Tunnelling and Underground Space Technology 80 (2018) 64–77

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

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

Prediction of karst for tunnelling using fuzzy assessment combined with T

geological investigations
⁎
Kai Zhanga,b, Dwayne D. Tannantb, Wenbo Zhengb, Shougen Chena, , Xinrong Tanc
a
Key Laboratory of Transportation Tunnel Engineering, Ministry of Education, Southwest Jiaotong University, Chengdu, Sichuan 610031, China
b
School of Engineering, University of British Columbia, Kelowna, BC V1V 1V7, Canada
c
China Railway Eryuan Engineering Group Limited Company, Chengdu, Sichuan 610031, China

A R T I C LE I N FO A B S T R A C T

Keywords: This paper presents a method for predicting karst features before and during tunnel construction. The prediction
Karst prediction of karst consists of two components: an initial karst prediction using a fuzzy assessment system to evaluate the
Tunnel underground karst state and an updating karst prediction where appropriate geological investigation methods
Fuzzy assessment system are selected based on the assessment of underground karst state. The investigation results are then used to
Underground karst state
update the underground karst state. The initial assessment system is based on a fuzzy comprehensive evaluation
Geological investigations
method. Nine influence factors are selected as the evaluation indices for the underground karst state, and each
index is quantitatively rated to four grades. The membership of the evaluation index is determined by using a
membership function, and the weights of these indices are distributed by using a fuzzy Analytical Hierarchy
Process. The fuzzy transform principle and maximum membership degree principle are applied to determine the
underground karst state level. Based on the assessment result, several techniques for geological investigation,
including the seismic reflection method, ground penetrating radar, infrared water detection, transient electro-
magnetic method, and advance probe boreholes, are recommended to predict the location, size, and distribution
of karst features ahead of tunnel faces. These geological investigations have different characteristics and can be
combined to improve the accuracy of the geological prediction. The appropriate combination of investigation
methods is selected using the assessed underground karst state, and the investigation results are also used as the
input to update the underground karst state. The proposed method can improve the prediction of karst in
tunnelling. An application of this method was performed in the Doupengshan tunnel project.

1. Introduction
Tunnel construction in karst terrain is fraught with problems asso-
ciated with the unexpected location, irregular geometry, and un-
predictable dimensions of karst structures (Alijia et al., 2013). Many
cases around the world have shown that karst is the major challenge for
tunnelling in karst areas since the presence of karst features can lead to
economic, safety-related, and environmental problems (Casagrande
et al., 2005; Filipponi, 2015; Zini et al., 2015). The prediction of karst
features is critical for the reduction of risk for tunnel construction.
Recently, several methods have been developed for predicting and
assessing the karst risk. Computer simulation of the expert decision-
making process was used to construct a karst disaster prediction system
(Zhang et al., 1993). A ground investigation method named Karst-ALEA
based on the knowledge of the speleogenesis processes and 3D mod-
eling was applied to assess the risk of karstic rock masses at the plan-
ning stage (Filipponi, 2009; Filipponi and Jeannin, 2010). Some
researchers used statistical data of karst features to assess karst hazards
(Kaufmann and Quinif, 2002; Waltham and Fookes, 2003; Galve et al.,
2011; Zini et al., 2015). These established assessment systems are useful
for predicting karst features, but they can only provide overall and
approximate guidance for construction. Underground karst structures
have complex characteristics and have random distributions in space.
For the safety of tunnel construction, it is essential to obtain specific
information about karst structures in advance in order to take coun-
termeasures during tunnelling.
Different site investigation and testing techniques can be used to
detect unfavourable geological structures in front of the tunnel face.
The investigation techniques in karst terrain mainly include two cate-
gories. The first category is geological analysis, including geological
surveys (Zini et al., 2015), advance drilling, and borehole tests
(Pesendorfer and Loew, 2004). The other category includes geophysical
methods such as the seismic reflection method (Asadollahi and
Foroozan, 2006; Alimoradi et al., 2008; Sun et al., 2008), ground

⁎
Corresponding author.
E-mail address: swjtu.civil@foxmail.com (S. Chen).

https://doi.org/10.1016/j.tust.2018.06.009
Received 10 November 2017; Received in revised form 12 April 2018; Accepted 8 June 2018
0886-7798/ © 2018 Elsevier Ltd. All rights reserved.
K. Zhang et al.
penetrating radar (Collins et al., 1994; Conyers and Goodman, 1997;
Chamberlain et al., 2000; Al-fares et al., 2002; Knödel et al., 2007), and
the transient electromagnetic method (Knödel et al., 2007;
Goldscheider and Drew, 2007). Because each technique/method has
shortcomings and restricted application conditions (Waltham and
Fookes, 2003; Li et al., 2008; Alijia et al., 2013), the prediction of karst
with a single technique/method may not be reliable. Thus, different
investigation techniques should be combined to increase the accuracy
of prediction.
This paper presents a comprehensive method to predict karst fea-
tures for tunnelling. The method contains two parts: (1) initial karst
prediction via assessment of underground karst state before tunnelling
and (2) an updated prediction of karst features by combining the pre-
vious assessment results with the geological site investigations that
occur during tunnelling. The initial evaluation system for assessing
underground karst state is constructed based on a fuzzy evaluation
method, and the updated karst prediction relies on the assessment re-
sults of the underground karst state and appropriately selected geolo-
gical investigations. The initial assessment result is used to guide the
planning of geological site investigations, and the results of the geolo-
gical investigations are used to update the initial assessment. This
comprehensive method can make the prediction of karst more efficient
and economical. A case study is provided to demonstrate the applica-
tion of this method for karst prediction.
2. Methodology for karst prediction
2.1. Initial karst prediction via assessment of underground karst state before
tunnelling
Geological data from maps and other sources can be used to assess
the underground karst state in advance of tunnel construction. The
underground karst state (UKS) is used to describe the current status of
the shape and the scale of the underground karst within a region. The
assessment of UKS can be regarded as an initial karst prediction. The
evaluation system for assessing the UKS constructed in this paper makes
use of a fuzzy evaluation method. The fuzzy method is an effective tool
to handle the uncertainties and subjectivities, and fuzzy boundaries (Shi
et al., 2014; Zhang et al., 2015; Lu et al., 2016; Liu et al.,2017). The
main components of the fuzzy evaluation of UKS are: (a) select eva-
luation indices, (b) establish index set and evaluation set, (c) construct
membership function and fuzzy evaluation matrix, (d) determine eva-
luation indices weights, and (c) produce a comprehensive evaluation.
2.1.1. Evaluation indices for underground karst state
Karstification is a geological process in soluble rock controlled by
the solubility and permeability of the rock and the mobility and ag-
gressivity of the groundwater (Sokolov, 1962). Geologic structures,
vegetation cover, climate and other factors also affect karst develop-
ment (LeGrand and Stringfield, 1973; Stringfield et al., 1979; Ford and
Williams, 2007; Li et al., 2013). The underground karst state can be
assessed from the status of the factors influencing karst development
(Stokes and Griffiths, 2000). Based on the existing research and the
statistical information of tunnelling cases in karst terrain, several fac-
tors are selected as the evaluation indices to assess the UKS.
(1) Formation lithology
Formation lithology is one of the major factors controlling the karst
development because the presence of soluble rock is a precondition for
karstification. It is more likely for karst to develop in formations with
strong solubility than those with weak solubility. The solubility coef-
ficient t has been recommended to evaluate the contribution of for-
mation lithology to karst development (Zhou et al., 2013). The li-
thology coefficient t is expressed as:
65
Tunnelling and Underground Space Technology 80 (2018) 64–77
3
t= ∑ Ai Bi = A1 B1 + A2 B2 + A3 B3 = 0.636B1 + 0.259B2 + 0.105B3
i=1 (1)
where A1, A2 and A3 represent the contribution of a different lithology
to karst development and are determined based on the statistical data
and expert judgement as suggested by Zhou et al. (2013). The values of
B1, B2 and B3, are the proportions of lithology with strong, medium, and
weak solubility respectively and meet B1 + B2 + B3 = 1 (Xu et al.,
2011). The division of lithologies with strong, medium, and weak so-
lubility are presented in Table 1.
Table 1
Solubility of carbonate rocks (after Mao et al., 2010; Zhou et al., 2013).
Solubility Weak Medium Strong
Formation Impure carbonate Pure carbonate rocks Pure
lithology rocks, or impure interbedded with carbonate
carbonate rocks impure carbonate rocks
interbedded with rocks or clastic rocks
clastic rocks
Impure carbonate rock includes argillaceous limestone, argillaceous dolomite,
siliceous limestone, siliceous dolomite. Pure carbonate rock includes limestone,
dolomite, dolomitic limestone, carbonaceous dolomite.
(2) Hydrodynamic condition
The presence and action of groundwater is the other precondition
for karstification. The hydrodynamic condition of groundwater can
control the karst development. The groundwater in a karst area is di-
vided into four vertical hydrodynamic zones with different character-
istics in karst development (Sokolov, 1967). Based on the general
characteristics of karst development in different hydrodynamic zones
and the specific groundwater features and runoff condition, the
groundwater hydrodynamic condition is quantitatively classified into
the four grades through expert evaluation.
(3) Geological structure conditions
Geological structures, including faults, folds, joints, and bedding
planes, are key to karst development because they host and guide the
underground solution conduit networks (Stringfield et al., 1979; Ford
and Williams, 2007). The width of fault fracture zones, the structural
features of folds, and the spacing of joints or bedding planes are se-
lected to evaluate the contribution of geological structures to karst
development. These three parameters are quantitatively classified into
the four grades according to four levels of underground karst state.
(4) Landform and physiognomy
Karst is regarded as a distinctive terrain developed in soluble rock
with landforms that relate to underground drainage (Waltham et al.,
2005). Negative landforms such as valleys, depressions or dolines in-
fluence the recharge of groundwater and can reflect the development
and occurrence characteristics of karst water. The proportion of the
ground surface with negative landforms is selected to assess the con-
tribution of landform and physiognomy to karst development, and it is
quantitatively divided according to four levels of underground karst
state (Li et al., 2013).
(5) Attitude of rock formation
The attitude of strata in a rock formation also influences the karst
development because it affects the permeability of a rock formation and
the seepage characteristics of groundwater. The permeability of a rock
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

