Professional Documents
Culture Documents
Technology Research
A Dynamic Modeling Approach to Predict Water Inflow during Tunnel Excavation in
Relatively Uniform Rock Masses
--Manuscript Draft--
Abstract: Karst landforms are widely distributed all over the world. With continuous improvement
of infrastructure construction, tunnelling in mountainous karst areas is inevitable. Due
to the complex hydrogeological conditions in karst areas, groundwater inflow during
tunnelling is usually difficult to predict and can cause huge economic losses and
casualties. Therefore, finding better ways to improve the prediction of tunnel water
inflow is of great significance to ensure the safety of tunnelling. In this study, three
commonly used modules (CHD, DRAIN and CFPM1) in MODFLOW are selected and
compared to predict the water inflow of tunnels. We find that the CFPM1 module
performs better in terms of considering both tunnel size and actual water inflow
dynamics during tunnel excavation, which is a continuous process rather than an
instantaneous process. Secondly, we build a hypothetical case and predict the tunnel
water inflows using above three different modules. We find that the CHD module yields
the largest initial and stable water inflow, followed by the DRAIN module, and the
CFPM1 yields the least. Besides, as the most sensitive factor in the CFPM1 module,
the influence of tunnel diameter on water inflow is discussed. Finally, taking the
Shizishan Tunnel as an actual field example, we compare the simulation results of
different modules with the measured tunnel inflow. In general, the prediction of the
CFPM1 module is the closest to the measured value, and the prediction accuracy is
the highest during the conventional tunnel construction process. However, when water
saturated karst features such as underground rivers, caves or faults are exposed
during tunnelling, the water level may change instantaneously, and the CHD module
usually performs better in terms of predicting the inflow under such conditions. Overall,
this study provides theoretical and numerical support for the safe operation of
tunnelling in karst regions.
Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Manuscript File Click here to view linked References
5 by
6 Zhongxia Li1*, Jing Xiao1, Junwei Wan1, Jianmei Cheng1, Haibo Feng2, Hongbin Zhan3*,
7 Kun Huang1
14 77843-3115, USA.
15 * Corresponding authors:
18
19
20
1
21 Abstract
22 Karst landforms are widely distributed all over the world. With continuous improvement
25 usually difficult to predict and can cause huge economic losses and casualties. Therefore,
26 finding better ways to improve the prediction of tunnel water inflow is of great significance to
27 ensure the safety of tunnelling. In this study, three commonly used modules (CHD, DRAIN
28 and CFPM1) in MODFLOW are selected and compared to predict the water inflow of tunnels.
29 We find that the CFPM1 module performs better in terms of considering both tunnel size and
30 actual water inflow dynamics during tunnel excavation, which is a continuous process rather
31 than an instantaneous process. Secondly, we build a hypothetical case and predict the tunnel
32 water inflows using above three different modules. We find that the CHD module yields the
33 largest initial and stable water inflow, followed by the DRAIN module, and the CFPM1 yields
34 the least. Besides, as the most sensitive factor in the CFPM1 module, the influence of tunnel
35 diameter on water inflow is discussed. Finally, taking the Shizishan Tunnel as an actual field
36 example, we compare the simulation results of different modules with the measured tunnel
37 inflow. In general, the prediction of the CFPM1 module is the closest to the measured value,
38 and the prediction accuracy is the highest during the conventional tunnel construction process.
39 However, when water saturated karst features such as underground rivers, caves or faults are
40 exposed during tunnelling, the water level may change instantaneously, and the CHD module
41 usually performs better in terms of predicting the inflow under such conditions. Overall, this
2
42 study provides theoretical and numerical support for the safe operation of tunnelling in karst
43 regions.
45 1. Introduction
46 Karst landforms can be seen almost everywhere around the global, covering 10% of the
47 Earth's surface (Phillips, 2016). China has one of the broadest distribution of karst landforms
48 in the world, covering an area of 3.46 million square kilometers of carbonate rock formations,
49 accounting for about 15 percent of the global karst distribution area (Cui et al., 2015; He et al.,
50 2010; Li et al., 2021). Karst landforms in China are mostly concentrated in Yunnan, Guizhou,
51 Sichuan, Chongqing and other areas in southwest China (Jin et al., 2013; Sun et al., 2020;
52 Waltham, 2009). With the launch of China's western development initiative, the construction
53 of infrastructure such as transportation and water conservancy in southwest China has been
54 steadily increased in recent decades, and tunnelling has become an important part of
55 construction of infrastructure in mountainous areas (Bu et al., 2017; Lv et al., 2020; Zheng and
56 He, 2021). In recent years, many tunnels have been successfully constructed in southwest
57 China, and inevitably they have to pass through karst formations (Liu et al., 2021b; Zhao and
58 Wang, 2020; Zheng and He, 2021), where karst depressions and gullies are widely distributed
59 and clearly visible on the surface. Besides, underground rivers and karst conduits are also
60 closely intertwined with bedrocks, faults and factures, generating a highly complex and
61 heterogeneous hydrogeological condition, in which water inflow is the most commonly seen
62 and disastrous geological hazards during tunnelling (Abusaada and Sauter, 2013; Bonacci and
3
63 roje-Bonacci, 2000). Many studies (including analytical, numerical, and empirical methods)
64 have been conducted for prediction of water inflow and risk assessment of karst tunnel over
65 the recent decades, and problems associated with groundwater inflow in tunnelling have still
67 Considering the fact that hydrogeological data are almost always lacking and uncertain in
68 karst regions, analytical and empirical methods are widely used for practical purposes (Coli
69 and Pinzani, 2014; Goodman et al., 1964; Polubarinova-Koch, 2015). By considering typical
70 flow patterns and specific boundary conditions, various formulas have been developed to
71 predict water inflow and to evaluate the sensitivity to system parameters (Chisyaki, 1984). For
72 instance, Goodman et al. (1964) first derived a steady-state analytical equation for the
73 prediction of water inflow in a circular tunnel drilled into a homogeneous semi-infinite aquifer.
74 Based on the work of Goodman et al. (1964), many modified models considering more
75 complex conditions have been proposed (Moon and Fernandez, 2010; Ying et al., 2018).
76 Recently, Shi et al. (2017) presented a prediction method of fissure water inflow in a high-risk
77 karst region and this method was successfully verified in engineering application. Li et al.
78 (2010) proposed a tunnel seismic prediction (TSP) method to detect typical geological
79 anomalies, such as karst caves. By adopting the hybrid grey wolf optimization (HGWO)
80 algorithm and the support vector regression (SVR) method, Liu et al. (2021a) proposed a
81 HGWO-SVR tunnel water inflow prediction model with satisfactory results. With the
82 advancement of numerical methods (in terms of computational accuracy and speed), it is now
83 standard to use numerical methods to study the problem of tunnel water inflow as well. Arjnoi
4
84 et al. (2009) used the finite element software ABAQUS to simulate a circular drainage tunnel
85 under the condition of steady groundwater flow. Li et al. (2009) adopted the code FLAC3D to
86 simulate the tunnel water inflow with an innovative water-gathering system. Chiu and Chia
87 (2012) used the MODFLOW-2005 to simulate groundwater flow of the Hsueh-Shan tunnel and
88 corrected the model using a nonlinear estimation technique. Compared with the analytical and
89 empirical methods, the numerical methods are more flexible and can be applied to calculate the
90 water inflow under complex situations, especially when geological structures such as faults and
91 karst caves are involved (Chen et al., 2020; Kaufmann and Romanov, 2020; Zhou et al., 2021).
