Engineering Geology 178 (2014) 15–27

Contents lists available at ScienceDirect

Engineering Geology
journal homepage: www.elsevier.com/locate/enggeo

Dewatering of a deep excavation undertaken in a layered soil

Estanislao Pujades a,b,⁎, Enric Vàzquez-Suñé a, Jesus Carrera a, Anna Jurado a,b
a
GHS, Institute of Environmental Assessment and Water Research (IDAEA), CSIC, Barcelona, Spain
b
GHS, Dept. Geotechnical Engineering and Geosciences, Universitat Politècnica de Catalunya, UPC-BarcelonaTech, Barcelona, Spain

a r t i c l e i n f o a b s t r a c t

Article history: In order to carry out deep excavations under the water table in urban environments, the safety of the work site
Received 25 November 2013 and of the adjacent buildings is a major cause for concern. One of the most common and effective methods of
Received in revised form 27 May 2014 undertaking these excavations involves combining the cut and cover method with a dewatering system. The
Accepted 6 June 2014
success of a construction depends on the stability of the excavation bottom, the effects produced outside the
Available online 19 June 2014
excavation by dewatering (soil movements) and/or the state of the enclosure (defects in the diaphragm
Keywords:
walls). This study proposes a realistic multidisciplinary procedure to address these issues. The work emphasizes
Pumping well the importance of soil characterisation and underlines the need to perform a Watertightness Assessment Test
Deep excavation (WTAT) before the excavation stage. The procedure was applied to the excavation of a deep shaft of the High
Dewatering Speed Train (HST) tunnel in Barcelona. An earlier geological characterisation at large scale ruled out the use of
Diaphragm walls deep pumping wells. However, a subsequent hydrogeological characterisation, which involved borehole logging,
Settlements grain size analyses, Natural Gamma Ray and pumping tests, revealed the presence of thin transmissive layers in-
Groundwater side the low hydraulic conductivity materials. The dewatering system was designed by considering different
model scenarios and the safest design was selected for the excavation. Depths of the enclosure and of the
pumping wells differed in accordance with the scenarios. The impacts (settlements due to pumping) and
the stability in each scenario were computed. The state of the enclosure underwent a WTAT before the start of
the excavation, but after constructing the enclosure, to verify its low permeability. The test consisted in pumping
inside the enclosure and monitoring the groundwater behaviour outside the enclosure. Numerical interpretation
of this test showed a defect in the diaphragm walls below the excavation bottom. Since this defect was not
repaired because of its location (below the bottom of the excavation), the dewatering system had to be
redesigned to ensure safety. Surface settlements, which were also a source of concern, were small. They were
computed using coupled hydro-mechanical models.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction Palmström, 2008). Incomplete characterisation complicates the predic-

Underground constructions in urban environments must be
sufficiently deep so as to avoid existing structures (tunnels, car parks,
basement and/or foundation of buildings) (Li and Yuan, 2012). They
are therefore built below the water table at locations where the ground-
water is not too deep, which could give rise to problems (El-Nahhas,
1999). As a result, it is essential to apply procedures that ensure
the safety of the excavation works in the presence of groundwater. A
number of underground constructions are the vertical shafts that
are used for maintenance during tunnel construction or as emergency
or ventilation exits during the operation of tunnels (Ni and Cheng,
2011).
A variety of techniques are used to undertake deep excavations
under the water table. The optimal choice depends on the characteris-
tics of the soil and on the hydrogeology of the site (Stille and
⁎ Corresponding author at: GHS, Institute of Environmental Assessment and Water
Research (IDAEA), CSIC, Barcelona, Spain. Tel.: +34 93 400 61 00; fax: +34 93 204 59 04.
E-mail address: estanislao.pujades@gmail.com (E. Pujades).
tion of the response of the medium to the construction and, therefore,
the design stage (Jurado et al., 2012). An efficient method consists in
combining the “cut and cover” method (Gulhati and Datta, 2005) with
deep pumping wells (Powers et al., 2007). Diaphragm walls avoid later-
al groundwater influx and ensure the verticality of the excavation walls
(Xanthakos et al., 1994) whereas deep pumping wells prevent water
from entering the excavation bottom, ensuring stable conditions. Bot-
tom uplift or liquefaction events are therefore averted (Pujades et al.,
2012a). However, deep pumping wells are not recommended when
the excavation is undertaken in a low hydraulic conductivity soil
(Cashman and Preene, 2001). In such cases, alternatives include eductor
wells, wick drains (when low rates of water must be extracted) or exca-
vation under undrained conditions (when the pressure drop caused by
the unloading of the soil ensures stable conditions) (Preene et al., 2004;
Powers et al., 2007). Nevertheless, these alternatives may pose a risk if
the soil has not been adequately characterised. Unexpected transmis-
sive layers within the low hydraulic conductivity materials may lead
to instability at the bottom of the excavation (Moore and Longworth,
1979; Ramaswamy, 1979; Davies, 1984). A detailed hydrogeological

