumping rate and start of gradual system decommissioning 550 to 370
Table 5. Features included in the groundwater models
Strong anisotropy of
surface chalk
Isotropic highly
permeable zone
Transition zone
Model 1 4 4 4
Model 2 4 4 3
Model 3 4 3 4
Model 4 3 4 4
5 10 15 20 25 30 35 40 455055 60 65 70 75 80
110
105
100
95
90
85
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
Boundary head
2 m OD
Boundary head
2 m OD
Reduced row
spacing
Reduced column
spacing
Boundary head
3 m OD
Fig. 12. The reduced spacing of cells around the excavation and
the boundaries for the model layers 3 to 12
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 643
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
appropriate for the chalk in this location, giving
0
.
047 < k
v
< 0
.
1 m/day. This is consistent with the level of
anisotropy applied in model 1. The models were largely
insensitive to any anisotropy of the base chalk over the same
range.
It is clear from the analysis that anisotropy can have an
important inuence on drawdowns inside excavations
bounded by cut-off walls. In this case the effect is particu-
larly pronounced as the diaphragm walls extend into the less
permeable base chalk, so that horizontal ow below the toe,
as well as ow around the toe, is limited. It is possible that
the anisotropy is largely determined by the orientation of
fractures, as suggested by Toynton (1983) for Norfolk chalk.
Bedding-parallel fractures, related to the deposition of the
chalk, occur at an angle approximately equal to the mean
plane of dip, and trace lengths can persist across the full
extent of the formation. In contrast, the trace length of
bedding-normal fractures is typically less than 1 m (Younger
& Elliot, 1995). Warren & Mortimore (2003) show that the
bedding planes at this location are near horizontal. Solution
development of such fractures may make anisotropy more
pronounced, particularly if the connectivity between bed-
ding-parallel fractures (by way of intersecting bedding-nor-
mal fractures) is low.
The remote drawdowns are highly sensitive to the hydrau-
lic conductivity of the highly permeable zone, which dom-
inates most of the hydrogeological system. In model 1 the
isotropic hydraulic conductivity of this feature is estimated
to be approximately 4800 m/day. Model 3, which omits the
highly permeable feature completely, signicantly overesti-
mates drawdowns centred on the retained cut sections by up
to 16 m at PI08. Excessive drawdowns in this area cause
increased drawdowns throughout the excavation (Fig. 17).
While it is possible to determine the approximate width of
the highly permeable zone using the specic capacity data
from individual wells (Fig. 10), the depth of the zone is less
clear. The base of the highly permeable zone must extend to
about 20 m OD because it was not cut off by the dia-
phragm wall that reached to 19 m OD at SRC1 and 17 m
OD at SRC2. There is some indication that the chalk core
quality at borehole SR5957 improves below 30 m OD (Fig.
11), implying that this may signify the base of the high-
permeability zone at this location. Fig. 20 shows that varying
the thickness of highly permeable material while maintaining
the same overall transmissivity has a minimal impact on the
t of the model. For these model runs the vertical hydraulic
conductivities were unchanged from those listed in Table 6
for model 1. The effect on the overall model t is shown in
terms of the sum of squares of residuals S
r
, as dened in
Table 6. If a residual of 0
.
5 m is acceptable for each of the
60 piezometers, S
r
should not exceed 15 for a close model
t, although this assumes that the residuals are distributed
evenly. For the base case (model 1) S
r
9
.
95 and for each
of the model runs in Fig. 20 S
r
, 15. The best t was
achieved using a depth of 23 m OD for the highly per-
meable zone, but this was inevitable given that the hydraulic
conductivies of the other zones were calibrated using this
conguration; improved ts for the other congurations
could be achieved if small changes were made to the
hydraulic conductivity of the terrace gravels, for example.
The internal drawdowns at standpipe piezometers PI07 and
PI08 show a maximum variation of 0
.
9 m for the different
zone congurations, and the mean residual for all the 60
modelled piezometers varies by just 0
.
38 m.
This analysis suggests that the hydraulic conductivity of
the highly permeable zone is in the range 23005850 m/day
220 210 200 190 180 170 160 150 140 130 120 110 100 90 80 70 60 0 10 20 30 40
12
1
2
3
4
5
6
7
8
9
10
11
1
2
3
4
5
6
7
8
9
10
11
Model
layer
Bottom elevation:
m OD
440 430 420 410 400 390 380 370 360 350 340 330 320 310 300 290 280 270 260 250 240 230 220 210 200
9
12
17
19
21
23
24
25
26
29
35
65
SCC3 SRC1
SRC2 SRC3
Distance from headwall: m
9
12
17
19
21
23
24
25
26
29
35
65
SLC SCC1
SCC2
Model
layer
Distance from headwall: m
Bottom elevation:
m OD
50
12
Fig. 13. Layer discretisation and the stepped diaphragm wall prole
644 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
2500 m
(b)
A A
65
26
23
17
12
9
3
Depth: m OD
Excavation (445 m)
Alluvium
Thames
boundary
Base chalk
Outcrop chalk
Surface chalk
HPZ
Cut-off walls
Tunnel portal
Transition zone
Terrace gravels
Surface chalk
Thames
A
Surface chalk
Tunnel portal
HPZ
T
ra
n
s
itio
n
z
o
n
e
T
h
a
m
e
s
A
(a)
H
P
Z
Surface chalk
0 500
m
Fig. 14. (a) Plan (217 to 223 m OD) and (b) cross-section
(AA) showing the hydrostratigraphic zones of the model
(HPZ denotes the highly permeable zone)
J
u
l
-
0
3
A
p
r
-
0
3
J
a
n
-
0
3
O
c
t
-
0
2
A
u
g
-
0
2
M
a
y
-
0
2
F
e
b
-
0
2
8
6
4
2
0
2
N
o
v
0
1

