Probabilistic hazard assessments are a fundamental tool for assessing the
threats posed by hazards to communities and are important for underpinning
evidence-based decision-making regarding risk mitigation activities. Indonesia has
been the focus of intense tsunami risk mitigation efforts following the 2004
Indian Ocean tsunami, but this has been largely concentrated on the Sunda
Arc with little attention to other tsunami prone areas of the country such
as eastern Indonesia. We present the first nationally consistent
probabilistic tsunami hazard assessment (PTHA) for Indonesia. This assessment
produces time-independent forecasts of tsunami hazards at the coast using data from
tsunami generated by local, regional and distant earthquake sources. The
methodology is based on the established monte carlo approach to probabilistic
seismic hazard assessment (PSHA) and has been adapted to tsunami. We account
for sources of epistemic and aleatory uncertainty in the analysis through the
use of logic trees and sampling probability density functions. For
short return periods (100 years) the highest tsunami hazard is the west coast
of Sumatra, south coast of Java and the north coast of Papua. For longer
return periods (500–2500 years), the tsunami hazard is highest along the
Sunda Arc, reflecting the larger maximum magnitudes. The
annual probability of experiencing a tsunami with a height of

Indonesia has the third highest population exposure to tsunami in the World,
with an estimated 5.5 million people at risk of once-in-500-years tsunami

Map showing the local sources (red lines, thin black text) used
in the PTHA and historical tsunami run-up observations (coloured circles)
from

Since the devastating Indian Ocean tsunami in 2004 there has been a large
amount of work focused on reducing the impact of future tsunami in
Indonesia. This has included the completion of detailed tsunami inundation
studies for high-risk locations such as Padang, Denpasar and Cilacap

To effectively address and mitigate the hazard posed by tsunami, accurate and
uniform information on tsunami hazard is required for the basis of risk
studies to prioritise the implementation of mitigation
measures. The probabilistic tsunami hazard assessment (PTHA) framework is one
such method that allows broad scale assessments of tsunami hazard

This paper presents the first national probabilistic tsunami hazard assessment for Indonesia. The outputs are long-term forecasts of tsunami hazard at the coast and are presented as tsunami hazard curves which describe the annual probability of exceeding a range of tsunami heights for sites around the coast of Indonesia. These hazard curves are then converted to hazard maps describing the tsunami height at the coast (1 m water depth) for a range of return periods and probability maps illustrating the annual probability of exceeding a set of given tsunami heights related to the warning thresholds of the Indonesian tsunami early warning system (InaTEWS). The tsunami hazard is also disaggregated to identify unit sources that contribute most to the tsunami hazard for a given coastal location, which can subsequently be used to select scenarios for detailed inundation modelling. Together, these results provide a nationally consistent assessment of tsunami hazard for Indonesia that can be used to underpin tsunami risk mitigation activities.

The aim of a PTHA is to calculate the probability of exceeding a set of tsunami heights at the coast or near shore. This is accomplished by superposition of results from simulations of unit source tsunami to increase computation efficiency.

The PTHA framework can be summarised as follows:

Define tsunami sources (fault geometries) to be
included in the analysis (Figs.

For each source discretize the fault into smaller sub-faults (Fig.

For each source create a synthetic earthquake catalogue based on the slip
rate of the fault and a magnitude-frequency
model (e.g. the exponential model of

For each subfault, calculate the seafloor deformation from 1 m of “unit
slip”
and propagate the tsunami from source to site (i.e. to the hazard points along
the coast Fig.

For each event in the synthetic catalogue, estimate the tsunami amplitude
at the sites by summing the tsunami waveforms of all individual subfaults that
make up that event, then scale the summed tsunami by the amount of slip for that
event, assuming uniform slip (Fig.

Sum the probabilities of individual events exceeding a range of pre-defined
tsunami amplitudes (Fig.

Plot probability of exceedance vs. tsunami amplitudes to generate a tsunami
hazard curve for each hazard site (Fig.

In this assessment the hazard is defined as the tsunami wave amplitude, where the term amplitude is identical to wave height, keeping with common convention in the tsunami hazard literature.

