Tunnelling and Underground Space Technology 65 (2017) 62–75

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

journal homepage: www.elsevier.com/locate/tust

Assessing the timing and magnitude of precipitation-induced seepage

into tunnels bored through fractured rock
Michael G. Sweetenham ⇑, Reed M. Maxwell, Paul M. Santi
Hydrologic Science and Engineering Program, Center for Underground Construction and Tunneling, Department of Geology and Geological Engineering, Colorado School of
Mines, Golden, CO, United States

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

Article history: Seepage into tunnels bored through fractured rock is a common occurrence that can cause significant
Received 3 July 2016 problems for the construction process, tunnel longevity, and regional hydrogeology. Predictions of
Received in revised form 8 February 2017 seepage using analytical solutions and numerical models are often inaccurate due to the inherent assump-
Accepted 13 February 2017
tions, volumetric averaging of fractures, and lack of important hydrogeological features. This study seeks to
Available online 3 March 2017
better understand tunnel infiltration processes through the application of a high-resolution, integrated
hydrologic model. First, a conceptual model is developed for this research using the factors shown by pre-
Keywords:
vious studies to control net infiltration and seepage. A stochastic fracture continuum is generated for bed-
Tunnels
Precipitation
rock using FRACK, which maps discrete fracture networks to a finite difference grid with heterogeneous,
Infiltration anisotropic permeability fields. An integrated hydrologic model, ParFlow is then used to investigate the con-
Seepage trol exhibited by factors such as climatic forcing; vegetation; soil type and depth; bedrock type; fracture
Fractured rock spacing; and tunnel depth on the timing and magnitude of seepage into tunnels. Simulations are run using
hourly meteorological forcing. Surface and subsurface properties are adjusted individually to investigate the
change in seepage response for varying hydrogeology and land cover. Results show that fracture spacing
and connectivity, bedrock type, and overburden are particularly important pieces in obtaining reliable seep-
age estimates. Higher fracture spacing causes higher total seepage at a more constant rate than a lesser
spacing, which exhibits a much larger range of fluctuation in seepage volumes. More permeable and porous
bedrock increases lag times while reducing seepage volumes that remain relatively constant over time.
Thicker and less conductive soils both increase lag times and reduce seepage magnitude. Tunnels, precip-
itation, infiltration, seepage, fractured rock.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Water resources and environment (Gargini et al., 2008; Gleeson

Designing tunnels requires creative applications of geotechnical
engineering to predict the responses of soil, rock, and water. The
response of soil and rock to tunneling are well understood, how-
ever the way in which water infiltrates into tunnels is difficult to
predict. In fractured rock, water flowing into tunnels is a common
occurrence that can cause significant problems. Potential impacts
on the following prompted increased attention to seepage over
the past few decades:

Tunnel and underground construction (Cesano et al., 2000; Chen
and Tolon, 2012; Fernandez and Moon, 2010a; Huang et al., 2011;
Kitterod et al., 2000; Li et al., 2009; Wang et al., 2017).
⇑ Corresponding author.
E-mail addresses: michael.sweetenham@gmail.com (M.G. Sweetenham), rmax-
well@mines.edu (R.M. Maxwell), psanti@mines.edu (P.M. Santi).
et al., 2009; Kim et al., 2005; Rademacher et al., 2003; Scanlon
et al., 2002, 2006; Seyfried et al., 2005; Shimojima et al.,
1993; Vincenzi et al., 2009).

Quality of nuclear waste repositories (Bagtzoglou and Cesano,
2007; Liu and Bodvarsson, 2001; Philip and Knight, 1989;
Reeves et al., 2008; Trautz and Wang, 2002).
As a result, a number of approaches have been used to predict
tunnel seepage beneath the water table. Analytical solutions have
been derived to estimate inflows under a variety of groundwater
scenarios and boundary conditions. These solutions, such as those
derived by Philip and Knight (1989) and Perrochet and Dematteis
(2007) contain inherent simplifying assumptions that include
homogeneity, isotropy, no recharge, and radially infinite boundary
conditions. Field-scale experiments and observations have also
been used to predict seepage. For example, Heuer (1995, 2005,
2012) created a relatively robust field method by which the

http://dx.doi.org/10.1016/j.tust.2017.02.003
0886-7798/Ó 2017 Elsevier Ltd. All rights reserved.
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75 63

histograms of hydraulic conductivity from packer test data are is novel in understanding tunnel seepage, we provide more com-
used to predict seepage volumes. There also exist a number of plete details of the methods below.
numerical models that simulate seepage in the saturated zone.
Many of these models, however, do not include the influence of 2.1. Parflow
the unsaturated zone, climatic forcing, vegetation, overburden,
and fractures, which are all crucial pieces in obtaining reliable ParFlow is a parallel, fully integrated, physical hydrologic model
estimates. that simulates three-dimensional variably saturated subsurface
Less research has focused on tunnels bored in the unsaturated flow and two-dimensional shallow overland flow. A brief descrip-
zone in fractured rock, which are not usually lined and can there- tion of the governing equations is provided below. For more
fore be more susceptible to seepage (Vincenzi et al., 2009). Empir- detailed explanations of ParFlow, refer to Ashby and Falgout
ical and observational methods have been utilized in those tunnels (1996), Jones and Woodward (2001), and Kollet and Maxwell
bored through the unsaturated zone for various fractured forma- (2006).
tions of igneous, metamorphic, and sedimentary rock in various Three-dimensional variably saturated subsurface flow is solved
climates to predict the geological conditions under which seepage using Richards’ (1931) equation:
might occur. This extensive body of research has resulted in some    
general conclusions about precipitation-induced infiltration and dwp dSw ðwp Þ
DSs Sw þ/ ¼ r  q þ qs ð2:1Þ
seepage including: (1) A lag time exists between precipitation dt dt
and seepage that is negatively correlated to the magnitude of the
in which q is given by:
event and positively correlated to the cavity depth (Dobson et al.,
2012; Shimojima et al., 1993; Rademacher et al., 2003). (2) This q ¼ K s ðxÞkr ðwp Þrðwp  zÞ ð2:2Þ
relationship is complicated by climate and wetting front condi- 1
tions, which vary the hydrologic properties of soil and fractured and Ss is the specific storage coefficient [L ], Sw is the degree of sat-
rock with time (Dobson et al., 2012; Rademacher et al., 2003; uration, wp is pressure head [L], t is time [T], / is the porosity [], qs
Shimojima et al., 1993). (3) Differing hydrogeologic properties also is a source sink term [T1], Ks(x) is saturated hydraulic conductivity
affects seepage rates: lower permeability and porosity of matrix, [LT1], kr is relative permeability [] and is a function of pressure
and lower capillarity all increase seepage (Trautz and Wang, head, wp , given by the van Genuchten (1980) relationships, where:
2002; Wang et al., 1999; Javadi et al., 2016). (4) Structural hetero- m 2
geneity of fractures and soils, and their orientation causes variable ð1  ðawp Þn1 ð1 þ ðawp Þn Þ Þ
kr ðwp Þ ¼ n
m ð2:3Þ
flow rates in response to spatially and temporally variable precip- ð1 þ ðawp Þ Þ 2

