Geopsy: Seismic Vibration Processing

Geopsy: A User-Friendly Open-
Source Tool Set for Ambient
Vibration Processing
Marc Wathelet*1
, Jean-Luc Chatelain1
, Cécile Cornou1
, Giuseppe Di Giulio2
,
Bertrand Guillier1
, Matthias Ohrnberger3
, and Alexandros Savvaidis4
Abstract
Cite this article as Wathelet, M., J.-
L. Chatelain, C. Cornou, G. D. Giulio,
B. Guillier, M. Ohrnberger, and A. Savvaidis
(2020). Geopsy: A User-Friendly Open-
Source Tool Set for Ambient Vibration
Processing, Seismol. Res. Lett. 91,
1878–1889, doi: 10.1785/0220190360.
Ambient vibrations are nowadays considerably used worldwide for numerous types of
engineering applications and scientific research. Geopsy and its companion tools are
part of that landscape. Since the first release of the program package in 2005, as out-
come of the European Union project Site Effects aSsessment from AMbient noisE,
Geopsy has become a mature multiplatform open-source package (released under
GNU Public License version 3) that has already been recognized as a reference tool
for analyzing ambient vibration data in the context of site characterization studies.
The community of users has grown from a core group of researchers up to thousands
of seismologists and engineers on every career level and on all continents. The versa-
tility of geopsy allows for the processing of all kinds of data needed in site characteri-
zation studies, that is, from single station single trace to three-component array
recordings. In all of the aforementioned cases, the steps from field acquisition to
the production of publication-ready figures are covered and supported by user-friendly
graphical user interfaces or corresponding command-line tools for the automation of
the complete processing chain. To avoid black-box usage, a number of lower-level tools
guarantee maximum flexibility in accessing and controlling processing results at any
stage of the analysis.
Introduction
Site characterization based on ambient vibration properties has
become increasingly popular during the past two decades,
mostly because it is a reliable low-cost alternative for obtaining
resonance frequencies for soils or buildings and VS profiles
compared with downhole or crosshole techniques (Garofalo,
Foti, Hollender, Bard, Cornou, Cox, Dechamp, et al., 2016).
Applications extend over a wide range: seismic bedrock depth
imaging (e.g., Ibs-von Seht and Wohlenberg, 1999; Parolai et al.,
2002; Özalaybey et al., 2011; Hellel et al., 2012), site effect assess-
ments (e.g., Lermo and Chávez-García, 1994; Guéguen et al.,
1998; Picozzi et al., 2009; Teague et al., 2018), buildings and
bridges (e.g., Crawford and Ward, 1964; Michel et al., 2010;
Guillier et al., 2014; Bindi et al., 2015), landslides (e.g.,
Meric et al., 2005; Burjánek et al., 2012; Zare et al., 2017), vol-
canoes (e.g., Sens-Schönfelder and Wegler, 2006; Perrier et al.,
2012; Bennington et al., 2015), and even Mars exploration
(Knapmeyer-Endrun et al., 2017). Similar to active source multi-
channel analysis of surface waves (Park et al., 1999), ambient
vibration processing relies mostly on the dispersive properties
of surface waves (Rayleigh and/or Love; Bonnefoy-Claudet et al.,
2006), but it carries additional complexities: several sources with
unknown properties and locations may be active at the same
time. Hence, unsupervised processing is certainly not the most
reliable way for extracting robust information. During the past
20 yr, several European Union projects focused on developing
new processing tools and on evaluating the limitations of ambi-
ent vibration methods: Site Effects aSsessment from AMbient
noisE (SESAME, an FP5 project number EVG1-CT2000-
00026, 2001–2004) and NEtwork of Research Infrastructures
for European Seismology, Joint Research Activity number 4
(an FP6 I3 project number RII3-CT-2006-026130, 2006–2010).
Bard et al. (2010) provide a detailed review of the amount of
work achieved during these projects. International blind tests
(Cornou et al., 2007; Boore and Asten, 2008; Cox et al., 2014;
Garofalo, Foti, Hollender, Bard, Cornou, Cox, Ohrnberger, et al.,
2016) showed that the estimation of surface-wave dispersion
curves is generally good. However, retrieving VS profiles from
1. Université Grenoble Alpes, Université Savoie Mont Blanc, CNRS, IRD, IFSTTAR,
ISTerre, Grenoble, France; 2. Istituto Nazionale di Geofisica e Vulcanologia, L’Aquila,
Italy; 3. University of Potsdam, Institute of Geoscience, Potsdam, Germany; 4. Bureau
of Economic Geology, The University of Texas at Austin, Austin, Texas, U.S.A.
*Corresponding author: marc.wathelet@univ-grenoble-alpes.fr
© Seismological Society of America
1878 Seismological Research Letters www.srl-online.org • Volume 91 • Number 3 • May 2020
Electronic Seismologist
Downloaded from https://pubs.geoscienceworld.org/ssa/srl/article-pdf/91/3/1878/4988673/srl-2019360.1.pdf
by CNRS_INSU user

