Time Series Analysis of Soil Radon Data

Using Multiple Linear Regression and

Artificial Neural Network in Seismic
Precursory Studies

S. Singh, H. P. Jaishi, R. P. Tiwari &

R. C. Tiwari

Pure and Applied Geophysics

pageoph

ISSN 0033-4553
Volume 174
Number 7

Pure Appl. Geophys. (2017)

174:2793-2802
DOI 10.1007/s00024-017-1556-4

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
International Publishing. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.

1 23
 Author's personal copy
Pure Appl. Geophys. 174 (2017), 2793–2802
 2017 Springer International Publishing
DOI 10.1007/s00024-017-1556-4 Pure and Applied Geophysics

Time Series Analysis of Soil Radon Data Using Multiple Linear Regression and Artificial
Neural Network in Seismic Precursory Studies
S. SINGH,1 H. P. JAISHI,1 R. P. TIWARI,2 and R. C. TIWARI1

Abstract—This paper reports the analysis of soil radon data half-life of 3.8 days, 220Rn (also called thoron),
recorded in the seismic zone-V, located in the northeastern part of
India (latitude 23.73N, longitude 92.73E). Continuous measure-
member of 232Th series, with an half-life of 54.55 s
ments of soil-gas emission along Chite fault in Mizoram (India) and 219Rn, member of 235U series, with an half-life of
were carried out with the replacement of solid-state nuclear track 3.928 s. Owing to its longer half-life, the variations
detectors at weekly interval. The present study was done for the
of its concentration in soil and fluid samples are
period from March 2013 to May 2015 using LR-115 Type II
detectors, manufactured by Kodak Pathe, France. In order to reduce considered as useful tools for prediction along active
the influence of meteorological parameters, statistical analysis tools fault zones (Chyi et al. 2005; Igarashi et al. 1995; Liu
such as multiple linear regression and artificial neural network have et al. 1984; Singh et al. 1999; Virk et al. 2001; Etiope
been used. Decrease in radon concentration was recorded prior to
some earthquakes that occurred during the observation period. and Lombardi 1995; Fu et al. 2005; Walia et al. 2005;
Some false anomalies were also recorded which may be attributed Külahci and Çiçek 2014). Furthermore, the variations
to the ongoing crustal deformation which was not major enough to in soil-gas radon concentration are not only con-
produce an earthquake.
trolled by the processes such as seismic activity,
Key words: Radon measurements, SSNTDs, MLR and ANN volcanic activity or deformation, but also by the
analysis, earthquake precursors, statistical algorithm, meteorolog- influence of meteorological parameters (Ghosh et al.
ical parameters.
2009; Stranden et al. 1984; Wattananikorn et al.
1998).
The proneness of northeast region of India is
attributed to the Indian plate and Eurasian plate in the
1. Introduction north and subduction tectonics along the Indo-
Myanmar range in the east (Dewey and Bird 1970).
During the last two decades, efforts on a large The surrounding area of this region has experienced
number of geophysical and geochemical measure- many large earthquakes during the past ten decades.
ments have been made in earthquake-prone regions of The geomorphologic map of Aizawl town (Fig. 1)
the world. Since radon is easily detectable, it has been forms a part of meridional ridges and narrow valley
mostly preferred as earthquake precursor. Moreover, (i.e., often controlled by faults). Its surroundings are
predication of seismic events is very complicated underlain by tertiary sedimentary rocks belonging to
because the physical mechanism of earthquakes and surma group of rocks of Oligocene–Miocene age.
the factors governing them are poorly understood The study area is mostly mountainous and hilly
(King 1986). Radon, with noble geochemical char- whose ranges traverse in the north–south direction
acteristics found in nature has three different separated by many north–south flowing rivers and
isotopes, i.e., 222Rn, member of 238U series, with an valleys. Because of the slope of its valleys it has
witnessed many landslides over the past few decades.
The study area is along Chite valley which occurs
1
east of Aizawl anticline ridge (Rawat and Parihar
Department of Physics, Mizoram University, Aizawl
796004, India. E-mail: [email protected]
2001). The Chite River flowing between Aizawl town
2
Department of Geology, Mizoram University, Aizawl and Zemabawk formed a broad valley approximately
796004, India.
 Author's personal copy
2794 S. Singh et al. Pure Appl. Geophys.