92 MODFLOW is one of the most commonly used groundwater flow numerical simulators,
93 and it has been widely used to predict the water inflow in tunnels as well. At present, the
94 DRAIN module and the prescribed head boundary condition (or the Dirichlet or constant-head
95 boundary or CHD module) in MODFLOW are the most commonly used methods to
96 characterize a tunnel (Rodriguez et al., 2008; Serrano-Hidalgo et al., 2021; Singh, 2010). The
97 CHD module implies that the central elevation of the tunnel is set as the known hydraulic head,
98 and groundwater in the nearby aquifer with its hydraulic head above the central elevation of
99 the tunnel will be drained towards the tunnel. The DRAIN module is usually used to predict
Q CD H zb , H >zb
(1)
Q 0, H zb
101 where Q is the discharge of groundwater into a drain (tunnel), H is the head of groundwater
102 near the drain, zb is the bottom elevation of the drain, CD is the drainage coefficient, which is
103 related to the length, width and hydraulic conductivity of the drain. Eq. (1) implies that the
5
104 drain only functions when the hydraulic head in the aquifer near the drain is higher than the
105 bottom elevation of the drain. If comparing the CHD and DRAIN modules, one can see that
106 water exchange between the tunnel and the surrounding aquifer in CHD is instantaneous,
107 implying that there is no resistance for draining the groundwater to the tunnel (or the drainage
108 coefficient is infinity). Such an infinite drainage coefficient means that the aquifer will response
109 to the tunnel excavation instantaneously, resulting in a large inflow to the tunnel at the initial
110 stage. In contrast to CHD, the DRAIN module involves a finite drainage coefficient, meaning
111 that drainage of groundwater to the tunnel will has some resistance, and the initial groundwater
112 inflow to the tunnel should be smaller than what will be predicted using CHD.
113 To deal with conduit flow in a porous system, the conduit flow process (CFP) was
114 developed by the U.S. Geological Survey (Shoemaker et al., 2008a) and incorporated in
115 MODFLOW. As conduit flow resembles flow in underground rivers and/or caves in karst
116 regions, the MODFLOW-CFP module has been widely used to simulate the hydrogeological
117 process in karst aquifers, including karst conduit evolution, spring discharge, and tracer tests
118 (for understanding the travel time and interconnected karst flow networks) (Bauer et al., 2003;
119 Birk et al., 2005; Karay and Hajnal, 2015; Liedl et al., 2003). There are three modes in
120 MODFLOW-CFP (Gallegos et al., 2013). The CFP mode 1 (CFPM1) is a hybrid model which
121 couples the conduit flow (in conduit) with the traditional groundwater flow (in porous media).
122 Both laminar and non-laminar flow in conduits are considered in this mode, depending on the
123 Reynold’s number (Teutsch and Sauter, 1998). The CFP mode 2 (CFPM2) simulates non-
124 laminar flow in a karst aquifer by inserting a preferential flow component and uses a critical
6
125 Reynolds number (Rec) to determine the transition between different groundwater flow states.
126 Reimann et al. (2013) improved the source code of MODFLOW-CFP so that it could manually
127 define adjustment parameters such as conduit associated drainable storage width and time
128 dependent input data etc., which further improved the utility of the MODFLOW-CFP module.
129 The CFP mode 3 (CFPM3) is a combination of CFPM1 and CFPM2 with all the above features.
130 The purpose of this study is to evaluate the ability of MODFLOW-CFP in simulating
131 groundwater inflow during tunnelling by selecting CFPM1, which is straightforward to use and
132 meets the need of our purpose. The simulation results of CFPM1 will be compared to the results
133 of two other conventionally used modules CHD and DRAIN in MODFLOW to test its accuracy
134 and applicability. To achieve the objectives, the results of water inflow with above three
135 different modules are predicted and compared. Later, we will use Shizishan Tunnel as a field
136 example to further test the advantages and disadvantages of the above three modules in
137 predicting water inflow. Overall, this research provides much needed theoretical and numerical
141 As the field case that will be addressed later concerns tunnelling in southwest China, we
142 will briefly discuss the features of karst aquifers in southwest China in the following. The karst
143 water-bearing media in southwest China are usually composed of bedrock, fractures, and karst
144 conduits. The hydrodynamic condition of groundwater flow in such karst regions is very
145 complex, with the co-existence of laminar and turbulent flow in karst conduits and the
7
146 alternating circulation of saturated flow and unsaturated flow (Ghasemizadeh et al., 2012; Zhao
147 et al., 2017). As a result, the traditional groundwater flow model primarily developed for
148 studying porous media flow is not suitable for describing the characteristics of karst
149 groundwater flow (Hubinger et al., 2016; Shu et al., 2020; Worthington and Soley, 2017). The
150 DRAIN and CHD modules in MODFLOW assign a series of grid cells to simulate the tunnel,
151 thus the geometric features of those cells should be closely related to the physical features of
152 the tunnel of concern. In particular, the vertical grid space should be similar to the diameter of
153 the circular tunnel, which is usually around tens of meters. This implies that when designing
154 the finite-different numerical mesh in MODFLOW to simulate a tunnel using either DRAIN or
155 CHD, one has to use an extremely fine grid mesh (around tens of meters) to simulate the tunnel
156 in a regional flow model. This is not always feasible to do numerically because of the following
157 reasons. Firstly, a typical tunnel (with a length of tens of kilometers) in the karst region of
158 southwest China usually drains a large area in the scale of tens to hundreds of square kilometers
159 horizontally and in the kilometer scale vertically, thus the grid spaces used in any numerical
160 models of simulating groundwater flow and tunnel inflow usually are in the scale of hundreds
161 of meters in the horizontal direction and tens to hundreds of meters in the vertical direction.