dispersion curve inversion, usually of the Rayleigh-wave funda-
mental mode only, remains a difficult problem with no unique
solution. A large part of the teams involved in these tests are
using the open-source software Geopsy, born in 2002 within
the SESAME project and primarily released in 2005. In
15 yr, it has become a reference tool worldwide for ambient
vibration processing. A Google Scholar search for “geopsy.org”
currently returns 742 entries (November 2019). A quick survey
made in September 2017 returned 491 entries, among which 203
were journal peer-reviewed papers using Geopsy.
Geopsy is a graphical desktop application dedicated to
seismic signal processing. geopsypack is the package used to dis-
tribute Geopsy and a number of companion tools (41 applica-
tions, 14 plugin libraries, and 22 core libraries, see Data and
Resources). All items are designed with object-oriented pro-
gramming, and the only language is C++. High-level graphical
interfaces are developed on top of Qt libraries (version 5.12, see
Data and Resources), which makes the package available to the
most common platforms: Linux, Microsoft Windows, and Mac
OS. Apart from the graphical user interface (GUI), all tools can
be used directly via the command-line interface, allowing
powerful scripting and the implementation of automatic analy-
sis schemes to bulk data sets. The recent release of geopsypack
(version 3) emphasizes this orientation with new command-line
tools implementing horizontal-to-vertical (H/V) spectral ratios
(also commonly termed HVSR) and array processing that were,
to date, available only inside the Geopsy graphical interface.
After a presentation of the Geopsy graphical environment,
this article focuses on the two main plugin tools used for ambi-
ent vibration processing: H/V and array tools. The correspond-
ing command-line tools are also detailed at the end.
General Framework
In this section, we briefly present the main features of Geopsy
GUI to work with ground-motion signals: import, associate,
view, and save. This section ends with a description of the spe-
cific plotting engine available with this package. More details
and tutorials can be found in the Geopsy on-line documenta-
tion (Wathelet et al., 2010).
Importing signals
Geopsy reads most standard formats commonly used for ambi-
ent vibration and active surface-wave imaging: CityShark2 and
MiniShark*, GeoSIG, GSE2*, GuralpGcf, MATLAB MAT*,
MiniSeed*, Passcal SegY, SAC*, SAF*, Seg2*, SegY*, SU*,
WAV*, and column text files*. Formats marked with an aster-
isk can also be exported. All formats are automatically recog-
nized from their content, except SU and GuralpGCF, for which
the correct file suffix is mandatory. For text formats in which
amplitude values are stored in columns, a custom parser can be
built to extract useful information from the headers (e.g., sam-
pling frequency, starting time, and position). Custom formats
are considered in the same way as standard formats when
trying to recognize a file format. Geopsy can be used to convert
seismic signals between different formats. For example,
geopsy my.mseed -export my_0001.sac
converts a single MiniSeed file into several SAC files (Goldstein
et al., 2003).
Several ways are available to import signals (Table 1).
Importing signals is generally very fast because only the header
information is read and stored in the internal database. Data
vectors are read only when required (e.g., for plotting signals
on the screen or for signal processing). The next section
presents how to associate signals before processing.
Associating and viewing signals
A signal is a continuous record of a single component between
an absolute starting time and an absolute ending time, inter-
nally represented by a single and continuous vector of values. If
some recorded samples are not valid (e.g., digitizer errors or
unplugged sensor), gaps can be added manually. A gap is
defined by a period of time during which no signal can be proc-
essed. Signals to process may be distributed over several files or
several data sources (e.g., recording device proprietary storage
or streaming). In addition, some files may contain signals to be
excluded from the process. Creating new files that contain only
the desired signals usually implies a format conversion and a
waste of disk space. Hence, the concept of file is not well suited.
Instead, we consider signals as the basic element.
Signals to process can be associated in a temporary or per-
manent way (group). Dragging signals from sources and
TABLE 1
Available Methods to Import Seismic Signals inside
Geopsy
Method Short Description
Application
arguments
Wildcards are accepted (e.g., geopsy WAU??/
WA.WAU*HH*)
Manual
selection
The classical dialog box for selecting files in a single
directory
File pattern Inside the graphical interface, files matching a pattern
even across several directories can be imported with a
single action
Storage media Direct read of proprietary storage media of some
acquisition devices (Cityshard cards and Campbell TOB3)
Directory
monitoring
Automatically import any new file created in a
directory. It can be coupled with an acquisition
application to process signals with Geopsy while being
on the field (e.g., Seismodule Controller Software from
geometrics)
Seedlink
connection
Connect to a Seedlink server and continuously
download available signals for the selected streams
Volume 91 • Number 3 • May 2020 • www.srl-online.org Seismological Research Letters 1879
by CNRS_INSU user

dropping them to one of the four destination viewers described
in the following (table, graphic, map, and chronogram; Fig. 1)
creates a temporary association. Sources can be the list of
imported files, the list of permanent associations (groups), or
an existing viewer, which represents a temporary association
of signals. At any time, it can be saved to become a permanent
association (group). A group is defined by a name, a list of sig-
nals, and a parent group. Folders can be created to help organize
groups. For example, a group can keep together the signals from
a target site or signals with the same components. Associated
signals can be processed with any of the implemented basic sig-
nal-processing tools, such as mean or trend removal, subtract,
multiply, filter, whiten, clip, phase shift, over-sample, decima-
tion, taper, cut, merge, three-component rotations, correlations,
convolution, and instrumental response removal. All processes
are logged as a sequence of JavaScript statements, which can be
executed on another signal association. For example, the follow-
ing code cuts and exports 30 s windows starting at 2:30 a.m. on 1
January 2000 from a list of three-component signals:
n=signalCount();
for (i=0; i<n; i++) {
restoreStep (“original”);
selectSignals (i, i);
cut (“[abs 20000101023000; delta 30s]”);
name=header (0, “Name”);
name+=“_”+header (0, “Component”);
exportFile (name+”.sac”);
}
More examples and details can be found in Wathelet et al.
(2010). Other more sophisticated processing tools are also
available and are implemented in separated plugins (see the
Plugin Tools section).
A table shows all properties of the signals with a tabular
format, one row per signal and one column per property.
The order and the list of the columns can be freely customized.
The various ways of editing property values are described in
the next section. The rows of a table support complex selec-
tions with SHIFT and CTRL keys (Fig. 1b). Moreover, signals
can be sorted with multiple sort keys so as to ease the selection.
Once selected, signals can be dragged and dropped into a new
viewer or into an existing one.
A graphic displays signal data vectors with an absolute or rel-
ative time axis. There is no limit to the number of signals to be
plotted. However, the length of a single signal cannot exceed the
limit of the physical memory. If the size required by all signals is
larger than a certain limit (by default, 80% of the physical
memory), the Geopsy internal swap mechanism prevents the
use of the operating system swap. The latter often alters the
computer performances and challenges the user patience.
(a)
(c)
(b) (d) (f)
(e)
Figure 1. Geopsy graphical interface showing (a) the list of
imported files, (b) the list of groups and the four ways of viewing
signals: (c) table, (d) graphic, (e) map, and (f) chronogram. The
topmost menu bar provides access to all actions that can be
performed on signals: input/output actions, editing signal
properties, creating new viewers, waveform operations, and
plugin tools. Just below, the icon tool bar contains a selection of
the most important actions organized in three sections: input/
output actions, creation of new viewers, and the list of plugin
tools. Data are taken from two ambient vibration arrays (circular
arrays from 5 to 15 m and from 15 to 45 m) recorded in
Mirandola, Italy, during the InterPACIFIC project (Garofalo, Foti,
Hollender, Bard, Cornou, Cox, Ohrnberger, et al., 2016).
by CNRS_INSU user

