Journal of Physics: Conference Series

PAPER • OPEN ACCESS Related content

- Horizontal Displacement Vector Analysis
Marked point process for modelling seismic activity in Ujong Muloh GPS Station (UMLH)
Sumatra Island on March 27 – April 25,
(case study in Sumatra and Java) 2012
S S Pamungkas, S Koesuma and Budi
Legowo
To cite this article: Hasih Pratiwi et al 2018 J. Phys.: Conf. Ser. 1022 012004 - The concept of geothermal exploration in
west Java based on geophysical data
Eddy Z Gaffar

- Different responses of chlorophyll-a

concentration and Sea Surface
View the article online for updates and enhancements. Temperature (SST) on southeasterly wind
blowing in the Sunda Strait
A Wirasatriya, Kunarso, L Maslukah et al.

This content was downloaded from IP address 103.94.190.18 on 11/10/2019 at 04:29

ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

Marked point process for modelling seismic activity (case

study in Sumatra and Java)

Hasih Pratiwi1, Lia Sulistya Rini2 and I Wayan Mangku3

1
Statistics Department, Universitas Sebelas Maret, Surakarta, Indonesia
2
Mathematics Department,Universitas Sebelas Maret, Surakarta, Indonesia
3
Mathematics Department, Institut Pertanian Bogor, Bogor, Indonesia

Corresponding email: 1hpratiwi@mipa.uns.ac.id, 2liasulistyarini@gmail.com,

3
wayan.mangku@gmail.com

Abstract. Earthquake is a natural phenomenon that is random, irregular in space and time.
Until now the forecast of earthquake occurrence at a location is still difficult to be estimated so
that the development of earthquake forecast methodology is still carried out both from
seismology aspect and stochastic aspect. To explain the random nature phenomena, both in
space and time, a point process approach can be used. There are two types of point processes:
temporal point process and spatial point process. The temporal point process relates to events
observed over time as a sequence of time, whereas the spatial point process describes the
location of objects in two or three dimensional spaces. The points on the point process can be
labelled with additional information called marks. A marked point process can be considered
as a pair (x, m) where x is the point of location and m is the mark attached to the point of that
location. This study aims to model marked point process indexed by time on earthquake data in
Sumatra Island and Java Island. This model can be used to analyse seismic activity through its
intensity function by considering the history process up to time before t. Based on data
obtained from U.S. Geological Survey from 1973 to 2017 with magnitude threshold 5, we
obtained maximum likelihood estimate for parameters of the intensity function. The estimation
of model parameters shows that the seismic activity in Sumatra Island is greater than Java
Island.

1. Introduction
Indonesia is in a position that is vulnerable to the threat of natural disasters. This is due to the
geographical position of Indonesia which is at the meeting of three large plates of earth namely the
Eurasian plate, the Indo-Australian plate and the Pasific plate. The encounter and movement of three
large earth's plates creates an active volcano ranks and earthquake potentials in most of the Indonesian
archipelago.
Sumatra Island and Java Island are two areas of Indonesia that are often affected earthquake.
Among the various regions in Sumatra, the areas with high earthquake potential are Aceh and West
Sumatra. Based on data from United State Geological Survey (USGS) in 1960 to June 2017 in Aceh
Province area, there are at least 2400 earthquakes with magnitude size above 4 on Richter scale. As

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

for the area of West Sumatra at the same time interval, there are at least 900 earthquakes with
magnitude measurements above 4 on the Richter scale. Since 1990 until June 2017, there have been
recorded 473 earthquakes that occurred in Java area with magnitude grater than or equal to 4 on
Richter scale. Examples of major earthquakes that occur on the Java Island include the Yogyakarta
earthquake which occurred on May 2006 with magnitude 5.9 on the Richter scale. A month later, an
earthquake accompanied by a tsunami occurred in Pangandaran, Ciamis Regency, West Java. Then in
September 2009, a major earthquake also occurred in Tasikmalaya Regency, West Java.
Point process is a major subject in statistical seismology. Point process is a part of stochastic
process that can be used to describe events in a particular pattern. In this process, earthquakes are
viewed as a random collection of points in a space, where each point represents the time or location of
an event [2]. The occurrences of earthquakes in certain locations can be viewed as points, whereas the
size associated with the occurrence of an earthquake is the magnitude or depth [10]. In this process,
the occurrence of the event is memory less and mutually independent from occurrences of other
events.
In seismological research, some models that can be used to describe the aftershock pattern are
Omori's law and Omori-Utsu's law. These mathematical models are not always accurate in modeling
earthquakes for long periods of time. The relation between mainshock and aftershocks can be
illustrated by other mathematical models which are the development of Omori-Utsu's Law [5,8,9].
Further research related to stochastic modeling to analyze seismic activity in a region with point
process approach has also been done by some experts [7,12]. In this article we discuss the marked
point process which is indexed by time. This process is presented by an intensity function which
considering the occurrence of mainshock and its aftershocks. Then we will estimate this function
through maximum likelihood method with considering the history process.