162 For such a large and thick aquifer, it is usually impractical to use an extremely fine numerical
163 cells (similar to the diameter of tunnel or tens of meters) in MODFLOW to characterize the
164 geometry of the tunnel. Secondly, even if it is possible to use extremely fine cells (like a few
165 meters to tens of meters) in MODFLOW, one has to progressively increase the sizes of
166 numerical cells away from the tunnel to suppress the numerical cut-off errors. This implies that
8
167 a large number of numerical cells have to be used eventually. The use of a large number of
168 numerical cells will substantially increase the computational cost thus reduce the performance
169 efficiency of the numerical model, it also requires a much larger number of input parameters
170 values (which are often unknown) for all the numerical cells and sometimes cause non-
172 In addition to the above numerical problems, several physical constrains also exist. For
173 instance, a unique feature of CHD module is that the hydraulic head near the tunnel is almost
174 instantly lowered to the elevation of the central node of the tunnel at the initial stage of
175 tunnelling, and the surrounding groundwater is drained into the tunnel quickly, resulting in a
176 massive initial water inflow. Compared with the CHD module (which implies that an infinitely
177 large drainage coefficient is used), the DRAIN module has a finite drainage coefficient (CD,
178 see Eq. (1)) to control the rate of water exchange, thus the initial and later time stable water
179 inflow rates are less than their counterparts predicted by the CHD module. The CFPM1 module
180 is based on the Darcian porous media flow in the aquifer coupled with laminar/turbulent
181 conduit flow in a discrete conduit network, which is closer to the physical reality of water flow
182 in a karst aquifer with the presence of tunnels. The conduit system is composed of a series of
183 nodes, each node can exchange water with the bedrock aquifer. The flow state of each node
184 can be described by the Kirchhoff's law (Shoemaker et al., 2008a) based on the mass
np
Q
i 1
ip Qss 0 (2)
186 where Qip represents the flow from conduit node point i to conduit node point n, Qss is the sum
9
187 of source sink items of all nodes. The Hagen-Poiseuille equation and Darcy-Weisbach equation
188 are respectively used to describe the laminar and turbulent conduit water flows as below:
d 4 g hc
Qip (3)
128l
hc gd 5
2.51
2
kc hc
Qip log (4)
2l 2 h gd 3 3.71d hc
c
l
189 where d is the diameter of conduit, g is acceleration due to gravity, ∆hc is the head loss of
190 conduits, υ is the kinematic viscosity coefficient, ∆l is the length of the conduit, kc is the average
191 roughness of the conduit, which is related to the shape of conduit wall. In addition, the CFPM1
192 uses a first-order approximation to calculate the flow exchange between the conduits and karst
K
ex = A (6)
b
194 where Qex is the flow exchanged between the conduit and the bedrock aquifer, αex is the
195 exchange coefficient of conduits, hc is the hydraulic head in the conduit, while hm is the
196 hydraulic head of bedrock aquifer near the conduit, b is the distance along the flow direction
197 between the conduit and bedrock, A is the cross-sectional area perpendicular to the flow
198 direction, K is the hydraulic conductivity of the bedrock aquifer adjacent to the conduit. When
199 Qex is negative, it indicates that the water flows from bedrock to karst conduits, and vice versa.
200 Eq. (5) is essentially the same equation used for describing the exchange of flow between
201 different domains in a dual-domain framework (Feehley et al., 2000; Flach et al., 2004).
10
202 2.2 Model construction
203 To compare the numerical simulation results of the three modules (CFPM1, CHD and
204 DRAIN) predicting tunnel water inflow, we have designed a hypothetical small-scale model
206
207 Figure 1. The schematic diagram of the hypothetical model. (b) The tunnel section has been
209 The hypothetical model is divided into two uniform layers, the upper layer (the first layer)
210 is an unconfined aquifer, and the lower layer (the second layer) is a confined aquifer, both of
211 which are described as 9 rows and 9 columns, the size of each grid is 10 m × 10 m × 10 m. The
212 north side of the model is a general head boundary, which serves as the recharge of the aquifer,
11
213 while the other boundaries are zero flux boundaries. The origin of the coordinate system is at
214 the left corner in Figure 1 at the base of the aquifer, the x and y axes are along the horizontal
216 We need to point out that the scale of this hypothetical model is much smaller than the scale
217 of the field case that will be demonstrated later in this study. The reason of using such a small-
218 scale model is to make it possible to utilize all three modules (CFPM1, CHD and DRAIN)
220 numerical model for a large-scale field problem (with tens or hundreds of square kilometers in
221 the horizontal scale and a few hundreds to kilometers in the vertical scale). The focus of such
222 a hypothetical case is to examine and compare the feasibility and credibility of these three
223 modules for assessing tunnel inflow. We have assigned the model parameters according to
224 many engineering examples. For instance, the horizontal hydraulic conductivity (Kx or Ky
225 which is the hydraulic conductivity values in the x or y directions, respectively) is 1.8 m/d,
226 while the vertical hydraulic conductivity (the z direction, Kz) is 0.18 m/d. The specific yield of
227 the unconfined aquifer is 0.08, the storage coefficient of the confined aquifer is 0.0001, and the
228 initial hydraulic head is 20 m. The reason to use a relatively small specific yield (0.08) here is
229 because the effective (or drainable) porosity of the carbonate bedrock aquifer that can
230 contribute to gravitational drainage of water is quite limited. The tunnel is located in the first
231 layer and it is along the y axis. We assumed that the length of the tunnel is 80 m, the diameter
232 is 5 m, and the central elevation of the tunnel is 15 m above the base of the aquifer (see Figure
233 1). The relevant parameters involved in CFPM1 are shown in Table 1.
12
234 Table 1. The parameters of the three different modules.
5 1 0.0001 25
CFPM1 ReU ReL Elevation (m) αex (m2/d)
4000 2000 15 3.53
Elevation(m) Conductance (m2/d)
DRAIN
15 18
Initial hydraulic head (m) Final hydraulic head (m)
CHD
15 15
235 Note: ReU is the upper critical number (laminar to turbulent), ReL is the lower critical number (turbulent
236 to laminar) (different Reynolds numbers correspond to different flow regimes, thus affecting the
237 prediction results of water inflow), R-height is the roughness height of the pipe, the elevation is above
239 To better observe the change of water inflow at the initial stage of simulation, modelling is
240 done with 4 cycles of stress over a sixteen-day period. The first stress period is one day, and
241 the second, third and fourth stress periods are each five days. Each stress period is divided into
242 five-time steps. Subsequently, the CFPM1, CHD and DRAIN modules are used to predict the
243 water inflow respectively. To study the water inflow at the initial stage of excavation, the first-
244 time step of the first stress period is selected as the initial water inflow, which is defined as 0.2
245 days. In addition, the fifth time step of the fourth stress period is defined as the stable water
246 inflow (as flow has achieved the steady-state status at that time), that is, the 16th day.
247 It is assumed that the tunnel is straight in the hypothetical case, so the tunnel tortuosity is
248 1. The water exchange coefficient (αex) is calculated by Eq (6) as 3.53 m2/d. As for other
249 parameters such as Reynolds number and average temperature, the default values listed in Table
250 1 are adopted. The DRAIN module requires two parameters (tunnel elevation and the
13
251 conductance), which can be calculated by Eq (1). In the CHD module, the initial and final
252 hydraulic heads are set to the tunnel central elevation. The relevant parameters of three different
253 models are summarized in Table 1. The simulation results predicted by CHD, DRAIN, CFPM1
255
256 Figure 2. The results of water inflow predicted by CHD, DRAIN, CFPM1 modules in
257 MODFLOW.
258 As can be seen from Figure 2, there are significant differences in the prediction results of
259 different models. The numerical simulation results can be divided into two types. One is the
260 results of CHD and DRAIN modules, which indicate that the initial water inflow is large and
261 then drops sharply with time, and eventually becomes constants; the other is the prediction
262 result of CFPM1 module, which indicates that the initial water inflow is almost equal to the
263 later time stable water inflow. To be specific, the CHD module predicts the largest initial and
264 stable water inflow, which are 452.07 m3/d and 249.58 m3/d, respectively, followed by DRAIN
265 module, which are 313.90 m3/d and 188.67 m3/d, respectively. The CFPM1 module predicts
14
266 the minimum water inflow, of which the initial water inflow is 105.36 m3/d and the stable water
267 inflow is slightly smaller at 99.89 m3/d. We can see that the water inflow changes rapidly within
268 a day of tunnel operation. After 16 days of tunnel excavation, the water inflow has reached a
269 stable state, and the water inflow at this time can be regarded as stable water inflow. To better
270 clarify the difference of water inflow predicted by three different methods, we compare the
271 drawdown planar distribution of the entire model area (9 rows and 9 columns) at the initial
272 time (0.2 day) and the stable time (16 day), as shown in Figure 3.