itation (Dobson et al., 2012; Flint et al., 2001; Gleeson et al., 2009;
Maxwell, 2010; Shimojima et al., 1993; Wang et al., 1999; Zhou in which a [L1] is inversely related to the air-entry pressure of the
et al., 2006) that results in fingering, preferential flow, and varying medium, n ¼ ð1  mÞ1 [] and is related to the distribution of pore
behavior at fracture intersections (Olofsson, 1994; Pruess, 1998; size; these values are determined empirically.
Zhou et al., 2006). (5) Preferential flow causes more steeply dipping Soil moisture is also a function of pressure head, wp , and is cal-
fractures to carry more water than fractures with a shallower dip, culated with:
which serve as connective pathways (Cesano et al., 2000; Gleeson
et al., 2009; Wang et al., 1999) and can impact seepage from as far Ssat  Sres
Sw ðwp Þ ¼ m þ Sres ð2:4Þ
away as one half to three times the cavity depth (Flint et al., 2001; ð1 þ ðawp Þn Þ
Rademacher et al., 2003; Wang et al., 1999).
Within these observations, there is an underlying theme that in which Ssat is relative saturated water content [] and Sres is the
infiltration and subsequent seepage most commonly occur where relative residual saturation.
a fracture network acting as a flow conduit is in connection with Two-dimensional shallow overland flow is integrated with sub-
an overlying source of water, be it a groundwater reservoir in per- surface flow by solving the kinematic wave approximation. This is
meable soil, a horizon of permeable conductive soil, or a location of then input to the overland flow boundary condition while preserv-
direct recharge (Gargini et al., 2008; Gleeson et al., 2009; Moon and ing continuity conditions of pressure and flux at the land-surface
Jeong, 2011; Olofsson, 1994; Rademacher et al., 2003; Scanlon boundary:
 
et al., 2005, 2006; Seyfried et al., 2005; Shimojima et al., 1993; dkws ; 0k
Tomonaga et al., 2017; Vincenzi et al., 2009, 2014). The crucial q ¼ K s ðxÞkr ðwp Þrðhs  zÞ ¼  r  ðv kws ; 0kÞ þ qr ðxÞ
dt
pieces governing these occurrences are land-atmosphere interac-
ð2:5Þ
tions (Maxwell, 2010), the land surface (Atchley and Maxwell,
2011; McCulley et al., 2004; Olofsson, 1994; Scanlon et al., 2002; where wS [L] is the surface ponding depth and is assumed to be wp
Wohling et al., 2011), subsurface heterogeneity (Kollet, 2009; [L] at the saturated ground surface (Kollet and Maxwell, 2006), v is
Maxwell, 2010), and preferential flow in soil and fractures depth-averaged surface water velocity vector [LT1], and qr is a
(Maxwell, 2010; Pruess, 1998; Shimojima et al., 1993). The source/sink term [LT1] (e.g. precipitation).
research presented here estimates how features like land cover, This assumes that:
fracture spacing, soil and geology affect the timing and magnitude
of precipitation-induced seepage to unlined tunnels using an inte- Sf ;i ¼ So;i ð2:6Þ
grated hydrologic model. in which So;i is the bed slope (gravity forcing), and is equal to Sf ;i , the
friction slope, and so Manning’s equation can be used to approxi-
mate flow with a depth-discharge relationship:
2. Conceptual models and methodology
1 2
Sf ;2i w3s
ParFlow, the Common Land Model (CLM), and FRACK were used v¼ ð2:7Þ
g
for modeling this system in an integrated manner. The governing
equations used within each model are listed in the following sec- in which g [LT1/3] is Manning’s coefficient (Kollet and Maxwell,
tions. As the use of integrated hydrologic models (such as ParFlow) 2006).
64 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75

2.2. Common land model (CLM) Probability distribution of the strike and dip of each fracture is
calculated using a fisher distribution as follows:
ParFlow is coupled with the Common Land Model (CLM) (Dai
j sinð/Þej cos ðf Þ
et al., 2003; Kollet and Maxwell, 2008; Maxwell and Miller, f ð/Þ ¼ ð2:8Þ
2005), a land surface model that requires inputs including ej  ej
land-surface characteristics, a multilayer snowpack model with in which / [] is deviation from mean, and j is the dispersion
compaction, vegetation properties, atmospheric forcing, and factor. A smaller value of j implies a wider distribution around
geographic coordinates. CLM is driven by observed meteorology the mean value.
(precipitation, temperature, solar radiation, wind, and humidity) Length is determined with:
and bypasses the need for simplifying assumptions at the upper
a1
boundary of the model by simulating precipitation, evapotranspi- ða  1Þlm
f ðl; lm ; aÞ ¼ a ð2:9Þ
ration, and root flux and passing these fluxes back to ParFlow l
(Maxwell and Miller, 2005). At each time step, mass and energy
in which l P lm , the minimum fracture length, and a is the power
balance error are calculated to ensure mass and energy conserva-
law exponent. When a is between one and three, both large and
tion. A description of the governing equations is provided in
small fractures contribute to the network connectivity.
Kollet and Maxwell (2008).
Hydraulic conductivity of fractures is then solved with:

2.3. Frack qgb2

Rock fractures are generated using FRACK (Reeves et al., 2008),
software that maps discrete fractures to a regularly spaced finite-
difference grid. Using a grid enables the solutions of matrix and
fracture flow when input to ParFlow by assigning contrasting per-
meability values between grid cells containing fractures and those
without. Random fractures are generated in FRACK based on user-
selected estimates of fracture density, height to length ratio, and
probability distributions of strike and dip, length, and aperture.
Anisotropic hydraulic conductivity is then calculated in each grid
cell according to the degree of fracture occupation. A brief descrip-
tion of the governing equations for the generation of a fracture net-
work is provided (Botros et al., 2008).
K¼ ð2:10Þ
12l
in which q is fluid density [ML3], g is gravity, b is hydraulic aper-
ture (lognormal distribution), and l is dynamic viscosity
[ML1 T1].
2.4. Model domain and setup
As the conceptual model being explored for this research is hypo-
thetical and not site-specific, the parameter values used for model-
ing were selected based upon the literature. Their use in the model is
discussed in the following sections. Each simulation varied one
parameter at a time from the base case, defined as thin, barren,

Fig. 1. Two-dimensional schematic of the domains for simulations that were run for which the parameters are discussed in detail. (a) Base case, (b) different soil type, (c)
thicker overburden, (d) vegetated, (e) different bedrock, (f) higher fracture spacing, (g) deeper tunnel. Not to scale.
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75 65