http://dx.doi.org/10.1016/j.enggeo.2014.06.007
0013-7952/© 2014 Elsevier B.V. All rights reserved.
16 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
characterisation is therefore essential. Borehole logging is the most
commonly used tool. However, fine layers of coarse sediments may
not be detected for various reasons: 1) coarse sediments tend to be
loose and may not be recovered from the core, 2) fine sediments of
upper layers may be dragged during drilling and may accumulate on
the outside of the core, concealing coarse sediments of deeper layers
and/or 3) fine layers of coarse sediments may be overlooked by the
geologist. Geophysical techniques (e.g. Ground Penetrating Radar,
Electrical Resistivity Tomography, Natural Gamma Ray,..), grain
size analyses and pumping tests should therefore be used for
hydrogeological characterisation. Soil characterisation also plays an
important role in determining the overconsolidation ratio (OCR) of
the soil, which is used to predict the shear strength and the pumping
settlements during dewatering. These settlements are small and elastic
in overconsolidated soils (Pujades et al., 2014). Moreover, a satisfactory
characterisation facilitates the setup of adequate numerical models to
select the most suitable dewatering system considering both the
stability of the bottom of the excavation and the impacts outside the
enclosure (e.g. pumping settlements).
A thorough assessment of the state of the enclosure is crucial to a
successful outcome of the excavation since diaphragm walls may suffer
from defects (Bruce et al., 1989; Knight et al., 1996; Pujades et al., 2012a;
Vilarrasa et al., 2012). Defects in the enclosure (gaps or open joints)
hamper the dewatering process giving rise to adverse consequences
(Pujades et al., 2012a). Defects located above the excavation may
cause inflows that drag sediments, leading to the formation of sink
holes outside the enclosure. However, if defects are located below the
excavation level, the water pressure will not be low enough, resulting
in unstable conditions. These situations pose serious risks that can
be significantly reduced by testing the enclosure. The test must be
Fig. 1. a) Geographical location of the study site. Plan views of the site, including the enclosure and the location, b) piezometers and pumping wells and, c) soil movement monitoring
points.
performed before excavation since the defects can be repaired by
injecting sealing substances or the dewatering system can be
redesigned. However, if they are detected during the excavation
stage, repair is more difficult and costly since pumping cannot be
interrupted (Pujades et al., 2012a). The existence of defects can be
determined by performing a Watertightness Assessment Test
(WTAT), which involves pumping inside the enclosure, monitoring
the groundwater behaviour and comparing the measured and the
predicted drawdowns. Defects can be detected by: 1) observing the
changes in flow behaviour as a result of pumping (Pujades et al.,
2012a), 2) applying drawdown type curves (Vilarrasa et al., 2012)
or 3) using numerical modelling.
The above discussion suggest that many factors such as characterisa-
tion, dewatering design, stability, impacts and assessment of the enclo-
sure, which may be overlooked, must be taken into account during deep
excavations in urban environments. These factors were borne in mind
when undertaking a deep excavation in Barcelona (Spain) during the
construction of the High Speed Train (HST) tunnel (Figure 1a). The ex-
cavation was undertaken at the crossroads of the Mallorca and Trinxant
streets, the site for the construction of one of the ventilation and
emergency circular shafts which were constructed every 700 m along
the tunnel. The tunnel was constructed by using a Tunnel Boring
Machine (TBM) and since the shaft was also used to carry out the main-
tenance works of the TBM, it was excavated before the construction of
the tunnel. Two main problems arose during the excavation of this
shaft. The first problem concerned the dewatering technique since an
earlier geological characterisation of the site (not sufficiently detailed)
suggested that the soil located at the bottom of the excavation had a
low hydraulic conductivity. The second difficulty concerned the state
of the enclosure since this suffered from a defect that was detected
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
following a WTAT. This test entailed pumping inside the enclosure and
monitoring the groundwater response at several piezometers located
around the shaft (Figure 1b). The type of defect could not be determined
because it was located below the bottom of the excavation bottom.
However, it may be postulated that the defect consisted in an open
joint between two panels of diaphragm walls or in one panel of
diaphragm wall that would not have reached the designed depth.
Other issues such as the stability of the excavation bottom and the
pumping settlements (a major cause for concern in urban environ-
ments) arose in the design and excavation stages. The former was guar-
anteed since the dewatering was designed meticulously by simulating
several dewatering scenarios and computing their stability analytically.
The latter were predicted numerically and monitored during the
construction. Numerical predictions of settlements were performed
from the enclosure to a distance of 500 m while soil displacements
were monitored by manual levelling at some points distributed around
the shaft (Figure 1c).
The present paper discusses the decisions taken and the works carried
out to excavate a shaft, which was completed successfully because of the
procedure followed. The methodology involved 1) hydrogeological
characterisation using different techniques, 2) detailed hydrogeological
and hydro-mechanical modelling, 3) monitoring of soil settlements and
drawdown during the construction and 4) assessment of the enclosure
by a WTAT.
Thus, the aim of this paper is threefold: 1) to propose a multidis-
ciplinary methodology to improve the construction of deep excava-
tions in urban areas, reducing risks and impacts, 2) to highlight the
importance of soil characterisation for selecting the most efficient
dewatering method when excavating under the water table and
3) to demonstrate that defects in deep enclosures can be easily
detected by a WTAT.
Fig. 2. Detailed geological profile of the site. Natural Gamma Ray register and borehole logging allowed us to identify 15 layers below the water table. The screens of the piezometers and
the pumping wells are also displayed. A simplified plan view is shown at the bottom on the right with the profile line. Q indicates the Quaternary materials and P, the Pliocene materials.
17
2. Materials and methods
2.1. Problem statement
2.1.1. General geographical, geological and hydrogeological description
The excavation was undertaken to construct one of the emergency
and ventilation shafts of the HST tunnel in Barcelona. This shaft
known as the “Trinxant shaft” is located at the crossroads between Ma-
llorca and Trinxant streets in the north of the city. The soil of Barcelona
is made up of Quaternary materials that overlie Tertiary materials. The
tunnel crosses predominantly the tertiary materials (Pliocene Age).
The Trinxant shaft crosses both the Quaternary, Q, (0 to 14 m depth)
and the Pliocene, P, (14 to 33 m depth) materials (Figure 2). The
Pliocene materials contain clays and grey marls (marine deposits) and
sequences of conglomerate with a sandy–clay matrix. The Pliocene is
characterised by an alternation of these lithologies. Fine sediments are
prevalent towards the bottom and the number and thickness of coarse
sediment layers increase at shallower depths.
The Quaternary materials in Barcelona are of two kinds: Pleistocene
and Holocene. The Pleistocene deposits composed of gravels with a clay
matrix at the bottom are alluvial fan deposits that are transformed into
sands with clay matrix seawards. These alluvial deposits are overlain by
aeolian deposits consisting of brown–yellow silts and calcareous
nodules. A calcrete stratum can be found at the top of this sequence.
This series may occur three times. The overlying Holocene sediments
consist of:
• Torrential, alluvial and foothills deposits made up of fine detritic ma-
terials (red clays or silts)
• Alluvial gravel and sands
• Coastal plain sands.
18 E. Pujades et al. / Engineering Geology 178 (2014) 15–27

Quaternary and Tertiary sediments can be regarded as a layered TBM. These chambers consisted of two enclosures delimited by 38 m
aquifer with a high vertical heterogeneity. The hydraulic conductivity deep diaphragm walls.
(k) of clay layers ranges from 0.001 to 0.01 m/d and that of sand and
gravel layers varies from 0.1 to 10 m/d. These values were derived 2.2. Basic concepts
from the numerous hydraulic tests performed during the HST tunnel
project and other projects (Pujades et al., 2012b, 2014). 2.2.1. Hydrogeological numerical models
A number of hydrogeological numerical models were setup using
the finite element code TRANSIN-IV (Medina et al., 2000; Medina and
2.1.2. Description of soil state
Carrera, 2003), with visual interface of VISUAL TRANSIN (UPC, 2003).
A thorough knowledge of the hydrogeological history and urbanisa-
Numerical models were used to determine the hydraulic parameters
tion of the soil is essential to predict its response to dewatering since the
of the soil and the state of the enclosure by automatic parameter estima-
structure and deformational characteristics of the soil depend on its
tion (also termed inverse problem or back analysis), employing the data
geological history. As in other European cities, the early stages of urban-
measured during the pumping tests. TRANSIN-IV performs the auto-
isation in Barcelona were accompanied by increased industrial activity.
matic estimation using the Levenberg–Marquardt algorithm (Carrera
Groundwater extraction by industry caused a significant lowering of
and Neuman, 1986a,b,c). Hydrogeological models were also used to
groundwater levels, which subsequently recovered due to the decline
simulate a number of dewatering scenarios in order to come up with
of industry (Vàzquez-Suñé et al., 2005). An important side effect of
the best alternative. The safety factor of each dewatering design was
groundwater fluctuations is the overconsolidation of the soil (Powers
computed analytically from the numerical results.
et al., 2007). Consequently, the soil in Barcelona is overconsolidated
Axisymmetric and multilayer numerical models were used depend-
and behaves elastically whenever groundwater oscillations do not ex-
ing on the geometry of each case (Figure 3a and b). The aquifer
ceed the maximum drawdown reached during the period of industrial
modelled was assumed as confined since the drawdown caused during
activity. Since this maximum drawdown was approximately 10–15 m
the pumping tests or during the dewatering period (outside the enclo-
at the study site, present day groundwater fluctuations (b2 m) usually
sure) was not large enough to drain the most conductive layers. Lateral
remain in the swelling (elastic) branch of the oedometric curve. If
boundaries were located at a sufficient distance from the site (2000 m)
drawdown required for excavation exceeds the maximum historical
to ensure that they were not affected by pumping. Top and bottom
drawdown, deformations are not elastic (i.e., reversible) but small
boundaries were modelled as no flow boundaries because: 1) the layers
since most deformations occur in the early stages of the oedometric
located at the top and at the bottom of the aquifer had low hydraulic
curve. This distinctive behaviour of the soil has been observed during
conductivities and 2) the drawdown caused in these layers by pumping
other deep constructions in Barcelona (Pujades et al., 2014).
was small. As a result, flow entering through these boundaries was min-
imal. Groundwater pumped comes from the lateral boundaries through
2.1.3. Construction characteristics the most transmissive layers. The flow rate or the drawdown was pre-
The Trinxant shaft was a 33 m deep circular excavation with a diam- scribed at the pumping well depending on the objective of each simula-
eter of 20 m (Figure 2). The water table was located at 10 m depth. The tion or the available data. Table 1 and Fig. 3a and b provide the main
excavation was undertaken by combining the cut and cover method characteristics of the numerical models used in this study.
with deep pumping wells. Diaphragm walls, which were 50 m deep,
were constructed using a hydro-mill trench cutter and pumping wells 2.2.2. Hydro-mechanical numerical models
were drilled using the OD method (Ferrer et al., 2008). The shaft had Coupled hydro-mechanical simulations were performed to predict
two outer chambers to facilitate the introduction and the exit of the the settlements outside the enclosure. We used the finite element

Fig. 3. Schematic description of the numerical models. a) Boundary conditions and mesh of the hydrogeological axisymmetric numerical model. b) General and detailed plan views (with
the boundary conditions) of the multilayered hydraulic model. Some piezometers are also displayed. c) Boundary conditions and mesh of the hydro-mechanical numerical model.
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27 19

Table 1
Main features of the numerical models (hydrogeological and hydro-mechanical) used to develop the research presented in this manuscript.