Time picking, zooming, and display normalization adjustments
are available. If data vectors are not in memory, they are loaded
from the original files, which may take some time during the
first display. The performances of this signal viewer are com-
pared with other viewing solutions later in a dedicated section.
Individual signals cannot be selected in a graphic; when dragging
signals from a graphic, all signals are considered.
A map is a 2D map of the sensor locations. Each signal is
located in space by its sensor coordinates. Geopsy supports
only metric coordinates because it is designed for small-scale
experiments (usually less than a few kilometer aperture), for
which earth curvature can be neglected. However, Universal
Transverse Mercator (UTM) coordinates can be used to keep
track of the true geographical position. In this case, a UTM
zone is also assigned to each signal. In a map, signals can
be selected in space by drawing a rectangle with the mouse.
Viewing a map is fast because data vectors are not accessed.
A chronogram is a tool to check the availability of signals in
time. Signals with the same name and the same receiver coordi-
nates are associated in a single entity. For each entity, green boxes
are displayed for the time ranges for which the signal is available.
For the missing parts, red boxes are displayed. Blue boxes show
available signals that overlap in time. Even for very long time
series (e.g., over months), viewing a chronogram is fast because
data vectors are not required. In a chronogram, signals can be
selected interactively by drawing rectangles with the mouse.
Signal properties
The main signal properties are sampling frequency, absolute start-
ing time, sensor location, component, and name. The name is an
arbitrary string to identify a signal. For a three-component sensor,
the three signals must share the same name, the same receiver
coordinates, and approximately the same time range (with at least
a 90% overlap). There are also other properties to store conver-
sion factors between counts, volts, and real units (m, m/s, or m/
s2
). A complete property list is available in the Geopsy online
documentation (Wathelet et al., 2010). The property values are
read from the header of the original signals when they are avail-
able. When file headers do not contain all of the necessary infor-
mation or contain wrong values, editing can be done directly in a
table before processing data. They can also be exported in a text
file, modified outside Geopsy, and reimported. An internal
JavaScript-like scripting engine can also be used to modify signal
properties. In the following example, the station names are
extracted from the file names for all signals in the current viewer:
for (i=0; i<signals.length; i++) {
signals[i].name=signals[i].shortFileName.
substring (3,8)
}
in which signals.length is the number of signals in the current
viewer and shortFileName is the file name without its absolute
path. A specific tool is available to modify receiver coordinates;
it can read text formats in which coordinates are arranged in
columns. Transformation to UTM from WGS84 latitude and
longitude is also provided.
Modifying property values is even better if it can be saved
for some later usages. Details about signal property storage are
given in the next sections.
Event table
Event table stores a list of seismic events defined by their loca-
tions and their absolute trigger times. It is built automatically
when importing Seg2 files, which are supposed to store signals
produced by an active source. It is generally triggered at the
beginning of the signals eventually with a pretrigger delay.
The table can be also edited or imported from a text file.
Events are not stored as a signal property because a source
may be recorded by several signals and a signal may record
several sources. Currently, dense arrays of sensors are gaining
popularity even for near-surface imaging. They usually provide
continuous records instead of triggered records. The event
table is used to define time windows to process and the cor-
responding source locations for active experiments.
Geopsy database
A Geopsy database gathers the list of signal files, the list of
groups, and the list of events. Each signal file contains a list
of signals with their properties but without data vectors. For
each file, the absolute file path is saved to access to the data
vectors. If paths are changed, missing files are detected and
the new paths are asked of the user. A dictionary linking
old and new paths is built for each session to avoid multiple
requests. A database file has a compressed XML format and is
named with a .gpy suffix. The XML content can be extracted
with the tar utility. To save time when opening a database,
original file headers are not read anymore.
Versatile plots
All plots produced by Geopsy and its companion applications
have the same structure and are managed by SciFigs library
(Wathelet et al., 2010). They are made of a stack of indepen-
dent layers that can be easily manipulated (reorder, save, pre-
pend, append, copy, paste, change opacity, or hide). All
painting operations are performed through parallel threads,
with three main advantages: (1) the application never freezes
during heavy painting operations, thus leaving the plain con-
trol to the user; (2) painting several plots at the same time uses
the full power of modern multicore architectures; and (3) plot
properties can be modified by the user with an instantaneous
influence over the final result. Layers can be copied across
applications from the Geopsy package, which greatly helps
when comparing results from various processes.
Plot properties can be fully customized inside the applica-
tion producing the plot or saved for future use (a compressed
by CNRS_INSU user