Date
W
a
t
e
r

l
e
v
e
l
:

m

O
D
0
100
200
300
400
500
600
700
T
o
t
a
l

f
l
o
w
:

l
/
s
SA5945 chalk
SA5945 gravel
SA5981 gravel
Total flow
Mean water
levels for
modelled
period
Fig. 16. Example piezometer hydrographs. Target drawdowns
are shown for the steady-state period from 10 August 2001 to 13
September 2002
0
5
10
15
20
25
30
0
Measured well flow: l/s
M
o
d
e
l
l
e
d

w
e
l
l

f
l
o
w
:

l
/
s
SLC SCC1
SCC2
SCC3 SRC1
SRC2 SCR3
30 25 20 15 10 5
Fig. 15. Comparison of modelled well ows with those applied
in all four models
Table 6. Statistical parameters used to compare the modelled and measured drawdown data sets, where x
i
is the measured and y
i
is
the modelled drawdown
Statistical parameter Calculation Comments
Mean residual error, M
M
1
n

n
i1
(x
i
y
i
)
Ideal value is 0
.
0 m. Negative values indicate
that model underestimates water levels;
positive values indicate an overestimation.
Negative and positive residuals can cancel
each other.
Absolute mean residual error,
MA
MA
1
n

n
i1
jx
i
y
i
j
Ideal value is 0
.
0 m. Gives indication of model
t irrespective of negative and positive signs.
Gradient, b, from linear
regression analysis
b

xy

x
2
Used to plot trendline on scattergram in the
form y bx, where b 1 is the ideal value.
Product moment correlation
coefcient, R
R
n

xy

x
2

x
_ _
2
_ _
n

y
2

y
_ _
z
2
_ _

R +1 is a perfect positive correlation

between modelled and measured water levels.
Sum of squares of residuals, S
r
S
r

n
i1
(x
i
y
i
)
2
Ideal value is 0
.
0. Used for sensitivity analysis.
Not inuenced by sign of errors.
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 645
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
(2
.
66 3 10
2
to 6
.
77 3 10
2
m/s), implying that the zone
consists of karstic chalk or a network of enlarged ssures.
Karstic behaviour of the chalk is characterised by high-
velocity ow through dissolution pipes, which provide path-
ways for the rapid transport of water (Banks et al., 1995;
MacDonald et al., 1998). Signicant conduits are common
in the area; during a regional hydrogeological study of the
chalk, such features were observed in a chalk quarry that
lies 2 km to the southwest of the site. Reeves (1979) dened
enlarged ssures as having an aperture greater than 3 mm
and a spacing from 2 to 20 m, which can give a bulk
hydraulic conductivity of the order of 1000 m/day
(1 3 10
2
m/s). Examination of chalk cores from borehole
SR5957 suggested a ssure spacing less than 60 mm with
apertures greater than 3 mm, which is consistent with the
even greater hydraulic conductivities apparently identied
here by the modelling studies.
A formal sensitivity analysis for model 1 conrms the
ndings already discussed. For the model 1 base case, the
parameter values in Table 7 were used. The impact of each
400 350 300 250 200 150 100 50
30
25
20
15
10
5
0
0
Distance from tunnel portal: m
W
a
t
e
r