Objective Type Mesh Boundaries

No. of Min. size Max. size Well(s) Opposite to the well External (multilayer)
elements (m) (m) (axisymmetric)

Pumping test Hydrogeological/ 13,000 0.1 50 Prescribed flow. Flow rate Prescribed drawdown. 0 m –
axisymmetric measured during the test
Dewatering design Hydrogeological/ 13,000 0.1 50 Prescribed drawdown. Different Prescribed drawdown. 0 m –
(stability) axisymmetric for each scenario (25 to 40 m)
Dewatering design Hydromechanical/ 4000 0.7 40 Prescribed drawdown (40 m) Prescribed drawdown. 0 m –
(settlements) axisymmetric
WTAT Hydrogeological/ 22,000 1 200 Prescribed drawdown. Different – Prescribed drawdown. 0 m
multilayer for each scenario (25 to 40 m)

numerical code CODE_BRIGHT (Olivella et al., 1994, 1996), which solves where σ V′ is the effective vertical stress, σV is the total vertical stress and
coupled hydro-mechanical problems in porous media. u is the water pressure. The total vertical stress is given by
The models were axisymmetric around the vertical well axis. Lateral
boundaries were at a sufficient distance (2000 m) from the site to σ V ¼ zγS ð4Þ
ensure that they were not affected by pumping and a prescribed
drawdown condition was imposed along the boundary opposite the and the water pressure by
pumping well. Top and bottom boundaries were modelled as no flow
boundaries. The drawdown was prescribed at the pumping well u ¼ hγ W ð5Þ
(Figure 3c). Mechanical boundary conditions involved restricting the
horizontal movements at the pumping well (symmetry axis) and at
where z is the depth of an observation point, γS the specific weight of
the lateral boundary (opposite the pumping well). The vertical
the soil, which was taken as 2 t/m3, and h is the piezometric head
movements were restricted at the bottom of the model. The main
above the observation point. Unstable conditions are reached when
characteristics of the hydro-mechanical numerical models are given in
σ V′ ≤ 0. This occurs at excavations where the dewatering system
Table 1 and in Fig. 3c.
(wells and diaphragm walls) is not designed properly. The dewatering
The variables used in the models were intrinsic permeability (k0),
system must ensure that σV N u or σV/u N 1 in the whole aquifer
Young's modulus (Ei) and Poisson's ratio of each layer (υi). Poisson's
below the excavation bottom. But in practice, owing to soil uncertainty,
ratio used was 0.35 for every layer. Intrinsic permeability was obtained
it is common to apply a safety factor (SF) to ensure stability. A SF of 1.2
from the results of the pumping test and Ei was calculated from the
was adopted,
compressibility of the soil in each layer (α i), applying
σV
3ð1−2υi Þ SF ¼ ≥1:2: ð6Þ
Ei ¼ ð1Þ u
αi

where αi was derived from the storage coefficient of each layer (Si), Since the soil is heterogeneous and since the fall in pressure is differ-
which was obtained from the pumping test, applying the equation ent in each geological layer, the SF was computed for each metre below
proposed by Jacob (1950) and cited by Ferris et al. (1962) for elastic the bottom of the excavation. The SF may appear small when compared
aquifers, with the recommendations given by the EUROCODE 7 (Frank et al.,
2004), which suggests that the hydraulic heave should be verified by
Si comparing the total stress and the pore water pressure multiplied by
αi ¼ −θi β ð2Þ the corresponding partial safety factors:
γ w Di

where θi is porosity, γW is the specific weight of the water, Di is the
thickness of the layer and β is water compressibility, which is much
smaller than typical soil compressibilities and is often ignored. In
fact, if β is not considered SSi = α iγW, where SSi is the specific storage
coefficient of the ith layer.
2.2.3. Bottom stability (safety factor)
The numerical results of several dewatering scenarios were
compared. The main difference between them was the depth of the
enclosure. All the dewatering scenarios had to ensure the stability of
the excavation bottom. Groundwater could give rise to two stability
problems at the bottom of the excavation: fluidisation and base heave.
Fluidisation occurs when the excavation is undertaken in unconfined
aquifers, and base heave takes place when the excavation is made
above a confined aquifer (Cashman and Preene, 2001). In practice, pos-
itive (i.e., compressive) effective stresses are necessary when dealing
with these two problems. The vertical effective stress can be calculated
by applying Terzaghi's equation in the vertical direction (Terzaghi and
Peck, 1948)
0
σ V ¼ σ V −u
γG;stb σ V ¼ γG;dst u ð7Þ
where γG,stb and γG,dst are the safety coefficients for stabilising and
destabilising actions, respectively. The Eurocode standards recommend
γG,stb = 0.9 and γG,dst = 1.35 for this case. But we used a lower global
safety factor of 1.2 for the following reasons:
a) The Eurocode recommendation is based on shallow pumping of the
excavation bottom and not on deep well pumping from the whole
soil profile below the shaft, which is the case under study. It is thus
possible to measure and control the value of the pore water pressure
inside the pumping well, thereby reducing the uncertainty of its
value.
b) Shear resistance such as cohesion and friction forces were not
considered in our calculations. Therefore, our calculations of the SF
could be considered conservative since these forces oppose the
water pressure effects.
Water pressure used to compute the SF was obtained numerically
from observation points located inside the enclosure but near the dia-
phragm walls. These points measured the water pressure at different
ð3Þ depths.
20 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
Fig. 4. a) Earlier coarse scale interpretation of the geology of the site. The scale of the study was not sufficient to identify the transmissive layers located below the excavation. b) Natural
Gamma Ray record obtained at the borehole B4. This sketch shows how the Natural Gamma Ray records were interpreted.
3. Application
3.1. Geological characterisation
The site had been geologically characterised as a part of the tunnel
project using the data of boreholes spaced about 100 to 150 m apart.
As a result, the study scale was too coarse (Figure 4a). Only one trans-
missive layer had been identified below the excavation bottom. Thus,
since most of the soil seemed to have low values of hydraulic conductiv-
ity, the need for deep conventional pumping wells was challenged.
Techniques employed to lower the water pressure in soils with low
values of hydraulic conductivity include wick drains, eductor wells or
excavation under undrained conditions. However, given the high risk
of these techniques (if the soil is not well characterised) and given
that conventional wells are a cost-effective technique, a detailed geolog-
ical study was performed to select the best alternative. Three tools were
used: borehole logging, grain size analyses and Natural Gamma Ray log-
ging. First, four boreholes were drilled (PZ-1, PZ-2, PZ-3 and B-4 in
Figs. 1b and 2) and their core, which was obtained unaltered, was
described. A number of layers of different lithologies were identified
(Table 2). Thereafter, soil samples were taken at some depths
(Table 2) for grain size analyses. Using these results and taking the
Wentworth grade scale (Wentworth, 1922) as a reference, the litholo-
gies observed in the borehole logging were verified (Table 2). Finally,
the Natural Gamma Ray was recorded for each borehole. This technique
allows us to observe the continuity of the layers and identify
the changes in the lithology (Cripps and McCann, 2000). Natural
Gamma Ray records were interpreted as follows. Since high values of
Natural Gamma Ray indicate the presence of clay, these values were
associated with layers composed of fine sediments or of coarse
sediments with a clay matrix. Layers made up of sand or gravel were as-
sumed when the values of Natural Gamma Ray were small (Figure 4b
provides a sketch to explain how the Natural Gamma Ray records
were interpreted).
See Fig. 2 for the geological interpretation of the site. The Quaternary
and Tertiary sediments are divided into several strata of different tex-
tural and lithological properties. Fifteen layers were differentiated
below the water table (Table 2). Most of the materials located below
the water table belong to the Pliocene Age (Layers 2 to 15). Despite
the predominance of fine sediments, some layers of coarse materials lo-
cated above the maximum excavation depth (Layers 9 and 11) enabled
us to facilitate dewatering by using deep conventional pumping wells.
However, this was not sufficient since it was necessary to know the
exact hydraulic conductivity of the layers in order to design the
dewatering method.
3.2. Hydrogeological characterisation
A pumping test was performed to estimate the hydraulic properties of
the different geological layers. Pumping tests involve pumping water and
monitoring the groundwater response at several observation points (pie-
zometers) located at different distances and screened at different depths.
The boreholes drilled to obtain the unaltered cores were used as piezom-
eters (PZ-1, PZ-2 and PZ-3) (Figures 1b and 2). Additionally, the head
data from a vibrant string piezometer was used (VW-1). The screen
of the piezometers was located at diverse depths to observe the re-
sponse at different layers. PZ-1 was screened between 54 and 57 m
depth, PZ-2 between 32 and 35 m depth, PZ-3 between 46 and 50 m
depth and VW-1 between 33 and 34 m depth. The testing well (WT)
was screened between 30 and 50 m depth (Figure 2). The pumping test
lasted five days. It involved two days of pumping and three days of recov-
ery. The average flow rate during the test, which was measured using an
electromagnetic flow-meter, was 5.6 l/s. Drawdown evolution was mon-
itored at the piezometers and at the pumping well. Head measures were
taken manually and automatically. The automatic data were taken every
minute using pressure sensors and the manual data were monitored with
a groundwater level probe at times distributed logarithmically. A cross
section of the drawdown distribution during the pumping test is shown
in Fig. 5. This drawdown distribution, which was obtained numerically,
corresponds to the time just before the cessation of pumping.
The hydraulic parameters, k and Ss of each layer, were obtained
(Table 2) by automatic parameter estimation performed with the
hydrogeological numerical model of Fig. 3a. The hydraulic conductivity
to start the estimation process was chosen in accordance with k, which
was obtained from the grain size results (Table 2) using Breddin's
method (Breddin, 1963). The main characteristics of the model (mesh,
boundary conditions…) are specified in Table 1.
Good fits were obtained for the drawdowns from the numerical
estimation at all observation points, PZ-1, PZ-2, PZ-3 and VW-1
(Figure 6). Estimated conductivities were consistent with the lithology.
Moreover, the effective transmissivity of the aquifer, which was calcu-
lated using Jacob's method (Cooper and Jacob, 1946), was about
170–200 m2/d, which compared well with that obtained when integrat-
ing the values of Table 2. The hydraulic conductivity values of the layers
located below the excavation bottom (N 0.1 m/d except Layers 12 and
13) revealed that the most suitable technique for performing the
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27 21