highly conductive soils overlying a densely fractured crystalline
bedrock with a domain size of 40 m by 40 m by 50 m and a tunnel
at 10 m depth. Varying one parameter at a time from the base case
simulation enables separation of each parameter’s control on seep-
age. A schematic of the various cases simulated is shown in Fig. 1.
Modeling inputs for each simulation is shown on Table 1.
The lateral domain extent was increased to 100 m in the x and y
directions for the simulation with a deeper tunnel because the
quantity of water capable of reaching the tunnel is dependent on
tunnel depth and fracture dip. As has been shown by Flint et al.
(2001), Rademacher et al. (2003), and Wang et al. (1999) precipita-
tion as far away laterally as three times the tunnel depth can cause
seepage. As the main fracture orientation for this research is spec-
ified as steeply dipping, the domain extends outward up to two
times the depth to bedrock to ensure boundary effects are isolated
from tunnel seepage estimates. A fine spatial resolution of 0.5 m
and 1 m were used to maintain computational efficiencies, yet still
capturing and representing behavior of heterogeneous subsurface
flow. Slopes at the top of the domain are varied between 0 and 2
degrees to allow for drainage of ponded water.
2.5. Climatic forcing
Many of the observed inflows to tunnels through fractured rock
occur in climates or geographic locations where a deep vadose
zone exists in rock considered relatively impermeable and of good
rock quality designation (RQD). These climates have been docu-
mented as semiarid (Pruess, 1998), and Mediterranean (Gargini
et al., 2008; Rademacher et al., 2003; Vincenzi et al., 2009). Forcing
data including shortwave and long wave radiation, temperature,
pressure, wind speed and direction, precipitation rate, and humid-
ity were obtained from the North American Land Data Assimilation
System (NLDAS) dataset for Denver, CO, a continental semiarid cli-
mate, and Portland, OR, a Mediterranean climate. Hourly meteoro-
logical forcing was used to drive the CLM and provide accurate
representation of diurnal cycles and intense precipitation events.
This forcing allowed investigation of the effects of non-linear
response on magnitude of seepage shown by Maxwell (2010),
and determination of the control exhibited by residual soil mois-
ture that affects the timing and magnitude of seepage. Using an
hourly time series increases simulation times, but maintains tem-
poral resolution. This enables comparison with response times
observed for varying magnitudes of precipitation by Dobson et al.
(2012) and Shimojima et al. (1993).
2.6. Overburden
Soil hydraulic conductivities and van Genuchten parameters of
these soils for model inputs are taken from the values experimen-
tally determined by Schaap and Leij (1998). The dependency on
geologic and climatic processes for the formation of soils also
induces spatial variability of hydraulic conductivity and soil mois-
ture that changes with scale (Green et al., 2009; Seyfried, 1998). At
some scale, spatial data for soil moisture is random and may be
portrayed with mean and variance (Seyfried, 1998). Green et al.
(2009) provide evidence that below 5 km scales, variability is sta-

Table 1
Modeling input parameters.

Parameter Base case Sandy loam 15 m sand Vegetated Sandstone/ 4 m fracture 25 m tunnel Portland
basalt spacing
Domain size (# cells) 640,000 640,000 640,000 640,000 640,000 640,000 500,000 640,000
Soil properties
Porosity 0.337 0.387 0.337 0.337 0.337 0.337 0.337 0.337
Theta r 0.058 0.049 0.058 0.058 0.058 0.058 0.058 0.058
% Sand 92.7 63.4 92.7 92.7 92.7 92.7 92.7 92.7
% Clay 2.9 11.1 2.9 2.9 2.9 2.9 2.9 2.9
Ksat (m/h) 0.426 0.0224 0.426 0.426 0.426 0.426 0.426 0.426
alpha (m-1) 4.79 2.63 4.79 4.79 4.79 4.79 4.79 4.79
n () 3.47 2 3.47 3.47 3.47 3.47 3.47 3.47
lambda x = lamda y (m) 15 15 15 15 15 15 15 15
lamda z 2 2 2 2 2 2 2 2
sigma 1 1 1 1 1 1 1 1
Vegetative cover
IGBP standards 16 16 16 7 16 16 16 16
Rock properties
Porosity 0.001 0.001 0.001 0.001 0.11 0.001 0.001 0.001
Specific storage (m-1) 0.058 0.058 0.058 0.058 0.049 0.058 0.058 0.058
Ksat (m/h) 9.00E-09 9.00E-09 9.00E-09 9.00E-09 1.08E-05 9.00E-09 9.00E-09 9.00E-09
alpha (m-1) 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
n () 3 3 3 3 3 3 3 3
lamda x = lamda y (m) 50 50 50 50 50 50 50 50
lamda z (m) 2 2 2 2 2 2 2 2
sigma 1 1 1 1 1 1 1 1
Primary/secondary fractures
Occurrence probability 0.66/0.34 0.66/0.34 0.66/0.34 0.66/0.34 0.66/0.34 0.66/0.34 0.66/0.34 0.66/0.34
Minimum fracture length (m) 3/2 3/2 3/2 3/2 3/2 3/2 3/2 3/2
Maximum fracture length (m) 10,000/ 10,000/ 10,000/ 10,000/ 10,000/10,000 10,000/10,000 10,000/ 10,000/
10,000 10,000 10,000 10,000 10,000 10,000
A (Truncated Pareto) 1.5/1.5 1.5/1.5 1.5/1.5 1.5/1.5 1.5/1.5 1.5/1.5 1.5/1.5 1.5/1.5
Dip mean (degrees) 120/30 120/30 120/30 120/30 120/30 120/30 120/30 120/30
Dip kappa (fisher distribution) 10/40 10/40 10/40 10/40 10/40 10/40 10/40 10/40
Strike mean (degrees) 135/135 135/135 135/135 135/135 135/135 135/135 135/135 135/135
Strike kappa (fisher distribution) 20/10 20/10 20/10 20/10 20/10 20/10 20/10 20/10
Aperture mean (microns) 2/3 2/3 2/3 2/3 2/3 2/3 2/3 2/3
Aperture standard deviation 3/2 3/2 3/2 3/2 3/2 3/2 3/2 3/2
(lognormal)
Alpha (van Genuchten) (m-1) 2/2 2/2 2/2 2/2 2/2 2/2 2/2 2/2
n (van Genuchten) () 3/3 3/3 3/3 3/3 3/3 3/3 3/3 3/3
66 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75
tistically stationary and that at hillslope scales (<200 m) infiltra-
tion may be higher than expected over larger scales. Seyfried and
Wilcox (1995) claim that this spatial variability at small scales
can be represented with stochastic methods. For these reasons,
the Turning Bands approach (Tompson et al., 1989) is used to
simulate hydraulic conductivity fields. This approach is common
for stochastic hydrogeology (Rubin, 2003) and represents the spa-
tial variation of saturated hydraulic conductivity as a continuous
stationary Gaussian random field in which:
ln K s ðxÞ ¼ F þ f ðxÞ ð2:15Þ
where flnK s ðxÞg ¼ F, ff ðxÞg ¼ 0, ff ðxÞ2 g ¼ r2F ¼ r2lnK , and
K g ¼ expðFÞ. Here (x) implies the mean of x, Kg is the mean hydraulic
conductivity, and r2 is its variance. A correlation structure, which
affects the degree of continuity in hydraulic conductivity, must also
be specified. The correlation structure is given by:
 
n soils.
Reff ðnÞ ¼ r2F exp  ð2:16Þ
k
where n [L] is the distance separating two hydraulic conductivity
values, k is the correlation length [L]. The values used for modeling
are those suggested by Rubin (2003) for alluvial soil where
kx ¼ ky ¼ 15 m, kz ¼ 2 m (Maxwell, 2010). Rubin (2003) suggests
using r2F ¼ 0:81, but for these simulations variance is increased to
1 to account for the lack of topography. Numerical turning bands
parameters were specified with the number of lines equal to 360
used with the grid spacing of 0.5 m. Normalized frequency incre-
ment of 0.2 was used with a Kmax of 100. The principal direction
of heterogeneity follows the slope of the land surface. These values
were also used in the application of turning bands to bedrock. Soil
thicknesses of 2 m and 15 m were applied in the model to account,
Fig. 2. Hydraulic conductivity realization for base case simulation.
respectively, for thin soils overlying fractured rock, and a potential
reservoir in overburden overlying fractured rock. Soil properties are
included in Table 1.
2.7. Vegetation
Vegetation was input according to percent cover and type.
Percent cover was shown by Seyfried and Wilcox (1995) to be at
a maximum of 50 when soils were greater than 1.5 m thick. When
soils are thinner than that, vegetative cover is reduced and
approaches zero. The International Geosphere-Biosphere Program
(IGBP) standards are used within CLM. Barren land (IGBP #16) is
run for all but one of the simulations, as recharge was seen to
amount to as much as 50% of precipitation (Seyfried and Wilcox,
1995). Shrubland (IGBP #7) was used in a different run to investi-
gate the effect of transitioning from barren soils to more vegetated
2.8. Bedrock
As seepage to unlined tunnels has been shown to occur in frac-
tured igneous, fractured metamorphic, and fractured sedimentary
rocks, bedrock was specified accordingly. The values for hydraulic
conductivity used were within the range for those listed in
Domenico and Schwarz (1990) for unfractured igneous and meta-
morphic rock, and low permeability porous rock. Porosity values
were taken from Heath (1983) for granite (corresponding to
igneous or metamorphic rock), and sandstone and basalt (for low
permeability porous rock). Turning bands was again used to
simulate heterogeneity in bedrock, the values for variance and
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75 67