For this assessment only tsunami generated by submarine earthquakes are
considered. While tsunami can be generated from other mechanisms,
such as submarine mass failures, submarine and sub-aerial volcanic eruptions
and meteorite impacts, it is difficult to associate probabilities of such
events, which prohibits including them in a probabilistic framework

Seismic sources of tsunami that could impact Indonesia are divided into
local, regional and distant sources. While it is assumed the hazard for
Indonesia will be dominated by local sources, regional and distant sources
are included to provide an all-encompassing assessment of the tsunami hazard.
Regional sources are those likely to provide short warning times (less
than a few hours), while distant sources are those that will have long
warning times (many hours). Furthermore, this also enables a database of
earthquake sources and tsunami events to be generated, which can be used for
future tsunami inundation modelling or integration into the existing InaTEWS

Map showing regional and distant sources (red lines) used in the PTHA. The black box is the Indonesia region where the 1 arcmin bathymetry was used.

Illustration showing the five main steps of PTHA.

Earthquake sources used in the PTHA. Local sources are taken from

20 local sources (Fig.

The dip of seismic sources is a critical parameter in any tsunami modelling,
since the resulting tsunami heights are sensitive to the fault dip

The regional and distant source model is represented by twelve subduction
megathrusts around the western and northern Pacific Ocean and the Makran
subduction megathrust in the northwestern Indian Ocean. The subduction zone
geometry and recurrence rates were taken from the PTHA for Australia

Each source is discretised into rectangular subfaults prior to the tsunami
summation calculation. The dimensions of the subfault represent the smallest
magnitude considered in the synthetic catalogue. For local subduction
sources, each subfault measures 40 km along strike by 25 km down dip, which
is approximately a

The use of unit sources have become increasingly popular for PTHA

Tsunami propagation modelling was carried out using the finite difference
code of

For each propagation simulation, tsunami time series were recorded at 3050 hazard points located at the 100 m water depth contour at approximately 5–10 km intervals around the Indonesia coastline. In water depths shallower than 100 m, non-linear terms in the shallow water wave equation can become important and the unit source approach can not be applied.

Since the goal of the study is to estimate the probabilistic tsunami hazard
at the coast, we extrapolate the tsunami height from the 100 m water depth
contour to the coast (1 m water depth) using Green's Law. This approach has
been used in Japan

Incorporating uncertainties is a critical component of any probabilistic
assessment. For this study there are two types of uncertainty that are dealt
with: aleatory uncertainty and epistemic uncertainty. Aleatory uncertainty is
the variability from randomness in a natural process and cannot be reduced
with increasing knowledge. In PTHA, an example would be the variability of
slip of a future tsunamigenic earthquake. Aleatory uncertainty can be
accounted for by sampling from a probability density function

Sources of epistemic uncertainty that are included in the PTHA are fault
segmentation, slip rate, magnitude-frequency distribution (MFD) type and
maximum magnitude. Variations in the values for these parameters are included
in logic trees for each source zone and represent different possible models
for the input parameters (Fig.

Schematic logic tree for seismic sources in the PTHA. The structure of the logic tree and weightings are for the Sunda Arc.

The Sunda Arc has a complex logic tree (Fig.

There are three main sources of aleatory uncertainty in the PTHA:
computational modelling uncertainty (

Modelling uncertainty can be described as errors in numerical models that are
used to model seafloor displacement and tsunami propagation. By comparing
tsunami waveforms from historical events to modelled waveforms of the event,
an estimate of this uncertainty can be quantified.

Uncertainty in the dip of the fault would normally be treated as an epistemic
uncertainty, since increased knowledge of the fault geometry, for example
from seismic reflection data, could reduce the uncertainty of the dip.
However, this would require recomputing the tsunami unit sources for each
variation in dip which would make this approach more computationally
intensive and may eliminate the benefit of using tsunami unit sources.
Instead, the dip uncertainty is included as an aleatory uncertainty.

The distribution of earthquake slip in future earthquakes is assumed to be a
random process. When the tsunami heights are calculated from the events in
the synthetic earthquake catalogue the slip is modelled as uniform across all
subfaults. To include this uncertainty in the tsunami unit source summation
would require a large number of slip distributions to be modelled for every
event in our catalogue (

The three sources of aleatory uncertainty are combined to compute a total
sigma term (

Tsunami hazard curves for the six tsunami zones as defined by

Tsunami hazard map showing maximum tsunami amplitudes at the coast
with a 1 in a 100 chance of being exceeded annually (

Tsunami hazard map showing maximum tsunami amplitudes at the coast
with a 1 in 500 chance of being exceeded annually (