l
e
v
e
l
:

m

O
D
Recorded level
Model 1
Model 2
Model 3
Model 4
W
1
8
Piezometer
W
1
4
W
0
1
P
I
0
4
a
P
I
1
5
W
3
1
P
I
0
6
P
I
2
1
P
I
0
7
P
I
2
2
P
I
0
8
P
I
0
9
Fig. 17. Drawdown prole inside the excavation for each of the
models. The recorded levels are shown with error bars of 6
0
.
5 m
0 5 10 15 20 25
30
25
20
15
10
5
0
30
Measured water level: m OD
M
o
d
e
l
l
e
d

w
a
t
e
r

l
e
v
e
l
:

m

O
D
Internal piezometers
External piezometers
Remote piezometers
0 5 10 15 20 25 30
Measured water level: m OD
M
o
d
e
l
l
e
d

w
a
t
e
r

l
e
v
e
l
:

m

O
D
30
25
20
15
10
5
0
Internal piezometers
External piezometers
Remote piezometers
0 5 10 15 20 25 30
Measured water level: m OD
M
o
d
e
l
l
e
d

w
a
t
e
r

l
e
v
e
l
:

m

O
D
0 5 10 15 20 25 30
Measured water level: m OD
M
o
d
e
l
l
e
d

w
a
t
e
r

l
e
v
e
l
:

m

O
D
(c) (d)
30
25
20
15
10
5
0
Internal piezometers
External piezometers
Remote piezometers
30
25
20
15
10
5
0
Internal piezometers
External piezometers
Remote piezometers
(a) (b)
Fig. 18. Comparison of measured and modelled piezometric heads for each model. The solid line indicates the ideal correlation and
the dashed lines represent an error of 60
.
5 m. (a) Model 1; (b) model 2; (c) model 3; (d) model 4
646 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
hydraulic conductivity zone was investigated in turn by
varying the hydraulic conductivity parameter, while main-
taining the other parameters unchanged. The relative sensi-
tivity of the model to each of the parameters is shown by
the gradient of the sensitivity curve (Fig. 21); the drawdown
is inuenced mainly by the horizontal hydraulic conductivity
of the highly permeable zone but is insensitive to the
horizontal hydraulic conductivity of the base chalk. The
anisotropy of the surface chalk, transition and base chalk
zones inuence the model t, as explained previously in
relation to the internal drawdown. The apparent insensitivity
of the model to the hydraulic conductivity of the outcrop
Table 7. Values of hydraulic conductivity (m/day) for each of the models, where k
h
is
horizontal hydraulic conductivity and k
v
is the vertical hydraulic conductivity
Hydrostratigraphic
units
Model 1 Model 2 Model 3 Model 4
k
h
k
v
k
h
k
v
k
h
k
v
k
h
k
v
Alluvium 0
.
1 0
.
01 0
.
1 0
.
01 0
.
1 0
.
01 0
.
1 0
.
01
Terrace gravels 150 150 150 150 150 150 150 150
Surface chalk 35 0
.
1 35 0
.
1 35 0
.
1 1
.
25 0
.
83
Transition zone 55 36
.
5 35 0
.
1 55 36
.
5 90 57
Outcrop chalk 50 5 50 5 50 5 50 5
Highly permeable
zone
4800 4800 4800 4800 55 36
.
5 4800 4800
Base chalk 2 0
.
02 2 0
.
02 2 0
.
02 2 0
.
02
Table 8. Summary of statistical data for the model simulations (M is mean of residual errors, MA is the mean of absolute residual
errors and R is the correlation coeffeicent, as dened in Table 6)
Internal External Remote Overall
Number of piezometers 12 16 32 60
Model 1 M: m 0
.
10 0
.
10 0
.
07 0
.
04
MA: m 0
.
31 0
.
28 0
.
34 0
.
32
Linear relationship y 0
.
988x y 1
.
012x y 1
.
013x y 1
.
000x
R +0
.
99 +0
.
94 +0
.
97 +0
.
99
Model 2 M: m 3
.
92 0
.
84 0
.
07 0
.
90
MA: m 3
.
93 1
.
12 0
.
34 1
.
20
Linear relationship y 1
.
326x y 1
.
106x y 0
.
987x y 1
.
176x
R +0
.
86 +0
.
84 +0
.
97 +0
.
94
Model 3 M: m 10
.
73 8
.
13 3
.
92 6
.
40
MA: m 10
.
73 8
.
13 3
.
92 6
.
40
Linear relationship y 1
.
715x y 1
.
923x y 1
.
764x y 1
.
778x
R 0
.
28 +0
.
63 +0
.
92 +0
.
82
Model 4 M: m 0
.
21 0
.
81 0
.
57 0
.
55
MA: m 0
.
50 0
.
81 0
.
65 0
.
67
Linear relationship y 1
.
001x y 1
.
090x y 1
.
104x y 1
.
053x
R +0
.
97 +0
.
84 +0
.
93 +0
.
96
1000000 100000 10000 1000
25
20
15
10
5
0
100
Anisotropy ratio, / k k
h v
W
a
t
e
r