273
274 Figure 3. Planar contour maps of water levels at different times for three different models.
15
275 The CHD module predicted the maximum drawdown and range (red area), followed by
276 DRAIN module, and CFP module predicted the minimum drawdown.
277 Figure 3 shows the planar distribution of drawdown of the hypothetical model, and three
278 different modules predict different results. We can see that the initial and the later time stable
279 drawdowns of the CHD module both reach 4 m. In contrast, the DRAIN module predicts a
280 smaller result, while the CFPM1 module predicts the smallest drawdown. This further confirms
281 that the CHD module has the greatest impact on the surrounding aquifer, and the predicted later
282 time stable water inflow is also the largest, followed by DRAIN, and CFPM1 yields the
283 smallest later time stable inflow. Compared with the other two modules, the CFPM1 module
284 predicts that both the initial water inflow and the later time stable water inflow are the smallest.
285 When comparing with CHD and DRAIN modules, the CFPM1 module involves parameters
286 such as tunnel diameter, hydraulic conductivity of the bedrock aquifer near the tunnel, tunnel
287 roughness and bending degree of tunnel, etc. A sensitivity study of Gallegos et al. (2013) using
288 CFPM1 found that the tunnel diameter was the most sensitive parameter. Therefore, we will
289 further explore the influence of different tunnel diameters on the initial and stable water inflows.
290 According to the cross-sectional area of tunnels defined by the International Tunnelling
291 Association (ITA), tunnels can be divided into five types: extremely small tunnel (2~3 m2),
292 small tunnel (3~10 m2), medium tunnel (10~50 m2), large tunnel (50~100 m2) and extremely
293 large tunnel (larger than 100 m2). The corresponding tunnel diameters are 1 m, 3 m, 6 m, 9 m
294 and 12 m respectively. The results of water inflow in tunnels with different tunnel diameters
16
296
297 Figure 4. The results of water inflow by CFPM1 modules with different tunnel diameters.
298 As can be seen from Figure 4 that the different tunnel diameters have different simulation
299 results. In general, with the increase of tunnel diameter, both the initial and the late time stable
300 water inflows increase. However, we can clearly observe two different curves in Figure 4.
301 When the tunnel diameter is small (1 m and 3 m), the initial water inflow is smaller than that
302 of the stable water inflow. When the tunnel diameter gradually increases (6 m, 9 m and 12 m),
303 the initial water inflow becomes larger than the stable water inflow, and the inflow-time
304 distribution shows an exponentially downward trend. When comparing the simulation results
305 of tunnels with different diameters of 1 m and 12 m, we find that the water inflows differ by
306 orders of magnitude, suggesting that it is critical to consider the physical size of the tunnel
308 To explore the relationship between the initial water inflow and the late time stable water
309 inflow, we select two representative tunnel diameters (3 m and 6 m) as examples. It can be seen
310 from Figure 4 that the water inflow changes most dramatically within a day. The tunnel cells
17
311 are numbered from 1 to 9 (see Figure 1), and the variation of hydraulic head at different cells
312 over time can be obtained. Besides, the hydraulic head changes of tunnel with diameters of 3
314
315 Figure 5. The hydraulic head at different positions of the tunnel changes with time.
316 Figure 5 shows that when the tunnel diameter is 3 m, the tunnel water inflow is relatively
317 small, and the aquifer recharge is greater than the tunnel discharge, implying that the difference
18
318 of aquifer recharge and tunnel discharge will be used to augment the aquifer storage. As can be
319 seen from Figure 5(a), the hydraulic head is the lowest when the simulation time is 0.2 d.
320 Meanwhile, the hydraulic head of cell No. 5 is 19.05 m. Compared with the prescribed
321 hydraulic head of 15 m at the tunnel whose elevation is set at 15 m (see Table 1), the hydraulic
322 head change at this time (in respect to 15 m) is the smallest, resulting in the smallest water
323 inflow. Then the hydraulic head change gradually increases with time, resulting in the gradual
324 increase of the tunnel water inflow, and finally reaches its asymptote or stable inflow. When
325 the tunnel diameter is 6 m, the tunnel water inflow is larger, and the aquifer recharge is smaller
326 than the tunnel discharge, suggesting that the aquifer storage will be depleted to supplement
327 the difference of tunnel discharge and aquifer recharge. For this case, we can see from Figure
328 5 (b) that the hydraulic head is the largest at 0.2 d. Similarly, the hydraulic head of cell No. 5
329 is 18.38 m. Compared with the fixed hydraulic head of 15 m at the tunnel, the hydraulic head
330 change at this time is the largest, resulting in the largest water inflow. Then the water level and
331 hydraulic head change gradually decreases with time, resulting in an exponential decrease of
332 the tunnel water inflow, which will finally reach its asymptotic value.
333 In summary, the predicted water inflow is different due to the different principles of the
334 three modules, in which CHD module has the largest water inflow, followed by DRAIN, and
335 CFPM1 is the smallest. In addition, the predicted water inflow of CFPM1 module is obviously
336 responsive to tunnel size, the curve shape of predicted water inflow is clearly different with
337 different tunnel sizes. In the following, we will select a field site to carry out the water inflow
338 predictions of three different models respectively and compare them with the actual water
19
339 inflow data to verify the application of different models in practical projects.
342 The study area is located in the Shizishan karst system in Dali, Yunnan Province, southwest
344
20
345 Figure 6. The satellite map of study area location. There are six karst water sub-systems
346 around the tunnel, and the study area is mainly located in the range of the sixth karst water
348 The Shizishan tunnel is 29.4 km in length, with a design buried depth of 200-450 m. Karst
349 is well developed in the study area, which can be divided into six relatively independent karst
350 subsystems (Ⅰto Ⅵ) (Fig. 6). The study area modeled in this investigation (blue area in Fig. 6)
351 mainly passes through sub-system Ⅵ (from DLⅡ 49+201 to DLⅡ 49+919), which has
352 sufficient monitoring data to constrain the basic hydrogeological conditions. The study site
353 covers an area of about 16 km2 and the geographical coordinates are 25°33′33.37″-25°36′25.72″
354 N, 100°30′15.92″-100°31′57.42″ E. The terrain of the study area is high in the middle with the
355 highest elevation of 2790 m above mean sea level (AMSL) and low in the north and south with
356 the lowest elevation of 2154 m AMSL. According to the meteorological data of China National
358 996.6 mm and the average annual temperature is 15.1℃. The study site has distinct dry and
359 rainy seasons and the annual precipitation mainly occurs from May to October.
