Research Unit #2: Hazard Analysis with Bayesian approach
- Goals: Seismic, Volcanic and Tsunami Hazard for the city of
Naples
- Responsible: Licia Faenza (INGV)
- Participants: Laura Sandri (INGV), Simona Pierdominici
(INGV), Sebastian Hainzl (GFZ), Alberto Michelini (INGV)
- Working groups:
WG2 (L. Faenza),
WG3 (L. Sandri),
WG4 (A. Grezio)
DESCRIPTION OF THE RESEARCH ACTIVITY OF RU2
The main goal of RU2 is the definition and quantification of the hazard
coming from natural events in the target area of the city of Naples, by
following the procedures and guidelines of RU1 for a homogeneous
definition
of Bayesian multi-risk. In particular, RU2 will be focused on the
assessment of hazard from earthquakes, volcanic eruptions and tsunamis,
paying special attention to the correct evaluation of uncertainties.
Since we will be using a Bayesian approach, for each hazardous
phenomenon
under study we will set up a prior probability distribution, based on
models and/or expert opinions. The observations relative to the past
occurrence of the hazardous phenomenon in the target area will form the
likelihood function. It is likely to suppose that these past data are
relative to all the possible factors that can affect the process causing
the occurrence of hazardous phenomenon (for example, a tsunami run-up
catalog most likely includes tsunami data due submarine earthquakes,
submarine landslides, or due to possible volcanic phenomena able to move
a great mass of water, and generate a tsunami wave). This implies that
particular attention has to be paid in order to analyze and include all
such factors, in the construction of the prior probability distribution.
Only in this circumstance, the prior estimate can be homogenous with
respect to the likelihood function, allowing their combination, via the
Bayes' theorem, to obtain a meaningful posterior distribution.
In the following, a detailed structure of the RU2 and the tasks of the
personnel are provided.
Task 1: Seismic Hazard
Approaching seismic hazard assessment through Bayesian strategies is
innovative. In literature, the techniques commonly applied are not
Bayesian, and can be distinguished into two broad classes, on the basis
of the models applied to describe earthquake distributions in term of
time, space and magnitude. One class is characterized by a probabilistic
modelling (this, in turn, is divided into two subclasses, based, one on
the so called "source approach", and the other one on the so called
"site
approach"). The second class is based on a deterministic modeling. In
this project, only the probabilistic procedures will be applied. Through
a Bayesian approach, we will assess the probabilistic seismic hazard by
integrating the assessments from source and site approaches, in order to
include in the calculation the different types of information that these
two methods bring.
Probabilistic seismic hazard, i.e. the quantification of the probability
of exceeding a given ground motion value in the exposure time window in
a given area, evaluated through the source approach, is usually made on
the basis of statistical modeling of earthquake occurrence and of
deterministic
modeling of ground motion through the so-called attenuation laws,
predicting
the propagation of seismic waves in the Earth.
Three factors concur to the probabilistic description of earthquake
occurrence:
their distributions in time, space and size. In this project, we will
apply
different models, already proposed by the scientific community, to
define these earthquake distributions. These models are based on
different
hypothesis on the seismogenetic process; this gives to our analysis a
much wider perspective of the problem, covering the possibility that a
single model is not able to entirely explain the physics of a stochastic
natural process.
In the standard probabilistic seismic hazard assessment, the temporal
earthquake occurrence is random (i.e., Poissonian), as it has been done
in the Italian seismic hazard map MPS04. In the project, we will apply
as well more realistic temporal distributions of the seismicity, to
include patterns that are widely accepted by the scientific community,
like the tendency of earthquakes to occur in clusters. The earthquake
spatial distribution will be represented by a spatial smoothing of past
seismicity. The last point to define is the earthquake size
distribution.
It is widely accepted that the earthquake size frequency relationship
follows a power law, the so-called Gutenberg-Richter (G-R) law. This
power law has universal applicability, to all spatial scales and on a
wide range of earthquake magnitude, and it has been used also in MPS04.
During this project, a new calibration of the G-R for the test area
using the data from the seismic catalog will be done. Moreover, a new
Bayesian method to infer the earthquake size distribution will be
applied.