correlation length are the same as those used by Maxwell (2010), similar to those used by Birkholzer et al. (2004), Finsterle et al.
and the other variables are the same as those specified above. (2003), and Kitterod et al. (2000). An example realization of
hydraulic conductivity is included as Fig. 2.

2.9. Fracture generation

Modeling fractured rock is often done in a volume averaging
fashion in an attempt to homogenize significantly heterogeneous
matrix and fracture permeabilities (Pruess, 1998; Pruess et al.,
1999). This can also average infiltration rates applied at the upper
boundary and underestimate wetting front arrival time (Pruess
et al., 1999). The most accurate techniques for modeling fracture
networks are with Discrete Fracture Network (DFN) models,
Stochastic Continuum (SC) models, and a combination of the two
(Liu and Bodvarsson, 2001; Ohman and Niemi, 2003; Huang
et al., 2011).
Here a stochastic continuum approach was taken, similar to
that by Botros et al. (2008), Maxwell (2010), and Reeves et al.
(2008) in which discrete fracture networks are generated and
mapped to a finite difference grid. This creates a saturated hydrau-
lic conductivity field that is heterogeneous and anisotropic, in
which the fracture network is represented as a porous medium
with hydraulic conductivity and other physical properties deter-
mined by random functions that are variable in space (Parashar
and Reeves, 2008). Turning bands was used for the mean hydraulic
conductivity of the rock mass and also applied to the fracture net-
work, to represent the different properties of fractures such as
asperity contacts, variable aperture, and roughness of fracture
walls (Pruess et al., 1999). Again, in all but one of the simulations
a fracture spacing of 0.8 m was specified, the other had a spacing of
4 m to see how this affected seepage. These values were chosen to
be consistent with engineering classifications of fracture spacing.
The parameters for fractures in the stochastic continuum model
are varied according to those used by Maxwell (2010) and are very
2.10. Tunnel
The tunnel dimensions were specified with a 6 m  6 m cross-
sectional area spanning the length of the domain. For consistency
with the circular or parabolic shape that tunnels are usually con-
structed with, porosity was specified as 0.785. The one-meter sur-
rounding the tunnel in all directions had a permeability reduction
that reduced the value to 65% of initial. This was to be consistent
with results found by Fernandez and Moon (2010a,b) for perme-
ability reduction due to a change in stress after boring in stiff rock.
Two different tunnel depths were investigated in modeling, one at
10 m and the other at 25 m.
2.11. Simulations
All simulations were run on the Colorado School of Mines
supercomputer Ra. Each of the simulations was distributed across
32 or 64 processors. Depending on the computational difficulty of
the simulations, solve times varied between two and seven days.
3. Results
The total volumetric seepage per meter into tunnels for all
seven simulations run with semiarid forcing is shown below on
Fig. 3 and the volumetric seepage per hour for those same simula-
tions is shown on Fig. 4. Volumetric seepage per hour for the
Mediterranean simulations is not shown due to difficulty solving.
Note that the noise on all subsequent plots is also due to difficulty
solving. Lag time to seepage can be clearly seen on the plots below,

Total Seepage into Tunnel in Response to Precipitation

8000 0

Precipitation 5

Base Case
6000
4 m Fracture Density
Low Permeability Porous
10
Vegetated
Total Seepage (L/m)

Precipitation (mm)

15 m Sand
25 m Tunnel Depth
4000 Sandy Loam 15

20

2000

25

0 30
1-Oct 31-Dec 31-Mar 30-Jun 29-Sep 29-Dec 30-Mar 29-Jun 28-Sep
Date

Fig. 3. Precipitation and total seepage into tunnel for the different runs forced with a semiarid climate.
68 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75

Seepage into Tunnel per Hour per Meter in Response to Precipitation

10 0

Precipitation
Base Case
4 m Fracture Density
1 10
Low Permeability Porous

Precipitation (mm)
Seepage (L/hr/m)

Vegetated
15 m Sand
25 m Tunnel Depth 15
Sandy Loam

0.1 20

25

0.01 30
1-Oct 31-Dec 31-Mar 30-Jun 29-Sep 29-Dec 30-Mar 29-Jun 28-Sep
Date

Fig. 4. Volumetric seepage into tunnel per hour per meter for simulations using semiarid forcing. Note this is a semi-log plot.

Seepage into Tunnels Beneath Sand and Sandy Loam

1.2 12

Base Case (Sand)

1 10
Sandy Loam
Seepage Into Tunnel (L/hr/m)

Precipitation
0.8 8
Precipitation (mm)

0.6 6

0.4 4

0.2 2

0 0
1-Dec 30-Jan 31-Mar 30-May
Date

Fig. 5. Seepage into tunnels beneath sand and sandy loam.

particularly where measurable seepage begins for each curve on conductive soils overlying densely fractured bedrock. This is then
Fig. 4. The difference in lag times between models is useful for gen- closely followed by the vegetated case. Interestingly, the next arri-
eral comparison and a quantitative analysis. The lack of spin up for val is at the deeper tunnel, exhibiting a lag time slightly shorter
the soil moisture, and the domain being mostly dry at the outset, is than a tunnel bored through lower fracture density. Seepage is
evident in that the time to lag is exaggerated. The first arrival, as next observed in the tunnel bored through a more porous and
predicted, occurs in the ‘base case’ tunnel with barren, thin, highly permeable medium. Changing from sand to sandy loam causes a
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75 69

Percent of Bedrock Storage Within Tunnel for 2 m Sand and 15 m Sand

25 30

Percent of Bedrock Storage Within Tunnel (%)

25
20
2m Sand
15 m Sand 20

Precipitation (mm)
15 Precipitation

15

10
10

5
5

0 0
1-Oct 9-Jan 18-Apr 27-Jul 4-Nov 12-Feb 23-May 31-Aug
Date

Fig. 6. Percent of total bedrock storage within tunnel with varying overburden depths.

