Accepted Manuscript

Geostatistical method for predicting RMR ahead of tunnel face

excavation using dynamically exposed geological information

Jianqin Chen, Xiaojun Li, Hehua Zhu

PII: S0013-7952(16)30849-3
DOI: doi: 10.1016/j.enggeo.2017.08.004
Reference: ENGEO 4613
To appear in: Engineering Geology
Received date: 21 December 2016
Revised date: 13 July 2017
Accepted date: 1 August 2017

Please cite this article as: Jianqin Chen, Xiaojun Li, Hehua Zhu , Geostatistical method for
predicting RMR ahead of tunnel face excavation using dynamically exposed geological
information. The address for the corresponding author was captured as affiliation for all
authors. Please check if appropriate. Engeo(2017), doi: 10.1016/j.enggeo.2017.08.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As
a service to our customers we are providing this early version of the manuscript. The
manuscript will undergo copyediting, typesetting, and review of the resulting proof before
it is published in its final form. Please note that during the production process errors may
be discovered which could affect the content, and all legal disclaimers that apply to the
journal pertain.
 ACCEPTED MANUSCRIPT

Geostatistical method for predicting RMR ahead of tunnel face

excavation using dynamically exposed geological information

Jianqin Chena, Xiaojun Lia,b,c,* , Hehua Zhua,c

a
Department of Geotechnical Engineering, College of Civil Engineering, Tongji University, 1239
Siping Road, Shanghai 200092, China
b
State Key Laboratory for Disaster Reduction in Civil Engineering, Tongji University, 1239 Siping

PT
Road, Shanghai 200092, China
c
College of Engineering, Tibet University, Tibet 850000, China
* Corresponding author

RI
SC
Abstract:

Rock Mass Rating (RMR) is a rock mass classication system that is often used
NU
to select appropriate excavation methods and rock support systems in tunnel projects.
MA

This paper presents a geostatistical method for predicting RMR values quantitatively

ahead of excavation of the tunnel face. The study makes full use of geological
ED

information exposed on excavated tunnel faces to capture the spatial correlation of

T

rock mass quality and later predicts the RMR value using the kriging method.
EP

Predictions are constantly updated during the tunnel construction process. The
C

advantages of the proposed method are as follows: (1) The RMR prediction
AC

uncertainty is quantified by accounting for spatial variability and model uncertainty;

therefore, the resulting prediction can consider the geological conditions of the worst

scenario; (2) The spatial variability of the geological condition is represented as a

variogram model that is updated by observation data on the new excavated faces, and

as the tunnel advances, the RMR prediction accuracy improves; and (3) The

periodicity of geological conditions can be considered in RMR prediction. The

1
 ACCEPTED MANUSCRIPT

proposed method is applied to a rock tunneling project, the Mingtang tunnel in Anhui

province, China. The method achieves approximately 80% prediction accuracy;

therefore, it has high potential as a tool for predicting RMR information ahead of

tunnel face excavation.

PT
Keywords: Rock Mass Rating Prediction; Spatial Variability; Model Uncertainty;

RI
Dynamic Updating

SC
NU
MA
T ED
C EP
AC

2
 ACCEPTED MANUSCRIPT

List of Tables
Table. 1 Measured RMR values provided by the excavated tunnel faces during

construction.

Table 2. Variogram models parameters of the components using two kriging

estimation methods.

PT
Table 3. Cross-validation criteria for two estimation methods.

RI
Table 4. Period of the variogram models for two kriging estimation methods using

SC
nested hole-effect model. NU
Table 5. Cross-validation criteria for two estimation methods considering periodicity.
MA
T ED
C EP
AC

3
 ACCEPTED MANUSCRIPT

List of Figures

Fig. 1. Location of the Mingtang Tunnel in China.

Fig. 2. (a) Three-dimensional geological model of the Mingtang Tunnel (granite is

represented by cyan, and gneiss is represented by yellow); and (b) Positions of

PT
the measured RMR values provided by excavated tunnel faces.

Fig. 3. Frequency distribution histogram of the Rock Mass Rating.

RI
Fig. 4. RMR estimation interval.

SC
Fig. 5. RMR prediction with increasing number of data; the actual positions of the
NU
circles are detailed in Table 1.
MA

Fig. 6. Variogram fitting at four positions (K21+170, K21+422, K21+612 and

K22+381); for each prediction position, three experimental variogram models

ED

are calculated with the number of data equal to eight, fourteen and twenty; for

each experimental variogram model, the bottom line represented by purple is

T
EP

fitted using the lower bound of 2f , the top line represented by green is fitted

using the upper bound of 2f and the middle line represented by blue is fitted
C
AC

by automatic fitting; (a) K21+170; (b) K21+422; (c) K21+612; and (d)

K22+381.

Fig. 7. Kriging prediction interval considering model uncertainty; the blue line and

black line are the estimation results using the upper and lower bounds of the

variogram models; the red line is the true value calculated by the in situ tunnel

face mapping; (a) K21+170; (b) K21+422; (c) K21+612; and (d) K22+381.

4
 ACCEPTED MANUSCRIPT

Fig. 8. Kriging variance at four prediction positions; the blue line and black line are

Kriging variance using the upper and lower bounds of the variogram models; (a)

K21+170; (b) K21+422; (c) K21+612; and (d) K22+381.