Table 2
Characteristics of the layers located below the water table. The table displays the depth where the samples for the grain size analyses were taken, the hydraulic conductivity obtained from
each analysis and the lithological description of the samples. It also shows all the detected layers located below the water table with the numerical results obtained from the pumping test
characterisation.

Layer Grain size sample depth Grain size K Top depth Bottom depth Numerical K Numerical Ss Description
(m) (m/d) (m) (m) (m/d)

LI 10 b0.008 10.0 11.0 0.5 1.00E−05 Fine sand with clayey–silty matrix
L2 11.5 0.008 to 0.06
11.75 0.008 to 0.06 11.0 14.0 0.01 9.99E−06 Sand with clayey–silty matrix
12.9 b0.008
L3 14.0 17.0 0.05 9.98E−06
L4 19.3 b0.008 17.0 28.0 0.7 9.89E−06 Sand with clayey–silty matrix
25.3 0.008 to 0.06
L5 28.0 30.0 0.0008 1.07E−05
L6 30.5 8 to 40 Sand (coarse, medium and fine) with silty–clav matrix
31.9 0.6 to 0.06 30.0 34.0 28 1.54E−05
33.1 0.6 to 0.06
L7 34 0.008 34.0 35.0 0.02 1.06E−05 Fine sand with clayey–silty matrix
L8 37.1 0.008 to 0.06 35.0 42.0 0.8 3.64E−05 Fine sand with clayey–silty matrix
L9 42.5 8 to 40 42.0 43.0 20 1.05E−05 Sand (coarse, medium and fine) with silty–clay
L10 45.1 0.008 to 0.06 43.0 46.0 6 1.20E−05 Fine sand with clayey–silty matrix
Lll 46.7 8 to 40 46.0 48.0 1.12E−05 Sand (coarse, medium and fine) with silty-clay matrix
47.5 8 to 40
L12 48.3 8 to 40
49.1 8 to 40
49.3 0.6 to 1.7 48.0 52.0 0.05 1.36E−05 Sand with variable proportion of silty–clayey matrix
49.5 0.008 to 0.06
49.9 0.6 to 1.7
50.5 0.008 to 0.06
L13 52.45 0.06 to 0.6 52.0 53.0 0.09 1.01E−05 Very fine sand with clayey–silty matrix
L14 53.0 55.0 0.8 1.02E−05
L15 55.3 0.008 55.0 65.0 0.1 1.05E−05 Silty with some of very fine sand
61.6 0.008