formation is typically anisotropic with the permeability coefficient where I1 to I9 are the evaluation indices as shown in Fig. 1 and Table 3.
along the interface of formation layers being much larger than that The fuzzy logical language is used to describe the likelihood for
perpendicular to the interface (Li et al., 2013). The recharge, runoff, karst features in an evaluated volume of rock. The evaluation set is
discharge, and infiltration characteristics of groundwater are also in- established as:
fluenced by the attitude of rock formations. Based on the analysis of the
V = {v1, v2, v3, v4} (3)
statistical cases, the strata inclination angle was divided into four
grades to assess the contribution to the karst development. To meet the where v1, v2, v3 and v4 are the four UKS levels as presented in Tables 2
data format in the mathematical methods, the strata inclination angle and 3.
was modified into the following forms: φ′1 = φ (0≤φ≤45), φ′2 = 90-
φ (45 ≤ φ ≤ 90), where φ is the strata inclination angle, φ′1 and φ′2 2.1.4. Membership function and fuzzy evaluation matrix
are modified angles, and the modified angle is uniformly expressed as Each evaluation index (Ii) has effects on all of the UKS levels (vj).
φ′ when φ′1 = φ′2 (Zhou et al., 2013). The effects can be expressed by the membership degrees, which can be
computed by the membership functions. The membership functions of
(6) Contact zone of soluble and insoluble rocks the evaluation indices are constructed as the following forms.
Assume a, b, c and d are the thresholds of the quantitative grading of
The groundwater penetration to initiate karst is often easier at a each evaluation index as shown in Table 3. When a < b < c < d, the
contact zone between dissolvable and insoluble rock than it is at bed- membership functions can be expressed as:
ding planes, joints, fissures within dissolvable rock (Ford and Williams,
2007). The contact zone of soluble and insoluble rock is often a place ⎧ 1 x⩽a
b−x
where karst initially develops. Based on the conditions of the layers and μ v1 (x ) = a<x⩽b
⎨b−a
interlayer fissures and the difference in the lithology of rocks in contact, ⎩ 0 x>b (4)
the contact zone is quantitatively classified into the four grades ac-
cording to expert evaluation (Li et al., 2013). ⎧ x −0 a x⩽a
⎪ a<x⩽b
b−a
(7) Surface karst features μ v2 (x ) =
⎨c−x b < x ⩽ c
⎪c−b
The connection of the water drainage systems on the surface to ⎩ 0 x>c (5)
those in the subsurface reflects the state of karstification underground
(Dreybrodt, 2012). Thus, the density of surface karst features is selected ⎧ 0 x⩽b
⎪ x−b
as an evaluation index to assess the influence of the surface karst on the ⎪c−b b < x ⩽ c
μ v3 (x ) =
underground karst development state. The surface karst feature density, ⎨d − x c < x ⩽ d
namely the number of karst features per square kilometre (Kaufmann ⎪d−c
⎪ 0 x>d (6)
⎩
and Quinif, 2002; Zini et al., 2015; Gao et al., 2016), is used as the
index and is divided into four grades according to four UKS levels.
⎧ x 0− c x⩽c
μ v4 (x ) = c<x⩽d
2.1.2. Levels for underground karst state and grading of evaluation indices ⎨d−c
The UKS was divided into four levels, v1, v2, v3 and v4, according to ⎩ 1 x>d (7)
the shape and scale of underground karst features and the status of karst where μ v1 (x ) , μ v2 (x ) , μ v3 (x ), and μ v4 (x ) are the membership degrees of
water system (Bao and Chen,1983; Gao et al., 2016), as shown in the evaluation index with regard to UKS levels v1, v2, v3 and v4 re-
Table 2. The evaluation indices were all graded according to four UKS spectively, and x is the quantitative value of each evaluation index.
levels. A hierarchy structure for assessing underground karst state was When a > b > c > d, the membership functions can be expressed
constructed as presented in Fig. 1. To meet the data format requirement as:
of the fuzzy evaluation, the evaluation indices at a u-criterion level
should be quantitatively graded. Based on previous studies (Li et al, ⎧ 1 x⩾a
x−b
2013; Ford and Williams, 2007; Mao et al., 2010; Zhou et al, 2013; μ v1 (x ) = b⩽x<a
⎨a−b
Kaufmann and Quinif, 2002; Zini et al., 2015; Gao et al., 2016) and ⎩ 0 x<b (8)
tunnelling cases in karst terrain, the nine evaluation indices are quan-
titatively determined. The quantitative grading standards for the eva- ⎧ a 0− x x⩾a
luation indices are presented in Table 3. The evaluation indices, u2, u4 ⎪ b⩽x<a
a−b
and u8, were quantitatively graded based on their ratings obtained μ v2 (x ) =
⎨ x−c c ⩽ x < b
through expert evaluation, and the rating standards of the three indices ⎪b−c
are presented in Tables 4–6 respectively. ⎩ 0 x<c (9)