Combining all the information from the source models and logic trees, a
synthetic catalogue is generated which represents the range of possible
earthquake magnitudes, locations and sources for every logic tree branch. The
catalogue was generated by iterating through each magnitude in the MFD, from
the minimum magnitude of

For each event in the synthetic catalogue, the tsunami hazard is calculated
at each hazard point along the coast by summing the tsunami unit source
contributions from the subfaults that make up that event and by scaling the
tsunami height by the average event slip. For each site, this results in a
list of tsunami heights (

By plotting the probability of exceeding a range of tsunami heights
(

The results from the PTHA are developed into hazard curves, hazard maps, probability maps and disaggregation maps. The maps are all derivative products of the hazard curves and can be used to provide information on different aspects of the tsunami hazard. For instance, the hazard maps illustrate the tsunami height expected at the coast over a given return period and are useful for understanding the size of tsunami expected for a community over a fixed time. Probability maps define the probability of exceeding a given tsunami height at the coast, which is linked to threat thresholds in the InaTEWS system. This is useful for assessing the annual probability of experiencing a tsunami warning, particularly for prioritising areas most likely to experience tsunami that are a threat to life safety. Disaggregation maps define the relative contribution of each source to the hazard at a particular site for a given return period or tsunami height. This information is extremely useful for identifying sources and selecting scenario tsunami events for detailed tsunami inundation modelling.

Tsunami hazard map showing maximum tsunami amplitudes at the coast
with a 1 in 2500 chance of being exceeded annually (

Hazard curves are developed for each hazard site located at the coast around
Indonesia. Hazard curves are the fundamental output from the PTHA. The hazard
curves are grouped together according to the tsunami zones in Indonesia as
defined by

The hazard curves in Western Indonesia show that the mean hazard for Sumatra and
Java is of a similar magnitude, however the spread of the hazard curves is
larger for Sumatra. This reflects the site location of the hazard curves in
the Sumatra zone, as some are located on the eastern coast of the Mentawai
and Nias islands that are protected from tsunami originating to the west
of these islands. Furthermore, the hazard curves on the west coast of the
offshore islands have higher hazard than the west coast of Sumatra and the
south coast of Java, as they are located adjacent to the Sunda Arc source. The
Java hazard curves have a distinct bulge above the 0.001 annual probability
(

In Eastern Indonesia, the tsunami hazard curves for the Banda, Papua and
Sulawesi zones are similar. The mean probability of exceeding 1 m coastal
tsunami height is similar (

The Kalimantan zone (including the north coast of Java and the east coast of
Sumatra) has the lowest tsunami hazard of any zone due to being remote from
any major tsunami sources besides the north Sulawesi megathrust and the
Makassar thrust in Sulawesi. These areas are also adjacent to shallow
(

The hazard curves can be used to prioritise which areas in Indonesia have the highest tsunami hazard. The results show that eastern Indonesia has a similar tsunami hazard to western Indonesia at higher annual exceedance probabilities, whereas western Indonesia has a higher tsunami hazard at lower probabilities of exceedances.

Hazard maps are generated for return periods of 100, 500 and 2500 years.
Figure

At the 500-year return period (Fig.

The hazard for the 2500-year return period (Fig.

Annual probability of experiencing a tsunami with a height at the
coast of

Hazard disaggregation for the 500-year return period at hazard
sites in Mataram, Lombok island

In Sumatra it is evident that the hazard varies significantly along the west coast of Sumatra at all return periods due to the presence or absence of offshore islands. Where offshore islands are present, the coastal areas of Sumatra have a lower hazard, approximately half that of the islands. This is due to two factors. Firstly the islands are located directly above the megathrust; if islands are present during the coseismic displacement, the displaced water volume will be less, resulting in a smaller tsunami on the west coast of Sumatra directly east of the islands (e.g. Padang city). Secondly, the islands act as a barrier for tsunami that have slip near the trench, which results in the west coast of the offshore islands receiving the largest tsunami waves and the west coast of Sumatra being more protected. With very short arrival times to the Mentawai and Nias islands and the very high tsunami hazard, this area clearly is of concern for tsunami mitigation efforts. In comparison, the south coast of Java has less variation along the coast due to the absence of offshore islands.