l
e
v
e
l

i
n

P
I
0
1
:

m

O
D
25
20
15
10
5
0
W
a
t
e
r

l
e
v
e
l

i
n

P
I
0
1
:

m

O
D
20 l/s
40 l/s
Recorded water level at 40 l/s
Recorded water level at 20 l/s
Fig. 19. The drawdown at the standpipe piezometer PI01 for
different anisotropy ratios of the surface chalk, when extracting
groundwater at two pumping rates from the cell SLC + SCC1
35 29 26 23
12
10
8
6
4
2
0
2
4
21
Depth of base of high permeability zone: m OD
W
a
t
e
r

l
e
v
e
l
:

m

O
D
0
10
20
30
40
50
60
70
80
S
u
m

o
f
s
q
u
a
r
e
s

o
f
r
e
s
i
d
u
a
l
s
,
S
r
Sum of squared residuals
Mean residual (M) for all piezometers
Water level at PI07 (recorded level 89 m OD)
Water level at PI08 (recorded level 88 m OD)
Fig. 20. The model t and internal drawdown in the retained
cut sections for different thicknesses of the highly permeable
zone. The transmissivity of the effective aquifer is constant for
each conguration
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 647
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
chalk means that the model could be simplied by treating
this zone as an extension of the surface chalk zone with the
same hydraulic conductivity values. The terrace gravels and
highly permeable zone would be expected to be substantially
isotropic in reality, and applying an anisotropy ratio of up to
50 has a negligible impact on the overall model t.
In the models presented in this paper it has been assumed
that the large variation in transmissivity through the excava-
tion is due to inhomogeneities in the surface chalk (17 to
26 m OD), and that the terrace gravels and base chalk
zones are homogeneous. It is possible that a good, or even
better, calibration could be achieved by a combination of
increasing the hydraulic conductivity of the terrace gravels
together with a counteracting decrease in the horizontal
permeabilities of the underlying chalk zones. In view of the
number of hydrostratigraphic zones and parameters it is not
possible to nd a unique solution, and this represents a
limitation of this type of model. However, the relative
differences in the hydraulic conductivity of the surface chalk
must still apply approximately, as demonstrated by the
simulations summarised in Fig. 22.
CONCLUSIONS AND IMPLICATIONS FOR PRACTICE
The sensitivity analysis presented here suggests that for a
mildly complex scheme, a numerical modelling approach is
unlikely to provide a unique solution giving the hydraulic
conductivity prole of each stratum zone identied. While
sensitivity analysis shows that the numerical modelling ap-
proach cannot be used to estimate uniquely the hydraulic
conductivity of each of the zones, it does provide a useful
framework for assembling and analysing complex data sets.
In this case the performance of the dewatering system
cannot be satisfactorily explained without the inclusion in
0
50
100
150
200
250
300
0
Parameter multiplier
(a)
S
u
m

o
f
s
q
u
a
r
e
s

o
f
r
e
s
i
d
u
a
l
s
,
S
r Terrace gravels
Surface chalk
Transition zone
High-permability zone
Outcrop chalk
Base chalk
Acceptable
error level
0
50
100
150
200
250
300
1
Anisotropy, / k k
h v
Terrace gravels
Surface chalk
Transition zone
High-permeability zone
Outcrop chalk
Base chalk
20 18 16 14 12 10 08 06 04 02
(b)
10000 1000 100 10
S
u
m