Fig. 9. Prediction interval at four prediction positions with consideration of model

uncertainty and spatial variability; the blue line is the upper bound of the

PT
estimation, the black line is the lower bound of estimation and the red line is the

RI
true value calculated by in situ tunnel face mapping; (a) K21+170; (b) K21+422;

SC
(c) K21+612; and (d) K22+381. NU
Fig. 10. RMR prediction intervals; (a) Model One: model uncertainty; and (b) Model

Two: model uncertainty and spatial variability.

MA

Fig. 11. Variogram model fitting; (a) RMRbasis; (b) joint condition (JC); (c) RQD; (d)

joint and bedding spacing (JS); (e) uniaxial compressive strength (UCS); and (f)
ED

groundwater condition (GW).

T

Fig. 12. Scatterplot of the measured values and predicted values for two prediction
EP

methods.
C

Fig. 13. Variogram model fitting considering periodicity; (a) RMRbasis; (b) joint
AC

condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial

compressive strength (UCS); and (f) groundwater condition (GW).

Fig. 14. Spectral analysis on RMRbasis and RMR components; (a) RMRbasis; (b) joint

condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial

compressive strength (UCS); and (f) groundwater condition (GW).

5
 ACCEPTED MANUSCRIPT

1. Introduction

Geological bodies are characterized by complexity, uncertainty and

heterogeneity after complex geological processes. Drilling investigations, in situ tests

and surface geophysical investigations in the exploratory phase of a tunneling project

are often limited and suffer from rough topography and difcult logistic conditions in

PT
mountainous terrain (Petronio et al., 2007). This may lead to large uncertainty and

RI
major geological risk in the construction phase of tunneling. Therefore, the rock mass

SC
classication systems known as empirical methods are still used in current
NU
engineering design and construction.

One of the most common classification systems is the Rock Mass Rating (RMR)
MA

that was developed by Bieniawski (1973, 1989). The RMR system incorporates six

parameters including uniaxial compressive strength (UCS) of rocks, Rock Quality

ED

Designation (RQD), joint and bedding spacing (JS), joint condition (JC), groundwater
T

condition (GW), and orientation of discontinuities with respect to the opening axis.
EP

Each of these parameters is classied and given a value or rating to express its
C

inuence on tunnel stability (Palmstrm, 2009). The RMR system is effective for
AC

evaluating the quality of rock mass surrounding the tunnel. It can be used to assist

engineers in proposing suitable excavation techniques and support systems in the

process of tunneling.

Regarding RMR prediction ahead of tunnel excavation, it is well known that

actual rock mass class may not be consistent with the design value because of the

6
 ACCEPTED MANUSCRIPT

variability of rock mass quality. Therefore, predicting the RMR value ahead of tunnel

excavation is crucial for selection of an appropriate rock support system, as well as to

plan a safe and economically efficient excavation method. Typically, RMR prediction

values range from 10 to 50 meters ahead of the tunnel face, and this is significant

because about one week is needed to adjust the excavation method to the geological

PT
conditions ahead. Currently, tunnel engineers usually rely on the seismic prediction

RI
method or ground penetration radar to predict rock mass quality and hazardous

SC
geological structures ahead of tunnel excavation. However, the inspection results of
NU
the seismic prediction method or ground penetration radar are qualitative, so they

must be interpreted by experienced geophysicists. The dynamic drill and blast

MA

excavation method provides a great opportunity for engineers and geologists to

observe and sample the rock mass on the tunnel faces. The RMR values can be
ED

calculated according to the exposed tunnel face. Moreover, tunnel construction is a

T

dynamic process. As the tunnel advances, new geological information is added

EP

constantly. However, this information has not been put into good use.
C

A geostatistical approach has been applied to estimate RMR in different kinds of

AC

site characterizations. For tunnel site characterization, most current research uses

drillhole data (Kaewkongkaew et al., 2015) to interpolate RMR values directly for the

whole domain in the exploratory phase of tunnel project. Ferrari et al. (2014)

performed the ordinary kriging estimation to a valley area. These studies focus mainly

on the exploratory phase of a project. This means that the point-wise measurements

7
 ACCEPTED MANUSCRIPT

from drill holes or geological eld surveys are used to interpolate the whole domain.

As the estimation process is conducted only once, the variogram model is constant.

However, new geological information could be gathered constantly as the tunnel

advances. Therefore, the variogram model can be updated by using observation data

obtained from newly excavated tunnel faces. In addition, model uncertainty that is

PT
introduced in the process of experimental variogram fitting is not considered in the

RI
conventional analysis, which generates prediction uncertainty. The lower bound of the

SC
RMR estimation value represents the possible worst scenario in tunneling. Tunnel
NU
excavation plan based on this result would be considered extremely conservative. The

upper bound of RMR estimation value represents the best possible scenario, so tunnel
MA

excavation plans based on this result would be considered a high-risk venture. In

addition, geological phenomenon often occurs repetitively over geological time,

ED

leading to repetitive or cyclic variations in the facies and petrophysical properties

T

(Pyrcz and Deutsch, 2014). The variogram fitting of RMR and its components must
EP

also consider this periodicity.

C

In summary, this research focuses on applying geostatistics to quantitatively

AC

predict RMR values ahead of the excavation face during tunnel construction. This

research offers several contributions: (1) The RMR prediction is based on dynamic

observations of tunnel faces during excavation; (2) The RMR prediction uncertainty is

quantified by taking into account model uncertainty and spatial variability; and (3)

The periodicity of geological conditions can be considered in the RMR prediction.