dewatering was deep pumping wells since the flow rates required
would be excessive for other techniques (eductor wells, wick drains
or excavation under undrained conditions). Hydrogeological character-
isation showed that, although only 11% of the saturated zone was
described as transmissive (Layers 1, 6, 9 and 11) in the geological inter-
pretation, most of the layers (74.5%) had values of hydraulic conductiv-
ity higher than 0.1 m/d. This suggested the presence of unidentified
transmissive layers intercalated inside the low transmissive materials.
Most of k values estimated from the pumping test interpretation
were generally consistent with those obtained from the grain size anal-
yses (Table 2). However, some differences were observed (Layers 1, 4, 6,
Fig. 5. Drawdown distribution during the pumping test. This distribution corresponds to
the time just before the cessation of pumping.
8, 10 and 12). The difference observed in Layer 12 was due to the fact
that the sample was contaminated by sediments of Layer 11, while the
main reason for the differences observed in Layers 1, 4, 6, 8 and 10
was that layers were not homogeneous. The k estimated from the
pumping test is the horizontal effective hydraulic conductivity (kH-EFF,
i.e., arithmetic mean of the k values of sublayers), which is far larger
than the k value obtained from aggregates grain size analyses (kGRA).
Differences observed are listed and explained in detail below:
• Sample from Layer 1 (kGRA b 0.008 m/d; kH-EFF = 0.5 m/d): This layer
comprises Pleistocene materials (gravel with a clay matrix). The sam-
ple was probably taken from a depth with a higher proportion of clay.
• Sample from Layer 4 (kGRA b 0.008–0.06 m/d; kH-EFF = 0.7 m/d): Layer
4 is composed of fine sediments and is 11 m thick. Fine transmissive
layers that increase the global value of kH-EFF of the layer may be
present.
• Two samples from the lower part of Layer 6 (kGRA b 0.06–0.6 m/d;
kH-EFF = 28 m/d): This layer has a high proportion of clay in the
middle (Figure 2). The samples were taken from this site.
• Sample from Layer 8 (kGRA b 0.008–0.06 m/d; kH-EFF = 0.8 m/d): Both
values (estimated with the data of the pumping test and obtained
from the grain size analyses) are relatively small. However, the
estimated value is higher. This layer has different proportions of clay
and sand. The sample was probably taken from a site with more clay.
• Sample from Layer 10 (kGRA b 0.008–0.06 m/d; kH-EFF = 6 m/d): The
sample was collected from the deepest part of Layer 10, which has
the highest proportion of clay.
• Three samples from the top of Layer 12 (kGRA b 0.6–40 m/d; kH-EFF =
0.05 m/d): As stated above, these samples were contaminated by
sand of Layer 11.
Sensitivity analyses were performed to demonstrate that the esti-
mated values were useful to predict the behaviour of the aquifer. The
pumping test was simulated taking into account three scenarios. The
hydraulic conductivity of the permeable layers was increased (scenario
1) and reduced (scenario 2) by one order of magnitude. Numerical
drawdown did not fit the measured drawdown during the pumping
22 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
Fig. 6. The plots show the drawdown evolution measured during the pumping test (dots) at the four piezometer versus the drawdown evolution simulated considering two different sce-
narios. In the former scenario, the pumping test was simulated using the hydraulic conductivities obtained from the numerical automatic calibration (black line). In the latter (dashed line),
the pumping test was simulated by considering the hydraulic conductivities obtained from the grain size analyses in the layers where these did not match with the estimated ones.
test in these two scenarios since the effective transmissivity of the aqui-
fer was considerably altered (Supplementary Fig. 1). The hydraulic con-
ductivity was also modified in the layers where the values estimated did
not match those obtained from the grain size analyses (scenario 3). In
these layers, the hydraulic conductivity obtained by Breddin's method
was used to simulate the pumping test. Again, computed drawdown
did not fit measurements (Figure 6), which indicated that the hydraulic
conductivities obtained from the automatic estimation proved more
useful to simulate the behaviour of the aquifer. Drawdown varied large-
ly when the hydraulic conductivity of the layers was modified. This fact
supported to use the hydraulic conductivities obtained numerically
from the pumping test to predict the behaviour of the groundwater
since the response of the aquifer to a pumping (e.g. pumping test)
could not be simulated using other values of hydraulic conductivity
(Figure 6 and on-line Appendix).
3.3. Enclosure and dewatering design
Once the numerical model was calibrated, a number of dewatering
scenarios were proposed and simulated. The simulations were per-
formed using an axisymmetric numerical model by prescribing the
drawdown in the pumping well. The depth of the diaphragm walls,
the screen of the pumping well and the drawdown in the well differed
in accordance with each simulation. The main features of the numerical
model used to design the dewatering system (mesh, boundary condi-
tions…) are shown in Table 1. Note that two premises had to be as-
sumed 1) the minimum depth of the diaphragm walls that guaranteed
the stability and the verticality of the walls and 2) the depth of the
pumping well, which should not exceed too much that of the enclosure
because of the concern over the pumping settlements. This was a conse-
quence of the initial numerical analyses, which predicted large pumping
settlements exceeding the warning limits defined by the Administrator
(a value of 5 mm was regarded as a warning limit for settlements).
However, the initial analyses overestimated the magnitude of the
pumping settlements since the adopted low values of stiffness obtained
mainly from pressurometers analysing the unloading–reloading phase.
The actual stiffness of the soil was recalculated using Eqs. (1) and (2)
with the storage coefficient of the soil (Pujades et al., 2014).
In the numerical simulations the well was located in the middle of
the enclosure (symmetry axis) and the diaphragm wall was situated
at a distance equal to the radius of the shaft. The results of each simula-
tion were obtained in steady state. The SF was calculated at different
depths below the excavation bottom applying Eq. (6). As an example,
Table 3 displays the results obtained considering 1) a pumping well
screened as far as 55 m and 2) a 50 m deep diaphragm wall, and 3) dif-
ferent drawdowns inside the pumping well. Shaded cells depict depths
where the SF is less than 1.2. A satisfactory dewatering design was
achieved combining diaphragm walls of 50 m depth, a pumping well
screened from the water table to 55 m depth, and 40 m of drawdown in-
side the pumping well. Numerical results indicated that conventional
deep pumping wells were able to drain the Pliocene materials because
of their hydraulic conductivity. The expected steady state flow rate
was 2 l/s, which could be pumped with one pumping well. However,
two pumping wells were recommended in case of breakdown.
Subsequently, the effects due to the dewatering outside the enclo-
sure (settlements) were determined. Settlements were computed
using a hydro-mechanical numerical model, whose features are
displayed in Table 1. Note that only the settlements caused by
dewatering were calculated. Fig. 7a shows settlement distribution in
depth at three distances from the enclosure (5, 10 and 20 m). Larger set-
tlements were observed in the deepest layers of the aquifer and in Layer
8. The deepest layers were not cut by the diaphragm walls with the re-
sult that the drawdown caused by pumping was greater. Layer 8 had the
highest value of compressibility. Note that the settlements were miti-
gated by the upper layers (Layers 1 to 5 in Table 2). Drawdown in
these layers was negligible (or non-existent in the unsaturated zone —
Layer 1) because the effective vertical hydraulic conductivity of the
aquifer was low. The mitigation in the surface direction is not uncom-
mon since similar situations occur when a confined aquifer is pumped
(the confining layers mitigate the surface settlements). Fig. 7b shows
the predicted settlement at the surface. Larger settlements were con-
centrated around the enclosure, which was consistent with the draw-
down distribution. Given the depth reached by the diaphragm walls,
the predicted drawdown outside the enclosure was small. As a result,
the settlements would not jeopardize the stability of nearby buildings
since the largest settlements would be around 0.5 mm.
3.4. Watertightness Assessment Test (WTAT)
Enclosures may have open joints or gaps produced during their con-
struction. These defects could adversely affect the functions of the en-
closures. If they are located above the excavation bottom, they may
cause sink holes and if they are situated below, they may hamper
dewatering, jeopardizing the stability of the bottom. It is therefore
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27 23