361 The study site is located in the southeastern part of the Qinghai-Tibet fault block, which is
362 also called the Chuan-Dian rhombic block. The Shizishan tunnel is located in the west wing of
363 Yangbaocun syncline. Due to the influence of stress extrusion, the strike of rock formation is
364 irregular, but the overall direction is along the southeast-northwest direction. The strata are
365 dominated by carbonate rocks with interspersed intrusive rocks. Granite and diabase are
21
366 exposed in the west and south of the study site, sand and shale are exposed in the east. The
367 main lithology in the study area is dolomite and dolomitic limestone. The surface of the aquifer
368 is mainly in the form of stone sprouts, dissolved gully, dissolved trough, and local karst caves.
369 According to the data of two boreholes (Z5 and Z6) whose locations can be seen in Figure 7,
370 we find that the deep karst is mainly dominated by moderately developed fissures.
371
372 Figure 7. Hydrogeology and geology overview of study area.
22
373 The aquifers in the study area are mainly divided into two types: a shallow fractured aquifer
374 type (with intrusive rocks) and a deep karst aquifer type (with the carbonate rocks). Based on
375 the geological survey and actual drilling data, the depth of the shallow fractured aquifer is about
376 40 m-60 m, with an average of 46 m, and the depth of the deep karst aquifer is mostly in 50 m-
377 100 m (with a small portion in 100 m-300 m). It is generally accepted that the deep fractured
378 aquifer has relatively low hydraulic conductivity with poorly developed fracture network.
379 However, the hydraulic conductivity will be significantly enhanced in regions when karst caves
381 To model groundwater flow in such a complex geological setting, it is often necessary to
382 simplify the geological conditions in the numerical simulations. In this study, the tunnel mainly
383 traverses the carbonate strata, and the simulated domain mainly includes dolomite and
384 dolomitic limestone. Because the weathered fissure may be hydraulically connected with karst
385 through fault zones, intrusive rocks such as granite and diabase are also incorporated in the
386 numerical model. Hydraulic properties that should be considered mainly include hydraulic
387 conductivity, specific yield, and specific storage. Different model parameters are assigned
388 according to the actual working conditions, and those parameters will be further calibrated
389 using dynamic water level data of two boreholes (Z5 and Z6) and spring discharge dynamic
390 data (Q6). A total of 6 boreholes were drilled along the tunnel, among which 1 pumping test
391 and 54 packer permeability tests were carried out in limestone, 16 groups of packer
392 permeability tests were carried out in the intrusive rocks. Based on the results of the pumping
393 test, the hydraulic conductivity of the limestone area is about 0.5 m/d. The packer permeability
23
394 test (or called the Lugeon tests) is used to calculate the development of fracture and its
395 permeability by injecting water into boreholes sealed with impermeable packers with relatively
396 high pore pressure, and the obtained permeability is usually expressed in unit of LU (1LU is
397 equivalent to 1*10-5 cm/s in terms of hydraulic conductivity) (Banks et al., 1992). The packer
398 permeability test at the study site showed that the maximum permeability value was 100 LU,
399 the minimum permeability value was 0.09 LU, and the average permeability value was 3.34
400 LU, exhibiting significant variation of permeability. The drastic change of permeability
401 obtained from the packer permeability test is not surprising, as this type of test usually yields
402 the local-scale permeability over a small domain sealed by the packers, and the significant
403 variation of permeability of different packer tests simply suggests that the permeability varies
404 greatly spatially, reflecting the strong heterogeneity that is typically seen in fractured carbonate
405 in the study site. The strong heterogeneity reflected in the packer tests is consistent with the
406 results of Xianglushan tunnel with similar hydrogeological conditions (Liu et al., 2021b).
407 In contrast to the packer permeability test, the pumping test usually influences a much
408 larger volume of aquifer, thus will yield the averaged permeability over the larger influenced
409 volume. Such an averaged permeability is more representative of the regional groundwater
410 flow, thus is employed in the following numerical modeling exercises. As for the intrusive rock
411 area, the maximum value of permeability obtained from the pumping test is 2.99 LU, the
412 minimum value of permeability is 2.05 LU, and the average value of permeability is 2.30 LU.
413 It is worth mentioning that the specific yield is only used in the unsteady simulation of
414 groundwater. In summary, the hydrogeological conditions of the study area are summarized in
24
415 Figure 7. The groundwater is replenished by atmospheric precipitation through the exposed
416 carbonate area in the north and west of the study area, and the groundwater flows from
417 northeast to southwest and from northwest to southwest and is finally discharged in the form
420 The Shizishan tunnel is along the northwest-southeast direction, and the elevation of the
421 tunnel is 1973 m AMSL, which is about 170 m below the groundwater discharge point, making
422 it prone for groundwater inflow (because it is sited deeply in the saturated groundwater flow
423 region). Therefore, it is of great significance if one can accurately predict the histogram of
424 tunnel water inflow as such information is critical for guiding tunnel construction and ensuring
425 safety of personnel during the tunnelling process. The profile along the tunnel is shown in
426 Figure 8.
25
427
428 Figure 8. The cross-sectional view of the study area, (a) profile of the study area, (b)
430 The purpose of this section is to quantitatively predict the water inflow through the
431 comparison of different numerical simulation models, and to compare the numerically
432 simulated results with field observed results for assessing the credibility of different numerical
433 models. The construction of the numerical model is divided into four steps. Firstly, the MAP
434 module in GMS (Groundwater Modeling System) software is used to create a hydrogeological
435 conceptual model. Secondly, the conceptual model is transformed into a numerical model by
437 conditions, setting stress period and time step. Thirdly, the numerical model is calibrated
438 according to the borehole water levels and flow rates. Finally, the model established in GMS
26
439 is translated into MODFLOW-2005, and the constructed model is imported into MODFLOW-
441 To better characterize the aquifer and tunnel but also to minimize the complexity of the
442 numerical model, the model is divided into three vertical layers (with variable thickness), and
443 each layer has 301 rows and 147 columns in the planar view with identical cells with the
444 dimension of 16 m by 16 m. Since the northwest and southwest sides of the domain have
445 intrusive rocks, thus the contact between carbonate rock and intrusive rock is defined as the
446 impervious boundary. The eastern part of the domain is controlled by fault structures, and the
447 mudstone and argillaceous siltstone overlie the carbonate karst aquifers. Therefore, the eastern
448 boundary (fault F103) can be treated as the impervious boundary as well.