8
 ACCEPTED MANUSCRIPT

This paper is organized by introducing the background of a rock tunneling

project in Section 2, presenting a geostatistical method for predicting RMR values

quantitatively ahead of the tunnel face in Section 3, discussing the application of the

method in Section 4, and presenting several conclusions in Section 5.

PT
2. Site Description

RI
To demonstrate application of the geostatistical method for RMR prediction for

SC
tunnel engineering, the Mingtang Tunnel is selected as a case study. Mingtang Tunnel
NU
is a rock tunnel situated in Yuexi County, Anhui Province, China (Fig. 1). This is part

of the Yue-wu highway that connects Wuhan city and Shanghai city. The tunnel was
MA

excavated using the drill and blast method. The total length of the tunnel is 7.548 km,

and the maximum burial depth is approximately 562 m. As shown in Fig. 2(a), the
ED

geology of the site area is mainly composed of granite and gneiss. Three major faults
T

transect rock formations in the tunnel alignment. The study area is in the interval
EP

between K17+599 and K24+708. During tunnel construction, geological information

C

is sampled from the excavation faces in forty-three discrete positions, and their
AC

corresponding rock mass quality is evaluated. The positions and measured RMR

values are shown in Table 1. Fig. 2(b) shows the positions of the tunnel face relative

to where the in situ tunnel face mapping is obtained.

3. Methodology

9
 ACCEPTED MANUSCRIPT

3.1 Photogrammetry to obtain RMR values

Dynamic drill and blast excavation provides a useful opportunity for engineers

and geologists to observe and sample rock mass. Most difficult, however, is obtaining

the discontinuity information safely and efficiently. In our study, RQD value, joint and

bedding spacing (JS) and joint condition (JC) (except inlling and weathering) of

PT
actual RMR values can be obtained using the photogrammetry-based tunnel face

RI
mapping technique (Haneberg, 2008; Sturzenegger et al., 2009; Chen et al., 2016; Li

SC
et al., 2016; Zhu et al., 2016). Uniaxial compressive strength (UCS) of rock mass is
NU
measured through the uniaxial compression test. Groundwater condition (GW),

inlling and weathering are obtained by field observations.

MA

3.2 Geostatistical method

Adjacent areas show similar geological conditions, so the RMR data sampled
ED

from locations closer together tend to be more similar than those far apart. This spatial
T

correlation can be quantied through a variogram, which is a measure of geological

EP

variability. Variogram increases as data become more dissimilar. Therefore, the

C

parameters of the variogram best fitting the data are first conducted using R software
AC

(R. Core Team, 2015) with the GSTAT package (Pebesma, 2004). Then, Kriging is

used to provide an unbiased estimation, with uncertain quantication.

3.2.1 Exploratory spatial data analysis

The RMR values range from 44 to 74 with a mean of 61.5 and a standard

deviation of 7.77. The frequency distribution is shown in Fig. 3. The normality of

10
 ACCEPTED MANUSCRIPT

RMR distribution is veried using the Shapiro-Wilk test (Shapiro and Wilk, 1965).

The Gaussian distribution of RMR is conrmed with a signicance level of 5%.

Moreover, trend analysis results show that there is no trend in the RMR. This

confirms the stationarity hypothesis of RMR in the studied domain.

3.2.2 Prediction with model uncertainty and spatial variability

PT
3.2.2.1 Model uncertainty

RI
Construction of the semivariogram, which is a mathematical model that captures

SC
the spatial correlation among data, is a very important step in any geostatistical
NU
analysis. The semivariogram is a measure of variability that increases as samples

become more dissimilar. The variogram is dened as the expected value of a squared
MA

difference (Isaaks and Srivastava, 1989):

2 h Var[Z ( x) Z ( x h)] E{[ Z ( x) Z ( x h)]2} (1)

ED

where Z is a stationary random function with known mean m and variance 2 ,

T

which is independent of location, so m( x) m and 2 ( x ) 2 for all locations x

EP

in the study area.

C

Spatial correlation is assessed by calculating the experimental variogram, then

AC

fitting with different parametric variogram models, such as the Spherical model,

Exponential model, and Gaussian model. There are four main approaches for

estimating the parameters of the variogram model: visual, least square regression

(Trangmar et al., 1985), weighted least squares (Cressie, 1985), and likehood methods

(Cressie, 1991). However, the process of variogram model fitting introduces some

11
 ACCEPTED MANUSCRIPT

degree of smoothing and subjectivity. These elements can introduce parametric

uncertainty (Rubin, 2003).

The method of least squares is used to fit the rising limb of the parametric

variogram to the experimental variogram, while holding the sill value equal to the

sample variance, so that range can be calculated. The sample variance is treated as a

PT
random variable. Hence, the uncertainty in determining the range is also considered

RI
by least square regression. Estimation of the sample variance is subject to uncertainty,

SC
which is estimated by Rehfeldt et al. (1992):

2 4
NU
Var ( 2f ) f (2)
N*
where 2f is the sample variance and N * is the number of independent
MA

measurements that can be obtained by dividing the tunnel into independent segments.