Figure 1
Location map of study area showing the geology of the region (modified after Singh et al. 2014)

600 m long and 45–60 m wide. Looking at the sce- entry of all radon, thoron and their progeny (Mayya
nario of its proneness the authors extended their et al. 1998). Therefore, in our present study we have
research work along Chite fault in Aizawl District, used only two compartments (i.e., pinhole compart-
Mizoram (India) (Jaishi et al. 2013, 2014a, b; Singh ment and filter mode compartment) to monitor radon/
et al. 2014). thoron concentration. The dosimeters loaded with
detector films were kept inside metallic cylinders at 3
locations with 3 detectors each in two locations and 4
2. Materials and Methods detectors in one. The final reading was the average of
all the 10 detectors. The hollow metallic cylinders
In the present research work, solid-state nuclear were kept 80 cm below the soil surface and 30 cm in
track detectors (SSNTDs) based dosimeters were diameter as shown in Fig. 2. The detector films were
used for continuous measurements of soil-gas radon/ kept exposed for a period of 7 days, and after the
thoron concentration along the Chite fault, Aizawl, exposure period they were retrieved and chemically
Mizoram. The cellulose nitrate films were loaded in etched using 2.5 N NaOH solution in a specially
the dosimeters which were cut into a size of designed constant-temperature bath model PSI-CTBI.
3 cm 9 3 cm. The dosimeters have three compart- The detector films were etched at a stable temperature
ments, i.e., membrane/pinhole compartment, filter of 60 C for approximately 60 min; after this process
mode compartment and bare mode. In case of pinhole the films were stripped off. The etching is done in
compartment, it measures radon only and allows order to enlarge the alpha track registered from the
more than 95% of radon gas to diffuse and suppress radon/thoron gas within LR-115 films. The stripped
thoron gas to less than 1% (Arafa 2002; Jönsson LR-115 detector films were further dried and counted
1981; Ramachandran et al. 1987). Whereas the filter using a spark counter to obtain the track densities in
mode allows both radon and thoron gas to diffuse into tracks per square centimeter. The track densities were
it, the entry of the progeny is blocked by placing a then converted into Becquerel per cubic meter as
filter paper. Regarding the bare mode, it allows the discussed in Singh et al. (2014). The meteorological
 Author's personal copy
Vol. 174, (2017) Time Series Analysis of Soil Radon Data Using Multiple Linear Regression 2795

Lid

Ground surface

Metal cylinder

Dosimeter

LR-115 film facing up

(thoron measurement)
80 cm

LR-115 film facing down

Stand (radon measurement)

20 cm
Filter

Soil gas
Figure 2
Measurements of radon/thoron in soil

data for this region were obtained from Indian software. A general MLR model can be formulated as
Meteorological Department-Regional Meteorological given in the following equation:
Center, Guwahati, India.
Y 0 ¼ a þ b1 X1 þ b 2 X2 þ    þ b n Xn ð1Þ
Here the predicted value of the dependent variable
3. Data Analysis is Y0 , a is the constant term and b1, b2, … bn
represents the regression coefficient of the indepen-
3.1. Multiple Linear Regressions dent variables X1, X2, …, Xn, respectively. An
important step in selecting the best model is to
In different branches of engineering and science,
determine how many variables and which particular
many problems are involved in analyzing the rela-
variables should be in the final model. We have
tionships between two or more variables. Multiple
performed stepwise regression which permits re-
linear regressions (MLR) is a statistical technique
examination at every step of the variables in previous
that is very significant for predicting the best
steps. A variable that enters at an early stage may
relationship between a dependent variable and sev-
become non-significant at later stages because of its
eral independent variables (Giacomino et al. 2011).
relationship with other variables subsequently added
MLR is based on least squares criterion: the model is
to model (Latt and Wittenberg 2014).
fitted in such a way that the sum of squares of
In this study, we performed MLR taking into
differences of observed and predicted values is
account soil radon concentration (Rn) as the
minimized. MLR analysis was performed using SPSS
 Author's personal copy
2796 S. Singh et al. Pure Appl. Geophys.

