See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/253341563

CRISIS2008: A Flexible Tool to Perform

Probabilistic Seismic Hazard Assessment

Article in Seismological Research Letters · May 2013

Impact Factor: 2.16 · DOI: 10.1785/0220120067

CITATIONS READS

14 580

4 authors, including:

Mario Ordaz Carlo Meletti

Universidad Nacional Autónoma de México National Institute of Geophysics and Volcano…
163 PUBLICATIONS 1,731 CITATIONS 61 PUBLICATIONS 856 CITATIONS

SEE PROFILE SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, Available from: Mario Ordaz
letting you access and read them immediately. Retrieved on: 04 July 2016
CRISIS2008: A Flexible Tool to Perform
Probabilistic Seismic Hazard Assessment
by M. Ordaz, F. Martinelli, V. D’Amico, and C. Meletti

INTRODUCTION CRISIS2007), both the GUI and the computation engine were
written in the same language.
In the frame of the Italian research project INGV-DPC S2 Finally, in the frame of the mentioned S2 project, starting
(http://nuovoprogettoesse2.stru.polimi.it/), funded by the from 2008, the program was split into two logical layers: core
Dipartimento della Protezione Civile (DPC; National Civil (CRISIS Core Library) and presentation (CRISIS2008). In
Protection Department) within the agreement 2007–2009, a addition, a new presentation layer was developed for accessing
tool for probabilistic seismic-hazard assessment (PSHA) was the same functionalities via web (CRISISWeb).
developed. The main goal of the project was to provide a flex- It is worth noting that CRISIS has been mainly written by
ible computational tool for PSHA; the requirements considered people that are, at the same time, PSHA practitioners. There-
essential for the success of the project included: fore, the development loop has been relatively short, and most
• ability to handle both stationary and non-stationary earth- of the modifications and improvements have been made to
quake time–occurrence models; satisfy the needs of the developers themselves.
• ability to use ground-motion prediction models (GMPMs)
that are not parametric equations but probabilistic foot-
New Features In CRISIS2008
prints of the intensities generated by earthquakes of
CRISIS is essentially based on the standard Esteva–Cornell
known magnitude and focal characteristics. Usually these
approach (Esteva, 1967; Cornell, 1968) to PSHA. Thus, it first
footprints are results of ground-motion simulations.
requires the definition of a set of seismic sources, each one with
Some commonly used programs (e.g., FRISK by McGuire uniform seismic rates of activity defined using specific recur-
[1978]; SEISRISK III by Bender and Perkins [1987]) and rence models (i.e., magnitude–frequency distributions). The
more recent and state-of-the-art tools (e.g., OpenSHA by Field sources are discretized into elements of appropriate size, whose
et al. [2003], http://www.opensha.org;OpenQuake, http:// seismicity rate is redistributed proportionally to their size.
openquake.org ) for PSHA were analyzed. We decided to focus Finally, the approach requires computing probabilistic esti-
on CRISIS2007, which was already a mature and well-known mates of the ground-motion intensities that are produced at
application (e.g., Kalyan Kumar and Dodagoudar, 2011; the site of analysis when an earthquake of known magnitude
Teraphan et al., 2011; D’Amico et al., 2012; see also http:// and location occurs using predictive models. For each discre-
ecapra.org/CRISIS-2007), but also suitable for additional devel- tized element, CRISIS evaluates the exceedance probability of
opment and evolution because its source code is freely available the required intensity values given that an earthquake of
on request. The computational tool resulted in an extensive known magnitude and occurrence probability took place. The
probabilities for all magnitudes and sources are then accumu-
redesign and renovation of the previous CRISIS2007 version.
lated, following probability rules, in order to compute the over-
CRISIS is a computer program for PSHA, originally de-
all exceedance probabilities.
veloped in the late 1980s using FORTRAN as programming
In the following lines we describe the main new elements
language (Ordaz, 1991). In this format, still without a graphical
introduced in CRISIS2008 with respect to former versions of
user interface (GUI), it was distributed as part of SEISAN tools the program.
(Ottemöller et al., 2011).
Ten years later, a GUI was constructed, generating what Output in Terms of Exceedance Probabilities Instead of
was called CRISIS99 (Ordaz, 1999). In this version, all the Exceedance Rates
graphic features were written in Visual Basic, but the compu- Previous CRISIS versions worked with ground-motion
tation engine remained a FORTRAN dynamic link library. intensity exceedance rates as measures of seismic hazard.
The reason for the use of mixed-language programming was CRISIS2008 gives seismic hazard only in terms of probabilities
that computations in Visual Basic were extremely slow. of exceedance of intensity values in given time frames. For
Around 2007 the program was upgraded in view of the instance, a valid measure of seismic hazard in the newer version
advantages offered by the object-oriented technologies. An is the probability of experiencing peak ground acceleration
object-oriented programming language was required and the (PGA) greater or equal than 0:2g in the next 50 years at a given
natural choice was Visual Basic.Net. In the new version (called location.