Table 3 broken down. Note that PZ-1′ was located outside the enclosure. As in
Safety factor at different depths calculated for four different scenarios. All the scenarios
the first test, head measures were taken manually after a logarithmic
have 50 m depth diaphragm walls and a 55 m depth pumping well. Drawdown in the
pumping well differs in accordance with each scenario. Stable conditions are achieved frequency and automatically every minute. The flow rate was mea-
with 40 m of drawdown in the pumping well. Flow rate is shown in the lower most line. sured by using an electromagnetic flow-meter at W1 and a Woltman
horizontal turbine flow-meter at W2. The averages of the flow rate
25 m 30 m 35 m 40 m
Depth were 4 and 3 l/s at W1 and at W2, respectively.
drawdown drawdown drawdown drawdown
Simultaneously, the drawdown of the test was predicted with a
33
Non saturated multilayer numerical model (Figure 3b). The model could not be
35 8.64
37 3.22
axisymmetric because there were two pumping wells. The number of
39 2.65 represented layers in the model was reduced from 15 to 9 to decrease
41 2.43 9.49 computation time (the number of elements had exceeded 35,000 in
43 2.32 5.37 the case of 15 layers). The multilayered model was set up by replacing
45 2.26 4.16 26.73 the layers with low values of k (where the horizontal flow is negligible)
47 2.20 3.57 9.37 above the bottom of the diaphragm walls by one-dimensional elements
49 1.59 1.96 2.54 3.63 that linked the rest of the layers (more conductive layers and layers
51 1.13 1.19 1.27 1.36
below the diaphragm walls). The characteristics of the numerical
53 1.10 1.14 1.18 1.23
model used are shown in Table 1.
55 1.14 1.18 1.22 1.27
1.24 1.28
Measured and predicted drawdowns did not fit (Figure 8), which in-
57 1.17 1.21
59 1.20 1.23 1.26 1.30
dicated that the enclosure had a defect. Three additional piezometers
61 1.23 1.26 1.29 1.32 (PZ-5, PZ-6 and PZ-7) screened at different depths were drilled (inside
63 1.26 1.28 1.31 1.34 the entrance and exit chambers) to locate the position of the defect. In
65 1.28 1.31 1.34 1.37 total, six piezometers were thus available: one screened below the dia-
Q (1/s) 1.25 1.5 1.75 2 phragm walls (PZ1′), two screened at the bottom of the diaphragm
walls (PZ3· and PZ6), two screened in the intermediate layers between
the bottom of the diaphragm walls and the water table (PZ5 and PZ7),
and one screened at the top of the saturated zone (PZ2) (Figures 1b
essential to determine the state of the enclosure before commencing the and 2). The distribution of drawdown in plan view enabled us to locate
excavation stage. The conditions of the enclosure can be determined by the defect (Figure 9) whereas the position of the screens allowed us to
using hydrogeological tools such as pumping tests (Pujades et al., approximate the depth of the defect. Larger drawdowns were observed
2012a; Vilarrasa et al., 2012). For this reason, a pumping test, which in- on the north side of the enclosure. Thus, it was deduced that the enclo-
volved pumping inside the enclosure and measuring the groundwater sure was open near PZ1′ and PZ3 (Figure 9). The defect was situated in
response outside, was carried out when the enclosure was completed. deep layers since the drawdown was higher at the deeper piezometers.
The pumping test lasted 7 days, four for pumping and three for recov- Moreover, defects in diaphragm walls usually occur at great depths
ery. Two pumping wells were used (W1 and W2), which were screened because construction problems are more common at these depths
from 10 to 55 m depth. The head was measured at piezometers PZ-1′, than at shallow depths. Further, numerical results were obtained by in-
PZ-2 and PZ-3 (Figures 1b and 2). VW-1 was not employed as it was corporating the defect into the model to corroborate our hypothesis.
out of order. PZ-1 was replaced by PZ-1′ because the former had also This was done giving the value of k of the aquifer to one panel of the
diaphragm walls, whose dimensions were 2 × 1.2 m, located on the
north side of the shaft. This was performed in the layers where the de-
fect was assumed to be (Layers 10 and 11 in Table 2). The results agreed
with the data measured (Figure 8).
The response time to the pumping (observed in the WTAT), which
increased with respect to the first pumping test, also confirmed the
location of the defect. This response time depends on the hydraulic
characteristics (k and SS) of the aquifer and on the path followed by
the groundwater. In the case of well constructed diaphragm walls,
groundwater needs to flow below the walls to reach the well. As a result,
the path followed is longer than in the case without walls. In the course
of the WTAT the response time increased by two orders of magnitude at
the piezometers located further from the defect (PZ-2), but increased
much less at the piezometers situated near the defect (PZ-3). This small-
er increase was attributed to the defect since the groundwater did not
need to flow below the walls to reach the well. Groundwater flowed
through the defect, reducing the response time.
The stability and the pumping effects were tested taking into ac-
count the defect because earlier dewatering predictions had assumed
an enclosure without defects. Two pumping wells (W1 and W2) were
necessary to lower the water pressure as required. One of them was lo-
cated close to the defect, which facilitated stable conditions in this area.
A decline in the effectiveness of the enclosure due to the defect is
reflected in the higher flow rate needed to ensure stability. The flow
rate rose from 2 to 6.6 l/s (+330%), suggesting an increase in the effec-
Fig. 7. a) Computed settlements versus depth at three distances from the enclosure (5, 10 tive transmissivity of the enclosure. Bottom stability was achieved by
and 20 m). b) Computed surface settlements versus distance from the enclosure.
Settlements in both figures were predicted by assuming the proposed dewatering system
prescribing 40 m of drawdown inside the two wells. Results are
(diaphragm walls of 50 m depth, a pumping well screened from the water table to 55 m shown in Table 4. Note that the SF was calculated in the middle of the
depth and 40 m of drawdown inside the pumping well). modelled layers because the water pressure could only be determined
24 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
Fig. 8. Evolution of measured (dots) and computed drawdowns during the Watertightness Assessment Test (WTAT). Numerical results are shown both with the defect in the enclosure
(continuous line) and without the defect (dashed line).
at these locations owing to the fact that the numerical model was
multilayered.
Numerical settlements were re-calculated considering the defect.
Given the difficulty of constructing a 3D hydro-mechanical model, the
settlements were obtained using an axisymmetric hydro-mechanical
model, the characteristics of which were the same as those of the
hydro-mechanical model used during the dewatering design stage
(see Section 3.3). To simulate the defect, the hydraulic conductivity of
the diaphragm walls located in the deep layers was increased from
0.00001 m/d to 0.5 m/d. This value was the effective horizontal k (of
the walls in the deep layers) assuming that the area affected by the de-
fect had a width of 2 m. Predicted settlements in the area close to the
defect were 1.3 mm at 10 m from the enclosure where the buildings
were located. The settlements on the other sides of the enclosure
were not calculated because outside drawdown caused by a defect is
concentrated around it (Pujades et al., 2012a; Vilarrasa et al., 2012).
The flow rate and settlements caused by dewatering were acceptable.
Therefore, the excavation continued as planned.
Fig. 9. Measured drawdown during the WTAT at the piezometers located around the en-
closure. Drawdown distribution indicated the existence of a defect on the north side of
the enclosure.
The defect was not repaired by injection because of its location
(below the excavation bottom) and because of the size of the settle-
ments. In addition, stable conditions were readily achieved. Moreover,
the injections would have interrupted the construction tasks. However,
if the defect had been located above the bottom, this would have been
repaired to avoid soil being dragged towards the excavation, which
would have produced a sinkhole.
3.5. Actual dewatering evolution
The head, the flow rate and the soil movements were measured
during the dewatering stage. The head was measured at piezometers
(PZ-1′, PZ-2, PZ-3, PZ-5, PZ-6 and PZ-7) and at the pumping wells
(W1 and W2) manually (Figure 1b). Measures were taken daily to verify
the dewatering of the excavation. The flow rate was measured at the
pumping wells by flow-meters. Since it was not possible to use an elec-
tromagnetic flow-meter given its fragility, the flow rate was monitored
using a Woltman horizontal turbine flow-metre. Soil movements were
monitored by manual levelling during the construction at some
monitoring points around the enclosure (Figure 1c).
Fig. 10 displays the head evolution during the dewatering stage
measured at the piezometers and at the pumping wells. The heads in-
side the pumping wells attained the adequate depth to ensure bottom
stability. A fall and a recovery due to the WTAT were observed in early