Starting from the prior hypothesis of the power law earthquake size
distribution, the method re-calibrates the probabilities of the
different
sizes through a likelihood function, designed on the basis of the local
size data of past earthquakes.
Once the earthquake probabilistic distribution is set up, the prior
distribution of hazard assessment will be evaluated by using MonteCarlo
catalog simulation; each catalog will be intended as a representation of
the real seismicity. To define the expected ground shaking, the ground
motion attenuation laws, calibrated for the target area and already
available in literature, will be applied. Finally, a new quantification
of site effects is contemplated in the project.
This result will constitute the prior model for the Bayesian approach.
The likelihood function will be defined through the data relative to
macroseismic information and, for the most recent earthquakes, through
shaking data, due to past events that have caused felt effects in
Neapolitan area.
The first step is a seismological study of the area capable to generate
damaging earthquakes and felt effects in Naples, in order to have a
complete view of tectonic and morphologic setting of the area. By
merging
the information from all the available databases, through the ShakeMap
software, implemented for Italy in general, and specifically for Naples,
we will produce an atlas of shaking maps for the past earthquakes in the
test area. The databases used will be CPTI - Parametric Catalog of
Italian
Earthquakes - one of most complete historical catalog worldwide; DBMI -
Italian Macroseismic DataBase - a rich source of macroseismic damage
effects caused by historical and instrumental earthquakes. The
macroseismic
damage is represented using the intensity scale, i.e., a discrete scale
that defines the level of damage in buildings due to seismic waves.
Concerning instrumental seismicity, we will use the ITACA database,
that has become available only recently; ITACA is a collection of all
the waveforms and peak values of the Italian earthquakes since 1976,
with minimum magnitude 3.5. For the definition of the geological
setting,
DISS - Database of Italy's Seismogenic Sources - and CSI - Catalog of
Instrumental Seismicity - will be analyzed.
Once the likelihood function will be built, it will be combined with the
prior distribution using a MonteCarlo simulation method; this will
provide a numerical evaluation of the posterior seismic hazard for the
city of Naples.
Since the seismic hazard assessment we propose is innovative in each
single part, the expertise from different research field is needed.
Indeed, it is the first time in scientific literature that a Bayesian
assessment for seismic hazard, able to integrate the source and site
approaches, is proposed. Concerning the source approach, it is the first
time that the MonteCarlo approach is implemented in Italy, while it has
been applied in other regions, in some cases by the personnel of this
project (Faenza et al, 2007, J Geophys Int 171:797-806); lastly, the
characterization of the seismic history of a site using ShakeMap code
is another novelty of this project.
Task 2: Volcanic Hazard
As mentioned so far, while Bayesian seismic hazard assessment is
innovatory,
some members of this project are already involved in Bayesian volcanic
hazard
assessment, having already produced a freeware code named BET_VH
(Bayesian Event Tree for Volcanic Hazard), based on a Bayesian event
tree.
BET_VH makes use of Bayesian inference to study the probability of all
concurrent events responsible for the occurrence of volcanic hazardous
phenomena, like tephra fall and invasion of pyroclastic density
currents.
The quantification of the probability that a given volcanic hazardous
phenomenon will occur can be subdivided in a logical sequence of events
(the tree), that have to occur because of the hazardous phenomenon
happens
in a certain point around the volcano. Therefore, it is necessary to
assign
a probability distribution to each of this connected event. This
probability
has to be conditioned to the occurrence of all the events that form the
previous part of the chain. BET_VH evaluates these conditioned
probabilities
through Bayesian inference, taking into account models (geological,
volcanological and theoretical) and past data. BET_VH has been designed
for the application at a single volcano, and our team has already
applied
it to evaluate the hazard due to tephra fall from Campi Flegrei. During
this project, BET_VH will be applied again to Campi Flegrei, for
evaluating the hazard due to invasion of pyroclastic density currents,
and ex-novo to the other two Neapolitan volcanoes, Vesuvius and Ischia,
for evaluating both the hazards from tephra fall and pyroclastic density
currents. Moreover, another interesting application, usually disregarded
in volcanic hazard assessment, is the evaluation of the overall volcanic
hazard in the target area, due to the three volcanoes. This probability
will be calculated by combining, with appropriate probabilistic weights,
the hazard from each single volcano.