doi: 10.1785/0220120067 Seismological Research Letters Volume 84, Number 3 May/June 2013 495
 This change was made in order to allow users to introduce made using the following time-dependent models: Brownian
in the computations probabilities of earthquake occurrences de- Passage Time (Matthews et al., 2002) and Double Branching
rived from non-Poissonian models. For non-Poissonian occur- model (Marzocchi and Lombardi, 2008).
rences the exceedance rate is not enough to fully characterize the
process and seismic hazard must be expressed in terms of the Ground-Motion Prediction Models
aforementioned exceedance probabilities. However, for Poisso- In general, GMPMs, also called attenuation relationships, estab-
nian occurrence processes, the exceedance probabilities in given lish probabilistic relations between earthquake characteristics
time frames can easily be converted into exceedance rates. (essentially magnitude) and shaking intensities at a site of in-
CRISIS2008 preserves the ability to produce maps of terest. These relations are probabilistic because, for given earth-
intensity values associated with a fixed exceedance probability quake characteristics, the intensities are regarded as random
in a given time frame. variables whose probability distribution is completely fixed
by the GMPM. In most of the cases this means that at least
Source Geometry Features the first two statistical moments (e.g., the median and the stan-
CRISIS accepts source geometries of the following three types: dard deviation of the natural logarithm, in the lognormal case)
(1) areas (polygons), (2) lines (polylines), and (3) points. The of the probability distribution must be provided by the GMPM.
point-source type, now called multi-point source, was en- CRISIS now recognizes three families of GMPMs:
hanced and now it is possible to associate with each point an 1. Attenuation tables. In these tables, relations between
arbitrarily oriented rupture plane, useful when working with earthquake characteristics and intensities at a site are given
GMPMs that use distance metrics for which rupture size is in terms of the following parameters: magnitude, struc-
relevant. tural period, source-site distance, and standard deviation
Also, when rupture size is relevant for attenuation com- of the natural logarithm of the predicted intensity mea-
putations, CRISIS allows the user to specify relations between sure. For the first statistical moment (usually the median
magnitudes and rupture sizes. In this regard, a new option has of a lognormal distribution), the attenuation relations are
been developed: if used, it causes the whole area or line to break matrices in which the rows run for the magnitude and the
every time an event takes place, regardless of earthquake mag- columns run for the distance. Note that when using
nitude. This option is intended to model characteristic- attenuation tables, the relations between magnitude,
earthquake behavior. distance, and intensity do not need to be of parametric
nature, because the intensity medians are given, point
Source Seismicity Features by point, for magnitude–distance combinations.
CRISIS2007 already included two varieties of Poissonian 2. Built-in models. These are popular models, published in
description of earthquake occurrences, which differed in their the literature, in which magnitude, distance, and intensity
magnitude–frequency distributions: modified Gutenberg– are usually probabilistically related by a set of formulas or
Richter curve and characteristic-earthquake model (Kiremid- parametric equations. There is a set of built-in models
jian and Anagnos, 1984). In CRISIS2008, besides these two ready to use in CRISIS and there is also the possibility of
Poissonian occurrence models, also non-Poissonian occur- adding new models. The list of GMPMs currently imple-
rences are admitted. These latter are described in non- mented in CRISIS Core is shown in Table 1.
parametric form, thus allowing computing hazard using even 3. Generalized models. Generalized attenuation models are
very complex time-dependent earthquake recurrences. non-parametric probabilistic descriptions of the ground
In the case of Poissonian occurrences for a given source, motions produced by individual earthquakes with known
the user provides the parameters that allow the computation of magnitude and location. In the context of CRISIS, a gen-
the probability of having a certain number of events within a eralized attenuation model is a collection of probabilistic
given magnitude range in the next T f years. footprints, one for each of the events considered in the
The new non-parametric seismicity model can be specified analysis. Each footprint gives, in probabilistic terms, the
by way of explicitly giving the probabilities of having geographical distribution of the intensities produced by
1; 2; …; N s earthquakes within a magnitude range, at a given this event.
location, during the next T f years. In this case, for a given For a given event, the footprint consists of several pairs of
location and a given value of T f , the next 50 years, say, the grids of values. Each pair of grids is associated to one of the
user must provide a matrix of probabilities in which the rows intensity measures for which hazard is to be computed.
run for magnitude range (e.g., earthquakes with magnitude CRISIS needs two grids for each intensity measure be-
from 4.0 to 4.2, 4.2 to 4.4, and so on) and the columns cause, as with other GMPMs, the intensity caused by the
run for number of events (the probability of having one event earthquake is considered probabilistic, so two statistical
in the next T f years, or two events, and so on). moments are required in order to fix a probability-density
The required probabilities are computed with means ex- function of the intensity at a particular location. These
ternal to CRISIS, which gives the code a great flexibility to footprints are constructed with means external to CRISIS,
accommodate results from any non-Poissonian occurrence and the corresponding grids are made available to CRISIS
model. So far, within the S2 project, applications have been in a pre-established format, along with the occurrence

