Scribd Logo
Upload

Welcome to Scribd!

  • 36 views

    TheMatchedFilterTutorial Oct 2017

    Uploaded by

    Ade Surya Putra
    The document provides instructions for using matched filter detection to detect earthquakes in continuous seismic data. It involves 5 main steps: 1. Preparing the continuous waveform data and template data, which includes filtering, normalizing headers, and computing SNR for the templates. 2. Running the detection on all continuous data using each template to obtain correlation traces and detect potential events. 3. Generating a catalog of detected events by stacking detections from all templates and removing duplicates. Magnitudes are also estimated. 4. Creating a simple figure comparing a detected event waveform to the template. 5. Measuring differential travel times between templates to refine the origin time estimation of detected events.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    TheMatchedFilterTutorial Oct 2017

    Uploaded by

    Ade Surya Putra
    36 views4 pages
    The document provides instructions for using matched filter detection to detect earthquakes in continuous seismic data. It involves 5 main steps: 1. Preparing the continuous waveform data and template data, which includes filtering, normalizing headers, and computing SNR for the templates. 2. Running the detection on all continuous data using each template to obtain correlation traces and detect potential events. 3. Generating a catalog of detected events by stacking detections from all templates and removing duplicates. Magnitudes are also estimated. 4. Creating a simple figure comparing a detected event waveform to the template. 5. Measuring differential travel times between templates to refine the origin time estimation of detected events.

    Original Title

    TheMatchedFilterTutorial_Oct_2017

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document provides instructions for using matched filter detection to detect earthquakes in continuous seismic data. It involves 5 main steps: 1. Preparing the continuous waveform data and template data, which includes filtering, normalizing headers, and computing SNR for the templates. 2. Running the detection on all continuous data using each template to obtain correlation traces and detect potential events. 3. