- © 2014 by the Seismological Society of America
Online Material: Computer code (.exe) for processing seismic records; example seismograms.
Using sonification for data analysis is an intensively developing area for scientists working in various fields from astronomy and seismology to psychology (Zwicker and Fastl, 1999; Rubinstein et al., 2007; Last and Gorelik, 2008; Hermann, 2009; Feder, 2012; www.sonification.de; last accessed May 2014). According to popular definition, “Sonification is the use of non‐speech audio to convey information or perceptualize data. Auditory perception has advantages in temporal, amplitude, and frequency resolution that open possibilities as an alternative or complement to visualization techniques” (Wikipedia). The sonification approach is promising in filtering large amounts of data, monitoring data streams, and communication (Feder, 2012). Our ears can serve as the processor of time series not only in popularization applications, but also as a serious scientific tool, because it allows detecting small changes in dynamical patterns of signal.
The simplest way to listen to seismic records is to compress record time by changing the sampling rate. Examples of this technique were presented in the 2012 issues of Seismological Research Letters (Kilb et al., 2012; Peng et al., 2012); authors suggested an automated method that allows effective communication with audiences having different backgrounds. In this paper, we present a different nonautomated approach using the popular Sony SoundForge package that employs more advanced features, such as iteratively applying different filters, speed‐up factors, selecting some parts of a record or gluing different records before sonification as well as to analyze long data files. This new tool, which we call SEISMOTOOL (available in the electronic supplement to this paper), can be used for both data analysis and data sonification and is a convenient way to watch, listen, process, and even analyze seismic records in an easy way.
The advantage of the program we present in this paper in comparison with the MATLAB program environments, such as the SeisSound package (Kilb et al. 2012), is offering the user the options to process long data files, splice together different recordings, extract recording segments, change compression rates, and filter the data records on the fly. The MATLAB environment used in SeisSound requires the data to be loaded in a block of MATLAB memory, which has a limited capacity. As a rule, digital seismic data files are long; for example, if the sampling rate of a record is 100 Hz, the one‐day seismogram includes 8,640,000 samples and takes 67,500 kilobytes of memory if the data type is double or integer. One feature of SEISMOTOOL is that it loads data into the computer memory directly; therefore, it can process longer data sets.
Another feature of SEISMOTOOL is that it can glue together seismograms, for example, merge hourly seismograms into daily ones, etc. The tool can also convert seismograms to acoustic signals (16‐bit WAV files), which allows one to listen to earthquakes and other phenomena recorded by seismometers. These acoustic signals (WAV sound files) created by SEISMOTOOL can be imported into many different programs (e.g., SoundForge, GoldWave, WavePad), enabling users to subsequently listen to seismograms and analyze them. The SeisSound program, distributed by the Incorporated Research Institutions for Seismology (IRIS) data products website at http://www.iris.edu/dms/products/seissound/ (last accessed May 2014), was built to be an automated way to sonify seismograms, whereas our SEISMOTOOL allows selection of different time scales (compression rates) and filtering parameters.
Our SEISMOTOOL package provides two output sound options: either a mono or stereo WAV file. In a stereo WAV file, for example, the first and second channels can contain the original and filtered seismograms. Alternatively, a stereo WAV file can be composed from seismograms recorded at two different stations. Using this approach, we can, for example, compare details of two seismograms recorded at different stations, estimate time delay of wave propagation between two stations, and so on.
Users also have the option to filter the data within a specified frequency range before producing a WAV file. For example, the SEISMOTOOL application has an option to apply a fourth‐order Butterworth band‐pass filter. To accomplish this, we extracted the Butterworth filter from the MATLAB signal toolbox and compiled a similar stand‐alone application. For computer systems that do not have MATLAB installed, the filtering and spectrogram creation requires only installing the MATLAB Run Time Library (version 7.10). We have successfully tested this package on a 32/64‐bit Windows OS XP and Seven. The package was not tested on the Mac but there is a possibility to run Windows‐based applications on Mac. (More information can be found at http://www.macworld.com; last accessed May 2014.) This library can be downloaded free from http://www.mathworks.com/products/compiler/ or http://www.ata-e.com/software/matlab-mcr (last accessed May 2014), or authors can transfer it on demand.
The SEISMOTOOL package uses the Sony SoundForge software for both data processing and listening to seismic data recordings. This program provides a broad range of manipulations (resampling, smoothing, normalizing, noise reduction, and so on) of a digital seismic recording, which, when used interactively, reveals fine details in the data that might not be readily apparent. In particular, SoundForge allows editing of WAV data, changing the time‐scale axis to highlight or compress details within seismograms, analyzing segments of recording, denoising, etc. This tool is very useful for fast identification of triggered tremor signals generated by passing wavetrains from remote strong earthquakes, which is now a hot topic in seismology (Miyazava and Brodsky, 2008; Hill and Prejean, 2009; Prejean and Hill, 2009; Hill, 2010; Chao et al., 2012, 2013; Peng et al., 2012; Gonzalez‐Huizar et al., 2012).