449 The top of the first layer is the surface elevation, and the upper boundary is the water table,
450 the bottom boundary is impermeable with a uniform elevation of 1800 m AMSL. The boundary
451 conditions, hydrogeological parameters, rainfall infiltration coefficient (which is defined as the
452 ratio of annual recharge to the actual precipitation) and observed well data of the model are
453 converted into feature objects in the MAP module and assigned to the grid. The obtained three-
27
455
456 Figure 9. The three-dimensional (3D) mesh subdivision of the study area.
457 We have calibrated the model by simulating steady flow in the dry season (from November
458 to April in the following year) and unsteady flow in the rainy season (from May to October at
459 the same year). The simulation of steady flow is mainly used to calibrate the hydraulic
460 conductivity of the study area because the changes of flow and water level are relatively small
461 during the dry season. Although the pumping test has been carried out on the main aquifers in
462 the study area, and the hydraulic conductivity obtained is about 0.5 m/d, such a hydraulic
463 conductivity parameter still need to be further calibrated because some hydraulic features (such
28
464 as fractures, faults, and some other heterogeneities) may be out of the cone of depression of the
465 pumping test, thus are not detected in the pumping test. Due to the small variation range of
466 spring discharge and borehole water level in dry season, spring discharge of Q6 and
467 groundwater level of Z5 and Z6 on April 14 are taken as the target values of steady flow
468 simulation for parameter inversion. We simulated the spring flow from Q6 through the Drain
469 module, and Z5 and Z6 are two water level observation points. We have calibrated the water
470 level of two boreholes (Z5 and Z6) and spring flow (Q6) through the simulation of steady flow
471 during the dry season, and the results are shown in Figure 10. The calibration is conducted by
472 minimizing the RMSE (Root Mean Square Error) between the measured value (either water
N 2
X m ,i X s ,i
1
RMSE (7)
N i 1
474 where Xm,i is the measured value, and Xs,i is the simulated value.
475
476 Figure 10. The steady flow simulation calibration of borehole and spring.
29
477 It can be seen from Figure 10 that the simulated value is close to the measured value, and
478 the errors of borehole water level and spring flow are both less than 5%. After the calibration
479 of the hydraulic conductivity from the steady state simulation, we can then use the measured
480 flow rate of the spring (Q6) and the water level changes of the observation borehole (Z5) to
481 calibrate the specific yield (Sy) and the specific storage (Ss) based on the unsteady state flow
482 simulation. Meanwhile, the aquifer parameters are identified by using the dynamic change of
483 borehole water level and spring discharge as the objective function. The final parameters of
484 the model (after calibration) are shown in Table 2. The procedure of calibration is similar to
485 what is done for the steady state flow using above Eq. (7).
30
488
489 Figure 11. Comparison curve between measured and simulated water level of Z5.
490 It can be seen from the measured and simulated results of Z5 (see Figure 11) that the curve
491 fitting effect is satisfactory, with a RMSE value of 0.665 m, which is merely 6.9% of the
492 measured variation of water level. Besides, the simulated results of Q6 fit the first half (see
493 Figure 12) better than the second half. However, the measured flow still changes greatly after
494 October 2021, while the simulated flow decreases stably after October 2021. We speculate that
495 this discrepancy is probably caused by the fact that the spring (Q6) is in a well-developed karst
496 drainage region with a lot of karst conduits and fissures that are not captured in our numerical
497 model. The RMSE of the spring discharge at Q6 is 0.253 L/s, and the deviation from the
498 measured flow is 17.9%. Overall, the simulation results are generally satisfactory.
31
499
500 Figure 12. Comparison curve between measured and simulated flow of Q6.
502 The above calibrated model is now utilized to simulate the operation of the Shizishan tunnel,
503 using the CHD, DRAIN and CFPM1 modules over the excavated tunnel length of
504 (DLII49+201~DLII49+919). The elevation of the Shizishan tunnel is 1973 m AMSL and its
505 diameter is 9.2 m. The water exchange coefficient calculated by Eq (6) is 1.82 m2/d. Other
506 relevant parameters are summarized in Table 1. In respect to the DRAIN module, the
507 conductance calculated by Eq (1) is 6.72 m2/d. It should be pointed out that DLII49+201~
509 DLII49+467~DLII49+671 is intrusive rock section. The simulated monthly tunnel inflows
510 using the CHD, DRAIN and CFPM1 modules are compared with the measured monthly tunnel
511 inflows and results are shown in Figure 13. When the tunnel is in an unsaturated state, the
512 CFPM1 module will predict the water inflow according to Eqs. (5) and (6). That is to say, the
32
513 water inflow mainly depends on the difference between the water level of the tunnel and the
514 bedrock. Besides, the water inflow is processed into the constant head boundary at the tunnel
515 outlet for both the saturated and unsaturated states. No matter if the tunnel is saturated or
517
518 Figure 13. Comparison of water inflow prediction results of different models.
519 We can see from Figure 13 that the prediction results of water inflow obtained by three
520 different methods are quite different: the predicted results of the CFPM1 module are the closest
521 to the actual water inflow, which are 646-2248 m3/d for the limestone section and 225-465 m3/d
522 for the intrusive rock section. The water inflow predicted by the DRAIN module is 1868-6163
523 m3/d in the limestone section and 370-772 m3/d in the intrusive rock section. We find that the
524 CHD module has the largest predicted inflow, with the simulated water inflow of the limestone
525 section ranging from 1939 to 8768 m3/d, and the intrusive rock section ranging from 720 to
526 1486 m3/d. In general, the actual water inflow of the limestone section is 840~2466 m3/d, and
527 the intrusive rock section is about 360~482 m3/d, which are basically between the predicted
33
528 value of the CFPM1 module and the DRAIN module, or closer to the predicted value of the
529 CFPM1 module. It is worth mentioning that the simulation is based on the condition of
530 tunnelling (plugging and excavating simultaneously). Since no working conditions such as
531 water-conducting faults or karst conduits are involved, the predicted results of the CFPM1
532 module are closer to the measured value. However, as the CHD module implies that the
533 groundwater level drops almost instantaneously to the elevation of the tunnel, the CHD module
534 is more suitable for situations when preferential flow pathways such as faults or karst conduits
535 cutting through the tunnel, and consequently it will predict the largest water inflow (due to the
536 presence of preferential flow for rapidly conducting flow to the tunnel). In addition to the tunnel
537 inflow, we further use three different modules to predict the drawdown caused by tunnelling,
539
540 Figure 14. The range of drawdown for three different modules.
34
541 As can be seen from Figure 14, different modules predict different drawdown results.
542 Generally speaking, when the drawdown is greater than 5 m, it will have a significant impact
543 on the nearby spring and groundwater resources, usually leading to undesirable ecological and
544 environmental problems (Mirnezami et al., 2018; Schmoll et al., 2006). Therefore, we chose 5
545 m as the maximum allowable drawdown in our following discussion. The CFPM1, DRAIN,
546 and CHD modules predict that the areas with the drawdown more than 5 m are about 0.87 km2,
547 1.10 km2, and 1.16 km2, respectively. The result predicted by the DRAIN module is 26.43%
548 larger than that obtained in the CFPM1 module, while the result predicted by the CHD module
549 is 33.33% larger than that predicted using the CFPM1 module. Generally speaking, it is better
550 to use the CFPM1 module to predict the water inflow during tunnelling than the DRAIN and
551 CHD (if preferential flow pathways are not a major concern in the system). However, when the
552 tunnelling occurs in regions with the presence of many preferential flow pathways such as
553 water-conducting faults or synclines, the groundwater level usually drops to the tunnel
554 elevation rapidly, and the CHD module may be a better choice in such circumstances. We must
555 point out that this study is based on a relatively simple setting of tunneling, thus the findings
556 reported here should be limited to such scenarios. In actual applications, the water inflow in
557 tunnel could be affected by many complex factors such as stratum lithology, geological
558 structure, and tunnel construction conditions, and influences of such complex factors must be
559 examined specifically, a task that is out of the scope of this study but deserves further
560 investigations.