2.1.3. Index set and evaluation set ⎧ 0 x⩾b

⎪ b−x
Based on the selected influence factors, the evaluation index set is ⎪b−c c ⩽ x < b
μ v3 (x ) =
constructed as: ⎨x−d d ⩽ x < c
⎪c−d
⎪ 0 x<d (10)
I = {I1, I2, I3, I4 , I5, I6, I7, I8, I9} (2) ⎩

Table 2
Division of the UKS levels (vi).
UKS level Very low (v1) Low (v2) Medium (v3) High (v4)

Description Karst features are mainly Karst features are mainly Karst features are mainly small-scale Karst features are mainly large-scale (≥50 m3)
dissolution fissures or pores; karst separate karst conduits; karst (< 50 m3) karst caves or separate karst karst caves, connected karst conduits or karst
water is mainly fissure water. water is mainly fissure water. conduits; karst water is mainly pipe water. rivers; karst water is mainly pipe water.

66
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Fig. 1. Hierarchical structure for assessing underground karst state.

W = [w1 w2 . .. wn] (13)

⎧ c −0 x x ⩾c
μ v4 (x ) = d⩽x<c
⎨c−d where wi (1 ≤ i ≤ n) is the weight of the evaluation index Ii.
⎩ 1 x<d (11)
In this paper, the weights of the evaluation indices are determined
Based on the membership degrees of the evaluation indices for the by the Fuzzy Analytical Hierarchy Process (FAHP), which was origin-
evaluation grades, an evaluation matrix U can be established in the ally developed by Chang (1996) to solve hierarchical problems. The
following form. FAHP is regarded as an efficient means of decision making and is used
for weight determination (Khademi Hamidi et al., 2010; Nezarat et al.,
μ ⋯ μ1m
⎡ 11 ⎤ 2015; Lu et al., 2016). The determination of weights for the indices
U=⎢ ⋱ ⎥
μ ⋯ μ using FAHP is conducted in a series of steps. A pairwise comparison
⎣ n1 nm ⎦ (12)
matrix is constructed, and the relative importance of one index over
where μij (1 ≤ i ≤ n; 1 ≤ j ≤ m) is the membership degree of the another is determined by expert judgement and is expressed by the
evaluation index (Ii) for the evaluation grade (vj). In this research, n is 9 triangular fuzzy numbers (TFN) shown in Table 7.
and m is 4. According to their contributions to the index at the adjacent upper
level, the indices at the lower criterion level were pairwise compared,
2.1.5. Evaluation indices weights and the corresponding comparison matrixes were constructed. Two
The weights of the evaluation indices can be expressed as a weight comparison matrixes are presented in Tables 8 and 9. The consistency
matrix shown as follows: ratio (CR) of the two comparison matrixes is less than 0.1, indicating

Table 3
Grading standards of evaluation indices.
Evaluation index UKS level

v1 v2 v3 v4

I1 Solubility coefficient of rock formation (t) 0–0.042 0.042–0.104 0.104–0.254 > 0.254

I2 Groundwater hydrodynamic condition > 85 70–85 60–70 0–60

I3 Width of fault fracture zone (m) 0–0.2 0.2–0.5 0.5–1.0 > 1.0
I4 Structural features of fold > 85 70–85 60–70 0–60
I5 Joint spacing (m) > 1.0 0.4–1.0 0.2–0.4 0–0.2
I6 Proportion of surface with the negative landforms (%) 0–20 20–40 40–60 > 60
I7 Modified strata inclination angle (φ′, φ′1 and φ′2 (°)) 0≤φ′1 < 5 5≤φ′2 < 10 10≤φ′ < 25 25≤φ′ < 45
I8 Contact state of dissolvable and insoluble rocks > 85 70–85 60–70 0–60
I9 Density of surface karst features (#/km2) 0–1 1–5 5–15 > 15

67
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Table 4
Rating and grading standards of groundwater hydrodynamic condition (u2).
Rating > 85 70–85 60–70 0–60

Groundwater Groundwater is mainly pore Groundwater is bedrock fissure Groundwater is bedrock fissure water Groundwater is mainly conduit
hydrodynamic water; runoff condition is poor; or water; runoff condition is and conduit water; runoff condition is water; runoff condition is very good;
condition the target area is located in the moderate; or the target area is good; or the target area is located in or the target area is located in the
deep circulation zone located in the aeration zone the full saturation zone seasonal fluctuation zone

Table 5 Table 7
Rating and grading standards for structural features of a fold (u4). Relative importance and corresponding triangular fuzzy number (after Nezarat
et al., 2015).
Rating > 85 70–85 60–70 0–60
Fuzzy Relative importance explanation TFN Inverse TFN
Structural No fold; or the Limbs of Hinge of flat Hinge of number
features of limbs of flat steep and and open steep and
fold and open folds tight folds folds tight folds ∼ Two factors contribute equally to the (1,1,1) (1,1,1)
1
objective
∼ One factor is slightly favoured over (1,3,5) (1/5,1/3,1)
3
that they met the consistency requirement (Saaty, 1979). The weights the other
∼ One factor is moderately favoured (3,5,7) (1/7,1/5,1/3)
of the indices were computed based on the corresponding comparison 5
over the other
matrix. The process used for weight calculation using FAHP is found in ∼
7 One factor is strongly favoured over (5,7,9) (1/9,1/7,1/5)
the literature (Chang, 1996; Nezarat et al., 2015; Lu et al., 2016), thus the other
∼ One factor is absolutely favoured over (7,9,11) (1/11,1/9,1/7)
the weight calculation process is not presented here, and only the re- 9
the other
sults are presented as shown in Tables 8 and 9. The weights of the ∼
2 Intermediate values between the two (1,2,4) (1/4,1/2,1)
indices at the I-criterion level were combined, and the results are pre- ∼ adjacent judgments
4 (2,4,6) (1/6,1/4,1/2)
sented in Table 10. ∼
6 (4,6,8) (1/8,1/6,1/4)
∼ (6,8,10) (1/10,1/8,1/6)
8
2.1.6. Comprehensive evaluation
The evaluation indices are evaluated by the fuzzy transform prin-
ciple to obtain the fuzzy comprehensive evaluation set of the UKS. The
fuzzy comprehensive evaluation matrix has the following form. assessment. The geological investigations results can then be used to
revise the assessment of UKS.
μ ⋯ μ1m
⎡ 11 ⎤
C = WU = [w1 w2 ⋯ wn] ⎢ ⋱ ⎥ = [c1 c2 ⋯ cm]
μ
⎣ n1 ⋯ μnm ⎦ (14) 2.2.1. Geological techniques to investigate karst features
Many investigation techniques have been used to investigate un-
Based on the comprehensive evaluation set, the maximum mem- favourable geological conditions ahead of a tunnel face. The commonly
bership degree principle is applied to discern the UKS level. The pre- used techniques include the seismic reflection method (SRM), ground
dicted UKS level, vs, can be obtained by the following formula: penetrating radar (GPR), and the transient electromagnetic method
m (TEM). Recently, an infrared water detector (IWD) has also been used to
vs = ⎧cl/ vl → max(cj )⎫ detect groundwater during tunnel construction in China. The infrared
⎨
⎩ j=1 ⎬
⎭ (15)
water detector records the infrared radiation field intensity emitted by
where cl is the maximum element in the fuzzy comprehensive evalua- the strata. The presence of water can be recognized by analyzing the
tion matrix (C), and vl is the corresponding UKS level in the evaluation changes in the infrared radiation field intensity received by the infrared
set (V). water detector (Wang et al., 2003; Xue et al., 2008; Lv and Chen, 2010).
Besides these geophysical investigation methods, advance probe bore-
2.2. Updating karst prediction by combining the initial prediction with hole drilling can directly detect adverse geological conditions. How-
results from geological investigations during tunnelling ever, the investigation ability of any single technique is not very reli-
able, because the interpretation of the investigation results contains
The UKS is closely related to the risk of the karst-related disasters. ambiguity and each technique has shortcomings and constraints as
To reduce the probability of karst-related disasters during tunnel con- summarized in Table 11.
struction, it is necessary to use geological investigations to improve
predictions the karst features ahead of the tunnel face and thus take 2.2.2. Updating karst prediction by combining the UKS assessment with
corresponding countermeasures. However, if the investigation techni- geological investigations
ques are used indiscriminately to predict karst features during the en- To improve the prediction of karst features ahead of the tunnel face,
tire construction process, the prediction might be inefficient and costly. different investigation techniques should be combined to benefit from
To make the prediction more efficient and economical, the choice of the advantages of different prediction techniques and overcome the
investigation techniques should be guided by the results of the initial shortcomings of a single technique. Considering the characteristics of

