- © Seismological Society of America
We present a new webservice, Syngine, running at the Incorporated Research Institutions for Seismology Data Management Center (IRIS‐DMC), that offers on‐demand and custom‐tailored seismograms served over HTTP. The free service produces full seismic waveforms, including effects like attenuation and anisotropy, that are calculated in commonly used spherically symmetric Earth models (preliminary reference Earth model [PREM], ak135‐f, IASP91). Users can freely adjust sources and receivers, retrieve seismograms from finite sources, convolve with arbitrary source time functions, and download Green’s functions suitable for moment tensor inversions. Syngine extracts and processes seismograms in as fast as fractions of a second, making it suitable for applications demanding short iteration times and a large number of waveforms. For the first time, researchers without large computational resources or specialized knowledge can easily access high‐quality, custom, broadband seismograms. In this article, we present the rational and basic principles of our method, including its limitations. Additionally, we demonstrate the features of Syngine and the included Earth models, showcase several applications, and discuss future possibilities.
Interactive Jupyter notebooks.
Synthetic seismograms are fundamental for a plethora of tasks in seismology, most notably for comparing to observed seismograms and thus testing hypotheses. A wide array of methods for calculating these synthetics have thus developed over the decades. They range from being very accurate but expensive to calculate to being very fast but using heavy approximations that might not be acceptable for a given purpose. Additionally, many seismological software packages are difficult to use, requiring significant expert knowledge to acquire trustworthy results, especially when significant changes to input parameters are required, such as with new Earth models.
In this article, we introduce Syngine (see Data and Resources), a webservice to calculate custom‐tailored, accurate seismograms sent to users upon request with return times as fast as fractions of a second. Syngine delivers full seismograms for spherically symmetric Earth models with anisotropic and viscoelastic rheologies, but the current set of databases does not include the effect of self‐gravitation and Earth rotation, which might affect long‐period applications. The source–receiver geometry can be arbitrarily chosen under the constraint that the receiver is at the surface of the Earth. Further features include seismograms from finite sources, convolutions with arbitrary source‐time functions (STFs), downloadable independent components to construct arbitrary source mechanisms with simple calculations, and windowing traces around seismic phases. A major focus of the project is usability, reliability, trustworthiness, and accuracy of the calculated seismograms.
The sole prerequisite for using Syngine is internet access. Thereby, users will not have to perform an expensive numerical simulation or store very large databases. This equips anyone, including researchers and groups that had no previous access to synthetic waveforms, either due to lack of computational facilities or knowledge, with access to broadband, high‐accuracy, full synthetic seismic waveforms.
This article is organized as follows: In the Methodology section, we introduce the mathematical, numerical, and technical methodology used to generate the synthetic seismograms. The Features section presents the features of the Syngine service, and the Available Earth Models section discusses the rationale behind the offered Earth models, providing guidance for their use. Finally, we discuss potential applications in the Applications section and the big picture in the Discussion section. The code to (re)create most figures is part of the Ⓔ electronic supplement available for this article in the form of interactive Jupyter notebooks (Pérez and Granger, 2007) and additionally on Seismo‐Live (see Data and Resources).
In this section, we present a short and intuitive introduction to the numerical and mathematical methods used to generate and extract the synthetic seismograms. The fundamental methodology has been described at length in the literature, and we will refer interested parties to relevant works where appropriate. We aim to provide readers the necessary knowledge for understanding the capabilities and also the limitations and trade‐offs of our scheme, enabling them to judge its applicability to arbitrary problems. In a nutshell, the response of the medium to two forces—a vertical and a horizontal force source at the surface of the Earth—is simulated and recorded in a 2D domain. By exploiting the reciprocity of the Green’s functions, it is possible to swap source and receiver and, assuming a spherically symmetric medium, reconstruct that response between any two points on the planet, given that the receiver is positioned at a fixed radius. In most instances, this will either be the surface of the Earth or the bottom of the oceans. The method employed by Instaseis (van Driel et al., 2015; see Data and Resources) assures that the seismic wavefield is a full solution to the elastic wave equation in 3D; the only limitation is the required spherical symmetry of both the elastic properties and the domain.
Generation of the Waveform Databases
The first step is to compute databases of accurate waveforms using AxiSEM. AxiSEM (Nissen‐Meyer et al., 2007, 2014; see Data and Resources) employs a spectral element scheme (e.g., Komatitsch and Tromp, 1999) to propagate global seismic waves in axially symmetric media. This assumption allows the analytic decomposition of the 3D wavefield into several 2D wavefields. Thus, only a 2D numerical problem must be solved, which is orders of magnitudes cheaper and therefore enables the calculation and storage of global wavefields at high frequencies. The final 3D wavefield, nonetheless, contains all effects of a viscoelastic rheology like attenuation (van Driel and Nissen‐Meyer, 2014a) and anisotropy (van Driel and Nissen‐Meyer, 2014b). An important limitation to keep in mind is that the seismograms only capture the relevant physics up to a period of about 100 s. At longer periods, effects such as gravitation and Earth rotation play an increasingly important role (see e.g., Komatitsch and Tromp, 2002) but are, at this point, not taken into account in AxiSEM. The overall effect of the neglected physics can be comparatively small but in some scenarios, it does matter, and it is thus crucial to be aware of it. The short‐period bound is dictated by the numerical mesh and can vary for each database. Strikingly, for a full database encapsulating all possible source–receiver geometries, only two simulations are required, due to reciprocity in the wave equation (van Driel et al., 2015): one for a vertical‐force source and one for a horizontal‐force source, both located at the surface of the planet. The displacement response for each is stored at every point of the numerical simulation grid up to a maximum depth. There is no limitation to the recorded depth range, but greater ranges result in significantly increased database sizes. As the recorded range determines the possible source depths, it is natural to limit it to the maximum depth of naturally occurring seismicity.
Exploiting the reciprocity now allows one to place a source anywhere in the recorded region in which the initial force sources correspond to the different components of a receiver. Recalling that we simulated in a spherically symmetric medium, we can place the reciprocal receiver anywhere on the planet and extract its response to sources throughout the recorded depth range. The stored displacements allow the on‐the‐fly reconstruction of the full strain tensor, and resulting seismograms therefore have all three components of a fully 3D wavefield. Storing the displacement on the grid of the numerical simulation and subsequent on‐the‐fly evaluation of the same basis functions as used in AxiSEM means that the spatial interpolation adds no additional error and is as accurate as the spectral‐element simulation for arbitrary source–receiver locations independent of the numerical grid. The wavefields are temporally downsampled to four samples per mesh period, which captures the spectral content of the wavefield down to around −80 dB with respect to velocity. Seismograms are later upsampled again with a tapered sinc function which approximates an optimal reconstruction filter. This extraction is performed by Instaseis, and the final seismograms are amended with meta information and serialized to common file formats by ObsPy (Megies et al., 2011; Krischer et al., 2015; see Data and Resources).
This combination of tools allows the extraction of broadband 3D seismograms propagating through spherically symmetric spheres in as little as fractions of a second with negligible spatial and temporal interpolation errors. The downsides of Instaseis as a stand‐alone tool are the demanding computational costs for each user to generate the initial AxiSEM databases and their unwieldy file sizes. Syngine, the topic of this article, tackles both these problems.
Syngine is a webservice offering convenient and fast access to synthetic seismograms, as described in the Seismogram Extraction section, over the HTTP protocol. Waveforms can be downloaded one at a time or in bulk.
Seismograms constitute the very core of Syngine. Each seismogram represents the response of the medium at one receiver to a particular source. Syngine is used by constructing a URL that encodes the source, the receiver, and additional optional parameters. Accessing this URL either with a script or a web browser triggers the server‐side extraction of the requested seismogram(s), which are then sent to the users. All available parameters are presented in detail in Table 1; a few of them are described in more detail in this section. Parameters are roughly grouped into source, receiver, and miscellaneous parameters.