Seepage into Tunnels Beneath Barren and Vegetated Soil

1.4 18

16
1.2
Barren
14
Shrub
1
Precipitation 12

Precipitation (mm)
Seepage (L/hr/m)

0.8 10

0.6 8

6
0.4
4

0.2
2

0 0
6-Mar 26-Mar 15-Apr 5-May 25-May 14-Jun 4-Jul 24-Jul 13-Aug
Date

Fig. 7. Seepage into tunnels beneath barren and vegetated soil.

significant increase in wetting front arrival time. The last appear- The Mediterranean simulations did not run to completion due
ance of tunnel moisture occurs where there is a much thicker to a great difference in moisture input to the domain and are dis-
overburden. cussed, but not included as figures. Seepage volumes observed dur-
Figs. 3 and 4 also provide insight into the magnitude of seepage ing the initial part of the simulations were an order of magnitude
and behavior that occurs between precipitation events. The last greater than those seen in the semi-arid runs.
two months, after the largest precipitation event, were excluded
from the plot for the base case and deeper tunnel because the no
flow boundary condition at the base of the domain caused the tun- 4. Discussion
nel to fill from beneath and distorted the results. Even without this
data, however, some clear distinctions can be seen between simu- Magnitude of precipitation events clearly exerts control on the
lations. Sustained seepage and peak seepage are two serious con- timing and magnitude of seepage to tunnels in a nonlinear fashion.
cerns here and the differences between these are discussed in There is a complex response exhibited that is controlled by climate,
more detail over the next few sections. temporally variable precipitation, and the subsequent wetting
70 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75

Seepage into Tunnel in Response to Precipitation

1.4 18

16
1.2 Igneous/Metamorphic

Low Permeability Porous 14

1
Precipitation
12

Precipitation (mm)
Seepage (L/hr/m)

0.8 10

0.6 8

6
0.4
4

0.2
2

0 0
5-Feb 15-May 23-Aug 1-Dec 11-Mar 19-Jun
Date

Fig. 8. Seepage into tunnels bored through igneous/metamorphic and low-permeability porous rock.

Seepage into Tunnels in Response to Precipitation

1.4 18

0.8 m Fracture Density 16

1.2
4 m Fracture Density
14
Precipitation
1
12

Precipitation (mm)
Seepage (L/hr/m)

0.8 10

0.6 8

6
0.4
4

0.2
2

0 0
16-Mar 15-May 14-Jul 12-Sep 11-Nov 10-Jan 11-Mar 10-May
Date

Fig. 9. Seepage into tunnels bored through different fracture spacing.

front conditions. A clear seasonality is exhibited in a semiarid cli-
mate over this cycle, in which the massive spring and summer pre-
cipitation events are causing peak seepage into the tunnel. These
large precipitation events are surrounded by other moderate
events that appear to be increasing the soil moisture content,
increasing the hydraulic conductivity of soils and fractures as they
approach saturated conditions, and thus reducing the lag time to
seepage. This behavior is also observed in a Mediterranean climate
during the fall and winter months over which the simulations ran.
These findings are consistent with those observed by Dobson et al.
(2012), Gleeson et al. (2009), Rademacher et al. (2003), and
Shimojima et al. (1993) and modeled Maxwell (2010), above-
average precipitation contributes exponentially more to recharge
than small amounts of precipitation.
There also appears to be a precipitation threshold, beneath
which background seepage does not change and above which lag
time to seepage decreases in a power-law fashion. The magnitude
of seepage is also dependent upon the precipitation threshold,
beneath which background seepage stays relatively constant and
above which seepage increases with precipitation in a non-linear
fashion.
Soil type exerts significant control on seepage. Lag time for a
sandy loam is significantly higher than for sand while the volume
of seepage is much smaller. Later times in Fig. 3 show the first sig-
nificant arrival of seepage through the sandy loam. This is likely
due to the lack of moisture in the sandy loam significantly reducing
the hydraulic conductivity for the first year of simulation. After this
first year, the soil becomes wet enough to permit the flow of water
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75
at a higher rate. Coarse-grained soils are providing more recharge
than finer grained soils. Not only does the fine-grained content of
soil decrease the rate of seepage, it also affects the volume. The
findings explain those observed by McCulley et al. (2004),
Olofsson (1994), Scanlon et al. (2002), and Wohling et al. (2011)
and are shown on Fig. 5. They are in part due to the reduction of
hydraulic conductivity keeping water near the surface and avail-
able for evapotranspiration, resulting in situations where antece-
dent conditions increase the time to seepage arrival for a larger
precipitation event than a smaller event similar to the findings of
Atchley and Maxwell (2011).
Soil thickness also exerts control over the timing of seepage but
less on the magnitude as shown on Fig. 3. This can be seen by com-
paring the total storage in soil and the percent of that which arrives
in bedrock that is contained within the tunnel (Fig. 6). Given the
response seen after the first year of simulation, it is expected that
the percent of bedrock storage within the tunnel for both soil types
would become similar. This is due to a reservoir developing in the
15 m of sand. The lag to seepage in this scenario is significantly
greater, the first appearance of seepage to the tunnel beneath
15 m of overburden being approximately 1 year after that with
2 m of overburden. This is most likely due to the domain being
dry when simulations start.
The impact of vegetation on seepage is not nearly as pronounced
as expected. When changing from barren soils to shrubs, seepage is
seen to decrease slightly as seen on Fig. 7. This is most likely due to
rapid drainage of sands and similar assigned properties such as root-
ing depth in the land-surface model CLM. This decrease is confirms
that reported by Sandvig and Phillips (2006), Scanlon et al. (2006),
and Seyfried et al. (2005) but is not nearly as apparent. The change
in storage of bedrock (net infiltration) however, is approximately
50% of precipitation for the barren simulation and is consistent with
that reported by Sandvig and Phillips (2006). This is also consistent
with that reported by Atchley and Maxwell (2011) in which soil
moisture after precipitation is initially controlled by vertical
hydraulic conductivity and until later times, during drying periods,
experiences little influence from vegetation.
The impact of vegetation can also be investigated through the
amount of water transpired. Unexpected behavior is observed in
50
45
Subsurface Storage Within Tunnel (%)
40
35
30
25
20
15
10
0 0
5-Feb
Fig. 10. Percent of subsurface storage within tunnel bored through different fracture spacing.
71
which the amount of transpiration is actually greater for the barren
domain than for that vegetated with shrubs. This implies that the
reduction in seepage magnitude is most likely due to interception.
Taking a closer look at the magnitude of precipitation and the
corresponding seepage for the vegetated case a precipitation
threshold of approximately 10 mm/48 h is observed below which
no increase in seepage is observed. As expected, the magnitude
of peak seepage increases with increasing precipitation and peaks
at approximately 1.225 L/h/m seven days after the 47 mm event.
Tunneling through fractured igneous or metamorphic rock as
opposed to fractured sandstone or basalt significantly increases
the magnitude of seepage (Fig. 8). It appears that seepage in the
low permeability porous simulation is occurring in a more volume
averaged fashion and this is probably due to exchange of water
between fractures and matrix of higher conductivity than in the
igneous/metamorphic simulation. Both simulations show a similar
background seepage of approximately 0.1 L/h/m, but the base case
exhibits much more extreme variations in seepage in response to
precipitation. This confirms the observations made by Trautz and
Wang (2002) and Wang et al. (1999) that lower permeability,
porosity, and capillarity of matrix all increase seepage to tunnels.
It can be seen that storage in the tunnel for the base case simula-
tion amounts to 22% of net infiltration, while the low permeability
porous simulation has only 5%.
It is clear that seepage into a tunnel bored through low perme-
ability porous rock takes longer to arrive for all precipitation
events and the magnitude is much less compared to the base case.
The precipitation threshold for increased seepage is also larger in
this simulation and no events below 20 mm/48 h are seen to have
any significant impact. Trends between precipitation and seepage
are not present. A fast arrival or seepage increase does not neces-
sarily occur with a larger precipitation event.
The fracture density is seen to have almost a counter-intuitive
result in which the more densely fractured rock results in less
seepage than the simulation with a less dense fracture network
(Fig. 9). This is an interesting finding and can be explained by the
ability of these fractures to transmit water. The more densely frac-
tured network creates a number of possible flow paths around the
tunnel through which the diversion of water can occur. However,
Percent of Subsurface Storage Within Tunnel
25
0.8 m Fracture Density
4 m Fracture Density 20
Precipitation
Precipitation (mm)
15
10
5
15-May 23-Aug 1-Dec 11-Mar 19-Jun
Date
72 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75

Fig. 11. ParFlow domain slicing that shows saturation after a large precipitation event for 0.8 m and 4 m fracture spacing.