561 Tunnel excavation (especially for long tunnels) is a slow process that usually takes a long
35
562 time. This means that the model representation of tunnels should be treated as a time-evolving
563 feature in the model (Zheng et al., 2021). This study mainly discussed the prediction of water
564 inflow in tunnel by three different models (DRAIN, CHD and CFPM1) to guide the safe
565 construction of tunnel, so did not specifically consider the time-evolving feature of tunnelling.
566 In addition, the prediction of tunnel water inflow could be a very complicated problem,
567 especially considering the possible complex geological structure conditions such as fractures,
568 faults, folds, dissolved rock mass and underground rivers in karst region, etc. Such complexity
569 is also not specifically considered in this study. However, this study provides a method of
570 reference for the prediction of water inflow in tunnel excavation process in a relatively uniform
571 rock mass in which the complex geological structures as mentioned above are secondary issues.
573 In this study, we have used a rather simple setting of tunnelling in the southwest karst region
574 of China as an example to compare the difference of three modules (CFPM1, CHD and DRAIN)
575 in simulating tunnel water inflow. We have found that the initial water inflow and the late time
576 stable water inflow predicted by the CHD module are the maximum, followed by the DRAIN
577 module, and CFPM1 module predicts the minimum water inflow. Besides, the diameter of the
578 tunnel has a significant impact on the prediction of tunnel water inflow. Through the numerical
579 simulation exercises of a hypothetical case and an actual field site (the Shizishan tunnel), we
580 have confirmed that the CFPM1 module performs the best in terms of predicting the tunnel
581 inflow, provided that the preferential flow paths such as water-conducting faults and fractures
582 do not cut through the tunnel. On the other hand, the CHD module may be more practical when
36
583 the preferential flow paths (such as syncline water-rich structures, water-conducting faults, and
584 karst caves, etc.) are presented. The results and methodology in this study provide theoretical
585 and numerical support for better prediction of the tunnel water inflow.
586 Acknowledgement
587 This study was supported by Yunnan Dianzhong Water Diversion Engineering (Grant No.
588 2021046355) and the Fundamental Research Funds for the Central Universities, China
590 References
591 Abusaada, M., Sauter, M., 2013. Studying the Flow Dynamics of a Karst Aquifer System with
592 an Equivalent Porous Medium Model. Groundwater, 51(4): 641-650.
593 Arjnoi, P., Jeong, J.H., Kim, C.Y., Park, K.H., 2009. Effect of drainage conditions on porewater
594 pressure distributions and lining stresses in drained tunnels. Tunnelling and
595 Underground Space Technology, 24(4): 376-389.
596 Banks, D., Solbjørg, M., Rohr-Torp, E., 1992. Permeability of fracture zones in a Precambrian
597 granite. Quarterly Journal of Engineering Geology, 25(4): 377-388.
598 Bauer, S., Liedl, R., Sauter, M., 2003. Modeling of karst aquifer genesis: Influence of exchange
599 flow. Water Resources Research, 39(10): 1285.
600 Birk, S., Geyer, T., Liedl, R., Sauter, M., 2005. Process‐based interpretation of tracer tests in
601 carbonate aquifers. Groundwater, 43(3): 381-388.
602 Bonacci, O., roje-Bonacci, T., 2000. Interpretation of groundwater level monitoring results in
603 karst aquifers: examples from the Dinaric karst. Hydrological Processes, 14: 2423-2438.
604 Bu, L. et al., 2017. Application of the comprehensive forecast system for water-bearing
605 structures in a karst tunnel: a case study. Bulletin of Engineering Geology the
606 Environment, 78: 357-373.
607 Chen, Y.-F. et al., 2020. Non-Darcian flow effect on discharge into a tunnel in karst aquifers.
608 International Journal of Rock Mechanics and Mining Sciences, 130: 104319.
609 Chisyaki, T., 1984. A study on confined flow of ground water through a tunnel. Groundwater,
610 22(2): 162-167.
611 Chiu, Y.-C., Chia, Y., 2012. The impact of groundwater discharge to the Hsueh-Shan tunnel on
612 the water resources in northern Taiwan. Hydrogeology journal, 20(8): 1599-1611.
613 Coli, M., Pinzani, A., 2014. Tunnelling and Hydrogeological Issues: A Short Review of the
614 Current State of the Art. Rock Mechanics and Rock Engineering, 47(3): 839-851.
615 Cui, Q.L., Wu, H.N., Shen, S.L., Xu, Y.S., Ye, G.L., 2015. Chinese karst geology and measures
616 to prevent geohazards during shield tunnelling in karst region with caves. Natural
37
617 Hazards, 77(1): 129-152.
618 Feehley, C.E., Zheng, C., Molz, F.J., 2000. A dual ‐domain mass transfer approach for
619 modeling solute transport in heterogeneous aquifers: Application to the
620 Macrodispersion Experiment (MADE) site. Water Resources Research, 36(9): 2501-
621 2515.
622 Flach, G.P., Crisman, S.A., Molz III, F.J., 2004. Comparison of single-domain and dual-domain
623 subsurface transport models. Ground Water, 42(6/7): 815.
624 Gallegos, J.J., Hu, B.X., Davis, H., 2013. Simulating flow in karst aquifers at laboratory and
625 sub-regional scales using MODFLOW-CFP. Hydrogeology journal, 21(8): 1749-1760.
626 Ghasemizadeh, R. et al., 2012. Groundwater flow and transport modeling of karst aquifers,
627 with particular reference to the North Coast Limestone aquifer system of Puerto Rico.
628 Hydrogeology journal, 20(8): 1441-1461.
629 Goodman, R.E., Moye, D.G., Van Schalkwyk, A., Javandel, I., 1964. Ground water inflows
630 during tunnel driving. College of Engineering, University of California.
631 He, K., Zhang, S., Wang, F., Du, W., 2010. The karst collapses induced by environmental
632 changes of the groundwater and their distribution rules in North China. Environmental
633 Geology, 61(5): 1075-1084.
634 Hubinger, B., Birk, S., Hergarten, S., 2016. A new equation solver for modeling turbulent flow
635 in coupled matrix‐conduit flow models. Groundwater, 54(4): 596-602.
636 Jin, D., Li, Y., Jin, H., 2013. An assessment of conservation effects in Shilin Karst of South
637 China Karst. Environmental Earth Sciences, 68(3): 821-832.
638 Karay, G., Hajnal, G., 2015. Modelling of groundwater flow in fractured rocks. Procedia
639 Environmental Sciences, 25: 142-149.
640 Kaufmann, G., Romanov, D., 2020. Modelling long-term and short-term evolution of karst in
641 vicinity of tunnels. Journal of Hydrology, 581: 124282.
642 Li, D. et al., 2009. Case studies of groundwater flow into tunnels and an innovative water-
643 gathering system for water drainage. Tunnelling and Underground Space Technology,
644 24(3): 260-268.
645 Li, S., Qi, Y., Li, Z., Li, H., Zhang, J., 2021. A novel treatment method and construction
646 technology of the pipeline gushing water geohazards in karst region. Tunnelling
647 Underground Space Technology, 113: 103939.
648 Li, S.C. et al., 2010. Predicting geological hazards during tunnel construction. Journal of rock
649 mechanics geotechnical engineering, 2: 232-242.