Sources always need a location specified by geographic latitude, longitude, depth values, and a mechanism. Mechanisms can be defined by passing the six independent moment tensor components, Mrr, Mtt, Mpp, Mrt, Mrp, and Mtp, or by giving strike, dip, rake, and the scalar seismic moment M0. Alternatively, the source mechanism can be a vectorial single force which is useful, for example, for noise studies (see e.g., Basini et al., 2013). For convenience, all of these parameters can be replaced by passing an event identifier from an event catalog. Syngine at the time of writing supports event‐source lookups in the Global Centroid Moment Tensor (Global CMT) catalog (Ekström et al., 2012) and the U.S. Geological Survey (USGS) finite‐fault model (FFM) database; other catalogs may be added in the future.
Receivers also require coordinates and must always be located at a fixed radius, depending on the database, usually the surface of the Earth. Thus, only latitude and longitude coordinates are required for a request. Network and station codes can be passed instead, and they will be used to locate coordinates of the corresponding real stations in the Incorporated Research Institutions for Seismology (IRIS) database. Wildcards can be applied to download a large number of synthetics at once. The following query, for example, will download seismograms for all stations starting with A from the virtual Global Seismographic Network (GSN) network: …&network=_GSN&station=A*&…
Convenience parameters, such as a data scaling factor or a custom file‐name label, are grouped under miscellaneous. The Earth model from which to extract seismograms is also given as a parameter. It specifies the velocity model, as well as the frequency content of the database. Additionally, any combinations of vertical, north, east, radial, and transverse receiver components can be requested either in displacement, velocity, or acceleration, each in SI units. Last but not least, the seismograms can be resampled to any sampling rate larger than the database sampling rate. The resampling employs a very high‐quality reconstruction algorithm and is careful to avoid edge effects; thus we encourage users to make use of this functionality.
Seismograms are returned either as a ZIP archive of SAC (Seismic Analysis Code; Helffrich et al., 2013) files or as a single miniSEED (IRIS, 2012) file. Both have all their important header values filled to aid in data organization and mimicking recorded data.
Exploiting reciprocity necessitates the use of a spherically symmetric planet unlike the real Earth, and Instaseis as well as AxiSEM internally use a spherical geocentric coordinate system. Most seismological applications and data sets, on the other hand, define geographical coordinates on the WGS84 ellipsoid (National Imagery and Mapping Agency, 2000), and we follow that convention and adapt it to our internally used spherical symmetry.
All geographical coordinates that users specify are assumed to be given in WGS84 and are, prior to seismogram extraction, converted to spherical, geocentric coordinates. This effectively is a zero‐th order ellipticity correction for comparison with real Earth data. Coordinates passed back to the users, for example, in the SAC file headers, are also in WGS84. This only affects the latitude, which is usually defined as the angle between the surface normal at a point and the surface normal at the equator at the same meridian. This differs for spherical and elliptical coordinate systems. The difference is usually fairly small and many applications are not affected by it, but it is vital to understand how Syngine handles coordinates.
Phase Relative Times
A multitude of applications only require information about certain seismic phases. To this end, the Syngine service supports the specification of start and end times relative to any phase arrival. We use a port of the TauP Toolkit (Crotwell et al., 1999) in ObsPy (Beyreuther et al., 2010), which employs the τ−p method (Buland and Chapman, 1983) used to calculate theoretical arrival times. It supports arbitrary phase names within its syntactical and semantic limitations and performs the calculations in a spherical coordinate system for better accuracy compared to earlier implementations. Generating phase‐windowed seismograms can greatly reduce the amount of data that needs to be transferred. Figure 1 shows a record section with phase relative time settings.
Syngine relies on synthetics calculated with the spectral element method. Sources for such simulations cannot be true delta functions, because this is numerically unstable. A narrow Gaussian smoothly resembling a delta peak is used instead. Considering the bandlimited nature of the signal, this is a good approximation. Also, each database contains a time series of the slip and slip rates used in AxiSEM to create it. Calculating seismograms with new STFs or slip‐rate functions requires the deconvolution of the database’s STF from the signal, followed by a convolution with the desired STF. Performing this in a single step yields numerical stability under some assumptions regarding the frequency content of both STFs: the new STF must not contain frequencies that the database STF does not have. We disallow STFs that have a finer sampling than the database STF, which contains all simulated frequencies and thus almost guarantees this. If the need arises we will implement some further stabilizing algorithms.
New seismograms uj having an STF sj are calculated from the database seismograms ui with the database STF si as follows: (1) and are the forward and inverse (fast) Fourier transforms, and w is a tapering function to make sure the initial seismogram ends with zero.
Syngine can perform this process, which we refer to as reconvolution, in a stable manner that also deals with subtle issues, such as time shifts. Users can either upload their own STF (see Table 2 and the Syngine documentation for instructions) or use the sourcewidth parameter which reconvolves the seismograms with a Gaussian defined as (2)with t being the time, t0 the offset, and a the source width. See Figure 2 for plots of this function, as well as its effect on seismograms. The peak of the Gaussian will be at the origin time; this results in no apparent time shift but some acausal effects. This, for example, can be used to approximate the triangular moment rate functions used in the Global CMT catalog (Ekström et al., 2012), in which case the sourcewidth parameter is twice the half‐duration specified in the catalog.
Although very convenient, it is oftentimes insufficient to describe seismic wavefields as originating from single points in space. Real earthquakes, especially larger ones, have rupture surfaces, and the point‐source approximation breaks down. The superposition principle applied to linear elastodynamics allows the calculation of finite‐source seismograms for kinematic sources via summation of a number of distributed point sources, each with their independent rise time. Finite‐source descriptions usually originate from kinematic source inversion, but finite‐source models coming from dynamic rupture simulations are emerging as well. The SRCMOD database (Mai and Thingbaijam, 2014) collects FFM, whereas other groups, such as the USGS, offer FFM solutions for download. Syngine supports the calculation of seismograms from FFMs. As of the writing of this article, it can use models described in the USGS “param” file format; other formats might be supported in the future.
The parameters are largely similar to the standard Syngine functionality for seismograms. Phase‐relative times are calculated from the sources’ hypocenters, for example, the patch with the first onset. A source is completely defined by specifying a USGS param file, no other source parameters are required. For the actual calculation of the finite‐source seismograms, we made some choices that may not be immediately obvious. Upon providing an FFM as a source, the following sequence of events occurs on the server side:
The slip rate of each point source is defined as an asymmetric cosine function with a certain rupture, rise, and fall time. We sample it at 10 Hz for a thousand seconds—this limits the maximum rise and fall times. Rise and fall times smaller than one second will be set to one second to make sure it can be accurately sampled. The cosine function is specific to the USGS param file, and the other FFM file formats might define a separate STF for each point which would be evaluated here.
Each sampled slip rate is zero‐padded with a number of samples at the beginning and the end (the additional time shift is accounted for later). This is done to avoid running into boundary issues with the following filter.
A fourth‐order Butterworth filter is applied twice (forward and backward) resulting in a zero phase filter. The corner frequency is the dominant frequency of the database. This assures that we do not introduce frequencies in the convolution that we cannot propagate in the numerical simulation.
The seismograms for all point sources are calculated, time‐shifted, convolved, and stacked.
The resulting seismograms are sent to the users in the same fashion as the point‐source seismograms. This has a profound impact on the seismograms (see Fig. 3 for plots of seismograms comparing a point source to a finite source).