496 Seismological Research Letters Volume 84, Number 3 May/June 2013

 Table 1
GMPMs Currently Implemented in CRISIS Core
Magnitude Distance Limit (km)
Reference Dataset Area Range and Distance Type
GMPMs for Active Tectonic Regions with Shallow Seismicity
Abrahamson and Silva (1997) World 4.4–7.4 ≤ 200r rup
Spudich et al. (1999) World (extensional regime eqs.) 5–7.2 ≤ 100r jb
Cauzzi and Faccioli (2008) World 5–7.2 ≤ 150r hyp
Akkar and Bommer (2010) Europe, Mediterranean, Middle East 5–7.6 ≤ 100r jb
Boore and Atkinson (2008) World (PEER NGA) 5–8 ≤ 200r jb
Campbell and Bozorgnia (2003) World 4.7–7.7 ≤ 60r seis
Sabetta and Pugliese (1996) Italy 4.6–6.8 ≤ 100r epi , r jb
Pasolini et al. (2008) (macroseismic intensity) Italy 4.4–7.4 ≤ 140r epi
GMPMs for Subduction Zones
Arroyo et al. (2010) Mexico > 5:5 ≤ 400r rup
Youngs et al. (1997) World 5–8.2 ≤ 500r rup
Atkinson and Boore (2003) World 5–8.3 ≤ 300r rup
Garcia et al. (2005) Central Mexico 5.2–7.4 ≤ 400r rup
All models are defined for moment magnitude M w , except Sabetta and Pugliese (1996), who adopt M s for magnitude ≥ 5:5 and M L
otherwise. r jb , Joyner–Boore distance; r rup , closest distance to the rupture plane; r seis , distance to seismogenic rupture; r hyp ,
hypocentral distance; r epi , epicentral distance.

probabilities of each of the individual earthquakes for straint is that the new classes must be compiled using one of the
which grids were computed. languages compatible with the main program, which is written
From this description, it is clear that it would be extremely in Visual Basic.Net. Therefore, all the languages of the .Net
difficult to perform a hazard study of regional (or larger) suite offered by Visual Studio.Net development environment
size using generalized attenuation models. Usually, a hazard (Visual Basic.Net, Visual C++, and C#) are acceptable.
model of regional size contains thousands of events, and the The detailed documentation of each function and prop-
task of geographically describing the intensities caused erty is embedded within the executable (as by standard devel-
by all of them in non-parametric form would be titanic. opment practice) and can be easily browsed.
Rather, generalized attenuation models will very likely be Attenuation tables, built-in models, and generalized mod-
used for local studies, for which the relevant earthquakes els are now particular cases of the more general GMPM CRISIS
are few and can be clearly identified. In this case, the grids interface.
of the required values (probabilistic geographical distribu-
tion of one or more intensity measures for each event) can Site Effects
be constructed using, for instance, advanced ground- CRISIS2008 permits inclusion of local site effects in hazard
motion simulation techniques (e.g., Stupazzini et al., 2009). computations. Site effects are given to CRISIS in terms of am-
plification factors that depend on site location, structural
New GMPM Architecture period, and ground-motion level (in order to account for soil
The GMPM architecture was completely revised. Now GMPMs non-linearity). Amplification factors are interpreted by CRI-
are all codified as classes which must implement a specific SIS in the following way. Suppose that during the hazard com-
programming interface (a software definition of methods and putations, CRISIS requires computing the median of the
attributes). Each GMPM class may have any number of differ- intensity at structural period T that would take place at site
ent parameters specific to the model (values, choices, flags, and S due to an earthquake of magnitude M originating at hypo-
external files); in this way a GMPM is completely defined by the center H. We will denote this intensity as I
S; T ; M; H . Nor-
relevant class (also called model template) together with the mally, I
S; T ; M; H is computed using the GMPM that the
values of the specific parameters. user has selected for the source to which H belongs.
GMPMs are now external software modules that are The value so computed is interpreted by CRISIS as the
hooked to the application at runtime. Additional custom median intensity without site effects. But if site effects are given,
GMPMs may be freely developed and integrated without having then the median intensity that CRISIS will use for the hazard
to recompile the core code. Therefore, the users can build new computations, IS, is the product of I
S; T ; M; H  by the
external classes implementing custom GMPM modules by amplification factor given by the user, which depends on site
adhering to the programming-interface specification, which location, structural period, and ground-motion level, I 0 . We will
is delivered as part of the executable program. The only con- denote this amplification factor as A
S; T ; I 0 . In other words,