XML format with .page suffix). Plots from various applica-
tions can be merged inside a single sheet with the lightweight
application figue. Compared with a direct export to an image
format such as PNG, JPEG, or SVG, all properties of a plot
can still be fully modified at any time (e.g., after a paper
review). The plot quality meets the standards required for
publication.
A fast and parallel signal display
Ambient vibration signals recorded with portable stations, par-
ticularly in urban environments, very often show a significant
deterioration in signal quality due to several reasons: bad cable
connection to sensor, baseline instabilities, amplitude satura-
tion, wind effects, nearby sources, power failure, and other
interferences. Hence, ambient vibration time-series analysis
always requires human control through an efficient signal dis-
play, for example, to reject transients before H/V processing
(SESAME, 2004). High-frequency sampling rate, large number
of stations, and long duration recordings lead to a high number
of samples to handle. For example, a 1-day recording at 100
samples per second with 10 three-component sensors, that
is, 259 million of samples, weighs 990 MB in Seismic
Analysis Code (SAC) format (Goldstein et al., 2003) and about
300 MB in miniSEED format.
To build signal plots, Geopsy is based on the layered plots
described in the previous section. The layer in charge of paint-
ing signals is designed for long time series. When the number
of samples is large, painting all high-frequency details of a sig-
nal is not efficient. Instead, Geopsy computes the minimum
and maximum values of the signal over the time range corre-
sponding to each pixel. A vertical bar is painted only between
those two values, which produces the same result as a full
painting of all signal samples. The time spent for displaying
signals is thus optimum. In addition, the size of the exported
vectorized images (SVG or PDF) is kept to reasonable values:
14 three-component stations during 6 hr at 100 samples per
second weigh only 366 KB in a PDF file. Other competing soft-
ware commonly used for signal control, such as MATLAB (see
Data and Resources) and PQLII (Mcnamara and Buland,
2004), do not offer the responsiveness obtained with Geopsy
plots.
To quantify Geopsy efficiency, we composed a data set
from a synthetic 1-day array of 10 three-component stations
recording at 100 samples per second (259 million of sam-
ples), displayed with Geopsy, PQLII, ObsPy (Beyreuther et al.,
2010), and MATLAB. The time required to plot the full data
set (Table 2) was counted manually on the same machine
(Intel Xeon CPU E5-2680 version 3 at 2.50 GHz). Geopsy
and ObsPy achieve almost the same performances for loading
and painting signals, but only Geopsy provides a smooth
interactivity to inspect them. The two other codes are not
able to handle this amount of data with a reasonable
interactivity.
Plugin Tools
Several modules have been developed to extend the function-
alities of the framework described in the previous sections.
They mostly propose sophisticated signal processing focused
on one particular application. The two most popular modules
dedicated to ambient vibrations, that is, H/V and array
processing, are presented later. Currently, 12 tools are avail-
able; their complete descriptions can be found in Wathelet et al.
(2010).
Tools are always attached to a viewer, which represents an
association of signals as discussed earlier. They provide a form
to specify processing and output parameters. They produce
either plots inside the graphical interface or output files to
be analyzed outside Geopsy.
Tools are compiled in separate dynamically loadable libra-
ries. New tools can be implemented in C++ by subclassing the
class GeopsyGuiInterface. Only two functions must be reimple-
mented: one for creating the action in the tool bar and another
one to create the parameter widget. The available plugin libra-
ries can serve as a template collection to inspire new tools.
Single station H/V
This is the very first Geopsy tool whose development started
during the SESAME project. The main purpose of this tool is to
compute the classical H/V of three-component ambient vibra-
tion recordings (Nakamura, 1989; Bonnefoy-Claudet et al.,
2006). Its core structure was inspired by FORTRAN codes
developed by Bard (1999). This module does not provide
advanced ways of computing H/V, such as, for example,
extracting Rayleigh ellipticity (Hobiger et al., 2009) or getting
spectral ratios compatible with the diffuse wavefield
assumption (Sánchez-Sesma et al., 2011; Tchawe et al., 2020).
Three-component signals are divided into time windows
over the entire record or over a selected portion of the record
(Fig. 2a). Time window size, optional antitrigger (short-term
average/long-term average), and various parameters (more
information in Wathelet et al., 2010) are set up manually or
loaded from a previous processing. Time windows can be
TABLE 2
Comparison of Signal-Plotting Applications
Software Version Interactive Paint Delay (s)
Geopsy 3.3.0 Smooth 1
ObsPy 1.1.1 (python 3.7.3) Basic 1
PQLII 2010.246 beta Basic 3
MATLAB R2019a Very slow 13
Approximate time needed to display 259 million of samples for some common
packages. The refresh time is provided here excluding the time needed to read
data from files. For PQLII, the time reported here is the time required to magnify
the full range.
by CNRS_INSU user

selected either manually or automatically, with or without
overlap, keeping or rejecting part of the samples. Generated
time windows may be removed manually, or the selection
may be inverted; they can also be loaded from a previous
processing. Once H/V curves are computed, time windows
can also be eliminated from the final plots, either manually
or automatically, using an H/V amplitude threshold inside a
given frequency range.
The horizontal spectrum is obtained using both horizontal
components, either with a squared average, the total power, or
the power along a given direction (Wathelet et al., 2010). The
averaged H/V curve is obtained by averaging H/V curves com-
puted for each selected time window. The most common
smoothing method is Konno and Ohmachi (1998) with
b-parameter value defined by the user, but other methods
are also available (Wathelet et al., 2010).
Individual H/V plots (Fig. 2b,c) display the average H/V
curve, its standard deviation, the peak frequency with the high-
est amplitude, and the corresponding standard deviation. The
peak frequency and its associated standard deviation are
obtained by averaging the frequency of the maximum found
on the H/V curve of each time window in a 10% interval
around the peak of the average H/V curve. Hence, it may
not always fit exactly with the peak of the average H/V curve.
The initial frequency range for searching peaks on the individ-
ual H/V curves can eventually be edited to fix misidentified
peaks or to evidence secondary peaks. Optionally, the H/V
curves for each time window can be shown with the same col-
ors as in the signal viewer of Figure 2a. This provides a quick
way to localize individual H/V curves on the time scale.
If the provided signals belong to more than one station (dis-
tinct names or coordinates), the aforementioned processing is
repeated for each station and a summary with at least two plots
is produced: the first one gathers all individual average H/V
curves (Fig. 3a), whereas the second shows the average curve
(and its standard deviation) of the H/V curves of the first plot
(Fig. 3b). The peak frequency average and its associated stan-
dard deviation are obtained by averaging peak frequencies of
all processed stations. If coordinates are assigned to the sta-
tions, a map is produced displaying H/V peak amplitudes
or peak frequencies at the location of each recording site
(Fig. 3c). Many options, detailed in Wathelet et al. (2010), exist
for customizing these maps.
Several tools are also available to further analyze a classical
H/V measurement. H/V rotate computes H/V in all directions
of the horizontal plane (Fig. 4). A strong azimuthal variation
may be the first indication of a nonuniform source distribution
and/or the presence of ground heterogeneities. Spectrum and
spectrum rotate provide smoothed spectra of individual com-
ponents with the same processing parameter as H/V. Damping
may help identify peak frequencies produced by anthropogenic
sources (Dunand et al., 2002). Other modules to extract
Rayleigh ellipticity (Hobiger et al., 2009) or H/V under the dif-
fuse wavefield assumption (Tchawe et al., 2020) might be
implemented in the future.
Array processing
The array-processing module provides frequency–wavenum-
ber (FK) techniques (Capon, 1969; Poggi and Fäh, 2010;
Wathelet et al., 2018) and spatial autocorrelation (SPAC) tech-
niques (Aki, 1957; Bettig et al., 2001; Köhler et al., 2007) for
vertical and three-component arrays. They are mostly dedi-
cated to ambient vibrations processing, but the FK techniques
can also be used for active sources. Only FK for ambient vibra-
tions is detailed in this short overview.
Inside Geopsy GUI, for a set of frequencies and for all sets of
time windows (block averaging, Capon, 1969), peaks of FK
maps (FK power vs. kx and ky) are automatically searched
and refined. Various properties of the maxima are saved in
a .max file: time, frequency, polarization, slowness, azimuth,
power, ellipticity, and noise. The last two are only available
for the Rayleigh three-component beamforming (RTBF) tech-
nique (Wathelet et al., 2018). Polarization can be either vertical
(single-component processing), Rayleigh, or Love (three-com-
ponent processing). Figure 5 shows how parameters can be
adjusted inside the graphical interface. A tool is also provided
(a)
(b) (c)
Figure 2. (a) Selected time windows for signals recorded at two
stations (CN01 and CN02). In this example, the antitrigger was
used. Horizontal-to-vertical (H/V) curves at stations (b) CN01 and
(c) CN02. H/V curve for each selected time window is shown with
the same color as the corresponding time window in (a). The
averaged H/V is shown with the black continuous line and the
standard deviation with the two dashed lines. The two gray areas
represent the averaged peak frequency and its standard
deviation. The peak frequency value is at the limit between the
dark and the light gray areas.
by CNRS_INSU user

