Journal of Structural Engineering No.

37-T2
Vol. 37, No. 2, June-July 2010 pp. 142–144

A short note on estimation of occurrence of earthquakes in Tamil Nadu region

T. Priya∗ , K. Balaji Rao *, ∗ , M. B. Anoop∗ , S. R. Balasubramanian∗ and Nagesh R. Iyer∗

∗
CSIR-Structural Engineering Research Centre, CSIR Campus, Taramani, Chennai - 600 113, India.
Received 15 July 2009; Accepted 10 August 2009

EARTHQUAKE PROCESS used a time homogeneous Poisson process model to predict

the probability of occurrence of earthquake of specified ma-
Poisson process model is commonly used for describing the gnitude range in a given period of time. Details regarding the
occurrence of rare events. Earthquakes have long been as- justification for the use of this model and its applications for
sumed to occur randomly in time in a given region. Also, the determination of reliability of structure/structural com-
the occurrence of earthquake is most commonly described ponents are presented in 7 . It is noted from literature that such
by a Poisson model. The Poisson process is a continuous pa- models are used for determination of probability of occur-
rameter discrete state space model. It is used in estimating rence of extreme earthquakes 8, 9 .
the probability of occurrence of number of earthquakes in a According to Poisson process model, the probability of
specified period of time. Being a counting process, the state occurrence of an earthquake with mean occurrence rate v i , in
space can have values from 0 to ∞. In Poisson process the the time period (0,t), n number of times is given by 6
number of occurrences in one time interval is independent (υi t)n
of the number that occurs in any other distinct time interval, P {Ni (t) = n} = e−υi t , n = 0, 1 · · ·
and the probability of occurrence during a very short time in- n!
terval is proportional to the length of the time interval. The BAYESIAN METHODOLOGY
times between the events of Poisson process are exponential-
ly distributed with no memory of time, size or location of any Bayesian methodology has been explained in literature 10 as
preceding event. For this and other reasons related to simpli- follows. Generally, accurate estimates of parameters require
city, easy to use and lack of sufficient data to support more large amount of data. When the observed data are limited, as
sophisticated models, the Poisson process model is most wi- is often the case in engineering, the statistical estimates have
dely used in Probabilistic Seismic Hazard Analysis 1 . to be supplemented (or may even be superseded) by judg-
The historical records in Tamil Nadu region shows that 19 mental information. With the classical statistical approach
earthquakes with magnitude 3-3.9, 12 earthquakes with ma- there is no provision for combining judgmental information
gnitude 4-4.9, and 14 earthquakes with magnitude 5-6 had with observational data in the estimation of the parameters.
occurred in the period of 196 years (1800-1996) 2-4. Further, The Bayesian method approaches the estimation pro-
from the data collected on historical earthquakes, it is noted blem by treating the unknown parameters of a distribution
that only one earthquake had occurred with magnitude 5-6 in as random variables. In this way, uncertainty associated with
Tamil Nadu region in the next twelve years (i.e., from 1996 the estimation of the parameters can be combined formally
to 2008) 5. This observation is used to check the estimation (through Bayes’ theorem) with the inherent variability of the
made based on data from 1800 to 1996. Further predictions basic random variable. With this approach, subjective judg-
for the time period 2008-2020 are also made. ments based on intuition, experience or indirect information
In this paper, in making probabilistic predictions on oc- is incorporated systematically with observed data to obtain
currences of earthquake the following three approaches are a balanced estimation. The Bayesian method is particularly
considered: helpful in cases where there is a strong basis for such judg-
ments.
1. Simple Poisson process model The occurrence of earthquakes may be modeled as a Pois-
2. Bayesian approach by assuming a diffused prior for the son process with mean occurrence rate υ (Benjamin, 1968) 11.
mean recurrence rate Supposing that historical record for Tamil Nadu region shows
3. Bayesian approach by assuming its conjugate distribution that n0 earthquakes have occurred in the past t 0 years, the li-
as a prior distribution kelihood function is then given by
L(υ) = P (no quakes in to years/υ)
POISSON PROCESS (υto )no −υto
= e ;υ ≥ 0 (1)
no !
The probability of occurrence of an earthquake using Pois-
son process model has been discussed in 6 . An important first If there is no other information for estimation of υ a uni-
step in risk analysis is hazard prediction. Balaji Rao et al 7 form diffuse prior may be assumed; this implies that f  (υ) is
* E-mail: [email protected]
(Discussion on this article must reach the editor before September 30, 2010)