Seismological Research Letters Volume 84, Number 3 May/June 2013 497

IS
S; T ; M; H  I
S; T ; M; H A
S; T ; I 0 : (1) Disaggregation
CRISIS can generate exceedance probabilities disaggregated by
Clearly, if no site effects are present, then A
S; T ; I 0   1. magnitude, distance, and epsilon value (Bazzurro and Cornell,
Note that the amplification factor only affects the median 1999). To generate a disaggregation chart, the user must select a
intensity, whereas the relevant uncertainty is not modified. site of analysis, an intensity measure (usually one spectral ordi-
Site effects should be provided on a regular grid, so the nate), an epsilon level, and either an intensity level or a value of
amplification values are interpolated if the site of interest does exceedance probability in a given time frame. For instance, the
not coincide with one of the given grid nodes. Figure 1 gives an user can compute a disaggregation chart for a given site, the spec-
example of the grid of amplification factors A for the spectral tral ordinate at period of 2 s, an epsilon level of 2, and an in-
ordinate with period T  2 s and ground-motion level, I 0  tensity level of 0:4g. Disaggregation charts are used to identify
0:001g (elastic soil behavior), for Mexico City. the magnitude–distance combination that produces the highest
contribution to hazard at a site. This combination is later used,
Logic-Tree Computations for example, to select accelerograms that are representative of the
The code has now enhanced capabilities to make logic-tree com- ground motions associated with a given hazard level.
putations. In the context of CRISIS, each branch of a logic tree Given the required parameters, CRISIS will generate a plot
is defined by the user as a separate data file, also referred as of the exceedance probability values that are produced by vari-
project in CRISIS user interface, along with a measure of the ous magnitude–distance combinations. CRISIS will compute
degree of belief that the analyst has on each of the branches being the corresponding probabilities of exceedance of ground-
the best one. Results from the different branches, along with the motion values integrating the corresponding density function
weights assigned to each branch, are combined as follows. up to epsilon times the standard deviation that characterizes
Assume that the probability of exceeding level y of an the appropriate GMPM.
intensity measure Y at a site, in the i-th time frame, according It is interesting to note that disaggregation charts are
to the j-th branch of a logic tree is P ij
Y > y. Assume also that computed by CRISIS on the fly, that is, immediately after
the probability of being the best one assigned to the j-th branch the user has selected all the required parameters. Because of
is wj , j  1; …; N (it is required that the N weights add up to the relatively fast numerical algorithms used by CRISIS, this
unity); then, the expected value of Pi
Y > y, once all branches computation usually takes a few seconds, so disaggregation
have been accounted for, is given by: analysis is rather agile; therefore, the user can easily experiment
with different bin sizes, intensity levels, and so on.
X
N
P i
Y > y  P ij
Y > ywj : (2) Help File
j1 A help file is available in different formats (on-line html and
pdf ), which allows for contextual help (in the desktop presen-
Results of the logic-tree combination can also be tation layer) and web access (in the web presentation layer). The
visualized. help file contains a brief theoretical section that presents the
basics of PSHA computations, as well as instructions on how
to practically use the program. In addition, the file contains
documentation on the most important aspects of the program.

Desktop Versus Client Server/Web Application

Software applications may be designed either to run as stand-
alone on a single computer or to allow two or more computers
to collaborate to produce the required results (distributed
applications). The first approach produces applications com-
monly called desktop. Desktop applications are simpler and
commonly used for most of the programs, and the non-expert
user may find them simpler.
However, for some applications, the constraint to be
bound to a single computer may be an issue, especially when
the computational resources (typically memory and CPU time)
are the limiting factors to produce the results, or when one
needs to share or publish data. In such cases, the benefits of a
different architecture could be considered against the increased
hardware and software costs and complexity.
There are several ways of developing distributed applica-
▴ Figure 1. Grid of amplification values due to site effects for tions and the preferred solution depends on a number of factors,
Mexico City. including the elaboration to delegate to the server, the available