to manually inspect sets of time blocks at any frequency. An FK
map is displayed with several options and a high-resolution
zoom, which dynamically samples the FK power function
(Fig. 5b).
Following the usage of Geopsy, two applications are pro-
vided to analyze the .max file: gphistogram and gpviewmax.
They both construct histograms of various data contained
in a .max file versus frequency.
The former offers more pos-
sibilities to dissect the results:
filtering outliers, separating
modes, and extracting statis-
tics. The last one is more static,
but it displays in a convenient
way all properties contained in
a .max file: dispersion, elliptic-
ity, noise, azimuth, and power
curves. An example is shown in
Figure 6 for a three-component
FK analysis. Once dispersion
curves and eventually elliptic-
ity curves are isolated for all
available modes, average
curves with their standard
deviation can be computed by
gphistogram. They can be pro-
vided as targets to dinver, an
application implementing a
neighborhood algorithm
(Wathelet, 2008) for the inver-
sion of dispersion curves to
retrieve VS and VP profiles.
Rayleigh and/or Love, funda-
mental and/or higher modes,
phase and/or group velocity,
and ellipticity curves can be
inverted together.
Core Tools
For heavy data sets or repeti-
tive tasks, the tools inside the
graphical interface might not
be suitable. Therefore, the
aforementioned tools have
their corresponding core appli-
cations that can be run in a
high-performance computing
(HPC) facility or on a server
with no graphical service. The
core libraries used by these
tools and their graphical coun-
terparts produce exactly the
same results. However, some
low-level parameters are not accessible inside the graphical
interface.
The signals to be processed are specified by passing a
Geopsy database file path and a list of groups as in the follow-
ing example:
geopsy-fk -db my.gpy -group-path mygroup
10
5
1
0.5
Frequency (Hz)
10
8
6
4
2
0
H/V
10
5
1
0.5
Frequency (Hz)
10
8
6
4
2
0
H/V
10
0
–10
–20
X (m)
20
10
0
–10
–20
Y
(m
)
0.84
0.80
0.76
0.72
H/V
peak
frequency
(Hz)
(a) (b)
(c)
Figure 3. H/V graphics summary obtained when at least two H/V curves are computed. (a) H/V curves
of all stations gathered on the same plot. (b) Averaged H/V curve of all individual curves of (a) and its
standard deviation. Legend as in Figure 2. (c) Map of H/V peak frequencies with the labels showing
the station names. Labels can also display the peak frequencies or the peak amplitudes.
by CNRS_INSU user

It starts FK processing with the default parameters on group
mygroup found in database file my.gpy. Parameters can be
adjusted directly in the command line or by providing a text
file with a list of keywords and their values. A template param-
eter file can be obtained by running
geopsy-fk -db my.gpy -param-example > my.param
A description of all available options for each application is
provided when “-h” is used as an argument.
All tools work with parallel threads whose number is set to
the number of available cores using processor affinity. For
example, on an HPC cluster managed by a task manager such
as OAR (Capit et al., 2005), the number of threads matches
exactly the number of reserved cores and each thread remains
always on the same core during the whole process so as to
deliver the best performances.
Table 3 provides the list of the currently available
core tools and a short description of the implemented
methods.
Conclusion
During the SESAME project (2001–2004), Geopsy was
meant and designed specifically for ambient vibration array
processing. Since then, it underwent huge and crucial transfor-
mations that made it a reference tool for all types of ambient
vibration processing, used worldwide in the engineering and
scientific communities.
Geopsy is a free software released under GNU Public
License version 3, and it can be installed on virtually all ver-
sions of Linux, Microsoft Windows, and Mac OS. The Geopsy
package and its documentation are available online (see Data
and Resources). Since 2005, 12 workshops have been organized
around the world to teach ambient vibration techniques and
Geopsy usage that include 20–30 participants (students, engi-
neers, and researchers) and last from 6 to 7 days. The next
sessions will be announced at Geopsy website (see Data and
Resources).
To date, the design, the implementation, and the mainte-
nance have been achieved by a very small group of enthusiastic
individuals. Operating systems, user needs, geophysical meth-
ods, and scripting languages in vogue (e.g., Python) are con-
stantly evolving. For a sustainable future for this project, a
continuous development effort should be made and better
shared among the community. Hence, readers and users are
kindly invited to participate.
Data and Resources
All data used in this article came from published sources
listed in the references. All figures were produced and all codes
snippets were tested with geopsypack version 3.3.0 available
at http://www.geopsy.org under GNU Public License (GPL)
version 3. Information about Qt libraries is available at https://
www.qt.io. MATLAB is available at https://www.mathworks.com/
products/matlab.html. All websites were last accessed in
February 2020.
Acknowledgments
The authors are grateful to Pierre-Yves Bard who led a number of
projects supporting site characterization with ambient vibrations in
the years 2000 that were particularly helpful during the startup of
Geopsy. The authors thank the numerous users who helped with
bug hunting and/or submitted their needs and suggestions. Geopsy
has been funded by Site Effects aSsessment from AMbient noisE
20
10
8
6
4
2
1
0.8
0.6
0.4
0.2
Frequency (Hz)
180
150
120
90
60
30
0
Azimuth
(degrees)
10
8
6
4
2
0
H/V amplitude
Figure 4. H/V amplitude distribution in the horizontal plane as a
function of the azimuth at station CN01 of Figure 2, calculated
from 0° to 180° every 10°, presented in a frequency–azimuth
graphic.
TABLE 3
Available Core Tools
Tool Name Short Description
geopsy-hv H/V spectra
geopsy-fk Conventional, high-resolution, vertical,
and three-component FK methods
geopsy-spac Vertical and three-component SPAC methods
FK, frequency–wavenumber; H/V, horizontal-to-vertical; SPAC, spatial
autocorrelation.
by CNRS_INSU user