Moment tensor inversions are performed by successively modifying the source mechanism to improve the best fit between data and synthetics until it has converged. This requires the generation of a large number of seismograms from a single origin with varying source mechanisms. A common way of doing this is to reconstruct seismograms as a linear combination of Green’s function from fundamental sources. Syngine supports returning Green’s functions, in the convention introduced by Minson and Dreger (2008), which are readily used by SeisComP3 and other software for moment tensor inversion. Seismograms for an arbitrary source mechanism can be calculated by a linear combination of waveforms from these four sources: (1) a vertical strike‐slip fault (SS), a vertical dip‐slip fault (DS), a dip‐slip fault with a dip of 45° (DD), and an explosive source (EP). Formulas for vertical (uz), radial (ut), and transverse components (ut) are then (3)(4)(5)in which Mrr, Mtt, Mpp, Mrt, Mrp, and Mtp are the moment tensor components, uX,ZZ, uX,DS, uX,DD, and uX,ES are seismograms from one of the elementary sources on the X component, and az is the source–receiver azimuth. See Minson and Dreger (2008) for a derivation and further explanation.
Syngine’s Green’s function service returns these elemental source seismograms for an arbitrary source–receiver geometry. Usage is very similar to requesting normal seismograms, except that the source–receiver geometry is specified by epicentral distance and source depth. Table 3 lists all parameters, and the Ⓔ electronic supplement contains a code example on how to use it. The Green’s functions returned by Syngine are bandlimited, with the delta peak being replaced by a narrow Gaussian moment‐rate function, as discussed in the Source‐Time Functions section.
Meta Information and Documentation
The service will evolve in the future, and it offers an interface to query available models and detailed information for each database. This information includes the frequency range within which each model can accurately deliver seismograms, the embedded source slip rate, and other information, such as the velocity model and the version of AxiSEM used to generate the model. If data are requested as zipped SAC files, it will also contain a log file describing the status for each seismogram requested, including errors for those that could not be generated.
The Syngine website (see Data and Resources) contains extensive documentation for the service with detailed information about each of the parameters, example queries, usage guides, tutorials, and an interactive URL Builder to aid in constructing custom queries. Furthermore, all databases are offered for download directly from the IRIS‐DMC. The heaviest users can therefore run their own Instaseis server to extract a very large number of seismograms without overloading the service and avoiding the latency of sending queries over the internet.
AVAILABLE EARTH MODELS
Earth models, no matter if 1D, 2D, or 3D, are constructed upon an inversion of seismic data, which requires a number of choices on data, modeling and inversion scheme. Inevitably, different models therefore satisfy different aspects of the inverse problem to a varying degree. For instance, in the spherically symmetric case, the preliminary reference Earth model (PREM) model has been constructed for a reference frequency and includes long‐period normal modes, whereas the IASP91 model is derived for body waves. Some models may include anisotropic elastic parameters or density, whereas others are isotropic. Even if laterally averaged, some may work better for continental paths, some for oceanic. As such, it is paramount for a generic webservice such as Syngine to accommodate the inevitable variety of existent models and therefore waveforms. Syngine can currently calculate seismograms propagating through seven widely used different Earth models. At the time of writing these are ak135f_1s, ak135f_2s, iasp91_2s, prem_a_2s, prem_i_2s, ak135f_5s, prem_a_5s, prem_a_10s, and prem_a_20s. The first part denotes the used Earth model, and the final number is the smallest period to which synthetics are accurately modeled. _i_ and _a_ denote isotropic and anisotropic variants. ak135‐f is an isotropic variant of the ak135 velocity model (Kennett et al., 1995), with the attenuation and density model taken from Montagner and Kennett (1996). IASP91 (Kennett and Engdahl, 1991) and PREM (Dziewonski and Anderson, 1981) are other well‐known Earth models; see Figure 4 for comparison of the models and Figure 5 for seismograms calculated in all the available models. Currently, all models are global and continental; we discuss future plans to include other models in the Conclusion and Perspective section.