September 2009. Pumping continued until mid September to detect
the defect by monitoring the drawdown at the additional piezometers.
Dewatering commenced when the excavation reached the water table
and continued until November 2009.
Fig. 11a displays the partial soil movements produced by the con-
struction of the diaphragm walls and Fig. 11b shows these movements
Table 4
Safety factor computed numerically assuming the defect in
the enclosure. Predicted flow rate is also shown in the
lower most line.
Depth 25 m drawdown
33 Nonsaturated
38.5
42.5
44.5 2.39
47 1.58
54 1.37
Q (1/s) 6.6
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
Fig. 10. Evolution of measured drawdown at the piezometers and at the pumping wells
during and after the dewatering stage.
during the excavation and dewatering stages. The settlements attribut-
ed to the construction of the walls were approximately 1 mm around
the enclosure except on the north side, where the soil dropped 2 mm.
This drop resulted from the problems on the north side during the con-
struction of the walls, which corroborated the presence of the defect.
The soil did not behave as expected during the excavation and the
dewatering stages since the observed pumping settlements did not fit
the predicted ones. The settlements were larger than the predicted
ones on the south and south-east sides and smaller on the north side.
No settlements were observed on the west side. Contrary to expecta-
tions the soil underwent a heave. Moreover, the expected shape of the
settlements was circular with a protuberance around the defect. But
the settlement distribution was not circular although the protuberance
could be intuited. The largest settlements were delimited by an elongat-
ed area that followed the direction of the street.
Evolution of soil movement at 8 monitoring points (Figure 1c) is
shown in Fig. 12a and b. Fig. 12a displays the soil movements measured
by monitoring points located on the north and west sides whereas
Fig. 12b shows the measures from the monitoring points located on
the south and south-east sides. The movements were similar at all the
monitoring points until the start of the excavation. Only HN-8 registered
larger settlements and HN-1 did not vary its position during the con-
struction of the walls. Similar observations are made in the plots of set-
tlement distribution (Figure 11). During the dewatering (second shaded
area in both plots) three types of behaviour were observed: points locat-
ed on the south and south-east sides dropped, points located on the
west side increased and points located on the north side registered a
moderate drop. When the pumping ceased, a small and rapid increase
Fig. 11. a) Plan view of the iso-displacements during the construction of the diaphragm walls. b) Plan view of the iso-displacements during the excavation and dewatering stages. Shaded
area is the zone where the soil heaved.
25
Fig. 12. Evolution of the soil movements measured at the monitoring points located on the
a) west and north sides and b) south and south-east sides. Plan view of the monitoring
points is in Fig. 1c.
was initially noted. Thereafter, two types of behaviour were observed:
the south and south-east sides increased during the recovery period
whereas the other sides dropped symmetrically.
The difference between the predicted and the observed soil displace-
ments was probably because the soil was assumed to be over-
consolidated. However, the soil could have been adversely affected and
could have lost its overconsolidation ratio at some locations around the
enclosure before the start of the excavation. It was postulated that the
soil had lost its overconsolidation ratio because of two liquefaction events
that occurred during the drilling of the pumping (testing) well. These led
to a large amount of sand being dragged to the surface. During the filling
of the gravel pack around the well, additional gravel was required to fill
the cavities brought about by the siphoning events. To fill the gravel
pack located between 30 and 32 m depth, 1 m3 of gravel was necessary
and to fill the cavity between 42 and 47 m depth, 4 m3 was needed. How-
ever, the alteration of the soil structure produced by these liquefaction
events was too local and only affected the area near the testing well. Al-
though these events contributed to the pumping settlements, it would
be exaggerated to attribute all the differences in the predictions to the
26 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
construction of the testing well. Moreover, the shape of the settlement
distribution was not caused by the siphoning events.
Nevertheless, the settlement distribution could be attributed to the
construction of a building located on the south-east side of the shaft
(Figure 1c). This building includes a 4 storey underground car park
reaching the water table (located at 10 m depth), and is therefore
deeper than the other buildings in the area. The excavation needed for
the construction of this car park would have brought about changes in
the soil strain leading to a reduction in the overconsolidation ratio of
the soil in the adjacent area. This area (south-east side of the enclosure)
settled more during dewatering provoking a rotation movement of the
soil, which produced the heave of the north and east sides. Note that
pumping settlements were similar in the area occupied by the building,
decreasing abruptly further away from this edifice. No heave was ob-
served on the north side since the dewatering settlements were larger
due to the defect but the observed settlements were smaller than the
predicted ones. As the overconsolidation ratio was reduced only in the
proximity of the building, the rest of the soil continued to behave elas-
tically. Consequently, when pumping ceased, the west side returned to
its initial position whereas the south-east side underwent a heave.
However, we do not have enough data to corroborate this hypothesis
unequivocally. Another factor which probably accounted for some of
the differences between the observed and the predicted behaviours
could be the uncertainty of the numerical models. Nevertheless, the du-
ration of the pumping test, which enabled us to determine the behav-
iour of a large part of the aquifer, and the sensitivity analysis indicated
that the errors in the numerical predictions were not excessive.
4. Discussion and conclusions
Our results highlight 1) the importance of characterising the soil by
combining different techniques (borehole logging, grain size analyses,
Natural Gamma Ray and pumping tests) and 2) the usefulness of hydrau-
lic tests to locate the presence of transmissive layers within low hydraulic
conductivity materials. Earlier geological studies of the site indicated the
suitability of dewatering techniques recommended for low permeability
materials (such materials were expected to be located at the bottom of
the excavation). However, a subsequent hydrogeological characterisation
revealed the presence of transmissive layers. Borehole logging, grain size
analyses and Natural Gamma Ray logging were used to determine the
characteristics of the soil. Moreover, a pumping test revealed that the
hydraulic conductivity exceeded 0.1 m/d in 75% of aquifer thickness,
which is not unusual. Soils and rocks described as materials with poor
permeability during well logging often contain layers of transmissive
materials. These layers lead to a significant transmissivity of the whole
formation. In our case, the nature of the materials tipped the balance
with the result that conventional deep pumping wells were used. If
another dewatering technique (focusing on dewater low hydraulic
conductivity materials) had been used, unstable conditions would have
developed during the excavation because of an insufficient drop in the
water pressure.
Despite their frequency in enclosures (Bruce et al., 1989; Knight
et al., 1996; Vilarrasa et al., 2012), defects are often overlooked.
Geophysical methods such as crosshole sonic logging (Paikowsky and
Chernauskas, 2003) are occasionally used to assess the state of the joints
along enclosures of diaphragm walls. However, these methods suffer
from two limitations: 1) only the concrete between two access tubes
can be assessed and 2) the results may be adversely influenced by a
variety of factors (White et al., 2008). Our results demonstrate the suit-
ability of performing a WTAT, which involves pumping tests before and
after constructing the enclosure, and comparing the monitored and pre-
dicted drawdowns. The whole enclosure can be tested by performing a
WTAT and the defects can be detected. In our study, numerical model-
ling was also used in combination with the groundwater measures
taken during the WTAT to detect the defect. Moreover, this test does
not interrupt the construction and the results are generally obtained
without delay. The WTAT, which was performed with piezometers
screened at different depths, also allowed us to identify the depth of a
defect. This was of paramount importance because repairs would have
been necessary if the defect had been detected above the excavation
bottom. In our case, it was detected below the bottom and the problem
was resolved by simply modifying the dewatering design. A major issue
of concern is the loss of efficiency of an enclosure containing a defect,