(a)
(b) (c)
1×10–5 1×10–4 1×10–3 1×10–2
FK power
Figure 5. Three-component frequency–wavenumber (FK)
processing example for an array of 15 three-component stations.
(a) Waveforms for the first five stations. The green rectangles
delineate the time blocks calculated for a frequency of 4.09 Hz
that can be adjusted in the parameter toolbox shown in (c). The
red rectangles are the 60 blocks (block averaging count factor in
(c) times 15 stations) used for computing the FK map shown in
(b). (b) Rayleigh three-component beamforming (RTBF) FK map
of the highlighted red blocks in (a). The colors represent the
maximum RTBF power obtained after varying the ellipticity for
each wavenumber point. Maxima are distributed over two circles,
corresponding to two propagation modes. The slowest is marked
by a black circle; its velocity value can be changed in the
underneath controls (251 m/s in this example). The gray dots are
located at the nodes of the starting grid used for searching
automatically the FK maxima. (c) Processing parameters form
divided into four tabs. Only the content of Processing tab is
shown. Time tab describes how time blocks are defined: starting
and ending time to process, length of blocks, and so on.
Processing tab specifies the frequency range of interest and how
the cross-spectrum matrix is computed. Grid search tab provides
ways to control the grid search of FK maxima. Status tab shows
the processing progress over all parallel threads after pressing the
Start button.
by CNRS_INSU user

(SESAME, an FP5 Project Number EVG1-CT-2000-00026, 2001–
2004), Sismovalp Intereg 3B project (2005), by the participants
to the first site characterization workshops in Grenoble (2005)
and Berlin (2006), by NEtwork of Research Infrastructures for
European Seismology project (NERIES-JRA4, an FP6 I3 Project
Number RII3-CT-2006-026130, 2006–2010), by Institut de Recherche
pour le Développement (IRD) since 2008, and by European Plate
Observing System-Implementation phase (EPOS-IP). The authors
also thank all institutions that hosted the site characterization work-
shops; a list is available at http://www.geopsy.org. Geopsy website is
hosted at Observatoire des Sciences de l'Univers de Grenoble (OSUG),
Université Grenoble Alpes, France. Finally, the insightful comments
of two anonymous reviewers helped raise the quality of the original
article.
References
Aki, K. (1957). Space and time spectra of stationary stochastic waves,
with special reference to microtremors, Bull. Earthq. Res. Inst. 35,
415–457.
10
5
Frequency (Hz)
100
(a) (b)
(c) (d)
(e)
0
500
Velocity
(m/s)
10
5
Frequency (Hz)
40
0
–40
Ellipticity
10
5
Frequency (Hz)
Incoherent
noise
10
5
Frequency (Hz)
300
200
100
0
Azimuth
(degrees)
10
5
Frequency (Hz)
1 × 10–1
5 × 10–2
1 × 10–2
5 × 10–3
1 × 10–3
5 × 10–4
1 × 101
5
5 × 101
Power
1000
100
10
1
0
Normalized probability density
Figure 6. Example of results obtained with gpviewmax for a three-
component FK processing. All plots show the probability density of
propertyvaluesmeasuredatthepeaksoftheFKmaps.(a)Dispersion
curve. Two modes are visible in this case. The black lines are located
at the resolution limits kmin/2 and kmin (Wathelet et al., 2008).
(b) Ellipticity curves corresponding to the two modes visible in (a).
(c) Propagation azimuth distribution. (d) FK power. (e) Incoherent
noise estimations as defined by Wathelet et al. (2018).
by CNRS_INSU user