Table 6
Rating and grading standards of the contact state of dissolvable and insoluble rocks (u8).
Rating > 85 70–85 60–70 0–60

Contact state of dissolvable Impure carbonate rocks in contact with Pure carbonate rocks interbedded with Pure carbonate rocks other Pure limestone in
and insoluble rocks insoluble rocks, or pure carbonate rocks in impure carbonate rocks or clastic rocks in than limestone in contact with contact with insoluble
contact with impure carbonate rocks contact with insoluble rocks insoluble rocks rocks

Refer to Table 1 for the definition of pure carbonate rock and impure carbonate rock.

68
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Table 8
Fuzzy pairwise comparison matrix for weight analysis of indices at F-criterion level.
Index F1 F2 F3 F4 F5 F6 F7 Weight

F1 (1,1,1) (1,2,4) (1,2,4) (1,3,5) (2,4,6) (1,3,5) (1,3,5) 0.208

F2 (1/4,1/2,1) (1,1,1) (1,2,4) (1,2,4) (1,3,5) (1,2,4) (1,2,4) 0.182
F3 (1/4,1/2,1) (1,2,4) (1,1,1) (1,2,4) (1,3,5) (1,3,5) (1,2,4) 0.177
F4 (1/5,1/3,1) (1/4,1/2,1) (1/4,1/2,1) (1,1,1) (1,2,4) (1,1,1) (1,1,1) 0.113
F5 (1/6,1/4,1/2) (1/5,1/3,1) (1/5,1/3,1) (1/4,1/2,1) (1,1,1) (1/4,1/2,1) (1/4,1/2,1) 0.076
F6 (1/5,1/3,1) (1/4,1/2,1) (1/5,1/3,1) (1,1,1) (1,2,4) (1,1,1) (1/4,1/2,1) 0.110
F7 (1/5,1/3,1) (1/4,1/2,1) (1/4,1/2,1) (1,1,1) (1,2,4) (1,2,4) (1,1,1) 0.134

CI = 0.028 (< 0.1), meets the consistency requirement.

Table 9 and high (v4). In areas where the UKS is low (v2) or medium (v3), the
Pairwise comparison matrix for weight analysis of indices I3, I4 and I5. use of three measuring lines is appropriate. While in the areas where
Index I3 I4 I5 Weight UKS is high (v3), the number of the measuring lines for GPR prediction
should be increased to five or more.
I3 (1,1,1) (1,2,4) (1,2,4) 0.430 IWD is used to predict whether there is karst water ahead of the
I4 (1/4,1/2,1) (1,1,1) (1,2,4) 0.358
tunnel face in areas where the UKS is medium (v3) or high (v4). If karst
I5 (1/4,1/2,1) (1/4,1/2,1) (1,1,1) 0.212
water has been confirmed in front of the tunnel face, TEM can be
CI = 0.046 (< 0.1), meets the consistency requirement. conducted to semi-quantitatively predict the location of the water-
bearing structures. In the areas with medium (v3) UKS, the blast holes
Table 10 can be lengthened to obtain the information of the karst features and
Weights of indices at F-criterion level and I-criterion level. discharge karst water. The length of a long blast hole should be more
than 5 m. While in the areas with high UKS, as the karst features would
Evaluation Weights at F-criterion level Integrated
index weight (wi)
be more developed, and the karst water will be more abundant, the
F1 F2 F3 F4 F5 F6 F7 advance probe boreholes with length of 30 m can be drilled to obtain
the information of the karst features, including the location, size of the
0.208 0.182 0.177 0.113 0.076 0.110 0.134 karst features, and the quantity of karst water. These probe boreholes
I1 0.208
can also function to drain karst water.
I2 0.182 However, the choice of the investigation techniques for karst pre-
I3 0.430 0.076 diction for each UKS level is not fixed. The prediction plan should be
I4 0.358 0.063 updated dynamically. According to the prediction results from the
I5 0.212 0.038
techniques previously conducted, the techniques to be subsequently
I6 0.113
I7 0.076 conducted should be adjusted. The UKS level should then be revised
I8 0.110 based on the geological prediction results. If current investigation re-
I9 0.134 sults indicate that the UKS level should be increased, the subsequent
investigation plan should take into consideration the updated UKS
level. When the investigation results indicate that the UKS level can be
different UKS levels as well as the applicability of different investiga- reduced, and this is also verified by other investigations, then the UKS
tion techniques, the choice of appropriate techniques is guided by the can be lowered.
previous assessment of the UKS.
SRM is often conducted during the whole tunnelling process. In the
2.3. Summary
areas where the UKS is very low (v1) or low (v2), the effective prediction
range (PR) of the SRM can be set as 150 m. In the areas where the karst
The prediction of karst in tunnelling contains two steps: the UKS
development state is medium (v3) or high (v4), the effective prediction
assessment based on a fuzzy evaluation before tunnelling, and an up-
range of the SRM should be reduced to 100 m.
dating prediction of karst ahead of the tunnel face by combining the
GPR is usually conducted to obtained more specific information of
UKS assessment with field investigation techniques. These two steps are
the abnormalities in areas where the UKS level is low (v2), medium (v3)
closely linked. The UKS assessment can guide the geological

Table 11
Geological investigation methods that can be used from within an advancing tunnel (Li et al., 2008; Wang et al., 2003).
Investigation method Maximum effective Characteristics
prediction range (m)
Advantages Shortcomings

SRM 150 Identify changes in rock lithology; reliably predict vertical Prediction accuracy of discontinuities with irregular interface
discontinuities and water is poor; time-consuming
TEM 30–300 Can detect the location and scale of faults and water-bearing The signal is susceptible to interference
structures
GPR 20–30 Can predict rock lithology changes; has relatively high Radar signal is susceptible to interference
ability to recognize fracture zones and water-bearing
structures; operation is simple
IWD 30 High ability to recognize insidious groundwater or water- Only sensitive to water but cannot obtain quantitative
bearing structures; operation is simple and quick parameters of the water-bearing structure
Advance probe 30 Directly obtain specific rock mass information; drain Prediction under complex geological conditions is poor; cannot
borehole groundwater from the ahead of the tunnel face obtain geological information between the boreholes; time-
consuming and costly

69
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Fig. 2. Karst prediction flow chart.

investigations, while the results from the geological investigations can

be used to update the UKS. The process for the two-step karst prediction
is presented in Fig. 2, where PR represents the prediction range of SRM
investigation, and NML represents the number of measuring lines to use
for GPR investigation. If karst features are predicted ahead of the tunnel
face, subsequent countermeasures can be conducted to ensure the safety
of the tunnel excavation. The field data observed during the tunnel
excavation can be used to verify the accuracy of the karst prediction.