which increases the effective transmissivity of the enclosure. The results
show that the flow rate required for stable conditions increased from 2
to 6.6 l/s (+330%).
The need for soil characterisation to design a suitable dewatering
system and the need to evaluate the state of the enclosure prior to exca-
vation underscore the significance of adopting a formal procedure for
deep excavations under the water table. Such a procedure must include
soil characterisation, dewatering numerical design, WTAT and monitor-
ing of the dewatering. The dewatering numerical design should entail
hydrogeological and hydro-mechanical modelling. Different scenarios
should be modelled and these should be compared taking into account
1) the stability of the bottom of the excavation and 2) the effects (settle-
ments) caused by the dewatering outside the enclosure. The most
suitable scenario can then be chosen to carry out the excavation.
The excavation of the Trinxant shaft demonstrated that pumping
settlements in overconsolidated soils are small and elastic. They can
be estimated with moderate error using stiffness parameters obtained
from the storage coefficient values of the aquifer (see Pujades et al.,
2014). However, in our study, pumping settlements could not be
predicted accurately since the initial stress distribution around the
shaft of Trinxant induced by the adjacent building was unknown. In
order to model the behaviour of the soil during the dewatering, a
three-dimensional hydro-mechanical numerical model and prior
knowledge of the initial stress distribution would have been necessary.
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.enggeo.2014.06.007.
Acknowledgements
The authors would like to acknowledge ADIF (Administration),
SACYR (construction company) and INTECSA-INARSA (technical assis-
tance), for their support throughout the hydrogeological monitoring of
the civil works. The authors were appointed by ADIF as external
advisors during the construction of the tunnel. Additional funding was
provided by the Spanish Ministry of Science and Innovation (MEPONE
project: BIA2010-20244); and the Generalitat de Catalunya (Grup
Consolidat de Recerca: Grup d'Hidrologia Subterrània, 2009-SGR-
1057). E. Pujades gratefully acknowledges the financial support from
the AGAUR (Generalitat de Catalunya) through “the grant for universi-
ties and research centres for the recruitment of new research personnel
(FI-DGR 2011–FI-DGR 2012)”.
References
Breddin, H., 1963. Die Grundribkarnten des Hydrogeologischen Kartenwekes
derWasserwistschafteverwattung von Norgrhein-Westfalen. Geologische
Mitteilungen 2 (4), 393–416.
Bruce, D., De Paoli, B., Mascardi, C., Mongilardi, E., 1989. Monitoring and quality control of
a 100 metre deep diaphragm wall. In: Burland, J., Mitchell, J. (Eds.), Pilling and Deep
Foundations. A.A. Balkema, Rotterdam, pp. 23–32.
Carrera, J., Neuman, S., 1986a. Estimation of aquifer parameters under transient and steady
state conditions: 1. Maximum likelihood method incorporating prior information.
Water Resour. Res. 22 (2), 199–210.
Carrera, J., Neuman, S., 1986b. Estimation of aquifer parameters under transient and
steady state conditions: 2. Uniqueness, stability and solution algorithms. Water
Resour. Res. 22 (2), 211–227.
Carrera, J., Neuman, S., 1986c. Estimation of aquifer parameters under transient and
steady state conditions: 3. Application to synthetic and field data. Water Resour.
Res. 22 (2), 228–242.
Cashman, P.M., Preene, M., 2001. Groundwater Lowering in Construction — A Practical
Guide. Spon Press, London.
Cooper, H.H., Jacob, C.E., 1946. A generalized graphical method for evaluating formation
constants and summarizing well-field history. Eos trans. AGU 27 (4), 526–534.
 E. Pujades et al. / Engineering Geology 178 (2014) 15–27
Cripps, A.C., McCann, D.M., 2000. The use of the natural gamma log in engineering geological
investigations. Eng. Geol. 55, 313–324.
Davies, 1984. Some geotechnical problems with foundations and basements in Singapore.
Proceedings of International Conference on Tall buildings, Singapore, pp. 22–26.
El-Nahhas, F.M., 1999. Soft ground tunnelling in Egypt: geotechnical challenges and
expectations. Tunn. Undergr. Space Technol. 14 (3), 245–256.
Ferrer, A., Davila, V., Sahuquillo, A., 2008. Descenso del nivel freático en centros urbanos.
Diseño e instalación de una red de control eficiente y sostenible. In: Carmen, M.,
Vallejos, A., Valverde, M., Lambán, L. (Eds.), El agua y las infraestructuras en el
medio subterráneo. Instituto geológico y minero de España, Madrid, pp. 335–353.
Ferris, J., Knowles, D., Brown, R., Stallman, R., 1962. Theory of Aquifer Tests. U.S. Geological
Survey Water Supply Paper 1536-E.
Frank, R., Baduin, C., Driscoll, R., Kavvadas, M., Krebs Ovesen, N., Orr, T., Schuppener, B.,
2004. Designer's Guide to EN1997-1, EUROCODE 7. Geotechnical Design-General
Rules. Thomas Telford, London.
Gulhati, S., Datta, M., 2005. Geotechnical Engineering. Tata McGraw Hill, Delhi.
Jacob, C.E., 1950. Flow of groundwater. In: Rousep, H. (Ed.), Engineering Hydraulics. John
Wiley, New York, pp. 321–386.
Jurado, A., De Gaspari, F., Vilarrasa, V., Bolster, D., Sánchez-Vila, X., Fernàndez-Garcia, D.,
Tartakovsky, D., 2012. Probabilistic analysis of groundwater-related risks at subsur-
face excavation sites. Eng. Geol. 125, 35–44.
Knight, D.J., Smith, G.L., Sutton, J.S., 1996. Sizewell B foundation dewatering-system
design, construction and performance monitoring. Géotechnique 46 (3), 473–490.
Li, X.G., Yuan, D.J., 2012. Response of a double-decked metro tunnel to shield driving of
twin closely under-crossing tunnels. Tunn. Undergr. Space Technol. 28, 18–30.
Medina, A., Carrera, J., 2003. Geostatistical inversion of coupled problems: dealing with
computational burden and different types of data. J. Hydrol. 281 (4), 251–264.
Medina, A., Alcolea, A., Carrera, J., Castro, L.F., 2000. Modelos de flujo y transporte en la
geosfera: Código TRANSIN IV (Flow and transport modelling in the geosphere: the
code TRANSIN IV). IV Jornadas de Investigación y Desarrollo Tecnológico de Gestión
de Residuos Redioactivos de ENRESA. Technical publication 9/2000, pp. 195–200.
Moore, J.F.A., Longworth, T., 1979. Hydraulic uplift at the base of a deep excavation in
Oxford Clay. Géotechnique 29 (1), 35–46.
Ni, J.C., Cheng, W., 2011. Shield machine disassembly in grouted soils outside the ventila-
tion shaft: a case history in Taipei Rapid Transit System (TRTS). Tunn. Undergr. Space
Technol. 26, 435–443.
Olivella, S., Carrera, J., Gens, A., Alonso, E.E., 1994. Non-isothermal multiphase flow of
brine and gas through saline media. Transp. Porous Media 15, 271–293.
Olivella, S., Gens, A., Carrera, J., Alonso, E.E., 1996. Numerical formulation for a simulator
(CODE_BRIGHT) for the coupled analysis of saline media. Eng. Comput. 13, 87–112.
27
Paikowsky, S., Chernauskas, L., 2003. Review of Deep Foundations Integrity Testing
Methods and Case Histories. BSCES-Geo-Institute Deep Foundations Seminar.
Powers, J.P., Corwin, A.B., Schmall, P.C., Kaeck, W.E., 2007. Construction Dewatering and
Groundwater Control — New Methods and Applications, third ed. John Wiley and
Sons, United States of America.
Preene, M., Roberts, T.O.L., Powrie, W., Dyer, M.R., 2004. Groundwater Control: Design and
Practice. CIRIA, London.
Pujades, E., Carrera, J., Vàzquez-Suñé, E., Jurado, A., Vilarrasa, V., Mascuñano-Salvador, E.,
2012a. Hydraulic characterization of diaphragm walls for cut and cover tunnelling.
Eng. Geol. 125, 1–10.
Pujades, E., Lopez, A., Carrera, J., Vàzquez-Suñé, E., Jurado, A., 2012b. Barrier effect of
underground structures on aquifers. Eng. Geol. 145–146, 41–49.
Pujades, E., Vàzquez-Suñé, E., Carrera, J., Vilarrasa, V., Jurado, A., De Simone, S., Ledesma, A.,
Ramos, G., Lloret, A., 2014. Deep enclosures versus pumping to reduce settlements
during shaft excavations. Eng. Geol. 169, 100–111.
Ramaswamy, S.D., 1979. Some aspects of practice and problems of foundations
in Singapore. Proceeding of 6th Asian Regional Conference on Soil Mechanics and
Foundation Engineering, Singapore.
Stille, H., Palmström, A., 2008. Ground behaviour and rock mass composition in
underground excavations. Tunn. Undergr. Space Technol. 23, 46–64.
Terzaghi, K., Peck, R.B., 1948. Soil Mechanics in Engineering Practice, second ed. John
Wiley and Sons, United States of America.
UPC, 2003. Codigo Visual Transin 1.1 R65. Developed in the Department of Geotechnical
Engineering and Geosciences (ETCG). UPC.
E.Vàzquez-Suñé, E., X.Sánchez-Vila, X., J.Carrera, J., 2005. Introductory review of specific
factors influencing urban groundwater, an emerging branch of hydrogeology, with
reference to Barcelona, SpainHydrogeol. J. 13, 522–533.
Vilarrasa, V., Carrera, J., Jurado, A., Pujades, E., Vazquez-Sune, E., 2012. A methodology for
characterizing the hydraulic effectiveness of an annular low-permeability barrier.
Eng. Geol. 120, 68–80.
Wentworth, C.K., 1922. A scale of grade and class terms for clastic sediments. J. Geol. 30,
377–392.
White, B., Nagy, M., Allin, R., 2008. Comparing cross-hole sonic logging and low-strain
integrity testing results. In: Santos, J.A. (Ed.), The 8th International Application of
Stress Wave Theory to Piles. IOS Press BV, Amsterdam.
Xanthakos, P., Abramson, L., Bruce, D., 1994. Ground Control and Improvement. John
Wiley and Sons, United States of America.