dependent variable and the meteorological parame- non-significant predictors of the dependent variable as
ters, viz., rainfall (Rf), relative humidity (Rh), the p value (significance) is more than a-value (0.05).
temperature (T) and pressure (P) as independent In backward regression method, only one predictor
variables. Here Rn has been regressed with Rf, Rh, variable is eliminated at a time (the predictor variable
T and P. All the predictor variables mentioned may with the highest p value is deleted first). Therefore,
not be a significant predictor of the dependent rainfall having the highest p value (0.755) is removed
variable. To sort out this problem and to build a from the model 1 and a second model is constructed
better model, we performed backward regression with T, Rh and P as independent variables. Model 2
method using SPSS statistical package. The back- shows that P is a non-significant predictor (i.e., the p
ward regression method starts with all the predictor value is more than 0.05). So, P is removed from the
variables, tests the significance of each predictor model and a third model is constructed with T and Rh as
variable using t test and then deletes the variable (if predictor variable. It can be seen in Model 3 that Rh is a
any) that improves the model the most by being non-significant predictor (since the p value is much
deleted. This process is repeated until the maximum more than 0.05). So, Rh is removed from the model and
number of steps is reached or no more independent a fourth model is built with T as a single predictor
variable qualifies for removal (Hocking 1976). The variable. It is now clear that T is a significant predictor
coefficients of the backward regression output for (p value \0.05) of Rn. Therefore, the raw score
radon are presented in Table 1. regression equation for model 4 can be written as
From the radon data, the backward regression 0
Rn ¼ 888:444 þ 88:283 ðTÞ ð2Þ
output has four distinct models (Model 1, Model 2,
Model 3 and Model 4). In the first model, it includes all We calculated the predictor value of measured
the predictor variables (Rf, Rh, T and P). Now, to check radon concentration (Rn’) using the above equation.
the significance of the predictor variables a t test is
performed at 95% confidence interval (i.e., the a value
3.2. Artificial Neural Network
is set at 0.05) and its corresponding probability is
shown in significance column of the table (Table 1). Artificial neural network (ANN) refers to a well-
Looking at the significance of each predictor variable known computational structure inspired by the oper-
in model 1, three predictor variables (Rf, Rh and P) are ation of the biological neural system, used widely in

Table 1
Coefficients of the backward regression output of radon

Model Unstandardized coefficients Standardized coefficient T Sig.

B Std. Error b

1 Constant -13,892.422 14,260.112 -0.974 0.332

Rf -0.323 1.032 -0.030 -0.313 0.755
T 102.204 20.410 0.605 5.008 0.000
RH 5.714 4.464 0.118 1.280 0.203
P 12.780 14.399 0.109 0.888 0.377
2 Constant -14,214.377 14,166.511 -1.003 0.318
T 100.783 19.819 0.596 5.085 0.000
RH 5.183 4.112 0.107 1.260 0.210
P 13.178 14.286 0.113 0.922 0.358
3 Constant -1152.375 415.438 -2.774 0.006
T 87.335 13.416 0.517 6.510 0.000
RH 3.849 3.847 0.079 1.001 0.319
4 Constant -888.444 320.955 -2.768 0.007
T 88.283 13.838 0.522 6.597 0.000
 Author's personal copy
Vol. 174, (2017) Time Series Analysis of Soil Radon Data Using Multiple Linear Regression 2797