142 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 2, JUNE-JULY 2010

independent of the values of υ and thus can be absorbed into region2-4 . The collected data was categorized based on the
the normalizing constant k. Then the posterior distribution of magnitude range of the earthquake. The magnitude of earth-
υ becomes quakes ranges from 3-3.9, 4-4.9, and 5-6 and the mean recur-
rence rate for each range was calculated. Out of these three
f  (υ) = kL(υ) magnitude ranges only the magnitude range of 5-6 is con-
(υto )no −υto sidered in this study, because among these three magnitude
=k e ;υ ≥ 0 (2) ranges, this magnitude range has potential to cause damage
no !
to the structures.
Upon normalization, k = t 0 The occurrence of earthquake was modeled as a Poisson
The probability of the event (E = n earthquakes in the process, and hence, the probability of occurrence and non-
next t years in Tamil Nadu region) is then given as follows: occurrence of earthquake with magnitude 5-6 in the twelve
years period starting from 1996 to 2008 was first calcula-
∞ ted using simple Poisson distribution. Then the Bayesian ap-
P (E) = P (E/υ)f  (υ)dυ proach is used to find the probability of occurrence and non-
occurrence of earthquakes from 1996 to 2008 in Tamil Nadu
0 region. This is done for two cases: by assuming a diffused
∞ prior and the second one by assuming its prescribed conjuga-
(υt)n −υt t0 (υt0 )no −υt
= e e dυ te distribution (viz., Gamma distribution) as the prior.
n! n0 ! The Probability of occurrence and non-occurrence of at
0
⎛ ⎞ least one earthquake with magnitude 5-6 from 1996-2008 in
∞
(t + t0 ) [υ(t + t0 )](n+n0 ) −υ(t+t0 )dυ ⎠ Tamil Nadu region is calculated and the results are presented
=⎝ e in Table 1.
(n + n0 )!
0 To compare the estimated values with the actual data, the
(n + no )! tn to no +1 historical earthquakes with magnitude in the range 5-6 for 12
(3) years from 1996-2008 is collected. From the data collected, it
n!no ! (t + t0 )n+no +1 is noted that only one earthquake, in 2001, had occurred with
magnitude 5-6 near Chennai over the time period 1996-2008
Since the integrand inside the parentheses is a gamma
in Tamil Nadu region 5.
density function, the integral is equal to 1, hence a result that
By comparing the results presented in Table 1 with the
was first derived by Benjamin (1968) 11 follows.
observed earthquake data, we can infer that, the probability
(n + no )! tn to no +1 of non-occurrence of earthquake with magnitude 5-6 in the
P (E) = next twelve years (1996-2008) obtained by using Bayesian
n!no ! (t + t0 )n+no +1 approach and assuming the Diffused distribution as the pri-
(n + no )! (t/to )n or distribution is 0.59. This value is the highest and close to
= (4) (1 − e−1 ) than others. And hence it can be concluded that
n!no ! (1 + t/to )n+no +1
Diffused distribution is the suitable prior in making rational
prediction in this study.
Use of Conjugate Distribution
TABLE 1
In deriving the posterior distribution of a parameter, consi- PROBABILITY OF OCCURRENCE AND NON-OCCURRENCE
derable mathematical simplification can be achieved if the OF EARTHQUAKE WITH MAGNITUDE 5-6 FROM
1996 TO 2008 IN TAMIL NADU REGION
distribution of the parameter is appropriately chosen with re-
spect to that of the underlying random variable. By choosing Probability of Probability of
Period Magnitude Types of prior non-occurrence occurrence in
a prior distribution that is a conjugate of the distribution of Range in next 12 years next 12 years
the underlying random variable, convenient posterior distri- (1996-2008) (1996-2008)
bution, which is usually of the same mathematical form as Simple Poisson 0.43 0.57
the prior is obtained. Process *(0.64) **(0.36)
It is to be noted that conjugate distribution is chosen so-
1996-2008 5-6 PRIOR 0.41 0.59
lely for mathematical convenience and simplicity. For a ran- Diffused *(0.65) **(0.35)
dom variable with a specified distribution, its conjugate prior
distribution may be adopted if there is no other basis for the PRIOR 0.45 0.55
choice of the prior distribution 10 . Gamma *(0.64) **(0.36)
distribution
By assuming the Gamma distribution as the prior for υ,
*Denotes the Probability of non-occurrence of exactly one earthquake with
the probability of event are calculated by obtaining the pri- magnitude in the range of 5-6 in Tamil Nadu region from 1996 to 2008.
or parameters using the mean and variance of the Gamma ** Denotes the Probability of occurrence of exactly one earthquake with
distribution. The posterior parameters of the Gamma distri- magnitude in the range of 5-6 in Tamil Nadu region from 1996 to 2008.
bution are calculated and the updated mean and variance are
obtained. By using the updated mean recurrence rate the pro- The probability of occurrence and non-occurrence of ex-
bability of event are calculated. The results are presented in actly one earthquake with magnitude in the range of 5-6 in
this paper. Tamil Nadu region from 1996 to 2008 is calculated. It is no-
ted that there is not much variations in the probability of
occurrence and non-occurrence of exactly one earthquake
PROCEDURE FOR PROBABILISTIC PREDICTION obtained using the methods presented in Table 1. And this
variation in the probability of occurrence and non-occurrence
The earthquake occurrence data over the time period of 196 will not be the same, in the case of obtaining the probability
years from 1800 to 1996 was collected for Tamil Nadu of more number of occurrences (i.e., more then one).

JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 2, JUNE-JULY 2010 143

 By using the available information that 15 earthquakes 2. Balaji Rao, K., Ramanjaneyulu, K., Manjuprasad, M.,
in magnitude range 5-6 had occurred in 208 years (i.e., from Gopalakrishnan, S and Appa Rao, T. V. S. R., “Regional
1800 to 2008) the predictions are made for next 12 years (i.e., Seismic Risk Analysis of Peninsular India”, Project Re-
from 2008 to 2020) and are given in Table 2. port No. RCS-MLP-06741-RR-99-1, May 1999, SERC,
Madras.
TABLE 2
PROBABILITY OF OCCURRENCE AND NON-OCCURRENCE 3. Chennai Metropolitan Development Authority, “Second
OF EARTHQUAKE WITH MAGNITUDE 5-6 IN
Master Plan for Chennai Metropolitan Area, 2026”, vol.
NEXT 12 YEARS (2008-2020)
III, Sectoral Background, pp 245–261.
Probability of Probability of
Period Magnitude Types of prior non-occurrence occurrence for
Range for next 12 years next 12 years 4. Rajarathanam, S., Ganapathy, G. P., and Muthukumar R.,
(2008-2020) (2008-2020) “Seismic Hazard Assessment Based on Attenuation Rela-
2008-2020 5-6 PRIOR tionship for Tamil Nadu state, India”, Centre for Disaster
Diffused 0.40 0.60 Mitigation and Management.
PRIOR 5. ANSS (Advanced National Seismic System), http://
Gamma 0.44 0.56
distribution www.ncedc.org/anns/catalog-search.html - last accessed
24/06/2009.
Simple Poisson
Process 0.42 0.58 6. Balaji Rao, .K, Appa Rao .T. V. S. R, “Risk Analy-
From the results presented in Table 2, it can be inferred sis for Cyclone-Prone Regions and Industrial Structu-
that in the next 12 years there is 56% to 60% of chance of res”, APSSRA 95, Proc. of Asian-Pacific Symposium on
occurrence of atleast one earthquake in magnitude range 5-6 Structural Reliability and its Application, University of
in Tamil Nadu region. Tokyo, Japan, November 12-14, 1995, pp 204–211.

CONCLUSION 7. Balaji Rao, .K, Thomas K. C. and Appa Rao .T. V. S.

R, “Stochastic modeling of cyclones”, Proc. 3rd Asia-
From the limited study carried out it is found that Diffused Pacific Symposium on Wind Engg., Hong Kong, Vol. 1,
distribution prior is the suitable prior for making rational pre- 1993, pp 543–548.
dictions about the occurrence of earthquakes in Tamil Nadu
region. Such studies shall be carried out for different regions, 8. Twisdale, L.A and Vickery, P. J , “Research on thunder-
for different reference periods in order to verify the general storm wind design parameters”, Jl. of Wind Engg. and
applicability of Diffused distribution as a prior distribution. Ind. Aero Dyn. Vol. 41, Nos. 1-3, 1992, PP 545–556.
Towards this aim, future work will be continued at SERC.
9. Wen, Y-K., “Wind Direction and Structural Reliability”,
Jl. of Struct. Engg., ASCE, Vol. 109, No.4, 1983, pp
ACKNOWLEDGEMENT
1028–1041.
This paper is being published with the kind permission of the 10. Ang, A. H.-S., and Tang, W. H., “Probability concepts in
Director, SERC, CSIR. Engineering Planning and Design”, Vol. 1, Basic prin-
ciples, 1975, John Wiley and Sons.
REFERENCES
11. Benjamin, J. R., “Probabilistic Models for Seismic Force
1. Kramer, S. L. “Geotechnical Earthquake Engineering”, Design”, Jl. of the Struct. Divn., ASCE, Vol. 9, No. ST5,
(3rd ed.), 1996, Pearson Education. May 1968, pp 1175–1196.

144 JOURNAL OF STRUCTURAL ENGINEERING, VOL. 37, NO. 2, JUNE-JULY 2010