Potential applications for the the Syngine service are numerous. We hope that providing an extremely low barrier for access to high‐quality synthetic seismograms will spark a number of new uses we have not yet considered. In general, Syngine is suitable for problems that require the calculation of a reasonable number of seismograms. The exact number depends on the required frequency content as well as the available internet connection. Studies that require hundreds of thousands of seismograms, for example, fully probabilistic source inversions, are better accommodated by Instaseis directly. This section showcases a few potential uses to demonstrate Syngine’s applicability to real‐world problems.
Algorithm Test Bed
Newly developed algorithms have to be tested and benchmarked against synthetic data with fully known and controlled properties. Most wave solvers have not been designed to be used by nonexperts, and the oftentimes inadequate usability of scientific software (Brown et al., 2015) brings about substantial efforts to actually acquire synthetic seismograms. Furthermore, it is very easy to misconfigure software, resulting in data that might initially look fine but that contain numerous subtle problems. Syngine and its stack depend on being thoroughly vetted and benchmarked against established solutions. Additionally, all input parameters are automatically examined and checked for consistency. Data acquired within the constraints presented in this article should be accurate. Algorithms that can be validated with Syngine include array processing, phase picking, event locators, backprojectors, and many others.
Data Quality Control
Assessing the quality of recorded data remains a major challenge for network operators and all practicing seismologists handling observed seismograms. Known data problems range from cross talk among recording channels and incorrect sensor orientations to timing errors and erroneous instrument characteristics. Some problems, such as missing or clipped data, are fairly simple to recognize. Others are much harder to detect. Efforts such as the IRIS Modular Utility for STAtistical kNowledge Gathering system (MUSTANG) project (see Data and Resources) are being pursued with the goal of measuring data characteristics that can be used to quantify data quality. Syngine offers the possibility for augmenting these quality measurements with comparisons of the observed data to synthetic seismograms for which all effects are fully known. Systematic and unexpected deviations from the synthetic seismograms can thus be found and investigated.
Examples are small time shifts that are typically introduced by clock errors. The described approach can be applied to various other quality metrics. This is an example demonstrating the feasibility of the approach, but it can well be expanded to be performed in a fully automatic fashion. Figure 6 visually illustrates the chosen strategy.
We compare observed and synthetic P‐phase seismograms with a cross‐correlation technique yielding time shifts with subsample accuracy (Deichmann and Garcia‐Fernandez, 1992). The real data’s instrument responses are deconvolved, and both synthetic and real data are band‐pass filtered with an eighth‐order zero‐phase Butterworth filter to a period band between 5.0 and 12.5 s. For each event, we subtract the mean time shift from the measured time shifts to account for effects such as Earth’s ellipticity, unknown and unmodeled 3D velocity structure, and faulty origin times and depths. These will affect all seismograms approximately equally, which is a fact exploited by many techniques (e.g., Waldhauser and Ellsworth, 2000). We call this the reference data set, and the results are average time shifts per station. Differences among these are due to topography and local velocity structure.
Measuring the time shifts for other events and comparing these to the reference data should result in very small differences if the station times are consistent. Otherwise, there is some mismatch and likely an error with the time information. Figure 6 represent both these cases.
These calculations, including the collection of synthetic seismograms from Syngine, can be performed in a matter of seconds, making it feasible to be integrated in a continuously running quality control system. The presented approach is a simplified proof of concept but already works quite well. Including more and especially local events, other phases, and more advanced processing and statistics will likely improve the assessment. Note the similarity between observed and synthetic seismograms, including matching amplitudes. The proposed technique could thus be extended to check for correct polarities and changing site effects.