3. Doupengshan tunnel case study

3.1. Overview of the tunnelling project

Fig. 3. Cross section of Doupengshan tunnel.
The Doupengshan tunnel is part of the Guiyang-Guangzhou Railway

70
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Fig. 4. Geological section at chainage 98,860–101,835 m of the Doupengshan tunnel.

and is located in Douyun City, Guizhou Province in China. The Table 13

Doupengshan tunnel is a single tube tunnel with two lanes. The de- Fuzzy comprehensive evaluation result of underground karst state.
signed cross-section has the area of 99.3 m2 and a perimeter of 37.3 m. Chainage (m) Evaluation set element UKS level (vs)
A cross-section sketch of the Doupengshan tunnel is shown in Fig. 3.
The Doupengshan tunnel is 7369 m long with a maximum depth of c1 c2 c3 c4
470 m. The entrance and exit chainages are 98,841 and 106,210 m re-
98,860–99,400 0.1717 0.2283 0.3593 0.2407 v3
spectively. 99,400–99,520 0.2648 0.1901 0.2792 0.2659 v3
The formations crossed by the Doupengshan tunnel line are lime- 99,520–99,775 0.2083 0.2562 0.2840 0.2515 v3
stone, dolomite, calcareous siltstone, shale, and quartz sandstone. The 99,775–99,860 0.1301 0.2466 0.2792 0.3441 v4
soluble carbonates containing limestone and dolomite are mainly found 99,860–100,120 0.1717 0.2379 0.3446 0.2458 v3
100,120–100,360 0.1282 0.1385 0.3948 0.3385 v3
at chainage 98,860–101,835 m. This section is located on the flank of
100,360–100,765 0.2149 0.2401 0.2734 0.2716 v3
the Xiaodong syncline and is crossed by three faults. The dip of the 100,765–100,880 0.0217 0.1869 0.3043 0.4870 v4
stratum ranges from 10° to 25°. The geological layout of the studied 100,880–101,250 0.2907 0.4204 0.2281 0.0608 v2
zone of the Doupengshan tunnel is shown in Fig. 4. 101,250–101,550 0.3472 0.4204 0.1817 0.0507 v2
101,550–101,765 0.7482 0.1469 0.0492 0.0557 v1
101,765–101,835 0.6134 0.1937 0.0241 0.1687 v1
3.2. Initial assessment of underground karst state

3.2.1. Data collection for evaluation indices tunnel is shown in Fig. 5.

Information was collected from geological surveys conducted before The assessment results showed that the UKS at chainage
the tunnel construction began, including geological boreholes and 99,775–99,860 m and 100,765–100,880 m was high. The UKS at chai-
geophysical measurements. The section from chainage 98,860 to nage 98,860–99,775 m and 99,860–100,765 m was medium. At chai-
101,835 m was divided into twelve sections corresponding to locations nage 100,880–101,550 m, the UKS was low, while at chainage
where the nine evaluation indices changed, and values of the evaluation 101,550–101,835 m, the UKS was very low. As section
indices are presented in Table 12. 100,765–100,880 m had the highest UKS level, this section is chosen to
demonstrate the updated karst prediction ahead of the tunnel face by
3.2.2. Assessment of underground karst state combining the UKS assessment with the combined geological in-
Based on the values of the evaluation indices, the memberships of vestigations.
the evaluation indices were computed using the membership functions
in Eqs. (4)–(11). The weight matrix of the evaluation indices can be 3.3. Updating karst prediction
expressed as W = [0.208, 0.182, 0.076, 0.063, 0.038, 0.113, 0.076,
0.110, 0.134] according to Table 10. The evaluation matrix was ob- 3.3.1. SRM measurements
tained following Eq. (14), and the UKS levels of the evaluated sections A seismic reflection survey was conducted using a TSP203 system
were determined according to Eq. (15). The evaluation results are (Amberg Measuring Technique, 2002) on the tunnel face at chainage
presented in Table 13, and a zoning map of UKS along the Doupengshan 100,765 m. The layout of the SRM measurements is shown in Fig. 6, and
the 2D plan of the SRM prediction result is shown in Fig. 7. The SRM
Table 12 prediction results show that rock mass parameters at chainage
Evaluation indices for twelve chainage sections. 100,765–100,885 m fluctuated. The compressional wave velocity
Chainage section (m) Evaluation index changed from 4100 to 5205 m/s and from 1700 to 2380 m/s respec-
tively; the velocity ratio of compressional and shear waves (Vp/Vs)
I1 I2 I3 I4 I5 I6 I7 I8 I9 changed from 1.50 to 2.04; Poisson’s ratio changed from 0.10 to 0.34;
98,860–99,400 0.222 65 0 90 0.200 15 12 65 10 density changed from 2.63 to 2.88 g/cm3, and Young’s modulus
99,400–99,520 0.237 65 0 90 0.125 5 10 75 10 changed from 64 to 106 GPa. The shear wave velocity, density, and
99,520–99,775 0.237 65 0 85 0.125 15 10 75 10 Young’s modulus significantly decreased at chainage
99,775–99,860 0.237 65 10 85 0.125 15 13 65 10 100,782–100,791 m, 100,796–100,802 m, 100,806–100,810 m, and
99,860–100,120 0.222 65 0 85 0.250 15 13 65 10
100,817–100,819 m. Based on the SRM measurement results, karst
100,120–100,360 0.222 65 15 80 0.125 5 13 60 10
100,360–100,765 0.237 65 0 80 0.200 10 14 75 10 caves or dissolution zones and karst water were inferred to exist at four
100,765–100,880 0.222 65 10 75 0.125 65 20 65 10 sections, and the UKS was conservatively assessed as level v4.
100,880–101,250 0.070 75 0 80 0.250 30 22 90 2
101,250–101,550 0.070 75 0 80 0.250 10 20 90 2
3.3.2. GPR measurements
101,550–101,765 0 80 0 80 0.250 10 21 100 0
101,765–101,835 0 80 5 80 0.125 20 25 100 0 As the SRM detection results indicated, abnormalities existed in the
strata ahead of the tunnel face at 100,765 m. GPR was subsequently

71
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Fig. 5. Zoning map for UKS along the tunnel.