signal processing, pattern recognition and other

applications. An ANN consists of a set of units
(neuron, nodes) and a set of weighted interconnec-
tions among them (links). The ANN architectures are
generally classified as (1) single-layer feedforward;
(2) multi-layer feedforward; (3) recurrent and (4)
lattice structure (Haykin 1994). Among this topology,
multi-layer feedforward neural networks, commonly
referred to as multi-layer perceptrons (MLPs), have
been used and applied successfully to solve some of
the difficult and diverse problems in several domains
including prediction and forecasting. MLP has a
layered architecture consisting of input, hidden and Figure 3
output layers. The input signals propagate through the The ANN topology for learning radon concentration dependency
on environmental parameters
network in a forward direction on a layer-by-layer
basis. The output of each layer is transferred to the
into three sets: the training set (70%), the cross-
input of neurons in the next layer through weighted
validation set (15%) and the test set (15%). The
links. The hidden layer aids in performing useful
training set has been used to train the network, the
complex computations by extracting progressively
validation set has been used for the purpose of
more meaningful features from the input layer. In
monitoring the training process and to guard against
case of training and weight adaptation, MLPs are
overtraining and the testing set has been used to judge
used in a supervised manner with a highly popular
the performance of the trained network. Furthermore,
algorithm known as the error back-propagation
the training was stopped just below threshold cross-
algorithm. The algorithm provides an established
validation error, i.e., when the cross-validation error
action for changing the weights in a back-propagation
was lowest.
network to assign the given input patterns, correctly
The ANN architecture with four input layers has
for a given training set of input–output pairs (Külahci
been created as shown in Fig. 3. The architecture is
et al. 2009; Sen and Birpinar 2004). In this study,
chosen based on the selection performance of differ-
Levenberg–Marquardt algorithm is applied as one of
ent networks. The four input nodes represent the
the back-propagation training methods, as it has the
meteorological parameters, viz., rainfall, temperature,
fastest convergence for medium-size ANNs (Zurada
relative humidity and pressure. The MLP contained
1992). This method is a modification of the classic
20 hidden layers and a single output which is the
Newton algorithm which is as follows:
estimated radon value. The estimated radon value
 1
vkþ1 ¼ vk  J T J þ lI J T e: ð3Þ was calculated on weekly interval starting from the
1st week to 118th week, i.e., from March 2013 to
Here, Newton’s method applied the approximate May 2015. One week measured data were fed to the
Hessian matrix, the scalar l is zero, but when l is network to calculate the estimated value of radon.
large, this becomes gradient descent with a small step The estimated radon value was then compared with
size. Therefore, l is decreased after each successful the measured radon value to identify a valid
step (reduction in performance function) and it is precursor.
increased only when a tentative step would increase
the performance function (The Mathworks Inc. USA
2007). In the present study, the ANN of the MLP type 4. Results and Discussion
was chosen for analyzing the soil-gas radon concen-
tration time series at Chite Fault, Aizawl, Mizoram. A We used the application of ANN to predict the
total of 118 datasets were trained, which were divided value of measured radon concentration during the
 Author's personal copy
2798 S. Singh et al. Pure Appl. Geophys.