Stability Testing in Source Inversion
The modularity of Syngine allows for quickly test seismological codes versus different velocity models. As an example, we show the result of an inversion for moment tensor and STF based on waveforms of P waves. The inversion is based on the method described in Sigloch and Nolet (2006) and Stähler et al. (2012). It alternates between updating the moment tensor and the STF by a joint deconvolution of all Green’s functions from the measured P waveforms. The waveforms are used in a 50‐s time window around the P arrival. The code does a grid search over all plausible depths (1–30 km) and chooses the one with lowest misfit. The code is available from GitHub (see Data and Resources). The inclusion of Instaseis allows it to switch between local waveform databases or the ones provided by Syngine. Because the waveform of the P‐pP‐sP wavetrain not only constrains the depth well but also depends strongly on the velocity model of the crust, different velocity models will result in different estimates for depth and STF. Inverting the same earthquake with different velocity models provides a qualitative estimate of the stability of the result, which is shown in Figure 7.
Beyond application as a research resource, Syngine serves as a valuable tool for education. Concepts explained by instructors, such as polarity of first arrivals, surface‐wave amplitudes depending on the hypocentral depth, or phase triplications, can be visually and interactively discovered and explored, not only by seismologists but also by undergraduate students. The only requirement is a working internet connection and the capability to read and plot seismograms. Figure 8 demonstrates two such possibilities. More generally, the intuitive and interactive nature of the tool opens the door to exploring this for school and museum projects.
For large earthquakes, backprojection rupture imaging (Ishii et al., 2005; Trabant et al., 2012) is a useful complement to finite‐fault modeling because the rupture velocity often trades off with length in the later approach, whereas it can be inferred independently using the former. Figure 9 highlights Syngine’s flexibility and ease of use by comparing backprojection results using P‐wave data with equivalent Syngine synthetics using the multisegment USGS FFM model as a source (see Data and Resources). Such a comparison can also provide insight into interpreting backprojection results when they are not clear due to insufficient station geometry or complicated Green’s functions. Downloading the 25 FFM synthetics needed for this example can take as little as three minutes with only a single URL request, using the appropriate USGS event id and network name or a user’s station list. This type of exercise can be easily repeated to help guide station configurations for future seismic networks or temporary experiments. Similarly, such an effort can be used to assess resolution issues with perturbations in either the source model or receiver array.
Services such as Syngine that remotely calculate on‐the‐fly whatever is needed for a particular application are very likely to become more prevalent in the future. Syngine is, to our knowledge, the first service of its kind that grants access to high‐quality and customizable synthetic seismograms for Earth with a simple web interface. ETH Zürich recently launched a service in the same vein to generate synthetic seismograms for various proposed Mars models (Ceylan et al., unpublished manuscript, 2017; see Data and Resources). Previously, this required high‐performance computing facilities and considerable technical knowledge, severely limiting its de facto availability within the research community.
The closest undertaking comparable to Syngine is the ShakeMovie project (Tromp et al., 2010). ShakeMovie produces more physically accurate synthetic data because it includes 3D Earth models, topography of the surface, and the effects of gravity and rotation. The major disadvantage is that ShakeMovie can only offer seismograms from a given list of earthquakes and receivers, and it is limited to comparatively low frequencies. Syngine, on the other hand, grants full flexibility in the source–receiver geometry, accommodates frequencies up to 1 Hz, and has numerous different models but is restricted to spherical symmetry. Because the 1D structure explains seismic data, especially body waves, rather well, there are a large number of cases where this limitation is acceptable, and the gained flexibility is a worthy trade‐off. The combination of capabilities provided by ShakeMovie and Syngine will remain computationally intractable for the foreseeable future.
CONCLUSION AND PERSPECTIVE
We present Syngine, a webservice for downloading on‐demand, customized, high‐frequency, fully 3D seismograms calculated through spherically symmetric, laterally averaged, anisotropic, and viscoelastic Earth models. Source and receiver parameters (locations and mechanism) can be freely chosen, as long as the receivers are located at the surface of the Earth. Extraction and calculation of seismograms takes less than a second, granting near‐instant results.