Seepage into Tunnels in Response to Precipitation

1.4 18

16
1.2

14
10 m Tunnel Depth
1
25 m Tunnel Depth 12

Precipitation (mm)
Seepage (L/hr.m)

Precipitation
0.8 10

0.6 8

6
0.4
4

0.2
2

0 0
5-Feb 26-Mar 15-May 4-Jul 23-Aug 12-Oct 1-Dec 20-Jan 11-Mar 30-Apr
Date & Time

Fig. 12. Seepage into tunnels bored at different depths.

in the less densely fractured network, when a large fracture hap-
pens to intersect the surface and the tunnel there is less chance
that this will be redirected by other fractures. This finding is con-
sistent with Zhou et al. (2006) who observed a positive correlation
between fracture spacing and infiltration, but a lack of correlation
between normalized seepage and fracture spacing. It is also inter-
esting that the highly fractured network causes more peak seepage
with slightly slower response to precipitation than the less dense
network. This peak seepage can be attributed to a fast arrival of
the wetting front through the well-connected network, which dis-
sipates shortly after its appearance. There also appears to be a sea-
sonal cycle in seepage for both simulations with the higher density
network showing more extreme highs and lows, while the lower
density is near the same order of magnitude.
Fig. 10 illustrates the effects of fracture density on the volume of
seepage. It shows almost 45% of net infiltration to be stored in the
tunnel for the 4 m fracture density. This is quite substantial, given
that the tunnel width is only 15% of the domain width, and its vol-
ume even less. This high storage percentage is likely a result of lat-
eral flow impacting seepage, as observed by Flint et al. (2001),
Rademacher et al. (2003), and Wang et al. (1999). The more highly
fractured network has approximately half the storage, which again
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75 73

Magnitude of Precipitation and Lag to Seepage

100

Lag to Seepage Arrival (days)

10

Base Case
Sandy Loam
1
Vegetated
Low Permeability Porous
4 m Fracture Spacing
25 m Deep Tunnel
0.1
0 5 10 15 20 25 30 35 40 45 50
Precipitation (mm/48hrs)

Fig. 13. Magnitude of precipitation and lag to seepage arrival for all cases except the 15 m overburden. Note the semi-log plot.

Magnitude of Precipitation and Associated Seepage

1.4

1.2
Magnitude of Peak Seepage (L/3hrs)

0.8

0.6

0.4 Base Case

Vegetated
0.2
Low Permeability Porous
4 m Fracture Spacing
0
0 5 10 15 20 25 30 35 40 45 50
Precipitation (mm/48hrs)

Fig. 14. Magnitude of precipitation and associated seepage for the four of seven simulations in which there was an observable increase.