Bard, P.-Y. (1999). Microtremor measurements: A tool for site effect
estimation? in K. Irikura, K. Kudo, H. Okada, and T. Sasatani
(Editors), State-of-the-art paper, Second International
Symposium on the Effects of Surface Geology on Seismic Motion,
Vol. 3, Balkema, Yokohama, 1–3 December 1998, 1251–1279.
Bard, P.-Y., H. Cadet, B. Endrun, M. Hobiger, F. Renalier, N.
Theodulidis, M. Ohrnberger, D. Fäh, F. Sabetta, P. Teves-Costa,
et al. (2010). From non-invasive site characterization to site
amplification: Recent advances in the use of ambient vibration mea-
surements, in Earthquake Engineering in Europe. Geotechnical,
Geological, and Earthquake Engineering, M. Garevski and A. Ansal
(Editors), Vol. 17, Springer, Dordrecht, The Netherlands,
105–123.
Bennington, N. L., M. Haney, S. De Angelis, C. H. Thurber, and J.
Freymueller (2015). Monitoring changes in seismic velocity related
to an ongoing rapid inflation event at Okmok volcano, Alaska, J.
Geophys. Res. 120, no. 8, 5664–5676.
Bettig, B., P.-Y. Bard, F. Scherbaum, J. Riepl, F. Cotton, C. Cornou,
and D. Hatzfeld (2001). Analysis of dense array noise measure-
ments using the modified spatial auto-correlation method
(SPAC): Application to the Grenoble area, Bolletino di Geofisica
Teorica ed Applicata 42, 281–304.
Beyreuther, M., R. Barsch, L. Krischer, T. Megies, Y. Behr, and J.
Wassermann (2010). ObsPy: A Python toolbox for seismology,
Seismol. Res. Lett. 81, no. 3, 530–533.
Bindi, D., B. Petrovic, S. Karapetrou, M. Manakou, T. Boxberger, D.
Raptakis, K. Pitilakis, and S. Parolai (2015). Seismic response of an
8-story RC-building from ambient vibration analysis, Bull. Earthq.
Eng. 13, no. 7, 2095–2120.
Bonnefoy-Claudet, S., C. Cornou, P.-Y. Bard, F. Cotton, and J. Kristek
(2006). H/V ratio: A tool for site effect evaluation. Results from 1D
noise simulations, Geophys. J. Int. 167, 827–837.
Boore, D. M., and M. W. Asten (2008). Comparisons of shear-wave
slowness in the Santa Clara valley, California, using blind interpre-
tations of data from invasive and noninvasive methods, Bull.
Seismol. Soc. Am. 98, no. 4, 1983–2003.
Burjánek, J., J. R. Moore, F. X. Yugsi Molina, and D. Fäh (2012).
Instrumental evidence of normal mode rock slope vibration,
Geophys. J. Int. 188, no. 2, 559–569.
Capit, N., G. Da Costa, Y. Georgiou, G. Huard, C. Martin, G. Mounié,
P. Neyron, and O. Richard (2005). A batch scheduler with high
level components, CCGrid 2005. IEEE International Symposium
on Cluster Computing and the Grid, 2005, 2, IEEE, 776–783.
Capon, J. (1969). High-resolution frequency-wavenumber spectrum
analysis, Proc. IEEE 57, no. 8, 1408–1418.
Cornou, C., M. Ohrnberger, D. M. Boore, K. Kudo, and P.-Y. Bard
(2007). Derivation of structural models from ambient vibration
array recordings: Results from an international blind test, Proc.
3rd International Symposium on the Effects of Surface Geology
on Seismic Motion, Grenoble, France, 30 August–1 September
2006.
Cox, B., C. Wood, and D. Teague (2014). Synthesis of the UTexas1
surface wave dataset blind-analysis study: Inter-analyst dispersion
and shear wave velocity uncertainty, Geo-Congress 2014: Geo-
Characterization and Modeling for Sustainability, 850–859.
Crawford, R., and H. S. Ward (1964). Determination of the natural
periods of buildings, Bull. Seismol. Soc. Am. 54, no. 6A, 1743–1756.
Dunand, F., P. Bard, J.-L. Chatelain, P. Guéguen, T. Vassail, and M.
Farsi (2002). Damping and frequency from Randomdec method
applied to in situ measurements of ambient vibrations:
Evidence for effective soil structure interaction, 12th European
conf. on earthquake engineering, London, 9–13 September 2002,
869 pp.
Garofalo, F., S. Foti, F. Hollender, P. Bard, C. Cornou, B. R. Cox, A.
Dechamp, M. Ohrnberger, V. Perron, D. Sicilia, et al. (2016).
InterPACIFIC project: Comparison of invasive and non-invasive
methods for seismic site characterization. Part II: Inter-compari-
son between surfacewave and borehole methods, Soil Dynam.
Earthq. Eng. 82, 241–254.
Garofalo, F., S. Foti, F. Hollender, P. Bard, C. Cornou, B. R. Cox, M.
Ohrnberger, D. Sicilia, M. Asten, G. Di Giulio, et al. (2016).
InterPACIFIC project: Comparison of invasive and non-invasive
methods for seismic site characterization. Part I: Intra-comparison
of surface wave methods, Soil Dynam. Earthq. Eng. 82, 222–240.
Goldstein, P., D. Dodge, M. Firpo, and L. Minner (2003). SAC2000:
Signal processing and analysis tools for seismologists and engi-
neers, in The IASPEI International Handbook of Earthquake
and Engineering Seismology, W.H.K. Lee, H. Kanamori, P. C.
Jennings, and C. Kisslinger (Editors), Vol. 81, Academic Press,
London, United Kingdom, 1613–1620.
Guéguen, P., J.-L. Chatelain, B. Guillier, H. Yepes, and J. Egred (1998).
Site effect and damage distribution in Pujili (Ecuador) after the
28 March 1996 earthquake, Soil Dynam. Earthq. Eng. 17, no. 5,
329–334.
Guillier, B., J.-L. Chatelain, H. Tavera, H. Perfettini, A. Ochoa, and B.
Herrera (2014). Establishing empirical period formula for RC
buildings in Lima, Peru: Evidence for the impact of both the
1974 Lima earthquake and the application of the Peruvian seismic
code on high-rise buildings, Seismol. Res. Lett. 85, no. 6, 1308–
1315.
Hellel, M., E. H. Oubaiche, J.-L. Chatelain, D. Machane, R. Bensalem,
B. Guillier, and G. Cheikhlounis (2012). Basement mapping with
single-station and array ambient vibration data: Delineating faults
under Boumerdes City, Algeria, Seismol. Res. Lett. 83, no. 5, 798–
805.
Hobiger, M., P.-Y. Bard, C. Cornou, and N. Le Bihan (2009). Single
station determination of Rayleigh wave ellipticity by using the
random decrement technique (RayDec), Geophys. Res. Lett. 36,
L14303, 1–5.
Ibs-von Seht, M., and J. Wohlenberg (1999). Microtremor measure-
ments used to map thickness of soft sediments, Bull. Seismol. Soc.
Am. 89, no. 1, 250–259.
Knapmeyer-Endrun, B., M. P. Golombek, and M. Ohrnberger (2017).
Rayleigh wave ellipticity modeling and inversion for shallow struc-
ture at the proposed InSight landing site in Elysium Planitia, Mars,
Space Sci. Rev. 211, nos. 1/4, 339–382.
Köhler, A., M. Ohrnberger, F. Scherbaum, M. Wathelet, and C.
Cornou (2007). Assessing the reliability of the modified three-
component spatial autocorrelation technique, Geophys. J. Int.
168, no. 2, 779.
Konno, K., and T. Ohmachi (1998). Ground-motion characteristics
estimated from spectral ratio betwwen horizontal and vertical
components of microtremor Bull. Seismol. Soc. Am. 88, no. 1,
228–214.
by CNRS_INSU user