498 Seismological Research Letters Volume 84, Number 3 May/June 2013

network infrastructure (between the cooperative computers), minimize the effort, and possibly preserve the long-used (and
and the software communication facilities (e.g., web server). therefore validated) existing code.
In the following, we introduce the requirements and tech- Among the possible solutions, we investigated the follow-
nical issues. The final product took into consideration also im- ing: (1) to completely rewrite the code in a new language;
portant aspects like usability, scalability, and computational otherwise to preserve the original language and modify the
performance. application either (2) towards a generic client–server architec-
ture, or (3) a client–server web application.
The User’s Needs Option (1) was to move the current program to a new
From the analysis of the then existing version of CRISIS (CRI- language, not dependent on the OS; the solution could either
SIS2007), also compared with other applications commonly need the user to recompile on each specific computer (like with
used (e.g., SEISRISK III, FRISK), some aspects emerged that C language) or to use a virtual machine (like Java or Python).
could be improved. The idea to convert the program to a completely new and dif-
ferent programming language (the application had already been
The independence of the Operating System (OS). CRISIS2007 subjected to such a process moving from FORTRAN to Visual
was developed using a proprietary language, Visual Basic.Net, Basic) was considered to require too much effort given the
and to run it required a computer having a Windows OS with complexity and the dimension of the source code. Additionally,
the appropriate version of the .Net Framework installed. there are always good probabilities of introducing errors during
Therefore, only computers running Windows, either as main the translation process.
OS, or as virtual OS, were enabled to run CRISIS2007. About option (2), it consists of splitting the solution into
The issue was particularly important for those users who two programs: the server and the client. The server offers the
have computers with non-commercial OS, like Linux. There service; the client contacts the server to request specific oper-
was no way to run the application on any non-Windows OS, ations. An architectural design is required to clearly split the
and it was not possible to recompile the application on such server code devoted to the heavy elaborations (PSHA compu-
systems. tations), from the client code devoted to the user interface.
Additionally, some code must be provided in order to make
Fewer troubles for code maintenance/upgrade distribution. A sec- the two components of the solution communicate.
ond issue was related to the distribution of further versions, We judged that this solution could have advantages on the
that is, upgrading. Each distribution required a new installation response time to the user: during the definition of the user input
package, which sometimes presented unexpected problems the server would not be involved (and the application response
(e.g., caused by specific user’s configuration or settings); it time would be very fast). The browsing of the results of the elab-
could take time and efforts to identify and solve them. The oration would still be only local to the client. The critical point is
solution required a minimum impact on upgrading. about the input data file and the output of the results: the as-
sociated data should be transferred between the client and the
Fewer resource requirements, especially CPU. CRISIS is an ap- server for every elaboration request and it could easily become
plication with low demand of resources (memory and CPU) an unacceptable amount of data, especially when considering
during the definition of the input data and the browsing of the the last-introduced feature of allowing non-parametric seismic-
results, but it requires intensive and long-lasting elaborations ity models, which involves huge data files. Also output data files
(ranging from minutes to hours or even days, depending on produced by the program can easily be large. Therefore, we rated
the type and size of input data, the complexity of the elabo- this solution as not applicable for our needs.
ration, and the number of computational points). Option (3) takes advantage of the web. The application we
The evolution of the application always resulted in more were developing was not aimed to face many concurrent clients
demanding versions in terms of resources for the hazard evalu- (at most tens of clients), but still the advantages offered by web
ation, not covered by the higher hardware speed commonly server applications remain. With this client–server approach,
available on personal computers from time to time. the client is a common web browser, which can be considered
The standalone version of CRISIS requires the computer part of the software infrastructure with no cost of develop-
where it is running not to be switched off for all the time the ment. The browser sends data to and retrieves data from a
hazard evaluation takes, and the users can perform only re- web server where the application is located. Following this ap-
duced activity on it. Thus, either a dedicated computer is de- proach the developer does not care which is the browser and
voted to CRISIS or the user’s activity is affected. the OS used as far as the data are returned to the client in a
The upgrading of the application towards a client–server format suitable for standard browsers. Therefore, the develop-
configuration, to delegate all the heavy computational work to ment is performed only on the server, is client-independent,
the server, was therefore investigated. and the client code is not affected by the application develop-
ment and maintenance.
Technical Options We also considered the option to unbundle the original
Because we were working on an upgrade, we tried at the same code into those portions devoted to the elaboration and those
time to provide the required functionality enhancements, devoted to the user interaction. The original standalone

Seismological Research Letters Volume 84, Number 3 May/June 2013 499

application could then be split into two logical layers: the core layer); a second activity was to create a new presentation layer
layer, in which the code for the elaboration is located, and the to offer a web interface that used the same core elaboration
presentation layer, in which is the code for the user interaction. classes. In this view, future developments could offer additional
At this point we needed an additional presentation layer for the presentation layers, for example, a web service, making use of
user interaction using a web browser: the core layer and the the already developed core layer.
new presentation layer resulted in the new web application.
Additionally, we could still have the desktop application by Specific CRISISWeb Features
merging the common core layer and the desktop presentation Once we identified the architecture based on two layers (core
layer (which required minor modifications). The desktop and and presentation), and defined the new presentation layer to
web applications will run the same code for the hazard evalu- develop (the web user interface), we investigated the benefits
ation, producing the same results, and therefore offering the that could be added to the users using the web interface with
researcher the freedom of choosing the preferred interface respect to the desktop application.
(a picture of the architecture of the application and the logical One requirement was the privacy and independence of each
layers involved is shown in Figure 2). user’s work about both input and output data. Whereas most of
The original code would remain in the original language the web applications get the users’ input from a web form (or a
(Visual Basic.Net) to minimize the coding effort during the set of forms) and return the elaboration results after a short time,
splitting of the original program into core and presentation this could not be true for an application like CRISIS, in which
layers. Microsoft provides an integrated development environ- the estimated elaboration times can be hours, or even days. Ad-
ment (IDE: a single tool in which the code can be written, ditionally, the user input is generally quite complex and it is nor-
compiled, and debugged), which can work with any language mal practice to prepare an elaboration set over a relatively long
based on .Net framework, that is, C#, C++, and Visual Basic. It period of time, often spanning multiple web sessions. The users
is also possible to develop each layer as a separate .Net project would also expect to have the opportunity to review the input
(possibly each layer written in different .Net language). Using data after the elaboration. For a desktop application these re-
the IDE, the integration between the presentation layer for the quirements are solved using the file system on the user’s com-
user interaction via web and the core layer, which together puter. With the adopted architectural solution, in which the
make the Web application, is simplified. Similarly, it is possible elaboration is performed on the web server, using files located
to integrate the desktop presentation layer with the core layer on the user’s computer is not feasible as it requires the file trans-
in order to make the desktop application. fer to the server for every elaboration request.
The above approach was identified as more sound for the We then pursued a solution in which input and output
project requirements, as it preserved the already solid code for files were hosted on the server. Still, the users can define
the most important and critical elaborations. For the new pre- the input parameters via the web interface. To comply with
sentation layer we decided to use the same programming lan- data privacy, the application was designed to have private fold-
guage used for the core layer, which appeared to have sufficient ers, one for each user, in which to save or load the input files for
functionalities for the requirements, and also the advantage of the elaboration and to store the results. To allow the associa-
eliminating any possible issue related to using different pro- tion between private folders on the server and the users, the
gramming languages (e.g., parameter passing rules). We iden- users had to be identified by the application; therefore, authen-
tified two development activities: one was to isolate the classes tication and authorization had to be used. We decided to use a
used to perform the elaborations from those used for the desk- mechanism shipped with the .Net framework to require the
top presentation and reorganize them into a layer (the core user’s authentication to access the site and we created two
groups: users and administrators.
The application was so conceived that all the users of
the same working group could let other members to use
some of their data files, for example, maps of the locations
of cities, or the GMPMs provided as tables. The adopted
solution was to create a special folder on the server, in which
all the users can access the files as input for their elaborations
(in read-only mode), and only administrators can upload or
delete files.
The identification of the users makes also possible to re-
turn personalized useful information, such as the expected time
to finish the submitted elaborations and possibly the log of the
encountered problems. The application keeps track of the sub-
mitted elaborations and their owner also after the user discon-
▴ Figure 2. Architectural structure of the application. On the left nects from the web, and it provides updated information when
the logical layers, and on the right the code blocks implemented. the user logs back into the system.