Fig. 6. Layout for SRM measurements.

applied to identify the karst features. When the tunnel face reached
chainage 100,765 m, a SIR-2000 radar system (Geophysical Survey
Systems, 2001a) was used to complete a GPR survey. Five measuring
lines (L1 to L5) were arranged on the tunnel face as shown in Fig. 8. An
antenna with a centre frequency of 100 MHz was used, and the number Fig. 8. Layout of measuring lines for the GPR measurements.
of data samples in a vertical scan was set to 512. After processing the
data using RADAN 6.6 (Geophysical Survey Systems, 2001b) software,

Fig. 7. Results from SRM on tunnel face at chainage 100765 m.

72
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Fig. 10. Layout of measuring lines for IWD measurements.

the GPR measurement results were obtained, and these are presented in
Fig. 9. The abnormalities are noted by numbers, and these were found
from 100,768 to 100,789 m. These were inferred to be caused by dis-
solution zones, karst cavities or conduits ahead of the tunnel face. As a
result, the UKS was still assessed as level v4.

3.3.3. IWD measurements

A HY303 infrared water detector was used to measure the infrared
radiation field intensity emitted by the strata ahead of the tunnel face at
chainage 100,765 m. Four measuring lines (L1-L4) were arranged on
the tunnel face at chainage 100,765 m with each measuring line having
five measuring points (from left to right and top to bottom respec-
tively). Five additional measuring lines (M1-M5) were arranged along
the tunnel. These measuring lines were 50 m long and had 10 equally
spaced measuring points. The layout of the measuring lines for IWD
measurements is presented in Fig. 10. The radiation field intensity
obtained by the infrared water detector on the tunnel face and along the
tunnel are presented in Fig. 11. The maximum difference in the radia-
tion field intensity monitored on the tunnel face is 13 μW/cm2

Fig. 9. GPR profiles acquired for measuring lines L1 to L5 on the tunnel face at
chainage 100,765 m.

Fig. 11. Infrared radiation field intensity measured by IWD measurements.

73
K. Zhang et al.
Fig. 12. Layout of measuring lines for TEM measurements.
(Fig. 11(a)), which is larger than the safe value (10 μW/cm2). It could
be inferred that there was probably a water-bearing structure in front of
the tunnel face. The radiation field intensities increased along the
tunnel, and the maximum difference reached to 15 μW/cm2
(Fig. 11(b)). From these values, it could be inferred that the water
content in the strata increased toward and ahead of the tunnel face.
According to the IWD measurements results, water was inferred to exist
at chainage 100,765–100,795 m. These results suggested that TEM
measurement should be conducted to furtherly confirm the location(s)
of water-bearing structures and the approximate amount of water.
3.3.4. TEM measurements
TEM measurements with a prediction range of 300 m were con-
ducted on the tunnel face at chainage 100,657 m to make a semi-
quantitative prediction of the karst water. The layout for the TEM
measurement is shown in Fig. 12. A map of apparent resistivity of strata
ahead of the tunnel face is shown in Fig. 13. A low resistivity ab-
normality zone with the apparent resistivity less than 1250 Ω·m was
located at chainage 100,755–100,807 m. This zone was inferred as a
fault zone. The apparent resistivity was noticeably low at chainage
100,757–100,782 m (less than 300 Ω·m), indicating that large-scale
water-bearing structures were present and water was abundant. Ac-
cording to the TEM measurement results, the UKS ahead of the tunnel
face at 100,765 m was assessed as level v4. This resulted in a decision to
drill advance probe boreholes to detect karst features. Note that while
the prediction range of the TEM was set to 300 m in this case, in areas
with high UKS level, TEM should be used for short-distance predictions,
and a more appropriate prediction range is 50 m.
3.3.5. Advance probe boreholes measurements
As karst features and water had been predicted to exist ahead of the
tunnel face at 100,765 m, boreholes were drilled into the tunnel face to
further verify the presence of karst features and to discharge karst water
from the strata. The layout of the boreholes on the tunnel face at
chainage 100,765 is shown in Fig. 14, where the numbering of the
boreholes corresponded to that of drilling. The results from the bore-
holes are presented in Table 14. While drilling the boreholes, the water
stopped flowing out from boreholes B1 and B2 when water started
flowing out from borehole B4 and B5. This phenomenon indicated that
the water-bearing structures ahead of the tunnel face were well
Fig. 13. Contour map of apparent resistivity of TEM measurements.
74
Tunnelling and Underground Space Technology 80 (2018) 64–77
Fig. 14. Layout of seven probe boreholes on the tunnel face at chainage
100,765 m.
connected. No water flowed out from boreholes B6 and B7 when they
were drilled three days after borehole B4 was completed. This phe-
nomenon indicated that the water table was stable and was lowered
below the elevation of B4. Based on the drilling results, the geology
encountered by the boreholes ahead of the tunnel face is presented in
Fig. 15. The inferred karst features and the karst water was consistent
with the initial assessment of UKS (level v4).
3.3.6. Integrated analysis and prediction result
The combination of five different investigation methods, SRM, TEM,
GPR, IWD, and advance probe boreholes, predicted the presence of
karst features at chainage 100,765–100,795 m with high confidence.
The seismic wave reflection indicated that the rock mass at chainage
100,765–100,795 m was blocky. Karst caves or dissolution zones were
inferred to be located at chainage 100,782–100,791 m from the seismic
wave measurements. The TEM and the infrared water detector in-
dicated that karst water was abundant in this section. The GPR mea-
surements provided further information about the location of each
abnormality. The borehole drilling indicated that a connected karst
conduit passed through the tunnel alignment in the lateral direction
and its width extended more than 6 m longitudinally. This karst conduit
was filled with sand and gravels, and the karst water was abundant and
not directly related to the surface water. Based on the information
provided by the prediction methods, the sketches of predicted karst
features for specific cross sections and a map of the predicted conduit
were created as shown in Figs. 16 and 17. The karst features and karst
water inferred by the different techniques were consistent with each
other and with the initial UKS assessment. These predictions were used
to modify the excavation method and the ground support for this sec-
tion of the tunnel.
3.4. Tunnel support needed to cross through the karst features
For the critical chainage of 100,765–100,795 m, it was deemed
necessary to install support in advance of the tunnel face. Steel pipe
forepoling and grouting was used. The forepoling consisted of 38 steel
pipes that were 20 m long. The pipe diameter was 180 mm, and they
were circumferentially spaced at 0.4 m. The pipes were installed with
an upward inclination of 3° in a range of 120° around the tunnel roof as
shown in Fig. 18(a). Cement grout with a water-cement ratio of 1:1 was
K. Zhang et al. Tunnelling and Underground Space Technology 80 (2018) 64–77

Table 14
Borehole parameters and observed data.
Number Depth (m) Horizontal angle (°) Vertical angle (°) Maximum flow rate (L/s) Water pressure (MPa) Total water discharge (m3/d)

B1 30 Leftward offset with 1° Upward offset with 10° 10 0.05 864

B2 30 Rightward offset with 1° Upward offset with 10° 7 0.06 604.8
B3 30 0° 0° 13 0.05 1123.2
B4 30 Leftward offset with 1° Downward offset with 5° 20 0.05 1728
B5 30 Leftward offset with 1° Downward offset with 5° 15 0.05 1296
B6 30 0° 0° – – –
B7 30 0° 0° – – –

pumped through the steel pipes. Supplementary pipe grouting was

conducted to ensure that the cement grout permeated adequately into
the dissolution fissures and consolidated the filling materials to form a
protective shell outside the primary support as shown in Fig. 18(b). A
grout with water-cement-water glass ratios of 1:1:0.5 was pumped
through small (42 mm diameter), 4.5 m long pipes. These were cir-
cumferentially spaced at 0.4 m with an upward inclination angle of 45°,
within the range of 150° on the tunnel roof (90° left of the midline and
60° to the right). The pipe grouting was conducted after every 5 m of
excavation, and the pipe-shield was installed after 18 m was excavated.
The tunnel excavation method was changed from the original design
of a top heading and bench method to three benches and reserving core
soil method. While excavating through the area where karst features
were predicted to be present, dissolution fracture zones and a pipelined
karst cave were observed. The cave had been filled with the grouting
materials. The range of the observed grouted areas was approximately
consistent with that of the predicted dissolution zones, karst caves, and
Fig. 15. Borehole geology ahead of the tunnel face from chainage 100,765 to the connected karst conduits, which indicated the prediction results
100,795 m. were credible. The field observations also verified that the assessment
of a UKS level (v4) was reasonable.