2. Marked Point Process

The point process is defined as a random collection of points located on a particular region. The points
of this process can be expressed as events, time of events, location of events, or time and location of
events [6]. The construction of a point process can be done by several approaches, one of them is
through a counting process. The state space and the parameter space of this counting process are non
negative integer sets and subsets of Rd with d ≥ 1. A counting process X(t) is defined as the number of
events occurring during time t. The counting process X(t) satisfies the following properties:
1) X(t) ≥ 0
2) X(t) is integer
3) If s<t, then X (s) ≤ X (t)
4) For s<t, then X(t) - X(s) denotes the number of events occurring in the interval (s, t].
The counting process has independent increments property. It means that the number of
occurrences between time s and t, X(t) - X(s), independent of the number of events occurred up to time
s. The counting process has stationary increments property if the distribution of the number of
occurrences at an interval depends only on the length of the interval. There are two types of point
processes, i.e. the temporal point process and spatial point process. The temporal point process relates
to events observed at a sequence of time, while the spatial point process relates to the location of the
object in a two or three-dimensional space [11].

2
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

The points of the point process are often labeled with additional information called marks. A
marked point process can be considered as a pair (X, m) where x is the point of location and m is the
mark or quantity attached to the point.
Definition 1. A marked point process in a space S with the mark in a space M is a point process Y at S
× M such that NY (K × M) < ∞, for all compact sets K ⊂ S.
This definition describes that the projected process (from unmarked points) is finite local [2].
Theorem 2. Let Y be the marked point process on S with marks in M. Let X be projected at S (from
unmarked points), then
1) X is a Poisson process in S with intensity μ, and given X, the mark attached to point in X
independently and identically distributed with joint distribution Q on M;

2) Y is a Poisson process in S x M with intensity measure μ  Q.

3. Conditional Intensity Function

An important concept of this temporal marked point process is the conditional intensity function on
the history of a time-dependent process. This concept describes the probability of an event in the
future which depends on previously information up to the present time. In seismology this function
can describe seismic activity in a region for a certain period of time. The ground intensity function is
the rate of events depend on time, and not only influenced by the present time but also influenced by
the previous events, or history process [3]. This function describes the instant Poisson rate as a
function that depends only on time. Let Nδ(t) be the number of occurrences at the time interval [t, t +
δ], the ground intensity function can be written as
1
 (t ,1 ,, m | H t )  lim P( N (t )  0 | H t )
 0 
where Nδ(t) is the number of occurrences at the time interval [t, t + δ).
Furthermore we discuss the intensity function on the marked point process indexed by time with
the magnitude as its marked. According to Kagan [4], the rate of aftershocks is defined as

C1
N (t ) 
t (1)
where C1 is parameter and t is time measured from the beginning of a mainshock. Nowadays a more
complex equation is used to estimate the rate of aftershocks:

C1
N (t ) 
(C2  t ) p (2)
where C1, C2, and p are parameters. Equation (2) is a modification from Omori’s Law (1) and called
modified Omori's Law or Omori-Utsu’s Law.
In the epidemic model, the number of individuals living at time t is controlled by the rate of
immigration, birth rate and death rate. In the case of earthquake, immigration refers to the occurrence
of independent events, whereas birth is associated with the sequence of events triggered by the
previous event. The process of birth and death depends on age for each individual x living at time t.