500 Seismological Research Letters Volume 84, Number 3 May/June 2013

A CRISISWeb Session: An Application to Italy depends on the source type: for areas and faults, the vertices
To create a set of input data (i.e., a CRISIS project), the user must be entered; for multipoint (as in the example of Figure 4),
needs to provide several elements. Information is grouped and a file with a specific format defines the geometry of the sources.
the user accesses each type by a different button on the Main Depending on the selected occurrence model, either the param-
Console (Figure 3). The application has also some internal eters in the case of Poissonian occurrence (i.e., Gutenberg–
checks to prevent the user’s overwrite of data from previous Richter or characteristic) or a specific file in the case of a
elaborations by mistake. The following are the input elements, non-parametric seismicity model must be provided. For each
which are available in specific pages accessible from the corre- source, one or more special regions can be defined. A special
sponding section of the main page. region is a geographic area for which it is possible to adopt a
The map-data page allows the inclusion into the project of different GMPM with respect to the general model for that
geographic elements, such as coastlines or city locations. seismic source (e.g., in case of soil condition in the target area
The sites for which hazard will be computed (set of scat- diverging from general rock condition). We did not use any
tered sites or regular grid) are defined in the computational special region in the example of Figure 4.
sites page. On the spectral ordinates page it is possible to define the
In the source zones page (Figure 4), the user handles the number and the range of the intensity levels and the structural
seismic-source zones, their geometry and seismicity parameters periods for which hazard is estimated.
(i.e., the occurrence model and the associated parameters). Dif- The global parameters page is used to define general
ferent types of sources are allowed in the same project. Depend- parameters for the elaboration, namely, the maximum integra-
ing on the source type (area, fault, or multipoint), coefficients tion distance, coefficients for the discretization of the sources,
that empirically relate magnitude with the rupture dimension the type of distance to be used in the disaggregation (e.g.,
(area or length) must be provided. The geometry definition Joyner–Boore, epicentral, and so on), the time frames for

▴ Figure 4. Definition of source geometry and seismicity

▴ Figure 3. Main console page. parameters.

Seismological Research Letters Volume 84, Number 3 May/June 2013 501

which the hazard is computed, and the type of output files to
produce.
The attenuation-model page allows defining the GMPMs
to be used for computation and relevant parameters. In the
example of Figure 5, the Boore and Atkinson (2008) model
is selected. After the selection of the model the specific options
are presented to the users for further choices. Several attenu-
ation models can be defined in the same project and assigned to
different source zones as far as they are referred to the same
intensity measure (acceleration, displacement, etc.). Lastly, site
effects can be included in the hazard computation by providing
the relevant file with amplification factors.
When the computation is finished, it is possible to browse
the results as maps: one can navigate the map (pan and zoom
operations) and select the hazard parameter to show. It is also
possible to set boundaries of the color scale and to request tex-
tual results for one parameter of the whole map. Hazard curves
and disaggregation graphs are also available on demand, both as
plots and text.
The map in Figure 6 shows the hazard values obtained for
Italy in terms of PGA with 10% probability of exceedance in
50 years using a seismicity model that is consistent with the
standard (Esteva–Cornell) Poissonian model adopted for
the reference Italian seismic-hazard map (MPS04, http://
zonesismiche.mi.ingv.it; Stucchi et al., 2011), and the Boore
and Atkinson (2008) GMPM for rock-site conditions. The
adopted seismicity model gives for each cell of a 0:1° × 0:1°
grid covering the whole of Italy the probability of occurrence
of exactly one event in 10, 30, and 50 years for each one of the
magnitude bins considered (12 bins ranging from M w 4.76 to
7.29). These probability values were derived within the S2
project (Meletti et al., 2009) from the seismicity rates of a sub-
set of the logic-tree branches defined in MPS04.