4. Conclusions

An evaluation system is established to assess the underground karst

Fig. 16. Predicted karst features for specific cross sections.
state (UKS) along a tunnel alignment based on a fuzzy evaluation
method. Nine factors controlling the karst development are selected as
evaluation indices for this evaluation system. Triangular membership
functions were constructed to compute the membership of the evalua-
tion indices, and a fuzzy analytical hierarchy process was applied to
distribute the weights for the evaluation indices. The UKS level was
determined based on the fuzzy transform principle and the maximum
membership degree principle. To improve the accuracy of the predic-
tion, a combination of different geological investigation techniques is
proposed to predict the karst features ahead of tunnel face. The selec-
tion of appropriate techniques to use for feld investigations should be
guided by the UKS assessment. The geological investigation results are
used as the input to revise the UKS level, and then if the UKS level
increases, further field investigations should be conducted. A combi-
nation of different field investigation techniques allows for corrobora-
tion and verification of the results, thereby improving the confidence in
the prediction of karst features. The updated prediction method can
make the prediction more efficient and accurate.
The proposed method for predicting karst features in this paper has
limitations. To assess the UKS levels using the fuzzy evaluation method,
the values of some indices are obtained by expert evaluation. The
weights of the evaluation indices are also based on expert judgement.
Thus, the assessment of the UKS level contains some subjectivity.
The assessment system was applied to assess the UKS along a por-
tion of the Doupengshan tunnel line. About sixty percent of the studied
tunnel section was assessed to have medium or high UKS, with the rest
having a low or very low UKS level. There is a close relationship be-
tween the UKS and the formation lithology with low UKS levels cor-
responding to lower solubility rock formations. Areas with high UKS
level were influenced by the faults and were more jointed. To