observation period. The method employed is useful to zone (68–141 km) given by Dobrovolsky et al.
differentiate those radon anomaly caused by geo- (1979):
physical process from those caused by various
D ¼ 10 exp 0:43M km; ð4Þ
meteorological parameters (Negarestani et al. 2002).
The ANN method was compared with the multiple where D is the epicentral distance and M is the
linear regression method (MLR). As indicated in magnitude of the earthquake. The spatial distribution
Fig. 4, the ANN method gives a better accuracy in of the selected earthquakes as per Eq. (4) is depicted
prediction than the MLR. This can be clearly in Fig. 5.
understood from the descriptive statistics of all the During the investigation period, a total of 16
parameters as described in Table 2. The correlation radon peaks were observed. Amongst these radon
coefficient of measured radon with respect to the peaks there are two peaks of radon concentration
predicted ANN value is more than that of the pre- which crosses above ?2SD, five peaks crossing
dicted MLR. The predicted value obtained by ANN ?1SD and nine peaks were beneath -1SD. From
method is then used to define a radon anomaly. A Fig. 6, it can be seen that all those positive radon
radon anomaly is usually defined as the value which peaks crossing ?2SD and ?1SD could not be cor-
exceeds the average value by ±2 standard deviation related with the seismic activity, since there was no
(Rannou 1989; Price et al. 1994; Guerra and Lom- such signature of any seismic events during that time
bardi 2001; Ghosh et al. 2009; Jaishi et al. 2014a, b). window. These positive peaks were considered as
The selection of earthquakes is always a difficult task false anomaly. Perhaps, these two increases in radon
and many earthquakes had occurred near the study concentration may have occurred due to ongoing
area during the investigation period (see Table 3); crustal deformation in the region which was not
however, we have selected only those earthquakes mature enough to produce an earthquake. Moreover,
that comes well within the earthquake precursory both these peaks occurred during heavy rainfall,

Figure 4
Time series of measured and predicted values of radon during the observation period
 Author's personal copy
Vol. 174, (2017) Time Series Analysis of Soil Radon Data Using Multiple Linear Regression 2799

Table 2
Descriptive statistics of measured radon, predicted radon (ANN), predicted radon (MLR) and meteorological parameters

Parameters No. of observations Mean Standard deviation (SD) Max. Min. Correlations

Measured radon (Bq/m3) 118 1204.63 613.26 3693.0 164 Zero-order Partial Part
Predicted radon (ANN) (Bq/m3) 118 1208.26 446.25 2811.01 224.45 – – –
Predicted radon (MLR) (Bq/m3) 118 1204.63 320.31 1685.00 488.33 – – –
Rainfall (mm) 118 40.76 56.92 311.1 0 0.208 -0.029 -0.025
Temperature (C) 118 23.71 3.63 29.15 15.60 0.522 0.426 0.398
Pressure (mb) 118 74.41 12.65 91.08 41.51 -0.349 0.083 0.071
Relative humidity (%) 118 959.48 5.25 968.13 931.91 0.116 0.120 0.102

Table 3
Details of earthquakes that occurred around the measuring site during the investigation period (source: www.imd.gov.in)

Date (MM/DD/YYYY) Lat Long Depth Magnitude (M) Region Epicenter distance (km)

03/02/2013 24.8N 92.2E 10 5.2 India (Karimganj) Bangladesh Border Region 138
12/30/2013 24.3N 93.2E 10 4.5 Churachandpur (Manipur) 68
01/29/2014 23.9N 93.9E 33 5.1 Myanmar–India (Manipur) Border Region 120
11/20/2014 23.8N 93.5E 33 5.6 Myanmar–India (Mizoram) Border Region 76
11/21/2014 23.6N 93.4E 46 4.7 India (Mizoram)–Myanmar Border Region 71
02/12/2015 24.2N 94.0E 90 5.0 Manipur (Chandel) 141

Figure 5
Spatial distribution of earthquakes that occurred around the study area during the observation period
 Author's personal copy
2800 S. Singh et al. Pure Appl. Geophys.

Figure 6
Plot showing the variation of measured radon, ANN radon and MLR radon during the investigation period along Chite Fault, Aizawl. The blue
vertical lines represent earthquakes along with their magnitude. The solid horizontal line represents the average value of radon concentration
(Avg.) and the dotted lines represent the deviation (SD) from the average value

where the water filters through a porous surface and 2000). Therefore, these positive peaks may be con-
affects the gas emanation thus preventing gas from sidered as false anomaly; whereas among the radon
migrating toward the surface. As a result, the radon peaks which were recorded below -1SD out of nine
values initially fall and then start rising over a period peaks only five of them could be correlated with the
of time (Kraner et al. 1964; Jaacks 1984; Virk et al. relevant seismic event.
 Author's personal copy
Vol. 174, (2017) Time Series Analysis of Soil Radon Data Using Multiple Linear Regression 2801