When assuming the sample variance to be chi-square distributed, the statistical

ED

distribution of the estimation error is normally distributed and an approximately 95%

T

confidence region can be constructed about 2f , as

EP

2f 2[Var( 2f )]1/2 2f 2f 2[Var( 2f )]1/2 (3)

C

Two theoretical variogram models can be fitted by using the lower and upper
AC

limits of 2f , and determining the corresponding ranges automatically. Kriging

estimation of these two theoretical variogram models generates the corresponding

upper and lower bounds of RMR values ( RMRupper and RMRlower ) that accounts for

the model uncertainty.

3.2.2.2 Spatial variability

12
 ACCEPTED MANUSCRIPT

In geostatistical estimation theory, the associated standard error for each

estimation is a measure of the possible variation in value, and hence the uncertainty of

the estimated value. If the standard error is small, then the possible statistical variation

is corresponding small; if it is large, then the possible variation is large.

When only considering the spatial variability, the estimation interval of RMR can

PT
be expressed as following with 68% confidence:

RI
RMR RMR RMRPred RMR RMR (4)

SC
where RMR is the mean and RMR is standard deviation of the kriging estimation.
NU
This prediction result only considers the spatial variability.

3.2.2.3 Model uncertainty and spatial variability

MA

As is shown in Fig. 4, when considering both the model uncertainty and spatial

variability, the estimation interval of RMR can be expressed as following:

ED

min{RMRlower lower , RMRupper upper } RMRPred max{RMRlower lower , RMRupper upper } (5)
T

where RMRupper and RMRlower are the kriging estimated values using the bounds of
EP

theoretical variogram models (defined in Eq. 3), and upper and lower are the
C

corresponding standard errors.

AC

3.2.3 RMR prediction

The spatial variability can be updated using observation data on newly excavated

tunnel faces. To evaluate the dynamic update effect of RMR prediction, different

number of observation data were used to construct the variogram model. Then, the

prediction results at the same position were compared.

13
 ACCEPTED MANUSCRIPT

As illustrated in Fig. 5, red circles represent the tunnel positions where

predictions are made and green circles represent the excavated tunnel positions with

field measured RMR values. The actual positions of the circles are detailed in Table 1.

The positions of the circles relative to the tunnel are shown in Fig. 2(b). There are

three clouds with different colors next to each of the red circles, and each cloud

PT
contains different numbers of observation data ranging from eight, fourteen and

RI
twenty. Three variogram models are constructed based on the observation data in the

SC
clouds and they predict the RMR values at the position of the red circle. The goal is to
NU
show how the RMR prediction changes as the number of measurements increase. The

prediction position advances to different red circles, such as tunnel excavation. For
MA

each prediction, the number of data surrounded by the clouds is the same.

3.2.4 Two kriging estimation methods: Direct Method and Indirect Method
ED

Two types of prediction strategies are conducted in our study. The first approach
T

is called the Direct Method. In this approach, the base RMR value is fit with a
EP

variogram model and then estimated directly. The base RMR is composed of five
C

parameters:
AC

RMRbasis UCS RQD JS JC GW (6)

The last parameter (orientation of discontinuities with respect to the opening axis) is

excluded from the kriging estimation because it depends on the characteristics of rock

discontinuities, as well as their relation to the tunnel structure, which is unknown

(Pinheiro et al., 2016).

14
 ACCEPTED MANUSCRIPT

The second approach is called the Indirect Method. In this approach, parameters

composed of RMR are estimated separately as individual variables. The final

estimated RMR is obtained by adding the ratings of each variable. The kriging

estimation Z ( RMR2 ) and kriging variance RMR

2
2
are represented as

Z ( RMR2 ) Z (UCS ) Z ( RQD) Z ( JS ) Z ( JC ) Z (GW ) (7)

PT
RMR
2
2
UCS
2
RQD
2
JS2 JC
2
GW
2

2cov(UCS , RQD ) 2cov(UCS , JS ) 2cov(UCS , JC ) 2cov(UCS , GW )

RI
(8)
2cov( RQD, JS ) 2cov( RQD, JC ) 2cov( RQD, GW )

SC
2cov( JS , JC ) 2cov( JS , GW ) 2cov( JC , GW )

where UCS
2
, RQD
2
, JS
2
, JC
2
and GW
2
are the kriging variances for each component.
NU
MA

4. Results and Discussion

4.1 Model uncertainty and spatial variability

ED

Fig. 6 shows the fitted results of variogram models at four positions, respectively.
T

For each prediction position, three experimental variogram models are calculated with
EP

the number of observation data equaling to eight, fourteen and twenty. The aim of
C

these models is to investigate how prediction accuracy changes with the number of
AC

observation data.

In addition, the experimental variogram model is fitted with three theoretical

variogram models. The bottom line (represented by purple) is generated using the

lower bound of 2f , the top line (represented by green) is generated using the upper

bound of 2f , and the middle line (represented by blue) is generated by automatic

fitting. The bottom line and top line almost encapsulate the variation range of the sill
15
 ACCEPTED MANUSCRIPT

of the variogram model.

As illustrated in Fig. 6, the variogram models change with increasing

measurement data. In Fig. 6(a), the sills of the variogram models decrease gradually.

The sills in Fig. 6(b) and Fig. 6(c) increase first and decrease later. In contrast, the

sills increase gradually in Fig. 6(d). Therefore, with newly added data, the variogram

PT
model is more consistent with the actual situation, meaning that the uncertainty of the

RI
variogram model decreases.