Lermo, J., and F. J. Chávez-García (1994). Are microtremors useful in
site response evaluation? Bull. Seismol. Soc. Am. 84, no. 5, 1350–
1364.
Mcnamara, D. E., and R. P. Buland (2004). Ambient noise levels in
continental United States, Bull. Seismol. Soc. Am. 94, no. 4, 1517–
1527.
Meric, O., S. Garambois, D. Jongmans, M. Wathelet, J.-L. Chatelain,
and J. M. Vengeon (2005). Application of geophysical methods for
the investigation of the large gravitational mass movement of
Séchilienne, France, Can. Geotech. J. 42, 1105–1115.
Michel, C., P. Guéguen, S. El Arem, J. Mazars, and P. Kotronis (2010).
Fullscale dynamic response of an RC building under weak seismic
motions using earthquake recordings, ambient vibrations and
modelling, Earthq. Eng. Struct. Dynam. 39, no. 4, 419–441.
Nakamura, Y. (1989). A method for dynamic characteristics estima-
tion of subsurface using microtremor on the ground surface, Q.
Rep. Railway Tech. Res. Inst. 30, no. 1, 25–33.
Özalaybey, S., E. Zor, S. Ergintav, and M. C. Tapırdamaz (2011).
Investigation of 3-D basin structures in the Izmit Bay area
(Turkey) by single-station microtremor and gravimetric methods,
Geophys. J. Int. 186, no. 2, 883–894.
Park, C., R. Miller, and J. Xia (1999). Multichannel analysis of surface
waves (MASW), Geophysics 64, no. 3, 800–808.
Parolai, S., P. Bormann, and C. Milkereit (2002). New relationships
between Vs, thickness of sediments, and resonnance frequency cal-
culated by H/V ratio of seismic noise for the Cologne area
(Germany), Bull. Seismol. Soc. Am. 92, 2521–2527.
Perrier, L., J.-P. Métaxian, J. Battaglia, and E. Garaebiti (2012).
Estimation of the near-surface velocity structure of the Yasur-
Yenkahe volcanic complex, Vanuatu, J. Volcanol. Geoth. Res.
227, 50–60.
Picozzi, M., A. Strollo, S. Parolai, E. Durukal, O. Özel, S. Karabulut,
J. Zschau, and M. Erdik (2009). Site characterization by seismic
noise in Istanbul, Turkey, Soil Dynam. Earthq. Eng. 29, no. 3,
469–482.
Poggi, V., and D. Fäh (2010). Estimating Rayleigh wave particle
motion from three-component array analysis of ambient vibra-
tions, Geophys. J. Int. 180, no. 1, 251–267.
Sánchez-Sesma, F. J., M. Rodríguez, U. Iturrarán-Viveros, F. Luzón,
M. Campillo, L. Margerin, A. García-Jerez, M. Suarez, M. A.
Santoyo, and A. Rodríguez-Castellanos (2011). A theory for
microtremor H/V spectral ratio: Application for a layered
medium, Geophys. J. Int. 186, 221–225.
Sens-Schönfelder, C., and U. Wegler (2006). Passive image interfer-
ometry and seasonal variations of seismic velocities at Merapi vol-
cano, Indonesia, Geophys. Res. Lett. 33, no. 21, doi: 10.1029/
2006GL027797.
SESAME (2004). Guidelines for the implementation of the H/V spectral
ratio technique on ambient vibrations: Measurements, processing,
and interpretations, SESAME European Research Project WP12—
D23.12, http://sesame.geopsy.org/Delivrables/Del-D23-HV_User_
Guidelines.pdf (last accessed 31 October 2019).
Tchawe, F. N., B. Froment, M. Campillo, and L. Margerin (2020). On
the use of the coda of seismic noise autocorrelations to compute H/
V spectral ratios, Geophys. J. Int. 220, 1956–1964.
Teague, D., B. Cox, B. Bradley, and L. Wotherspoon (2018).
Development of deep shear wave velocity profiles with estimates
of uncertainty in the complex inter-bedded geology of Christchurch,
New Zealand, Earthq. Spectra 34, no. 2, 639–672.
Wathelet, M. (2008). An improved neighborhood algorithm:
Parameter conditions and dynamic scaling, Geophys. Res. Lett.
35, L09301, doi: 10.1029/2008GL033256.
Wathelet, M., P.-Y. Bard, J.-L. Chatelain, C. Cornou, G. Di Giulio, D.
Fäh, B. Guillier, S. Haleimikail, C. Michel, M. Ohrnberger, et al.
(2010). Geopsy on-line documentation, available at http://www
.geopsy.org/wiki (last accessed October 2019).
Wathelet, M., B. Guillier, C. Cornou, P. Roux, and M. Ohrnberger
(2018). Rayleigh wave three-component beam forming: Signed
ellipticity assessment from high-resolution frequency-wavenum-
ber processing of ambient vibration arrays, Geophys. J. Int. 215,
no. 1, 507–523.
Wathelet, M., D. Jongmans, M. Ohrnberger, and S. Bonnefoy-Claudet
(2008). Array performances for ambient vibrations on a shallow
structure and consequences over Vs inversion, J. Seismol. 12,
no. 1, 1–19.
Zare, M., E. Haghshenas, and M. Jafari (2017). Interpretation of dynamic
response of a very complex landslide (Latian-Tehran) based on ambi-
ent noise investigation, Soil Dynam. Earthq. Eng. 100, 559–572.
Manuscript received 26 November 2019
Published online 8 April 2020
by CNRS_INSU user
View publication stats
View publication stats

Geopsy: Seismic Vibration Processing

More Related Content

What's hot (20)

Similar to Geopsy: Seismic Vibration Processing (20)

More from Ali Osman Öncel (20)

Recently uploaded (20)

Geopsy: Seismic Vibration Processing