75
K. Zhang et al.
Fig. 17. Map of predicted karst features at chainage 100,765–100,795 m (at elevation 942.3 m).
Fig. 18. Pipe forepoling and supplementary pipe grouting on tunnel face.
demonstrate the prediction of karst using fuzzy assessment combined
with geological investigations, karst features ahead of the tunnel face
were predicted in an area with high UKS level. The characteristics of the
karst features were analyzed and inferred based on the results of SRM,
GPR, IWD, TEM, and advance probe boreholes. The updated prediction
results agreed with the field observations of karst features after the
tunnel passed through the area. When predicting karst during tunnel-
ling, the selection of an appropriate technique to use for investigation
depends on previously investigations. The prediction results of one
technique not only influences the subsequent prediction but also affects
the updating of the UKS assessment.
Acknowledgments
This work was support by the Fundamental Research Funds for the
Central Universities of China [grant number SWJTU11ZT33] and the
Program for Innovative Research Team in University of Ministry of
Education of China [grant number IRT0955]. The financial support
from the China Scholarship Council to the first author is also gratefully
acknowledged.
References
Amberg Measuring Technique, 2002. TSP 203 Processing. Amberg Co., Switzerland.
Al-fares, W., Bakalowicz, M., Guérin, R., Dukhan, M., 2002. Analysis of the karst aquifer
structure of the Lamalou area (Hérault, France) with ground penetrating radar. J.
Appl. Geophys. 51 (2), 97–106.
Alijia, S., Torrijo, F.J., Quinta-Ferreira, M., 2013. Geological engineering problems as-
sociated with tunnel construction in karst rock mass: the case of Gavarres tunnel
(Spain). Eng. Geol. 157, 103–111.
76
Tunnelling and Underground Space Technology 80 (2018) 64–77
Alimoradi, A., Moradzadeh, A., Naderi, R., Salehi, M.Z., Etemadi, A., 2008. Prediction of
geological hazardous zones in front of a tunnel face using TSP-203 and artificial
neural networks. Tunn. Undergr. Space Technol. 23 (6), 711–717.
Asadollahi, P., Foroozan, R., 2006. Comparison of the evaluated rock mass properties
from the TSP system and the RMR classification (Semnan tunnel, Iran). Tunn.
Undergr. Space Technol. 21 (3) 236-236.
Bao, H.L., Chen, G.L., 1983. Karst Engineering Geology. China Railway Publishing House,
Beijing (in Chinese).
Casagrande, G., Cucchi, F., Zini, L., 2005. Hazard connected to railway tunnel con-
struction in karstic area: applied geomorphological and hydrogeological surveys. Nat.
Hazards Earth Syst. Sci. 5 (2), 243–250.
Chamberlain, A.T., Sellers, W., Proctor, C., Coard, R., 2000. Cave detection in limestone
using ground penetrating radar. J. Archaeol. Sci. 27 (10), 957–964.
Chang, D.Y., 1996. Applications of the extent analysis method on fuzzy AHP. Eur. J. Oper.
Res. 95 (3), 649–655.
Collins, M.E., Cum, M., Hanninen, P., 1994. Using ground-penetrating radar to investigate
a subsurface karst landscape in north-central Florida. Geoderma 61 (1–2), 1–15.
Conyers, L.B., Goodman, D., 1997. Ground-Penetrating Radar: an Introduction for
Archaeologists. AltaMira Press, Walnut Creek, California.
Dreybrodt, W., 2012. Processes in Karst Systems: Physics, Chemistry, and Geology.
Springer Science & Business Media.
Filipponi, M., 2009. Spatial analysis of karst conduit networks and determination of
parameters controlling the speleogenesis along preferential lithostratigraphic hor-
izons. Doctoral dissertation, École Polytechnique Fédérale de Lausanne, Lausanne,
Switzerland, pp. 130.
Filipponi, M., 2015. Karst risk assessment for underground engineering: Comparison of
the KarstALEA method with a random karst distribution approach. In: Lollino, G.,
Manconi, A., Guzzetti, F., Culshaw, M., Bobrowsky, P., Luino, F. (Eds.), Engineering
Geology Society and Territory, Springer International Publishing, pp. 603–607.
Filipponi, M., Jeannin, P.Y., 2010. Karst-ALEA: a scientific based karst risk assessment for
underground engineering. In: Andreo, B., Carrasco, F., Duran, J.J., LaMoreaux, J.W.
(Eds.), Advances in Research in Karst Media. Environmental Earth Sciences. Springer,
Berlin, Heidelberg, pp. 435–440.
Ford, D., Williams, D.P., 2007. Karst Hydrogeology and Geomorphology. John Wiley &
Sons, Chichester.
Gao, W., Qi, J., Xu, M., Li, Y., Wang, N., 2016. Influence of fractal characteristics of a
surface karst landform on tunnel construction. Modern Tunnell. Technol. 2, 35–43 (in
K. Zhang et al.
Chinese).
Galve, J.P., Remondo, J., Gutiérrez, F., 2011. Improving sinkhole hazard models in-
corporating magnitude–frequency relationships and nearest neighbor analysis.
Geomorphology 134 (1), 157–170.
Geophysical Survey Systems, 2001a. < http://www.geophysical.com/Documentation/
Manuals/SIR2000%20Manual_RevB.PDF>.
Geophysical Survey Systems, 2001b. < http://www.geophysical.com/software.htm>.
Goldscheider, N., Drew, D., 2007. Methods in karst hydrogeology. International
Contributions to Hydrogeology. Taylor & Francis, London.
Kaufmann, O., Quinif, Y., 2002. Geohazard map of cover-collapse sinkholes in the
‘Tournaisis’ area, southern Belgium. Eng. Geol. 65 (2), 117–124.
Khademi Hamidi, J., Shahriar, K., Rezai, B., Rostami, J., Bejari, H., 2010. Risk assessment
based selection of rock TBM for adverse geological conditions using Fuzzy-AHP. Bull.
Eng. Geol. Environ. 69 (4), 523–532.
Knödel, K., Lange, G., Voigt, H.J., 2007. Environmental Geology: Handbook of Field
Methods and Case Studies. Springer Science & Business Media.
LeGrand, H.E., Stringfield, V.T., 1973. Karst hydrology - a review. J. Hydrol. 20 (2),
97–120.
Li, S., Xue, Y., Zhang, Q., Li, S., Li, L., Sun, K., Ge, Y., Su, M., Zhong, S., Li, X., 2008. Key
technology study on comprehensive prediction and early-warning of geological ha-
zards during tunnel construction in high-risk karst areas. Chin. J. Rock Mech. Eng. 27
(7), 1297–1307 (in Chinese).
Li, S.C., Zhou, Z.Q., Li, L.P., Xu, Z.H., Zhang, Q.Q., Shi, S.S., 2013. Risk assessment of
water inrush in karst tunnels based on attribute synthetic evaluation system. Tunn.
Undergr. Space Technol. 38, 50–58.
Liu, Y., Huang, X., Duan, J., Zhang, H., 2017. The assessment of traffic accident risk based
on grey relational analysis and fuzzy comprehensive evaluation method. Nat. Hazards
88 (3), 1409–1422.
Lu, Z., Wu, L., Zhuang, X., Rabczuk, 2016. Quantitative assessment of engineering geo-
logical suitability for multilayer urban underground space. Tunn. Undergr. Space
Technol. 59, 65–76.
Lv, Q., Chen, J., 2010. Infrared acquisition technology applied to karst tunnels. Modern
Tunnell. Technol. 47 (4), 45–49 (in Chinese).
Mao, B.Y., Xu, M., Jiang, L.W., 2010. Preliminary study of risk assessment of water and
mud inrush in karst tunnel. Carsologica Sinica 29 (2), 183–189 (in Chinese).
Nezarat, H., Sereshki, F., Ataei, M., 2015. Ranking of geological risks in mechanized
tunnelling by using Fuzzy Analytical Hierarchy Process (FAHP). Tunn. Undergr.
Space Technol. 31 (50), 358–364.
Pesendorfer, M., Loew, S., 2004. Hydrogeologic exploration during excavation of the
Lötschberg base tunnel (AlpTransit Switzerland). Eng. Geol. Infrastruct. Plann.
Europe 347–358.
Saaty, T.L., 1979. Applications of analytical hierarchies. Math. Comput. Simul 21 (1),
1–20.
77
Tunnelling and Underground Space Technology 80 (2018) 64–77
Shi, S.S., Li, S.C., Li, L.P., Zhou, Z.Q., Wang, J., 2014. Advance optimized classification
and application of surrounding rock based on fuzzy analytic hierarchy process and
tunnel seismic prediction. Autom. Constr. 37, 217–222.
Sokolov, D.S., 1962. Fundamental Conditions for the Development of Karst.
Gosgeoltekhizdat, Moscow.
Sokolov, D.S., 1967. Hydrodynamic zoning of karst water. Hydrologie des roches
fissurées, AIHS–Unesco. Louvain 1, 204–207.
Stokes, T.R., Griffiths, P., 2000. A Preliminary Discussion of Karst Inventory Systems and
13 Principles (KISP) for British Columbia. BC Ministry of Forests, Lands and Natural
14 Resource Operations. < https://www.for.gov.bc.ca/hfd/pubs/docs/wp/wp51.
htm>. (accessed 15 May 2010).
Stringfield, V.T., Rapp, J.R., Anders, R.B., 1979. Effect of karst and geological structure on
the circulation of water and permeability in carbonate aquifers. J. Hydrol. 43,
313–332.
Sun, K.G., Li, S.C., Zhang, Q.S., Xue, Y.G., Li, S.C., Xu, Z.H., 2008. Application of the TSP
geological forecast method of a mountain tunnel in a karst zone. J. Shandong
University (Eng. Sci.) 38 (1), 1–6 (in Chinese).
Wang, Y., Chen, Q., Wei, Y.Y., Wang, H., 2003. Application of infrared acquisition
technology in prediction of water gushing in Yuanliangshan tunnel. Chin. J. Rock
Mech. Eng. 22 (5), 855–857 (in Chinese).
Waltham, A.C., Fookes, P.G., 2003. Engineering classification of karst ground conditions.
Q. J. Eng. Geol. Hydrogeol. 36 (2), 101–118.
Waltham, T., Bell, G.F., Culshaw, M., 2005. Sinkholes and Subsidence: Karst and
Cavernous Rocks in Engineering and Construction. Springer Science & Business
Media, Berlin.
Xu, Z.H., Li, S.C., Li, L.P., Hou, J.G., Sui, B., Shi, S.S., 2011. Risk assessment of water or
mud inrush of karst tunnels based on analytic hierarchy process. Rock Soil Mech. 6,
1757–1765 (in Chinese).
Xue, Y.G., Li, S.C., Zhang, Q.S., Li, S.C., Su, M.X., Liu, B., Liu, Q., 2008. Geological pre-
diction of karst-fractured groundwater in tunnel informational construction. Rock
Soil Mech. 29 (12), 3360–3364 (in Chinese).
Zhang, G.H., Jiao, Y.Y., Chen, L.B., Wang, H., Li, S.C., 2015. Analytical model for as-
sessing collapse risk during mountain tunnel construction. Can. Geotech. J. 53 (2),
326–342.
Zhang, Q., Tian, S., Mo, Y., Dong, X., Hao, S., 1993. An expert system for prediction of
karst disaster in excavation of tunnels or underground structures through a carbonate
rock area. Tunn. Undergr. Space Technol. 8 (3), 373–378.
Zhou, Z.Q., Li, S.C., Li, L.P., Shi, S.S., Song, S.G., Wang, K., 2013. Attribute recognition
model of fatalness assessment of water inrush in karst tunnels. Rock Soil Mech. 34
(3), 818–826 (in Chinese).
Zini, L., Chiara, C., Franco, C., 2015. The challenge of tunnelling through Mediterranean
karst aquifers: the case study of Trieste (Italy). Environ. Earth Sci. 74 (1), 281–295.