The first two decreases in radon concentration correlated with relevant earthquakes. These three
were observed on 11/29/13 and 12/13/13; these radon negative anomalies may have been caused by
minima could be correlated to the earthquake of admixing of radon-poor water from another aquifer
magnitude M4.5 which occurred on 12/30/13 of the through cracks created by the earthquake-related
observation period (i.e., 31 and 17 days prior to the strain changes along the study region. However,
seismic events). The third decrease in radon con- much more data recorded by various geophysical and
centration was registered on 2/4/14 and this may be geochemical instruments with good time resolution
correlated to the earthquake of M5.1 which occurred and rigorous statistical data analysis are needed to
on 1/29/14 (i.e., 6 days before the anomaly was understand the underlying earthquake mechanism,
observed or we can consider this peak as post-cursor also to discover more true earthquake precursors.
anomaly). The fourth anomalous change in radon
concentration (-1STD) was recorded on 11/7/14
which was followed by two seismic events taking Acknowledgements
place on 11/20/14 and 11/21/14 with a magnitude of
M5.6 and M4.7. This is considered as precursor This work has been funded by the Ministry of Earth
anomaly with a precursor time of 13 and 14 days Sciences (MoES), Govt. of India, New Delhi, in the
prior to the events. Their epicentral distances from form of Major project vide Sanction Order No.
the monitoring site were 76 and 71 km. The fifth MoES/P.O.(Seismo)/1(167)/2013 dated 10.12.2013.
anomaly was registered on 2/6/15 followed by an
earthquake of M5.0 (on 2/12/15, i.e., 6 days after the REFERENCES
observed anomaly). These decreases in radon con-
centration may have been caused either by admixing Arafa, W. (2002). Permeability of radon-222 through some mate-
of radon-poor water from another aquifer through rials. Radiation Measurements, 35, 207–211.
Chyi, L. L., Quick, T. J., Yang, T. F., & Chen, C. H. (2005). Soil
cracks created by the earthquake-related strain gas radon spectra and earthquakes. Terrestrial, Atmospheric and
changes or else may be due to accumulation of tec- Oceanic Sciences, 16, 763–774.
tonic strain and a slow, steady increase of stress under Dewey, J. F., & Bird, J. M. (1970). Mountain belts and the new
global tectonics. Journal of Geophysical Research, 75,
geological condition. These types of decrease in
2625–2647.
radon concentration have been reported by certain Dobrovolsky, I. P., Zubkov, S. I., & Miachkin, V. I. (1979). Esti-
authors in the past (King and Wakita 1981; Kuo et al. mation of the size of earthquake preparation zones. Pure and
2010, Külahci and Şen 2014). Applied Geophysics, 117, 1025–1044.
Etiope, G., & Lombardi, S. (1995). Evidence for radon transport by
carrier gas through faulted clays in Italy. Journal of Radioana-
lytical and Nuclear Chemistry, 193, 291–300.
5. Conclusion Fu, C. C., Yang, T. F., Walia, V., & Cheng, C. H. (2005).
Reconnaissance of soil gas composition over the buried fault and
fracture zone in southern Taiwan. Geochemical Journal, 39,
During the 2 years of measurement along the 427–439.
Chite fault, the study revealed that 60% of the Ghosh, D., Deb, A., & Sengupta, R. (2009). Anomalous radon
emission as precursor of earthquakes. Journal of Applied Geo-
anomalies observed seem to be correlated with the
physics, 69, 67–81.
relevant earthquakes. In our study, we observed that Giacomino, A., Abollino, O., Malandrino, M., & Mentasti, E.
anomalous decrease in radon was correlated with (2011). The role chemometrics in single and sequential extrac-
seismic events. The decrease in radon concentration tion assays: A review. Part II. Cluster analysis, multiple linear
regression, mixture resolution, experimental design and other
prior to the seismic events had been reported only techniques. Analytica Chimica Acta, 688, 122–139.
very few times in literature. Implementation of sta- Guerra, M., & Lombardi, S. (2001). Soil-gas method for tracing
tistical data analysis such as multiple linear neotectonic faults in clay basin: The Pisticci field (southern
Italy). Tectonophysics, 339, 511–522.
regression and artificial neural network is quite Haykin, S. (1994). Neural networks: A comprehensive foundation.
promising in earthquake-related precursory studies. New Jersey: Prentice-Hall Inc.
Out of the five radon anomalies, three anomalies (two Hocking, R. R. (1976). The analysis and selection of variables in
linear regression. Biometrics, 32, 1–49.
crossing -2SD and one crossing -1SD) were
 Author's personal copy
2802 S. Singh et al. Pure Appl. Geophys.