suggests that the fractures are more connected and able to divert
water from the tunnel. These observations highlight the need for
appropriate site exploration and characterization to identify large
extensional fractures, which can have profound effects on tunnel
seepage.
Again, a precipitation threshold exists at approximately
10 mm/48 h, beneath which increased seepage is not seen. It is inter-
esting to note that this seepage is occurring at a much more constant
rate than any of the other simulations. This is because the overbur-
den is becoming more saturated in response to these precipitation
events because the matrix conductivity is so low and there are fewer
high conductivity fractures to flow through. Those fractures, which
are connected to the reservoir forming in the overburden and net-
work, are also becoming saturated and are therefore permitting flow
of water at a rate more similar to the saturated hydraulic conductiv-
ity. Fig. 11 illustrates this observation.
Again the larger precipitation events are arriving at the tunnel
much faster, but there is little correlation between precipitation
and seepage volumes.
The magnitude of seepage into tunnels for the two different
depths modeled is shown on Fig. 12 and follows a different result
than that expected. The deeper tunnel is actually receiving less
seepage than a shallower tunnel. This effect is most likely because
the higher fracture density is diverting more water than would be
expected for one large extensional feature that is more likely to be
encountered with increasing depth and is different than predicted
by Cesano et al. (2000). The timing of this seepage is predictable in
that the lag time is greater for a deeper tunnel. Interestingly, this is
the only simulation in which seepage to the tunnel returns to zero.
A precipitation threshold is once again present, but for this
simulation it appears to be 20 mm/48 h. This could be an explana-
tion for the seepage volume returning to zero during the winter
74 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75
Table 2
Modeling results.
Results summary
Parameter varied: from – to Lag Reason
timing
Sand – sandy loam * Reduced K slows wetting front advancement
2 m – 15 m soil * Thicker overburden for wetting front to move through
Barren – vegetated * Evapotranspiration reduces soil saturation during dry
down and thus reduces K
Igneous/metamorphic – * Higher conductivity of bedrock reduces effect of fractures
sandstone/basalt
0.8 m – 4 m fracture spacing + Fractures remain saturated and therefore have higher K
10 m – 25 m tunnel depth in * Greater distance to travel
bedrock
months. Again, higher magnitude events are arriving faster than
the smaller ones. The magnitude of the event appears to increase
the amount of seepage: this behavior is clear in Fig. 11. This is pos-
sibly due to wetting of the soil and fractures by large early events,
with subsequent events associated with shorter lag time with lar-
ger seepage magnitude due to the increased hydraulic
conductivity.
The total precipitation over two days and lag to seepage arrival
for each simulation is plotted on Fig. 13. On this plot, it is evident
that the fastest appearance of seepage occurs in the case with 4 m
fracture spacing due to the saturation of fractures staying relatively
high and the formation of a reservoir. The base case comes next
and is closely followed by the vegetated simulation. The 25 m deep
tunnel and low permeability porous rock have larger and similar
lag times to seepage, but the sandy loam is significantly higher.
None of the precipitation events is seen to arrive in less that
12 h: some take longer than ten days to be observed.
Fig. 14 shows the precipitation and associated peak seepage for
the four simulations in which it could be separated. The low per-
meability porous simulation has significantly lower peak seepage
than the 4 m fracture spacing. The two highest values for peak
seepage are observed in the base case and vegetated simulations.
The effect of each parameter on the timing and magnitude of
precipitation arrival is shown in Table 2. This table summarizes
common features which are important for tunnel design; mainly
the arrival time of sustained or varying seepage magnitudes. The
importance of sustained versus fluctuating seepage and the timing
of its arrival is dependent on the tunnel purpose.
5. Conclusions
This research uses a fully integrated model with heterogeneous,
anisotropic permeability fields forced by real meteorological data
to simulate precipitation-induced seepage into a tunnel for eight
different scenarios. These eight scenarios are hypothetical and rep-
resent a tunnel bored through a realistic bedrock geological setting.
The hypothetical domain was constructed at fine spatial resolution
with detailed fracture distributions and subsurface heterogeneity.
The magnitude of precipitation events that triggered seepage to
tunnels, the timing at which that seepage occurred, and the mag-
nitude of seepage were analyzed within the simulation results.
This study shows that timing and magnitude of seepage into
tunnels is highly dependent on climatic, ecologic, geologic, and
hydrogeologic variables. A clear non-linearity is shown in the rela-
tionship between the magnitude of precipitation and seepage.
However, this relationship is subject to the timing of the event
and the antecedent conditions, and it exhibits threshold-like
behavior. A precipitation threshold below which the background
seepage is consistent or decreasing is shown to exist across a num-
Peak Reason
magnitude
+ Reduced K keeps water available at surface for
evapotranspiration
+ More soil storage
+ Interception of vegetation
+ Higher storage and conductivity of bedrock
+ Reservoir forms in overburden which keeps
seepage more constant
+ Increased diversion away from tunnel with
depth
ber of simulations. Above this precipitation threshold, seepage
arrival times tend to scale exponentially with event size. Seepage
magnitude is found to be highly dependent on the hydrogeologic
geologic setting. Table 2 shows that while some scenarios had a
steady rate of seepage that is less influenced by temporally vari-
able precipitation, others show fluctuations between storms. The
difference between this steady rate seepage and extreme fluctua-
tion is very important to consider in tunnel design.
This study provides insight for those tunneling through frac-
tured rock as to the conditions under which seepage is most likely
to occur and the range of volumetric extremes that can be
expected. It also highlights the need to characterize the regional
geology in an appropriate manner if one is to obtain reliable esti-
mates when conducting a preliminary investigation. The potential
for using tunnels for water resources in these environments,
although not explicitly stated, is also a possible outcome.
Coupled, fully integrated hydrologic models are powerful tools
that can be used in conjunction with available data to estimate
hydrologic response. Future work should focus on seepage
responses to precipitation in other climates, with other combina-
tions of the parameters varied for this research, in tunnels bored
beneath the water table through fractured media, and using Monte
Carlo simulation on a real site to determine the capability of these
methods to produce accurate estimates and the associated
uncertainty.
Acknowledgements
Funding for this study was provided by the Center for Under-
ground Construction and Tunneling, Colorado School of Mines,
for which the authors are grateful. The authors are also grateful
for the anonymous reviewers for their constructive and thoughtful
comments.
References
Ashby, S., Falgout, R., 1996. A parallel multigrid preconditioned conjugate gradient
algorithm for groundwater flow simulations. Nucl. Sci. Eng. 124 (1), 145–159.
Atchley, A., Maxwell, R., 2011. Influences of subsurface heterogeneity and
vegetation cover on soil moisture, surface temperature, and
evapotranspiration at hillslope scales. Hydrogeol. J. 19, 289–305.
Bagtzoglou, A., Cesano, D., 2007. Dripping into unsaturated rock underground
excavations - literature review and geologic and hydrogeologic setting
description. Environ. Geol. 51, 1285–1294.
Birkholzer, J., Mukhopadhyay, S., Tsang, Y., 2004. Modeling seepage into heated
waste emplacement tunnels in unsaturated fractured rock. Vadose Zone J., 819–
836
Botros, F., Hassan, A., Reeves, D., Pohll, G., 2008. On mapping fracture networks onto
continuum. Water Resour. Res. 44.
Cesano, D., Olofsson, B., Bagtzoglou, A., 2000. Parameters regulating groundwater
inflows into hard rock tunnels – a statistical study of the bolmen tunnel in
southern Sweden. Tunnelling Underground Space Technol., 153–165
 M.G. Sweetenham et al. / Tunnelling and Underground Space Technology 65 (2017) 62–75
Chen, R., Tolon, F., 2012. Fracture cluster modeling for groundwater inflow
prediction into rock tunnels using geostatistics. Geocongress, 2372–2381.
Dai, Y., Zeng, X., Dickinson, R., Baker, I., Bonan, G., Bosilovich, M., 2003. The common
land model. Am. Meteorol. Soc., 1013–1023
Dobson, P., Ghezzehei, T., Cook, P., Rodriguez-Pineda, J., Villalba, L., De la Garza, R.,
2012. Heterogeneous seepage at the Nopal I natural analogue site, Chihuahua,
Mexico. Hydrogeol. J. 20, 155–166.
Domenico, P., Schwarz, F., 1990. Physical and Chemical Hydrogeology. John Wiley &
Sons, New York.