Tsunami probability maps, which illustrate the probability of exceeding a
given tsunami height, are less common in PTHA but are useful for prioritising
tsunami mitigation activities based on the likelihood of having a significant
tsunami. The InaTEWS sets its warning thresholds at 0.5 m for a
tsunami warning and 3.0 m for a major tsunami warning

Figure

The probability of experiencing a tsunami with a height at the coast greater
than 3.0 m, which would trigger a major tsunami warning, varies along
the Sunda Arc depending on whether offshore islands are present or not. The
probability that the Mentawai and Nias islands would experience a 3.0 m or
greater tsunami in any given year is 5 %, compared to less than 1 % for
the west coast of Sumatra, including the city of Padang. However, it is noted
that the tsunami hazard in Padang is considered to be much higher in the
short-term if time dependence is captured

These results highlight that areas in eastern Indonesia such as north Papua and north Sulawesi have similar probabilities of experiencing a major tsunami warning as western Indonesia; however, east of Lombok no tsunami scenarios exist in the InaTEWS database. The results presented here warrant an extension of the InaTEWS to develop a scenario database for eastern Indonesia.

The PTHA methodology aggregates all possible tsunami events to generate
tsunami hazard curves. The tsunami hazard can also be disaggregated by source
and magnitude. Tsunami hazard disaggregation maps are developed to quantify
how much each source contributes to the hazard for a single site and a single
return period. They are useful for scenario event selection when detailed
inundation modelling is undertaken. Figure

For sites along the Sunda Arc the contribution is solely from the Sunda Arc megathrust. This contrasts with sites in eastern Indonesia where there are more local and regional sources, and the tsunami hazard comes from more than one source. Any detailed inundation modelling should use disaggregation information for event selection to ensure that all significant source zones are covered in the analysis.

This study represents the first attempt at developing a nationally consistent
probabilistic tsunami hazard assessment for Indonesia, one of the most
tsunami-prone countries in the World

Until now, the majority of tsunami risk mitigation activities have been
located along the Sunda Arc, due to a wealth of data on the earthquake history
of the Sumatra megathrust

The PTHA assumes earthquakes that generate tsunami follow a Poisson
process; that is, they are time-independent. This assumption is fundamental in
PSHA and PTHA and over the long-term is usually a valid approach. However,
earthquake faults have been shown to interact on short time scales, and
earthquake sequences are now being recognised

A number of limitations are evident from our PTHA which will be improved upon
in future model updates. The assessment presently only considers earthquake
sources of tsunami and not other sources, such as submarine landslides
(earthquake triggered or independent of earthquakes), volcanic collapse, or
volcanic products such as pyroclastic flows. All of these
sources have generated significant tsunami in Indonesia in the past 200
years. These events include the 1815 Tambora eruption that caused large
pyroclastic flows that generated large tsunami

Since the 2004 Indian Ocean tsunami, the majority of tsunami hazard studies
have focussed on Sumatra and to a lesser extent Java. A result of this focus
is that the recurrence rate of previous earthquakes has been well-studied,
using coral uplift data above the trench

Uncertainty arising from random slip distributions is incorporated into the PTHA by means of sampling a probability density function. A better approach may be to include heterogeneous slip by assigning different slip values to subfaults that are used for each event. While this approach would be more computationally intensive than the current implementation, it would produce a more direct means of incorporating slip uncertainty.

Finally, an area of further research is the inclusion of “tsunami earthquakes” in PTHA. Tsunami earthquakes are earthquakes with large tsunami relative to the magnitude of the event. At present tsunami earthquakes are not treated any differently in the PTHA. However, there are ways of including these in future PTHA studies, for example by allowing increase slip for earthquakes located on the shallow portion of the plate interface in areas where tsunami earthquakes are known to occur (e.g. Java Trench and Mentawai islands).

Since a PTHA is a statistical forecast of the future tsunami hazard, testing
should be applied to rate the performance of these models

This study developed the first national probabilistic tsunami hazard assessment for Indonesia and provides forecasts of long-term tsunami hazard
at the coastline from earthquake sources. Results show that the annual
probability of experiencing a tsunami with a height of

This work was supported by the Indonesian National Disaster Management Agency
(BNPB) and the Australian government through the Australia-Indonesia Facility
for Disaster Reduction (AIFDR). Early versions of the manuscript benefited
from comments by David Burbidge, Gareth Davies, David Robinson and
John Schneider. Some figures were generated using the generic mapping tools
(GMT) package of