Igarashi, G., Sacki, S., Takahata, N., Sumikawa, K., Tasaka, S., areas using a twin-cup dosimeter with three track detectors.
Sasaki, Y., et al. (1995). Groundwater radon anomaly before the Radiation Protection Dosimetry, 77, 177–184.
Kobe earthquake in Japan. Science, 269, 60–61. Negarestani, A., Setayeshi, S., Ghannadi-Maragheh, M., & Akashe,
Jaacks, J. A. (1984). Meteorological influence upon mercury, radon B. (2002). Layered neural networks based analysis of radon
and helium soil gas emission. Ph.D. Thesis, Colorado School of concentration and environmental parameters in earthquake pre-
mines. diction. Journal of Environmental Radioactivity, 62, 225–233.
Jaishi, H. P., Singh, S., Tiwari, R. P., & Tiwari, R. C. (2013). Price, J. G., Christensen, L., Hess, R., La Pointe, D. D., Ramelli, A.
Radon and thoron anomalies along Mat fault in Mizoram, India. R., Desilets, M., et al. (1994). Radon in outdoor air in Nevada.
Journal of Earth System Science, 122, 1507–1513. Health Physics, 66, 433–438.
Jaishi, H. P., Singh, S., Tiwari, R. P., & Tiwari, R. C. (2014a). Ramachandran, T. V., Lalit, B. Y., & Mishra, U. C. (1987). Mea-
Correlation of radon anomalies with seismic events along Mat surement of radon permeability through some membranes.
fault in Serchhip District, Mizoram, India. Applied Radiation and International Journal of Radiation Applications and Instrumen-
Isotopes, 86, 79–84. tation Part D Nuclear Tracks and Radiation Measurements, 13,
Jaishi, H. P., Singh, S., Tiwari, R. P., & Tiwari, R. C. (2014b). 81–84.
Temporal variation of soil radon and thoron concentrations in Rannou, A. (1989). The bare detector and results of indoor radon
Mizoram (India), associated with earthquakes. Natural Hazards, survey in France. In Proceedings of International Workshop on
72, 443–454. Radon monitoring in Radioprotection, Environmental Radioac-
Jönsson, G. (1981). The angular sensitivity of Kodak LR-film to tivity and Earth Science (pp. 145–222).
alpha particles. Nuclear Instruments and Methods in Physics Rawat, U. S., & Parihar, C. P. S. (2001). Geoenvironment of
Research, 190, 407–414. Mizoram and its implication in development. In Proceedings of
King, C. Y. (1986). Gas geochemistry applied to earthquake pre- National Symposium–Role of Earth Sciences in Integrated
diction: An overview. Journal of Geophysical Research, 91, Development and Related Societal Issues, GSISP–65 (I),
12269–12281. 185–190.
King, C.-Y., & Wakita, H. (1981). Anomalous radon changes in an Sen, Z., & Birpinar, M. E. (2004). Water resources assessment,
artesian well and possible relation to earthquake (abstract). IACWRA 2004 Lecture Notes (p. 100). Istanbul: Water Engi-
Earthquake Notes, 52(1), 71. neering Research and Development Center.
Kraner, H. W., Schroeder, G. L., & Evans, R. D. (1964). Mea- Singh, S., Jaishi, H.P., Tiwari, R.P., Tiwari, R.C. (2014). Variations
surement of the effects of atmospheric variables on radon-222 of soil radon concentrations along Chite Fault in Aizawl district,