SC
Fig. 7 shows the Kriging estimation result with a different number of observation
NU
data. The blue line and black line are the estimation results using the upper and lower

bounds of the variogram models. The red line is the true value calculated by the in
MA

situ tunnel face mapping. The prediction interval considers the variogram model

uncertainty described in Section 3.2.2. The prediction interval covers the true value
ED

gradually with an increase of the measured data, which indicates that prediction
T

accuracy increases.
EP

Fig. 8 shows the corresponding kriging variance using upper and lower bounds
C

of variogram models. The upper and lower bounds of kriging variance are used to
AC

quantify spatial variability. The final RMR estimation interval allows for both model

uncertainty and spatial variability, as shown in Fig. 9. When the spatial variability is

incorporated into the prediction result, the prediction interval increases to cover the

true value.

The prediction result of the eight positions is summarized in Fig. 10. Model One

16
 ACCEPTED MANUSCRIPT

refers to prediction that considers only model uncertainty. Model Two refers to

prediction that considers both model uncertainty and spatial variability. Fig. 10(a)

shows the RMR prediction intervals for Model One. Fig. 10(b) shows the prediction

intervals for Model Two. In Fig. 10, the prediction intervals decrease with an increase

of measurement data, especially at the position K21+170. For the position K21+170,

PT
when the number of measurement data is eight, the upper bound of 2f is large,

RI
which leads to a large RMR estimation interval with the lower bound of the estimated

SC
RMR value close to zero. When the number of measurement data increases to twenty,
NU
the upper bound of 2f decreases, which leads to a smaller RMR estimation interval

with the lower bound of estimated RMR value approximating to the true value.
MA

Reduction of the prediction interval does not decrease the prediction accuracy.

4.2 Comparison of the Direct Method and Indirect Method

ED

For the Indirect Method, the components of RMR are estimated separately as
T

individual variables and fitted with separate variogram models from Fig. 11(b) to Fig.
EP

11(f). The parameters of these variogram models are listed in Table 2. For the
C

variogram model of each component, the range of GW is the largest, the ranges of
AC

UCS are secondary, and the ranges of JC, RQD and JS are the smallest. Because

ground water can flow in the fractures of rock mass, the influence area is larger when

comparing the fracture parameters (JC, RQD and JS) and strength parameter (UCS).

However, the fractures are generated by tectonic actions and thus show larger

variability and heterogeneities in the domain. Moreover, the sill of the GW is largest

17
 ACCEPTED MANUSCRIPT

because three faults traverse the Mingtang tunnel and dramatically change the

groundwater conditions. There is abundant ground water neighboring the faults, while

there is limited ground water in the regions far away from the tunnel.

In the Direct Method, the basic RMR value (defined in equation 6) is fitted with

a variogram model, as is shown in Fig. 11(a). The sill of RMRbasis (67.00) is affected

PT
by the largest sill of its component, which is GW with the value of 55.55. In addition,

RI
the range of RMRbasis (171.33) is affected by the smallest sill of its component,

SC
which is JC with a value of 175.45. NU
Fig. 12 shows the scatterplot between measured values and predicted values for

two estimation methods. It reveals a near- linear positive relationship between the
MA

measured value and predicted value for the two models. The correlation coefficients

for the Indirect Method and Direct Method are 0.73 and 0.7, respectively. Three
ED

performance metrics, the mean error (ME), the root mean squared error (RMSE) and
T

the mean standardized prediction error (MSPE), are used to evaluate the overall
EP

predictive accuracy. The ME should be close to zero. The RMSE measures the
C

difference between the predicted values and measured values, and this should be as
AC

small as possible. The MSPE is standardized by dividing ME by the kriging variance.

An accurate model should have an MSPE close to zero. As shown in Table 3, the

Indirect Method has the smaller absolute values of ME and RMSE, with the MSPE

value being closer to zero. Therefore, the Indirect Method has better estimation

accuracy for the case study.

18
 ACCEPTED MANUSCRIPT

4.3 Periodicity in the variogram models

As is shown in Fig. 13, the experimental variograms exhibit a sinusoidal wave

that conveys the cyclicity of the underlying geological phenomenon. This

phenomenon is sometimes referred to as the hole effect (Journel and Huijbregts, 1978;

Webster and Oliver, 2007). Hole effect is representative of a periodic phenomenon

PT
and is characterized by undulations in the variogram. Therefore, nested variograms

RI
composed of the Spherical model, Periodic model and Hole-effect model are

SC
constructed to fit the experimental variogram models of the whole RMR and each
NU
component of RMR.

Spectral analysis (Lomb, 1976) is used to detect cyclicities in the RMRbasis and
MA

RMR component. As shown in Fig. 14(a), the power spectral density of RMRbasis is

dominated by three significant peaks at frequencies of 0.0003, 0.0006 and 0.0021 that
ED

correspond to periods of 1421.8, 507.8 and 122.6 meters, respectively. In the RMR
T

component, there are also strong cyclicities indicated by peaks in the power spectral
EP

density. As shown in Table 4, the periods computed by the variogram model are
C

similar to the periods computed by the spectral analysis. This cyclicity suggests a
AC

strong inuence of tectonic actions of the geological body on the quality of rock mass.

Cross validation is performed to test the fit of the nested models to the data.

Their correlation coefficients both increase to 0.76. Three performance metrics are

used to evaluate the overall predictive accuracy. As is shown in Table 5, the Indirect

and Direct Methods have almost similar estimation accuracies for ME, RMSE and

19
 ACCEPTED MANUSCRIPT

MSPE values after considering the periodicity in the variogram models. Furthermore,

prediction results of the Indirect Method, with the exception of RMSE, is better after

incorporating the periodicity model, in comparison to Table 3. This indicates that

integrating periodicity into the variogram increases prediction accuracy. However,

when the Indirect Method considers periodicity, the variability of the prediction

PT
increases as each component is predicted separately.