METHODS: HOW TO RUN THE SEISMOTOOL PROGRAM
When running the SEISMOTOOL application, the user has four independent options to choose from (Fig. 1; Table 1). Each option is independent of the others, which allows the operations to be executed separately or sequentially, depending on the project goals. If the goal is to simply produce a WAV file of filtered data within a specific frequency band, only the filter option (option 3) needs to be applied. More complex work that requires filtering, merging, and extracting data before producing a WAV file can also be accomplished using all four data processing options. A full list of the SEISMOTOOL options and examples can be found in the Appendix.
Note that the final output of the program will contain a WAV file without exact indication of how the sound file corresponds the seismogram; here we use the idea that they both have the same duration and start and end at the similar times. This approach differs from the SeisSound codes, which include a time marker on the seismogram and spectrogram, making it easy to discern how the sound and visuals are related; for an example, see http://www.iris.edu//files/products/seissound/data/Movie/3337497_II_TLY_BHZ_2012_04_11_083838_vco_720.mp4; last accessed May 2014). SEISMOTOOL could be improved by further development to include exact time markers for WAV files.
SEISMOTOOL APPLICATION TO RECORDINGS OF THE TOHOKU, JAPAN, EARTHQUAKE IN GEORGIA
Tectonic tremors represent a new class of seismic events related to the discovery of the phenomenon of dynamic triggering (Obara, 2003; Rubinstein et al., 2007; Miyazava and Brodsky, 2008; Hill and Prejean, 2009; Prejean and Hill, 2009; Hill, 2010; Gonzalez‐Huizar et al., 2012; Chao et al. 2012, 2013; Obara and Matsuzawa, 2013). According to these authors, tremor signals have dominant frequencies in the 1–10 Hz range, last for tens of minutes, and propagate at a speed close to that of a shear wave. Spatially, tremor signals are most frequently triggered in hydrothermal areas, but these signals have also been observed in nonhydrothermal areas as well (Prejean and Hill, 2009). For example, the great Mw 9 Tohoku, Japan, earthquake triggered (presumably) local seismic events (Figs. 4–8) in Georgia (Caucasus). Georgia is a continental collision area, separated from Japan by 7800 km. We used the SEISMOTOOL package to analyze seismogram records at two broadband seismic stations located in Oni (on the south slope of the Greater Caucasus) and Tbilisi (in the valley of the Kura River). The stations are separated by 130 km. The records were converted to the WAV format files using option 4 in the SEISMOTOOL package. High‐pass (0.5–20 Hz) filtered signals show that the strongest triggered events at both sites correspond to the arrival of P waves. The sequence of triggered events is quite similar at both stations.
The comparison of the different components of records (N, E, and Z) shows (1) the vertical component (Z) of the wavetrain generates tremors only at the arrival of P and Rayleigh waves (Figs. 4 and 5), and (2) on the horizontal components besides the P wave, triggered events are also generated by Love and, more intensively, Rayleigh waves (Figs. 6 and 7; only the N component is presented, as the E component follows the same pattern).
As recorded both at the Oni and Tbilisi stations, the tremor rate (number of local tremors per hour) before, during, and after Tohoku reveals a clear maximum during the strong earthquake wavetrain passage, including the coda (Fig. 8). This correlation suggests that tremor signals were triggered by the passage of the Tohoku large amplitude seismic waves. We also observe that the anomalously high tremor rate continues for ∼6–8 h. As is shown in Hill and Prejean (2009), the delay between triggered events and arrival of the first Love and Rayleigh waves varies from seconds to several days, and some tremors appear during low‐frequency coda (fig. 6 in Hill and Prejean¸ 2009). The large delay is thought to be connected with seismohydraulic effects—that is, activation of tremors by relatively slow pore water pressure reaction to the wavetrain impact (Miyazava and Brodsky, 2008; Hill and Prejean, 2009; Prejean and Hill, 2009)
Power spectrum analysis of the triggered tremors shows that the maximal energy is released in the 0.4–0.8 Hz frequency range. The tremor spectrum differs very much from the power spectrum of the broadband recording of the Tohoku earthquake, which indicates that the maximal power in Georgia was released at much lower frequencies, in the range 0.01–0.1 Hz. This indicates that very‐low‐frequency forcing is necessary for triggering tremors.
It should be noted that teleseismic waves of strong aftershocks can create a seismic signal that mimics the repetitive features often observed in dynamically triggered local tremor events. The difficulty is that these signals can have some common frequencies. Analysis of the Tohoku earthquake record in Tbilisi carried out by Zhigang Peng shows that the timing of some of the peaks obtained after high‐pass (5–20 Hz) filtering of the Tbilisi record (which is a bit different from our band‐pass range of 0.5–20 Hz, Figs. 4–7) are close to the arrival times of teleseismic waves from strong aftershocks (Mw 7.4, 7.7, and 7.5). This is one of what may be many examples indicating that one needs to conduct a careful analysis of the data in order to distinguish dynamically triggered events from teleseismic wavetrains of remote strong aftershocks.