CONCLUSIONS
We have presented an updated version of CRISIS, a tool for
PSHA , for which significant improvements concern both com-
putational and technological features. ▴ Figure 5. Selection of attenuation model and specific model
From the computational point of view, the new ways of parameters.
data definition implemented in CRISIS2008 offer to the user
great flexibility in the selection of the features used to produce
the estimates, such as the definition of source geometries, earth- user computational resources; program code updated without
quake recurrence models (i.e., also non-Poissonian occurrence distribution to the users). The core layer now accepts GMPMs
is now allowed), and GMPMs. It is also possible to include in as external libraries, allowing for custom-made models without
the computation site effects and to perform logic-tree compu- code recompilation.
tations. Outputs are expressed in terms of exceedance proba- In terms of flexibility and opportunities presented to the
bilities of ground shaking in given time frames; also user, the new features of CRISIS2008 make it now comparable
disaggregation by magnitude, distance, and epsilon is provided. to other state-of-the-art commercial or free programs for
From the technological point of view, the code architec- PSHA . Furthermore, results of the code have been checked suc-
ture was reviewed, and split into two logical layers: core and cessfully with the benchmarks proposed in a project sponsored
user interface. Two versions of the program are now available, by the Pacific Earthquake Engineering Research Center (PEER)
which differentiate for the user interface only. The desktop in- and documented in Verification of Probabilistic Seismic Hazard
terface is an evolution of the previous one (CRISIS2007). The Analysis Computer Programs by Thomas et al. (2010). The val-
web interface was newly developed, offering all the associated idation report can be downloaded from https://dl.dropbox
advantages (e.g., independency from the user OS and from the .com/u/29431467/ValidationTESTS.pdf.

502 Seismological Research Letters Volume 84, Number 3 May/June 2013

▴ Figure 6. PGA (in g) with 10% probability of exceedance in 50 years in Italy computed using a seismicity model consistent with the
MPS04 Italian seismic-hazard map and the Boore and Atkinson (2008) GMPM for rock conditions (see selected options in Figs. 4 and 5).

ACKNOWLEDGMENTS AND FINAL NOTE by DPC do not represent its official opinion and
policies.
This research has benefited from funding provided by the The authors wish to thank all the people involved in the
Italian Presidenza del Consiglio dei Ministri, Dipartimento S2 project who contributed their feedback to the improvement
della Protezione Civile (DPC). Scientific papers funded of the program.