RI
SC
5. Conclusions NU
This paper presents a geostatistical method for predicting RMR values

quantitatively ahead of tunnel face excavation by making full use of the geological
MA

information collected from the tunnel face and considering the dynamic construction

process of tunneling. RMR prediction accuracy is evaluated by considering spatial

ED

variability and model uncertainty, newly added geological data, and periodicity of the
T

geological condition.
EP

First, RMR prediction uncertainty is quantified by accounting for spatial

C

variability and model uncertainty. Model uncertainty is derived from the variogram
AC

model fitting process, and spatial variability originates from variation in the estimated

value. The resulting RMR prediction interval incorporates the best scenario case and

worst scenario case. The tunnel excavation method is based on the limits of the

interval representing different levels of risk. Furthermore, considering the model and

spatial variability also increases the prediction accuracy.

20
 ACCEPTED MANUSCRIPT

Second, RMR prediction accuracy increases with newly added geological data.

Model uncertainty decreases with newly added geological data.

Finally, RMR prediction is implemented in two approaches: the Direct Method

and Indirect Method. The Direct Method estimates the RMR value directly, while the

Indirect Method estimates the components constituting the RMR system separately. In

PT
our case study, the Indirect Method shows better estimation accuracy than the Direct

RI
Method. Moreover, the nested variograms composed of Spherical model, Periodic

SC
model and Hole-effect models perform relatively well in terms of prediction accuracy
NU
for both the Indirect Method and Direct Method. When the Indirect Method considers

periodicity, variability increases because each component is predicted separately.

MA

Acknowledgments
ED

The authors gratefully acknowledge support from the Natural Science

T

Foundation of China (NSFC 41272289, 41130751), the Science and Technology Plan
EP

Project of the Ministry of Transport of China (2013318J02120), and the Fundamental

C

Research Funds for Central Universities. Special thanks are due to Prof. Yoram Rubin
AC

from University of California, Berkeley for his valuable instructions regarding the

geostatistical method.

References

Alimoradi, A., Moradzadeh, A., Naderi, R., Salehi, M.Z., Etemadi, A., 2008.

21
 ACCEPTED MANUSCRIPT

Prediction of geological hazardous zones in front of a tunnel face using TSP-203

and artificial neural networks. Tunnelling and Underground Space Technology

23(6), 711-717.

Amberg Measuring Technique, 2002. TSP 203 Processing. Amberg Co.,Switzerland.

Bieniawski, Z.T., 1973, Engineering classification of jointed rock masses: Transaction

PT
of the South African Institution of Civil Engineers, v. 15, pp. 335-344.

RI
Bieniawski, Z.T., 1989, Engineering rock mass classifications: a complete manual for

SC
engineers and geologists in mining, civil, and petroleum engineering, New York,
NU
Wiley, xii, 251pp.

Cressie, N., 1985. Fitting variogram models by weighted least squares.Journal of the
MA

International Association for Mathematical Geology, 17(5), 563-586.

Cressie, N., 1991. Statistics for spatial data. John Wiley & Sons, NY, pp.900.
ED

Chen, J., Zhu, H., Li, X., 2016. Automatic extraction of discontinuity orientation from
T

rock mass surface 3D point cloud. Computers & Geosciences 95, 18-31.
EP

Choi, J.Y., Lee, C.I., 2007. An estimation of rock mass rating using 3D- indicator
C

kriging approach with uncertainty assessment of rock mass classication. In:

AC

Proceedings of the 11th congress of the international society for rock mechanics,

Lisbon, vol. 2, pp. 1285-88.

Choi, Y., Yoon, S.Y., Park, H.D., 2009. Tunneling analyst: a 3D GIS extension for

rock mass classification and fault zone analysis in tunneling. Computers &

Geosciences 35(6), 1322-1333.

22
 ACCEPTED MANUSCRIPT

Exadaktylos, G., Stavropoulou, M., 2008. A specific upscaling theory of rock mass

parameters exhibiting spatial variability: analytical relations and computational

scheme. International Journal of Rock Mechanics and Mining Sciences 45(7),

1102-1125.

Ferrari, F., Apuani, T., Giani, G.P., 2014. Rock Mass Rating spatial estimation by

PT
geostatistical analysis. International Journal of Rock Mechanics and Mining

RI
Sciences 70, 162-176.

SC
Gholami, R., Rasouli, V., Alimoradi, A., 2013. Improved RMR rock mass
NU
classification using artificial intelligence algorithms. Rock mechanics and rock

engineering 46(5), 1199-1209.

MA

Haneberg, W.C., 2008. Using close range terrestrial digital photogrammetry for 3-D

rock slope modeling and discontinuity mapping in the United States. Bulletin of
ED

Engineering Geology and the Environment 67(4), 457-469.

T

Isaaks, E.H., Srivastava, R.M., 2001. An introduction to applied geostatistics. 1989.

EP

New York, USA: Oxford University Press. Jones DR, A taxonomy of global
C

optimization methods based on response surfaces. Journal of Global Optimization,

AC

23, 345-383.

Journel, A.G, Huijbregts, C.J., 1978. Mining Geostat. NewYork, NY: Academic Press,

Inc.