3
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

For the next interval (t, t + dt), there are probability of one birth, g(x)dt, and probability of one death,
h(x)dt. Birth and death are independent of each other. If the birth process considers immigration at rate
μ per unit time and there is no death, h (x) = 0 then the process has the following conditional intensity:

(t) =  + ti<t g(t-ti) (3)

where ti is the ith occurrence time and N(t) is the number of occurrences, {ti}, in (0, t).
Equation (3) can be extended to the multivariate point process, {tim}, and the conditional intensity
function for the discrete magnitude value of j and m is

j(t) = j + ΣmΣ ti<t gjm (t-tim ) (4)

If we assume that gjm(t) = c(m)gj(t) and the point process N(t) = ΣmNm(t) then the conditional intensity
is given by λ(t) = Σjλj(t) and

(t) = μ + Σti<t c(mi) g(t-ti) (5)

where ti is the occurrence time of N (t), mi is the magnitude corresponding to ti, and g(t) = Σj gj(t).
Based on the Omori-Utsu law (2) and the epidemic type model (5), we require the following
assumptions:
1) The ground seismic rate in a region is constant μ, which means that the ground seismic occurs
according to the stationary Poisson process with constant rate μ.
2) All earthquake events, including aftershocks, may cause subsequent aftershocks, and any
subsequent aftershocks generate the second subsequent independently. The number of
aftershocks of a major earthquake with magnitude M follows Poisson distribution with average
c(M):

c(M) = eα(M-M0) (6)

where M0 is the magnitude threshold while  and α are parameters.

3) Aftershocks of a mainshock with magnitude M occurring at certain time intervals appropriate
with Poisson process with rate c(M)g(t-t0) with g(t) is a normalized Omori function at a certain
time so that ∫ g(t)dt = 1 and is expressed as

p 1 t
g (t )  (1  )  p (7)
c c
where c and p are parameters, while t0 is the occurrence time of the mainshock.
4) Magnitude distribution is independent with rate of occurrence. By using the explicit form of
the Gutenberg-Richter relationship, the probability density function of the magnitude is
f(M) = βeβ(M-M0)
where β corresponds with b, β = 2.30b.
By substituting c(mi) with equation (6) and g(t - ti) with equation (7) in equation (5), the conditional
intensity function of marked point process is

4
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

t  ti  p
 (t )     t t e ( M M ) (1 
i 0
)
i
c (8)

4. Seismicity Modeling for Sumatra Island and Java Island

In this section we estimate the conditional intensity function for earthquake data in Sumatra and Java.
The earthquake data is a secondary data sourced from the United States Geological Survey (USGS).
The earthquake data contains times of ith earthquake occurrence, ti, and magnitude of ith earthquake
occurrence, mi. We use earthquake data occurred from January 1973 to April 2017 with a magnitude
of ≥ 5 and a depth of ≤ 70 km.
4.1 Sumatra Earthquake Data
In the Sumatra Island for the period time January 1973 to April 2017 with magnitude of ≥ 5 and depth
of ≤ 70 km there are 2,114 earthquakes. Based on the maximum likelihood estimate method, the
conditional intensity function (8) for Sumatra earthquake data can be written as

Figure 1. Plot of Magnitude and Time of Sumatra Earthquake Data

5
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

Figure 2. Plot of Logarithmic of Conditional Intensity Function for Sumatra Earthquake Data
Plot of magnitude and time of Sumatra earthquake data is presented in Figure 1 while the plot of
logarithmic of conditional intensity function is presented in Figure 2. Based on Figure 1 it can be seen
that on March 11, 2012, September 12, 2007, March 28, 2005, March 11, 2012 and December 26,
2004 there was an earthquake with large magnitude occurring in off the west coast of northern
Sumatra, southern Sumatra, northern Sumatra, and off the west coast of northern Sumatra. Based on
Figure 2 it appears that these areas also have high intensity functions.

4.2 Java Earthquake Data

In Java Island for the time period from January 1973 to April 2017 with magnitude ≥ 5 and depth ≤ 70
km there are 488 earthquakes. Based on maximum likelihood estimate method, the intensity function
of marked point process for Java earthquake data can be written as:

Plot of magnitude and time of Java earthquake data is presented in Figure 3 whereas plot of the
logarithmic of conditional intensity function is presented in Figure 4. Based on Figure 3 it can be seen
that on September 7, 1974, April 3, 2011, October 25, 2000, September 2, 2009 and July 17, 2006,
there was an earthquake with large magnitude occurring in Southern Java and the Sunda Strait. Based
on Figure 4, it is seen that in those areas the conditional intensity function for Java earthquake data is
also high.