o
f
s
q
u
a
r
e
s

o
f
r
e
s
i
d
u
a
l
s
,
S
r
Acceptable
error level
Fig. 21. Sensitivity curves for model 1: (a) horizontal hydraulic
conductivity (k
h
) is varied from base case (Table 7) by the
multiplication factor; (b) vertical hydraulic conductivity (k
v
) is
varied while k
h
is kept at the base case value to give a variable
anisotropy ratio
TG SC TZ HPZ
1
TG SC TZ HPZ
2
TG SC TZ HPZ
3
TG SC TZ HPZ
4
TG SC TZ HPZ
5
1
0
0
1
5
0
2
0
0
2
5
0
3
0
0
4
0
3
5
3
0
2
5
2
0
6
0
5
5
5
0
4
5
4
0
5
2
0
0
4
8
0
0
4
4
0
0
4
0
0
0
3
6
0
0
1
10
100
1000
10000
Simulation
k
h
:

m
/
d
a
y
0
20
40
60
80
100
S
u
m

o
f
s
q
u
a
r
e
s

o
f
r
e
s
i
d
u
a
l
s
,
S
r
Sum of squares of residuals
Fig. 22. Horizontal hydraulic conductivities (k
h
) used to achieve good calibrations of model 1 (i.e. S
r
< 15). k
h
of the terrace gravels (TG), surface chalk (SC), the transition zone (TZ) and the highly permeable zone
(HPZ) have been varied, while the outcrop chalk, base chalk and alluvium were kept constant at the base
case (Table 5). Anisotropy was applied at the same ratios as used for the base case
648 BEVAN, POWRIE AND ROBERTS
Delivered by ICEVirtualLibrary.com to:
IP: 212.49.235.19
On: Wed, 18 Aug 2010 08:13:56
the model of large-scale inhomogeneities within a 445 m
long excavation including
(a) a zone of anisotropic chalk with a ratio k
h
/k
v
of 350
750
(b) a zone of isotropic highly permeable chalk with a
k
h
k
v
of 2
.
66 3 10
-2
to 6
.
77 3 10
2
m/s
(c) a transition zone between the anisotropic chalk and
highly permeable chalk.
Modelling shows that anisotropy (k
h
/k
v
) has a signicant
impact on groundwater ows into the excavation bounded by
cut-off walls and therefore needs to be considered carefully
during the design of a dewatering system.
It is common practice for modelling studies to be carried
out during the design of a dewatering system. However, the
success of the design is strongly dependent on the concep-
tual understanding of the geology and the accuracy of the
input parameters. Generally, the input parameters are as-
sumed to be representative of the site as a whole, but the
importance of local variations should not be underestimated.
Features such as highly permeable zones may be identied
by more thorough site investigation, including the more
widespread use of pumping tests. However, the inuence of
anisotropy is harder to establish, primarily because it has its
most signicant effect once a horizontal ow barrier is in
place, which is generally not the case during site investiga-
tion. The cost benets that result from achieving an im-
proved optimisation of a dewatering scheme design must be
weighed against the increased costs of more comprehensive
site investigation and interpretation. If there is sufcient
exibility in the construction schedule, modern methods of
data collection and review make it possible to take an
observational method approach, as implemented at the
CTRL Thames tunnel, with success.
ACKNOWLEDGEMENTS
The Thames tunnel permanent works were designed by
Rail Link Engineering for client London and Continental
Railways. The main contractor was Hochtief Murphy Joint
Venture. The specialist dewatering contractor was WJ
Groundwater Limited. The study described in this paper was
funded through the engineering doctorate programme of the
Engineering and Physical Sciences Research Council in col-
laboration with WJ Groundwater Limited and the Environ-
ment Agency.
NOTATION
C factor in Sichardts formula
h
2
/h
1
ratio of mean tidal amplitudes of two piezometers
k
h
horizontal hydraulic conductivity
k
v
vertical hydraulic conductivity
L
o
distance of inuence
R correlation coefcient
S storage coefcient
S
r
sum of squares of residuals
T transmissivity
t
o
tidal period
x
2
x
1
horizontal distance between two piezometers
x
i