Kaewkongkaew, K., Phien-Wej, N., Kham-ai, D., 2015. Prediction of rock mass along

tunnels by geostatistics. KSCE Journal of Civil Engineering 19(1), 81-90.

23
 ACCEPTED MANUSCRIPT

Lomb, N.R., 1976. Least-squares frequency analysis of unequally spaced data.

Astrophysics and space science, 39(2), 447-462.

La Pointe, P.R., 1980, January. Analysis of the spatial variation in rock mass

properties through geostatistics. In The 21st US Symposium on Rock Mechanics

(USRMS). American Rock Mechanics Association.

PT
Palmstrm, A., 2009. Combining the RMR, Q, and RMi classification systems.

RI
Tunnelling and Underground Space Technology, 24(4), 491-492.

SC
Pebesma, E.J., 2004. Multivariable geostatistics in S: the gstat package. Computers &
NU
Geosciences, 30(7), 683-691.

Pyrcz, M.J. and Deutsch, C.V., 2003. The whole story on the hole effect.Geostatistical
MA

Association of Australasia, Newsletter, 18, pp.3-5.

Pyrcz, M.J. and Deutsch, C.V., 2014. Geostatistical reservoir modeling. Oxford
ED

university press, New York, NY, pp.87.

T

R Core Team: R: A Language and Environment for Statistical Computing, R

EP

Foundation for Statistical Computing, Vienna, Austria, available at:

C

https://www.R-project.org/ (last access: 16 December 2015), 2015.

AC

Santos, V., da Silva, A.P.F., Brito, M.G., 2015. Prediction of RMR Ahead Excavation

Front in D&B Tunnelling. In Engineering Geology for Society and

Territory-Volume 6 (pp. 415-419). Springer International Publishing.

Stavropoulou, M., Exadaktylos, G., Saratsis, G., 2007. A combined three-dimensional

geological- geostatistical- numerical model of underground excavations in rock.

24
 ACCEPTED MANUSCRIPT

Rock mechanics and rock engineering 40(3), 213-243.

Trangmar, B.B., Yost, R.S. and Uehara, G., 1986. Application of geostatistics to

spatial studies of soil properties. Advances in agronomy, 38, 45-94.

You, K., Lee, J. S., 2006. Estimation of rock mass classes using the 3-dimensional

multiple indicator kriging technique. Tunnelling and Underground Space

PT
Technology 21(3), 229-229.

RI
Zhu, H., Wu, W., Chen, J., Ma, G., Liu, X., Zhuang, X., 2016. Integration of three

SC
dimensional discontinuous deformation
NU analysis (DDA) with binocular

photogrammetry for stability analysis of tunnels in blocky rockmass. Tunnelling

and Underground Space Technology 51, 30-40.

MA

Li, X., Chen, J., Zhu, H., 2016. A new method for automated discontinuity trace

mapping on rock mass 3D surface model. Computers & Geosciences 89, 118-131.
ED

Petronio, L., Poletto, F., Schleifer, A., 2007. Interface prediction ahead of the
T

excavation front by the tunnel-seismic-while-drilling (TSWD) method.

EP

Geophysics 72(4), 39-44.

C

Egaa, M., Ortiz, J.M., 2013. Assessment of RMR and its uncertainty by using
AC

geostatistical simulation in a mining project. Journal of GeoEngineering 8(3),

83-90.

Rehfeldt, K.R., Boggs, J.M., Gelhar, L.W., 1992. Field study of dispersion in a

heterogeneous aquifer: 3. Geostatistical analysis of hydraulic conductivity. Water

Resources Research 28(12), 3309-3324.

25
 ACCEPTED MANUSCRIPT

Sturzenegger, M., Stead, D., 2009. Quantifying discontinuity orientation and

persistence on high mountain rock slopes and large landslides using terrestrial

remote sensing techniques. Natural Hazards and Earth System Science 9(2),

267-287.

Pinheiro, M., Vallejos, J., Miranda, T., Emery, X., 2016. Geostatistical simulation to

PT
map the spatial heterogeneity of geomechanical parameters: A case study with rock

RI
mass rating. Engineering Geology 205, 93-103.

SC
Webster, R., Oliver, M.A., 2007. Geostatistics for environmental scientists. John
NU
Wiley & Sons.

Zhang, L., Einstein, H.H., 1998. Estimating the mean trace length of rock
MA

discontinuities. Rock Mechanics and Rock Engineering 31(4), 217-235.

Rubin, Y., 2003. Applied stochastic hydrogeology. Oxford University Press, New
ED

York, NY, 56pp.

T

Shaphiro, S.S., Wilk, M.B., 1965. An analysis of variance test for

EP

normality. Biometrika, 52(3), 591-611.

C
AC

26
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
Fig. 1. Location of the Mingtang Tunnel in China.
MA
T ED
C EP
AC

Fig. 2. (a) Three-dimensional geological model of the Mingtang Tunnel (granite is

represented by cyan, and gneiss is represented by yellow); and (b) Positions of the

measured RMR values provided by excavated tunnel faces.

27
 ACCEPTED MANUSCRIPT

PT
RI
SC
Fig. 3. Frequency distribution histogram of the Rock Mass Rating.
NU
MA
T ED
C EP
AC

Fig. 4. RMR estimation interval.

28
 ACCEPTED MANUSCRIPT

Fig. 5. RMR prediction with increasing number of data; the actual positions of

PT
the circles are detailed in Table 1.