Fernandez, G., Moon, J., 2010a. Excavation-induced hydraulic conductivity
reduction around a tunnel - Part 1: Guideline for estimate of a ground water
inflow rate. Tunnelling Underground Space Technol. 25, 560–566.
Fernandez, G., Moon, J., 2010b. Excavation-induced hydraulic conductivity
reduction around a tunnel – Part 2: Verification of proposed method using
numerical modeling. Tunnelling Underground Space Technol. 25, 567–574.
Finsterle, S., Ahlers, C., Trautz, R., Cook, P., 2003. Inverse and predictive modeling of
seepage into underground openings. J. Contam. Hydrol., 89–109
Flint, A., Flint, L., Bodvarsson, G., Kwicklis, E., Fabryka-Martin, J., 2001. Evolution of
the conceptual model of unsaturated zone hydrology at Yucca Mountain,
Nevada. J. Hydrol. 247, 1–30.
Gargini, A., Vincenzi, V., Piccinini, L., Zuppi, G., Canuti, P., 2008. Groundwater flow
systems in turbidites of the Northern Apennines (Italy): natural discharge and
high speed railway tunnel drainage. Hydrogeol. J., 1577–1599
Green, T., Dunn, G., Erskine, R., Salas, J., Ahuja, L., 2009. Fractal analyses of steady
infiltration and terrain on an undulating agricultural field. Vadose Zone J. 8,
310–320.
Gleeson, T., Novakowski, K., Kyser, T., 2009. Extremely rapid and localized recharge
to a fractured rock aquifer. J. Hydrol. 376, 496–509.
Heath, R., 1983. Basic Ground-Water Hydrology. United States Geological Survey, U.
S.G.S. Reston, VA.
Heuer, R., 1995. Estimating rock tunnel water inflow. In: RETC Proceedings. I.M.
Random House Trade.
Heuer, R., 2005. Estimating rock tunnel water inflow-II. In: 2005 RETC Proceedings.
Society of Mining, Metallurgy, and Exploration Inc, Seattle.
Heuer, R., 2012. Predicting Groundwater Inflows. Tunneling Short Course. Center for
Underground Construction and Tunneling, Golden.
Huang, Y., Yu, Z., Zhou, Z., 2011. Simulating groundwater inflow in the underground
tunnel with a coupled fracture-matrix model. J. Hydrologic Eng.
Javadi, M., Sharifzadeh, M., Shahriar, K., 2016. Uncertainty analysis of groundwater
inflow into underground excavations by stochastic discontinuum method: Case
study of Siah Bisheh pumped storage project, Iran. Tunnelling Underground
Space Technol. 51, 424–438.
Jones, J., Woodward, C., 2001. Newton-Krylov-multigrid solvers for large-scale,
highly heterogeneous, variably saturated flow problems. Adv. Water Resour. 24,
763–774.
Kim, S., Hyun, Y., Lee, K., 2005. Time series modeling for evaluation of groundwater
discharge rates into an urban subway system. Geosci. J., 15–22
Kitterod, N., Colleuille, H., Wong, W., Pedersen, T., 2000. Simulation of groundwater
discharge into a tunnel in fractured rock and numerical analysis of leakage
remediation, Romeriksporten tunnel, Norway. Hydrogeol. J., 480–493
Kollet, S., Maxwell, R., 2008. Capturing the influence of groundwater dynamics on
land surface processes using an integrated, distributed watershed model. Water
Resour. Res. 44, 18.
Kollet, S., Maxwell, R., 2006. Integrated surface-groundwater flow modeling: a free-
surface overland flow boundary condition in a parallel groundwater flow
model. Adv. Water Resour. 29, 945–958.
Kollet, S.J., 2009. Influence of soil heterogeneity on evapotranspiration under
shallow water table conditions: transient, stochastic simulations. Environ. Res.
Lett. 4, 9.
Li, D., Li, X., Li, C., Huang, B., Gong, F., Zhang, W., 2009. Case studies of groundwater
flow into tunnels and an innovative water-gathering system for water drainage.
Tunnelling Underground Space Technol., 260–268
Liu, H., Bodvarsson, G., 2001. Constitutive relations for unsaturated flow in a
fracture network. J. Hydrol. 252, 116–125.
Maxwell, R., 2010. Infiltration in arid environments: spatial patterns between
subsurface heterogeneity and water-energy balances. Vadose Zone J. 9, 970–
983.
Maxwell, R., Miller, N., 2005. Development of a coupled land surface and
groundwater model. J. Hydrometeorol. 6, 233–247.
McCulley, R., Jobbagy, E., Pockman, W., Jackson, R., 2004. Nutrient uptake as a
contributing explanation for deep rooting in arid and semi-arid ecosystems.
Oecologia 141, 620–628.
Moon, J., Jeong, S., 2011. Effect of highly pervious geological features on ground-
water flow into a tunnel. Eng. Geol. 117, 207–216.
Ohman, J., Niemi, A., 2003. Upscaling of fracture hydraulics by means of an oriented
correlated stochastic continuum model. Water Resour. Res. 39 (10), 11.
75
Olofsson, B., 1994. Flow of groundwater from soil to crystalline rock. Appl.
Hydrogeol., 71–83
Parashar, R., Reeves, D.M., 2008. Computation of flow and transport in fracture
networks on a continuum grid. Modflow and More Conference Proceedings/
Abstracts.
Perrochet, P., Dematteis, A., 2007. Modeling transient discharge into a tunnel drilled
in a heterogeneous formation. Groundwater, 786–790.
Philip, J., Knight, J., 1989. Unsaturated seepage and subterranean holes: conspectus,
and exclusion problem for circular cylindrical cavities. Water Resour. Res. 25,
16–28.
Pruess, K., 1998. On water seepage and fast preferential flow in heterogeneous,
unsaturated rock fractures. J. Contam. Hydrol. 30, 333–362.
Pruess, K., Faybishenko, B., Bodvarsson, G., 1999. Alternative concepts and
approaches for modeling flow and transport in thick unsaturated zones of
fractured rocks. J. Contam. Hydrol. 38, 281–322.
Rademacher, L., Clark, J., Boles, J., 2003. Groundwater residence times and flow
paths in fractured rock determined using environmental tracers in the Mission
Tunnel; Santa Barbara County, California, USA. Environ. Geol. 43, 557–567.
Reeves, D., Benson, D., Meerschaert, M., 2008. Influence of fracture statistics on
advective transport and implications for geologic repositories. Water Resour.
Res. 44 (W08405), 1–13.
Reeves, D.M., Zhang, Y., Pohll, G., Benson, D., 2008. FRACK: a freeware flow and
transport suite for fractured media. Modflow and More Conference
Proceedings/Abstracts.
Richards, L., 1931. Capillary conduction of liquids through porous mediums. Physics
1, 318–333.
Rubin, Y., 2003. Applied Stochastic Hydrogeology. Oxford University Press, New
York, New York.
Sandvig, R., Phillips, F., 2006. Ecohydrological controls on soil moisture fluxes in arid
to semiarid vadose zones. Water Resour. Res. 42, 1–20.
Scanlon, B., Healy, R., Cook, P., 2002. Choosing appropriate techniques for
quantifying groundwater recharge. Hydrogeol. J. 10, 2002.
Scanlon, B., Keese, K., Flint, A., Flint, L., Gaye, C., Edmunds, W., 2006. Global synthesis
of groundwater recharge in semiarid and arid regions. Hydrol. Process. 20,
3335–3370.
Scanlon, B., Levitt, D., Reedy, R., Keese, K., Sully, M., 2005. Ecological controls on
water-cycle response to climate variability in deserts. PNAS 102, 6033–6038.
Schaap, M., Leij, F., 1998. Database-related accuracy and uncertainty of pedotransfer
functions. Soil Sci. 163, 765–779.
Seyfried, M., 1998. Spatial variability constraints to modeling soil water at different
scales. Geoderma 85, 231–254.
Seyfried, M., Wilcox, B., 1995. Scale and the nature of spatial variability: field examples
having implications for hydrologic modeling. Water Resour. Res. 31, 173–184.
Seyfried, M., Schwinning, S., Walvoord, M., Pockman, W., Newman, B., Jackson, R.,
2005. Ecohydrological control of deep drainage in arid and semiarid regions.
Ecology 86, 277–297.
Shimojima, E., Tanaka, T., Yoshioka, R., Hoso, Y., 1993. Seepage into a mountain
tunnel and rain infiltration. J. Hydrol. 147, 121–151.
Tomonaga, Y., Marzocchi, R., Pera, S., Pfeifer, H.R., Kipfer, R., Decrouy, L.,
Vennemann, T., 2017. Using noble-gas and stable-isotope data to determine
groundwater origin and flow regimes: application to the Ceneri Base Tunnel
(Switzerland). J. Hydrol. 545, 395–409.
Tompson, A., Ababou, R., Gelhar, L., 1989. Implementation of the three-dimensional
turning bands random field generator. Water Resour. Res. 25 (10), 2227–2243.
Trautz, R., Wang, J., 2002. Seepage into an underground opening constructed in
unsaturated fractured rock under evaporative conditions. Water Resour. Res.
38. 6–1 – 6–13.
van Genuchten, M., 1980. A closed-form equation for predicting the hydraulic
conductivity of unsaturated soils. Soil Sci. Soc. Am. 44, 891–898.
Vincenzi, V., Gargini, A., Goldscheider, N., 2009. Using tracer tests and hydrological
observations to evaluate effects of tunnel drainage on groundwater and surface
waters in the Northern Apennines (Italy). Hydrogeol. J., 135–150
Vincenzi, V., Gargini, A., Goldschneider, N., Piccinini, L., 2014. Differential
hydrogeological effects of draining tunnels through the Northern Apennines,
Italy. Rock Mech. Rock Eng. 47, 947–965.
Wang, T.-T., Zhan, S.-S., Chen, C.-H., Su, W.-C., 2017. Characterizing fractures to
mitigate inrush of water into a shaft using hydrogeological approaches.
Tunnelling Underground Space Technol. 61, 205–220.
Wang, J., Trautz, R., Cook, P., Finsterle, S., James, A., Birkholzer, J., 1999. Field tests
and model analyses of seepage into drift. J. Contam. Hydrol. 38, 323–347.
Wohling, D., Leaney, F., Crosbie, R., 2011. Improving confidence in deep drainage
estimates for arid and semi-arid areas using multiple linear regression with
percent clay content and rainfall. Hydrol. Earth Syst. Sci. Discuss. 8, 4535–4557.
Zhou, Z., Salve, R., Liu, H., Wang, J., Hudson, D., 2006. Analysis of a mesoscale
infiltration and water seepage test in unsaturated fractured rock: spatial
variabilities and discrete fracture patterns. J. Contam. Hydrol. 87, 96–122.