6
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

Figure 3. Plot of Magnitude and Time for Java Earthquake Data

Figure 4. The Plot of Logarithmic of Conditional Intensity Function for Java Earthquake Data

4.3 Seismic Activity

In this section seismic activity on Sumatra and Java islands is compared by considering all parameters
of their intensity function. Table 1 shows that parameter value  of Sumatra earthquake data is greater
than Java which means that the base seismic rate of Sumatra Island greater than Java Island.
The parameter value  of Java earthquake data is greater than Sumatra which means that the
productivity of aftershocks in Java is higher than Sumatra, while the parameter value  of Sumatra
earthquake data is greater than Java which means that the efficiency of earthquake with certain
magnitude in producing aftershocks in Sumatra is higher than Java. Furthermore, the parameter value
c of Sumatra earthquake data is greater than Java which means that in time scale, the decay rate of
aftershocks in Sumatra is higher than Java, and parameter p of Java earthquake data is greater than
Sumatra which means that the decay rate of aftershocks in Java is higher than Sumatra.

7
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

Table 1. Parameter Estimate of Conditional Intensity Function for Earthquake Data

in Sumatra and Java
Parameter Sumatra Java

 0.0291 0.0152

 0.2240 10.8592

 2.4417 0.0875

c 0.0199 0.0043

p 0.9930 1.0484

5. Conclusion
Intensity function on a marked point process can modeling seismic activity in a region. By using
maximum likelihood estimate method we can apply this model to determine earthquake characteristic
in Sumatra and Java islands. Base seismic rate in Sumatra is greater than the one in Java. Although
decay rate of aftershocks in time in Sumatra is higher than the one in Java, but impact of earthquake
occurrences in Sumatra is greater than the one in Java. This process have not yet considered the spatial
component for its intensity function, so the improvement of intensity function (5) can be done to
model seismic activity in a region by considering magnitude, time occurrence, and location.

6. References
[1] Console R, M Murru and G Falcone 2010 Retrospective forecasting of m≥4.0 earthquakes in
new zealand, Pure and Applied Geophysics 167
[2] Daley D J and D Vere-Jones, 2007 An introduction to the theory of point process, volume II:
general theory and structure, (Springer: New York)
[3] Harte D 2010 Pt process: An R package for modelling marked point processes indeed by time,
J. of Stat. Soft. 35(8)
[4] Kagan Y Y 2014 Errthquakes: models, statistics, testable forecasts (John Wiley and Sons, Ltd)
[5] Kumazawa T and Y Ogata 2014 Nonstationary ETAS models for nonstandard earthquakes The
Annals of Applied Statistics 8(3)
[6] Ogata Y 1988 Statistical models for earthquake occurrences and residual analysis for point
processes J. of the Amer. Stat. Assoc. 83 9–27
[7] Ogata Y 1999 Seismicity analysis through point process modelling: A review Pure and Applied
Geophysics 155 471–507
[8] Ogata Y and H Tsuruoka 2016 Statistical monitoring of aftershock sequences: A sase study of
the 2015 Mw 7.8 Gorkha, Nepal Earthquake, Earth, Planets and Space
[9] Omi T, Y Ogata, Y Hirata and K Aihara 2014 Estimating the ETAS model from an early
aftershock sequence Geophys.Res. Lett. 50 850–57
[10] Pratiwi H, I Slamet, D R S Saputro and Respatiwulan 2017 Self-exciting point process in
modeling earthquake occurrences Journal of Physics: Conference Series 855 1-7
[11] Schoenberg F P 2000 Introduction to Point Processes (UCLA: New York)
[12] Zhuang J 2000 Statistical modelling of seismicity patterns before and after the 1990 Oct 5 Cape
Palliser Earthquake, New Zealand New Zealand Journal of Geology & Geophysics 43 447–
60

8
ICoSMEE IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1022 (2018)
1234567890 ‘’“” 012004 doi:10.1088/1742-6596/1022/1/012004

Acknowledgments
The authors would like to thank Ministry of Research, Technology, and Higher Education of the
Republic of Indonesia and Universitas Sebelas Maret for supporting this research through Grant of
Pasca Doctor 2017.