RI
SC
NU
MA
T ED
EP

Fig. 6. Variogram fitting at four positions; (a) K21+170; (b) K21+422; (c) K21+612;
C

and (d) K22+381.

AC

29
 ACCEPTED MANUSCRIPT

PT
RI
SC
Fig. 7. Kriging prediction interval considering model uncertainty; (a) K21+170;
NU
(b) K21+422; (c) K21+612; and (d) K22+381.
MA
T ED
C EP
AC

Fig. 8. Kriging variance at four prediction positions; (a) K21+170; (b) K21+422;

(c) K21+612; and (d) K22+381.

30
 ACCEPTED MANUSCRIPT

PT
RI
SC
Fig. 9. Prediction interval at four prediction positions with consideration of
NU
model uncertainty and spatial variability; (a) K21+170; (b) K21+422; (c) K21+612;

and (d) K22+381.

MA
T ED
C EP
AC

31
 ACCEPTED MANUSCRIPT

PT
RI
SC
NU
MA
T ED
EP

Fig. 10. RMR prediction intervals; (a) Model One: model uncertainty; and (b)
C

Model Two: model uncertainty and spatial variability.

AC

32
 ACCEPTED MANUSCRIPT

PT
RI
Fig. 11. Variogram model fitting; (a) RMRbasis; (b) joint condition (JC); (c)

SC
RQD; (d) joint and bedding spacing (JS); (e) uniaxial compressive strength (UCS);

and (f) groundwater condition (GW).

NU
MA
T ED
C EP
AC

Fig. 12. Scatterplot of the measured values and predicted values for two

prediction methods.

33
 ACCEPTED MANUSCRIPT

PT
RI
Fig. 13. Variogram model fitting considering periodicity; (a) RMRbasis; (b) joint

SC
condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial

compressive strength (UCS); and (f) groundwater condition (GW).

NU
MA
T ED
C EP

Fig. 14. Spectral analysis on RMRbasis and RMR components; (a) RMRbasis; (b) joint
AC

condition (JC); (c) RQD; (d) joint and bedding spacing (JS); (e) uniaxial

compressive strength (UCS); and (f) groundwater condition (GW).

34
 ACCEPTED MANUSCRIPT

Table. 1 Measured RMR values provided by the excavated tunnel faces during
construction.
Number Position Measured RMR value
1 K17+599 64
2 K17+620 56
3 K17+640 61
4 K17+681 59
5 K19+676 59
6 K20+124 54

PT
7 K20+205 63
8 K20+241 69
9 K20+347 66

RI
10 K20+375 66
11 K20+402 62

SC
12 K20+432 68
13 K20+456 68
NU
14 K20+475 74
15 K20+583 60
16 K20+650 74
MA

17 K20+880 74
18 K20+905 74
19 K20+930 72
20 K20+960 74
ED

21 K20+977 60
22 K21+037 50
T

23 K21+062 60
EP

24 K21+170 57
25 K21+200 57
26 K21+405 64
C

27 K21+422 64
AC

28 K21+503 44
29 K21+547 45
30 K21+612 47
31 K21+955 54
32 K21+987 52
33 K22+381 61
34 K22+533 59
35 K22+575 57
36 K22+881 52
37 K22+939 58
38 K23+790 68

35
 ACCEPTED MANUSCRIPT

39 K24+090 65
40 K24+120 65
41 K24+149 65
42 K24+177 65
43 K24+708 59

Table 2. Variogram models parameters of the components using two kriging

estimation methods.

PT
Parameters in Direct Indirect Method
estimation Method

RI
Methods
Model parameters RMRbasis JC RQD JS UCS GW

SC
Model Sph* Sph* Sph* Sph* Sph* Sph*
Sill 67.00 16.55 1.81 5.17 1.04 55.55
NU
Range 171.33 175.45 260 163.68 350 1107.85

Nugget 0 0 0 0 0 0
MA

*Sph stands for Spherical model.

ED

Table 3. Cross-validation criteria for two estimation methods.

Method
Indirect Method Direct Method
T

Criterion
EP

ME 0.0558 -0.2035
RMSE 1.1179 5.4462
C

MSPE 0.0038 -0.0082

AC

Table 4. Period of the variogram models for two kriging estimation methods
using nested hole-effect model.
Parameters in Direct Indirect Method
estimation Method
Methods
Period (m) RMRbasis JC RQD JS UCS GW

Variogram model 225 280 500 450 700 1500

36
 ACCEPTED MANUSCRIPT

Spectral analysis 123 284 711 296 592 1185

Table 5. Cross-validation criteria for two estimation methods considering

periodicity.
Method Indirect Method Direct Method
Criterion
ME 0.0125 -0.0865
RMSE 5.0961 4.9934

PT
MSPE 0.0011 -0.0051

RI
SC
NU
MA
T ED
C EP
AC

37
 ACCEPTED MANUSCRIPT

Highlights

>A geostatistical method for predicting RMR values quantitatively ahead of

the tunnel face which makes full use of geological information exposed on
tunnel face is presented.
>The spatial variability of geological condition is updated by observation data
on the new excavated faces.

PT
The RMR prediction uncertainty is quantified by accounting for spatial
variability and model uncertainty.
The prediction of RMR values allows for periodicity of geological conditions.

RI
SC
NU
MA
T ED
C EP
AC

38