measured water level
y
i
modelled water level
h drawdown
REFERENCES
Banks, D., Davies, C. & Davies, W. (1995). The chalk as a karstic
aquifer evidence from a tracer test at Stanford-Dingley,
Berkshire, UK. Q. J. Engng Geol. 28, No. S1, S31S38.
BGS (British Geological Survey) (1997). Geological map of the
Lower Thames Valley. London: British Geological Survey.
Clayton, C. R. I., Matthews, M. C. & Simons, N. E. (1995). Site
investigation. Cambridge: Blackwell Science.
Ferris, J. G. (1951). Cyclic uctutaions of water level as a basis for
determining aquifer transmissivity. Int. Assoc. Sci. Hydrol. 33,
148155.
Gibbard, P. L. (1994). The Pleistocene history of the Lower Thames
valley. Cambridge: Cambridge University Press.
Ineson, J. (1962). A hydrogeological study of the permeability of
Chalk. J. Inst. Water Eng. 16, 449463.
Leiper, Q., Roberts, T. & Russell, D. (2000). Geotechnical engineer-
ing for the Medway Tunnel and approaches. Proc. Instn Civ.
Engrs, Transport 141, No. 1, 3542.
Lord, J. A., Clayton, C. R. I. & Mortimore, R. N. (2002). Engineer-
ing in chalk, CIRIA Report C574. London: Construction Indus-
try Research and Information Association.
MacDonald, A. M., Brewerton, L. J. & Allen, D. J. (1998).
Evidence for rapid groundwater ow and karst-type behaviour in
the Chalk of southern England. In Groundwater pollution,
recharge and vulnerability (ed. N. S. Robins), pp. 95106.
London: Geological Society.
McDonald, M. G. & Harbaugh, A. W. (1988). A modular three-
dimensional nite-difference groud-water ow model. Reston,
Virginia: United States Geological Survey.
Marsland, A. & Randolph, M. F. (1978). A study of the variation
and effects of water pressures in pervious strata underlying
Crayford Marshes. Geotechnique 28, No. 4, 435464.
Marsland, A. (1986). The ood plain deposits of the Lower Thames.
Q. J. Engng Geol. Hydrogeol. 19, No. 3, 233247.
Powrie, W. & Roberts, T. O. L. (1995). Case history of a dewatering
and recharge system in chalk. Geotechnique 45, No. 4, 599
609.
Preene, M., Roberts, T. O. L., Powrie, W. & Dyer, M. R. (2000).
Groundwater control: design and practice, CIRIA Report C515.
London: Construction Industry Research and Information Asso-
ciation.
Price, M., Downing, R. A. & Edmunds, W. M. (1993). The Chalk
as an aquifer. In The hydrogeology of the chalk of North-West
Europe (ed. R. A. Downing), pp. 3558. Oxford: Clarendon
Press.
Price, M. (1987). Fluid ow in the Chalk of England. In Fluid ow
in sedimentary basins and aquifers, Geological Society Special
Publication No. 34, pp. 141156. London: The Geological
Society.
Reeves, M. J. (1979). Recharge and pollution of the English chalk:
some possible mechanisms. Engng Geol. 14, No. 4, 231240.
Rhoades, R. G. & Sinacori, B. W. (1941). Patterns of groundwater
ow and solution. J. Geol. 49, No. 8, 785794.
Roberts, T. O. L. & Preene, M. (1990). Case studies of construction
dewatering in chalk. Chalk pp. 571575. London: Thomas
Telford.
Spink, T. W. (2002). The CIRIA Chalk description and classication
scheme. Q. J. Engng Geol. Hydrogeol. 35, No. 4, 363369.
Toynton, R. (1983). The relation between fracture patterns and
hydraulic anisotropy in the Norfolk Chalk, England. Q. J. Engng
Geol. 16, No. 3, 169185.
Warren, C. D. & Mortimore, R. N. (2003). Chalk engineering
geology Channel Tunnel Rail Link and North Downs Tunnel.
Q. J. Engng Geol. Hydrogeol. 36, No. 1, 1734.
White, J. K. & Roberts, T. O. L. (1993). The signicance of
groundwater tidal uctutations. In Groundwater problems in
urban areas (ed. W. B. Wilkinson), pp. 3142. London: Thomas
Telford.
Younger, P. L. & Elliot, T. (1995). Chalk fracture system character-
istics: implications for ow and solute transport. Q. J. Engng
Geol. 28, S39S50.
INFLUENCE OF LARGE-SCALE INHOMOGENEITIES ON A CONSTRUCTION DEWATERING SYSTEM 649