Seismological Research Letters Volume 84, Number 3 May/June 2013 503

 We also wish to thank the anonymous reviewer for the Marzocchi, W., and A. M. Lombardi (2008). A double branching model
useful suggestions that helped to improve the paper. for earthquake occurrence, J. Geophys. Res. 113, B08317, doi:
10.1029/2007JB005472.
The desktop application can be obtained for academic and Matthews, M. V., W. L. Ellsworth, and P. A. Reasenberg (2002). A Brow-
scientific purposes only, after a request by e-mail to both the nian model for recurrent earthquakes, Bull. Seismol. Soc. Am. 92,
coordinators of the S2 project, Professor Ezio Faccioli 2233–2250.
(faccioli@stru.polimi.it) and Dr. Warner Marzocchi McGuire, R. K. (1978). FRISK: A computer program for seismic risk
(warner.marzocchi@ingv.it). analysis using faults as earthquake sources, U.S. Geol. Surv. Open-
File Rept. 76-67, 90 pp.
The web application is still hosted on the prototype server, Meletti, C., V. D’Amico, and F. Martinelli (2009). Module for ER
which is designed for a small number of users. A more powerful model based on Poisson applied to ZS9, Progetto INGV-DPC
server will be made available only after a formal authorization S2, Deliverable D2.1, http://nuovoprogettoesse2.stru.polimi.it/
by DPC and in case of interest by the scientific community. Deliverables.html.
Ordaz, M. (1991). CRISIS. Brief description of program CRISIS, Inter-
nal report, Institute of Solid Earth Physics, University of Bergen,
REFERENCES Norway, 16 pp.
Ordaz, M. (1999). User’s manual for program CRISIS-99, Technical
Abrahamson, N. A., and W. J. Silva (1997). Empirical response spectral report, Universidad Nacional Autonoma de Mexico, Mexico City.
attenuation relations for shallow crustal earthquakes, Seismol. Res. Ottemöller, L., P. Voss, and J. Havskov (2011). Seisan earthquake analysis
Lett. 68, 94–127. software for Windows, Solaris, Linux and MacOsX; http://seis.geus
Akkar, S., and J. J. Bommer (2010). Empirical equations for the .net/software/seisan/seisan.html.
prediction of PGA, PGV, and spectral accelerations in Europe, Pasolini, C., D. Albarello, P. Gasperini, V. D’Amico, and B. Lolli (2008).
the Mediterranean region, and the Middle East, Seismol. Res. Lett. The attenuation of seismic intensity in Italy part II: Modeling and
81, 195–206. validation, Bull. Seismol. Soc. Am. 98, 692–708.
Arroyo, D., D. García, M. Ordaz, M. A. Mora, and S. K. Singh (2010). Sabetta, F., and A. Pugliese (1996). Estimation of response spectra and
Strong ground-motion relations for Mexican interplate earthquakes, simulation of nonstationary earthquake ground motions, Bull.
J. Seismol. 14, 769–785. Seismol. Soc. Am. 86, 337–352.
Atkinson, G. M., and D. M. Boore (2003). Empirical ground-motion re- Spudich, P., W. B. Joyner, A. G. Lindh, D. M. Boore, B. M. Margaris, and
lations for subduction-zone earthquakes and their application to J. B. Fletcher (1999). SEA99: A revised ground motion prediction
Cascadia and other regions, Bull. Seismol. Soc. Am. 93, 1703–1729. relation for use in extensional tectonic regimes, Bull. Seismol. Soc.
Bazzurro, P., and C. A. Cornell (1999). Disaggregation of seismic hazard, Am. 89, 1156–1170.
Bull. Seismol. Soc. Am. 89, 501–520. Stucchi, M., C. Meletti, V. Montaldo, H. Crowley, G. M. Calvi, and
Bender, B., and D. M. Perkins (1987). SEISRISK III: A computer program E. Boschi (2011). Seismic hazard assessment (2003–2009) for
for seismic hazard estimation, U.S. Geol. Surv. Bull. 1772, 48 pp. the Italian building code, Bull. Seismol. Soc. Am. 101, 1885–1911.
Boore, D. M., and G. M. Atkinson (2008). Ground-motion prediction Stupazzini, M., R. Paolucci, and H. Igel (2009). Near-fault earthquake
equations for the average horizontal component of PGA, PGV, and ground motion simulation in the Grenoble Valley by a high-
5%-damped PSA at spectral periods between 0.01 s and 10.0 s, performance spectral element code, Bull. Seismol. Soc. Am. 99,
Earthq. Spectra 24, 99–138. 286–301.
Campbell, K. W., and Y. Bozorgnia (2003). Updated near-source Teraphan, O., J. Douglas, R. Sigbjörnsson, and C. G. Lai (2011). Assess-
ground-motion (attenuation) relations for the horizontal and ment of ground motion variability and its effects on seismic hazard
vertical components of peak ground acceleration and acceleration analysis: A case study for Iceland, Bull. Earthq. Eng. 4, 931–953.
response spectra, Bull. Seismol. Soc. Am. 93, 314–331. Thomas, P., I. Wong, and N. Abrahamson (2010). Verification of
Cauzzi, C., and E. Faccioli (2008). Broadband (0.05 to 20 s) prediction probabilistic seismic hazard analysis computer programs, PEER Re-
of displacement response spectra based on worldwide digital records, port 2010/106, Pacific Earthquake Engineering Research Center,
J. Seismol. 12, 453–475. College of Engineering, University of California, Berkeley, 186 pp.
Cornell, C. A. (1968). Engineering seismic risk analysis, Bull. Seismol. Soc. Youngs, R. R., S. J. Chiou, W. J. Silva, and J. R. Humphrey (1997). Strong
Am. 58, 1583–1606. ground motion attenuation relationships for subduction zone
D’Amico, V., C. Meletti, and F. Martinelli (2012). Probabilistic seismic earthquakes, Seismol. Res. Lett. 68, 58–73.
hazard assessment in the high-risk area of south-eastern Sicily
(Italy), Boll. Geof. Teor. Appl. 53, 19–36.
Esteva, L. (1967). Criteria for the construction of spectra for seismic M. Ordaz
design, in 3rd Pan-American Symposium of Structures, Caracas, Instituto de Ingeniería
Venezuela, 3–8 July (in Spanish). Universidad Nacional Autonóma de México
Field, E. H., T. H. Jordan, and C. A. Cornell (2003). OpenSHA: A Mexico City, Mexico
developing community-modeling environment for seismic hazard mordazs@iingen.unam.mx
analysis, Seismol. Res. Lett. 74, 406–419.
García, D., S. K. Singh, M. Herráiz, M. Ordaz, and J. F. Pacheco (2005).
Inslab earthquakes of Central Mexico: Peak ground-motion param- F. Martinelli
eters and response spectra, Bull. Seismol. Soc. Am. 95, 2272–2282. V. D’Amico
Kalyan Kumar, G., and G. R. Dodagoudar (2011). Seismic input C. Meletti
motion for Kanchipuram, South India, Int. J. Earth Sci. Eng. 4, Istituto Nazionale di Geofisica e Vulcanologia
6 SPL, 189–192.
Kiremidjian, A. S., and T. Anagnos (1984). Stochastic slip-predictable Via della Faggiola 32
model for earthquake occurrences, Bull. Seismol. Soc. Am. 74, 56126 Pisa, Italy
739–755. francesco.martinelli@pi.ingv.it

504 Seismological Research Letters Volume 84, Number 3 May/June 2013