We also carried out a physical modeling of dynamic triggering of tremors, namely, laboratory experiments on the slider‐spring system with superimposed periodic electromagnetic (EM) or mechanical forcing. The forcings were very weak in comparison with the main dragging force of the spring (Chelidze et al., 2006, 2010). Acoustic emission bursts were used as markers of slip events. When the forcing intensity exceeds some threshold value, due to non‐linear coupling, a remarkable synchronization of slip‐generated acoustic signals with a weak periodic forcing is observed. There is strong resemblance between our experimental results on EM or mechanical synchronization of stick slip by weak periodic forcing (Chelidze et al., 2006, 2010) and the phenomenon of tectonic tremor correlation with Love or Rayleigh wavetrains, observed after filtering records of large‐scale natural events (Hill, 2010; Chao et al. 2012, 2013).
This means that the phenomenon of synchronization (Pikovsky et al., 2003) has a universal character and dynamical triggering of seismicity can be successfully modeled in laboratory.
We present the program SEISMOTOOL, which can be used for education and popularization purposes and for developing a sonification approach to seismic data mining. It provides new options for data analysis: processing long data files, gluing recordings, extracting recording segments, changing compression rate, and filtering records using a multitude of recipes of an elaborated program package, provided by Sony SoundForge. The program can be a useful addition to the existing SeisSound package (Kilb et al., 2012; Peng et al., 2012). The tool is useful for a fast identification of triggered and synchronized tremors generated by passing wavetrains from remote strong earthquakes, which is now a hot topic in seismology. It shows that triggering and synchronization of seismicity can be modeled by laboratory experiments on the stick‐slip process under weak periodic forcing. Further developing of SEISMOTOOL to include time indicators that will connect exactly the original broadband record and produced WAV files is desirable.
The authors acknowledge the financial support of European Centre’s Geodynamical Hazards of High Dams, operating in the frame of the Open Partial Agreement on Major Disasters at the Council of Europe and Rustaveli Scientific Foundation of Georgia (Grant Number FR/567/9‐140/12). We highly appreciate generous efforts of Debi Kilb, who spared a lot of time to improve the quality and clarity of statements of this paper.
The downloadable SEISMOTOOL package includes sample data (the folder seism with hourly seismogram files). These can be merged into a daily record to make sure that the program is working correctly. Attention: The seisfilt.exe and seisspectro.exe programs must be located in the same folder with SEISMO.exe. To run options 3 and 4, MATLAB or MATLAB Run Time Library should be installed in the Windows Operating System (OS).
SEISMOTOOL OPTIONS AND EXAMPLES
Option 1: Extract Segment
This operation is used when we want to extract some segment from the seismogram.
Click on the button Extract segment.
In the new dialog window, select an input file with the seismogram.
In the field First point, indicate the first position in a seismogram from which the segment to be extracted begins.
In the field Last point, indicate the last position of the segment to be extracted in the input seismogram.
Click on Save segment to, and type the output file name for a selected segment.
Click on the button Extract, and wait for the message “DONE” in the Status memo.
Example: From the file input.dat, we need to extract segment (20000:120000) and save it to a file segment.dat. It is a daily seismogram, which is created by merging hourly seismograms (see option 2).
Option 2: Glue Seismograms (An Option to Merge Seismic Records and Process Long Seismograms)
Click on the Output File button, and indicate an output file name. (The directory of the output file and directory of the input seismogram files must be different.)
Click on the Glue seismograms button, and indicate the folder with the seismogram record files.
Example: We have directory seism with an hourly records of seismograms, and we want to glue these seismograms to obtain one day seismogram saved to the output file tbl110311.dat (Fig. A1). Notation: The output file should be saved as tbl110311.dat in any directory except the seism directory.
Option 3: Band‐Pass Filter (Fig. 2)
Click on the Input Signal button, and select a file with a seismogram.
Click on the Output Signal button, and indicate a file name for the filtered seismogram.
In the Fc1 and Fc2 fields, indicate a band‐pass frequency range (in Hz). Notation: Fc1 and Fc2 should be less than half of the record sampling rate.
In the Sampling Rate field, indicate the sampling rate (in Hz) of a seismic signal;
Click on the Filter button, and wait for the end of the operation. If everything is working correctly, a command window will appear and, after its closing, a file with filtered seismogram will appear.
Optional: Plot spectrograms of the original and filtered seismograms.
Example: We want to filter a seismogram in the 0.5–20 H frequency range. The seismogram is recorded at the sampling rate of 100 Hz and saved in the file input.dat. The filtered seismogram will be saved to the file filt.dat.
Option 4: Convert a Seismogram to a WAV File (Fig. 3)
Click on the Add channel 1 button, and select an input file with a seismogram.
Click on the Add channel 2 button, and select an input file with a filtered seismogram. If channel 2 is not selected, only mono file will be created.
In the Sampling Rate field, indicate the sampling rate (in Hz) of a seismic signal.
Click on the To WAV button, and indicate the output WAV file name. Wait for operations to finish.
Example: We want to convert original and filtered seismograms saved in files input.dat and filt.dat to stereo WAV file with the name sound.wav. The first channel will represent the original seismogram, and the second channel will represent the filtered one. The seismogram is recorded with the sampling rate of 100 Hz.