The present state of Syngine is its first realization, and it comes with a number of different global Earth models and seismograms in various frequency bands. Future plans include adding more models, some also with water layers, various lithospheric structures, and regional‐specific, even higher‐frequency models/databases. The web interface is sufficiently generic to accommodate synthetics calculated by other means, for example, via normal‐mode summation if gravity effects are of interest. Syngine is currently an IRIS‐DMC service, but we welcome implementations from other institutions, ideally with compatible interfaces. All software components required to use a Syngine webservice are openly available.
To increase reproducibility, the code to (re)create all figures except Figures 7 and 9 is part of the Ⓔ electronic supplement in the form of interactive Jupyter notebooks and is also on Seismo‐Live (see Data and Resources).
DATA AND RESOURCES
Observed waveform and some used event data can be obtained from the Incorporated Research Institutions for Seismology (IRIS) Data Management Center (DMC) at http://www.iris.edu (last accessed March 2017). This includes waveform data from the Global Seismographic Network (GSN—IRIS/U.S. Geological Survey [USGS], doi: 10.7914/SN/IU, last accessed March 2017), the Southern California Seismic Network (Caltech, doi: 10.7914/SN/CI, last accessed March 2017), and the Australian National Seismograph Network. Other earthquake data have been retrieved from the USGS earthquake catalog at http://earthquake.usgs.gov/fdsnws/event/1/ (last accessed March 2017), and the finite‐source data for Figure 3 are from the USGS event page at https://earthquake.usgs.gov/earthquakes/eventpage/us20002926#finite-fault (last accessed March 2017). Further information about services and software used and mentioned in the article can be found at the following websites: the Syngine website (http://ds.iris.edu/ds/products/syngine/; http://service.iris.edu/irisws/syngine/1/query?network=IU&station=ANMO&components=Z&eventid=GCMT:M110302J, both last accessed March 2017), the Seismo‐Live website (http://seismo-live.org, last accessed March 2017), the Instaseis website (http://instaseis.net, last accessed March 2017), the AxiSEM website (http://axisem.info, last accessed March 2017), the ObsPy website (http://obspy.org, last accessed March 2017), and the IRIS MUSTANG website (http://service.iris.edu/mustang, last accessed March 2017). The code used to perform the inversion in the Stability Testing in Source Inversion section is available from https://github.com/seismology/stfinv (last accessed March 2017). The other data can be accessed with URI quakeml:us.anss.org/focalmechanism/10006g7d/mwr. The further information on the synthetic seismograms for various proposed Mars models can be found at the unpublished manuscript by S. Ceylan, M. van Driel, F. Euchner, A. Khan, J. Clinton, L. Krischer, M. Böse, and D. Giardini (2017). From initial models of seismicity, structure and noise to synthetic seismograms for Mars, submitted to Space Sci. Rev.
We thank Editor Zhigang Peng, as well as Mark Panning and an anonymous reviewer for their thoughtful and constructive comments, which helped improve the article. L. Krischer was partially supported by the European Union (EU)-FP7 VERCE project (number 283543) and also acknowledges support from the EU-funded European Plate Observing System (EPOS) project. M. van Driel was supported by grants from the Swiss National Science Foundation (SNF-ANR project 157133 “Seismology on Mars”), S. Stähler by a grant from the Deutsche Forschungsgemeinschaft (project SI1538/2-1 “RHUM-RUM”), and T. Nissen‐Meyer by EU’s Horizon 2020 Research and Innovation Programme under the Marie Sklodowska-Curie Grant 641943. Collaborative visits between some of the authors have been generously supported by eCOST (European Cooperation in Science and Technology) Action ES1401-TIDES. Development and implementation of Syngine at the Incorporated Research Institutions for Seismology Data Management Center (IRIS-DMC) was supported by U.S. National Science Foundation Awards EAR-1063471 and EAR-1261681.