flux and soil-gas concentrations. In L. A. S. Adams & W. Mizoram, India. Radiation Protection Dosimetry. doi:10.1093/
M. Lowder (Eds.), Symposium Proceedings of Natural Radiation rpd/ncu221.
Environment, Houston, Texas (pp. 191–214), 10–13 April 1963. Singh, M., Kumar, M., Jain, R. K., & Chatrath, R. P. (1999). Radon
University of Chicago Press: Chicago. in ground water related to seismic events. Radiation Measure-
Külahci, F., & Çiçek, Ş. (2014). Time-series analysis of water and ments, 30, 465–469.
soil radon anomalies to identify micro-macro-earthqaukes. Ara- Stranden, E., Kolstad, A. K., & Lind, B. (1984). The influence of
bian Journal of Geosciences, 8(7), 5239–5246. moisture and temperature on radon exhalation. Radiation Pro-
Külahci, F., Inceoz, M., Dogru, M., Aksoy, E., & Baykara, O. tection Dosimetry, 7, 55–58.
(2009). Artificial neural network model for earthquake prediction The Mathworks Inc. (2007). Version R2007a. New Jersey: Prentice
with radon monitoring. Applied Radiation and Isotopes, 67, Hall.
212–219. Virk, H. S., Walia, V., & Kumar, N. (2001). Radon/helium pre-
Külahci, F., & Şen, Z. (2014). On the correction of spatial and cursory signals of Chamoli earthquakes, Garhwal Himalaya,
statistical uncertainties in systematic measurements of 222Rn for India. Journal of Geodynamics, 31, 201–210.
earthquake prediction. Surveys in Geophysics, 35(2), 449–478. Virk, H. S., Walia, V., Sharma, A. K., Kumar, N., & Kumar, R.
Kuo, T., Su, C., Chang, C., Lin, C., Cheng, W., Liang, H., et al. (2000). Correlation of radon with microsiesmic events in Kangra
(2010). Application of recurrent radon precursors for forecasting and Chamba Valleys of N-W Himalaya. Geofisica Internacional,
large earthquakes (Mw [ 6.0) near Antung, Taiwan. Radiation 39, 221–227.
Measurements, 45, 1049–1054. Walia, V., Virk, H. S., Yang, T. F., Mahajan, S., Walia, M., &
Latt, Z. Z., & Wittenberg, H. (2014). Improving flood forecasting Bajwa, B. S. (2005). Earthquake prediction studies using radon
in a developing country: A comparative study of stepwise mul- as a precursor in NW Himalayas, India: A case study. Terrestrial,
tiple linear regression and artificial neural network. Water Atmospheric and Oceanic Sciences, 16, 775–804.
Resources Management, 28, 2109–2128. Wattananikorn, K., Kanaree, M., & Wiboolsake, S. (1998). Soil gas
Liu, K. K., Yui, T. F., Tasi, Y. B., & Teng, T. L. (1984). Variation radon as an earthquake precursor: Some considerations on data
of radon content in groundwater and possible correlation with improvement. Radiation Measurements, 29, 593–598.
seismic activities in the northern Taiwan. Pure and Applied Zurada, J. M. (1992). Introduction to artificial neural systems (p.
Geophysics, 122, 231–244. 683). St. Paul: West Publishing Company.
Mayya, Y. S., Eappen, K. P., & Nambi, K. S. V. (1998).
Methodology for mixed field inhalation dosimetry in monazite

(Received March 5, 2016, revised April 15, 2017, accepted April 19, 2017, Published online April 24, 2017)