650 Liedl, R., Sauter, M., Hückinghaus, D., Clemens, T., Teutsch, G., 2003. Simulation of the
651 development of karst aquifers using a coupled continuum pipe flow model. Water
652 Resources Research, 39(3): 1057.
653 Liu, D., Xu, Q., Tang, Y., Jian, Y., 2021a. Prediction of water inrush in long-lasting shutdown
654 karst tunnels based on the HGWO-SVR model. IEEE Access, 9: 6368-6378.
655 Liu, R., Wang, J., Zhan, H., Chen, Z., Zheng, S., 2021b. Influence of thick karst vadose zone
656 on aquifer recharge in karst formations. Journal of Hydrology, 592: 125791.
657 Lv, Y., Jiang, Y., Hu, W., Cao, M., Mao, Y., 2020. A review of the effects of tunnel excavation
658 on the hydrology, ecology, and environment in karst areas: Current status, challenges,
38
659 and perspectives. Journal of Hydrology, 586: 124891.
660 Mirnezami, S.J., Bagheri, A., Maleki, A., 2018. Inaction of society on the drawdown of
661 groundwater resources: a case study of Rafsanjan plain in Iran. Water Altern, 11(3):
662 725-748.
663 Moon, J., Fernandez, G., 2010. Effect of excavation-induced groundwater level drawdown on
664 tunnel inflow in a jointed rock mass. Engineering Geology, 110(3-4): 33-42.
665 Phillips, J.D., 2016. Landforms as extended composite phenotypes. Earth Surface Processes
666 Landforms, 41(1): 16-26.
667 Polubarinova-Koch, P.I., 2015. Theory of ground water movement. Princeton University Press.
668 Reimann, T. et al., 2013. Addition and enhancement of flow and transport processes to the
669 MODFLOW-2005 conduit flow process, 2013 NGWA Summit—The National and
670 International Conference on Groundwater. Ground Water Assoc, TU Dresden, Dresden,
671 Germany.
672 Rodriguez, L.B., Cello, P.A., Vionnet, C.A., Goodrich, D., 2008. Fully conservative coupling
673 of HEC-RAS with MODFLOW to simulate stream–aquifer interactions in a drainage
674 basin. Journal of Hydrology, 353(1-2): 129-142.
675 Schmoll, O., Howard, G., Chilton, J., Chorus, I., 2006. Protecting groundwater for health:
676 managing the quality of drinking-water sources. World Health Organization.
677 Serrano-Hidalgo, C., Guardiola-Albert, C., Heredia, J., Elorza Tenreiro, F.J., Naranjo-
678 Fernández, N., 2021. Selecting Suitable MODFLOW Packages to Model Pond–
679 Groundwater Relations Using a Regional Model. Water, 13(8): 1111.
680 Shi, S. et al., 2017. Application of comprehensive prediction method of water inrush hazards
681 induced by unfavourable geological body in high risk karst tunnel: a case study.
682 Geomatics Natural Hazards and Risk, 8: 1407-1423.
683 Shoemaker, W.B., Cunningham, K.J., Kuniansky, E.L., Dixon, J., 2008a. Effects of turbulence
684 on hydraulic heads and parameter sensitivities in preferential groundwater flow layers.
685 Water resources research, 44(3): W03501.
686 Shoemaker, W.B., Kuniansky, E.L., Birk, S., Bauer, S., Swain, E.D., 2008b. Documentation of
687 a conduit flow process (CFP) for MODFLOW-2005, 6. US Department of the Interior,
688 US Geological Survey Reston, Va.
689 Shu, L. et al., 2020. Laboratory and numerical simulations of spatio-temporal variability of
690 water exchange between the fissures and conduits in a karstic aquifer. Journal of
691 Hydrology, 590: 125219.
692 Singh, S.K., 2010. Modeling the Transient Pumping from Two Aquifers Using MODFLOW.
693 Journal of irrigation and drainage engineering, 136(4): 276-281.
694 Sun, W., Song, J., Yang, W., Zheng, Y., Kuang, D., 2020. Distribution of carbonate rocks and
695 variation analysis of karst water resources in China. Carbonates Evaporites, 35(4): 121.
696 Teutsch, G., Sauter, M., 1998. Distributed parameter modelling approaches in karst-
697 hydrological investigations. Bulletin d'hydrogéologie(16): 99-109.
698 Waltham, T., 2009. The karst lands of southern China. Geology Today, 25(6): 232-238.
699 Worthington, S.R., Soley, R.W., 2017. Identifying turbulent flow in carbonate aquifers. Journal
700 of hydrology, 552: 70-80.
39
701 Ying, H.-w., Zhu, C.-w., Shen, H.-w., Gong, X.-n., 2018. Semi-analytical solution for
702 groundwater ingress into lined tunnel. Tunnelling and Underground Space Technology,
703 76: 43-47.
704 Zhao, H., Wang, L., 2020. Application of AMT in the Investigation of New Loushanguan
705 Tunnel of Chongqing Guizhou Railway. IOP Conference Series Earth Environmental
706 Science, 531: 012065.
707 Zhao, L. et al., 2017. Discussion and application of simulation methods for karst conduit flow
708 based on MODFLOW. Carsologica Sinica, 36(3): 346-351.
709 Zheng, X. et al., 2021. Evaluation of hydrogeological impact of tunnel engineering in a karst
710 aquifer by coupled discrete-continuum numerical simulations. 597: 125765.
711 Zheng, Y., He, S., 2021. Characteristics, challenges and countermeasures of giant karst cave:
712 A case study of Yujingshan tunnel in high-speed railway. Tunnelling Underground
713 Space Technology, 114: 103988.
714 Zhou, J.-Q. et al., 2021. A semi-empirical model for water inflow into a tunnel in fractured-
715 rock aquifers considering non-Darcian flow. Journal of Hydrology, 597: 126149.
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
40
743 Figure caption
744 Figure 1. The schematic diagram of the hypothetical model. (b) The tunnel section has been
746 Figure 2. The results of water inflow predicted by CHD, DRAIN, CFPM1 modules in
747 MODFLOW.
748 Figure 3. Planar contour maps of water levels at different times for three different models.
749 The CHD module predicted the maximum drawdown and range (red area), followed by
750 DRAIN module, and CFP module predicted the minimum drawdown.
751 Figure 4. The results of water inflow by CFPM1 modules with different tunnel diameters.
752 Figure 5. The hydraulic head at different positions of the tunnel changes with time.
753 Figure 6. The satellite map of study area location. There are six karst water sub-systems
754 around the tunnel, and the study area is mainly located in the range of the sixth karst water
757 Figure 8. The cross-sectional view of the study area, (a) profile of the study area, (b) profile
759 Figure 9. The three-dimensional (3D) mesh subdivision of the study area.
760 Figure 10. The steady flow simulation calibration of borehole and spring.
761 Figure 11. Comparison curve between measured and simulated water level of Z5.
762 Figure 12. Comparison curve between measured and simulated flow of Q6.
763 Figure 13. Comparison of water inflow prediction results of different models.
764 Figure 14. The range of drawdown for three different modules.
41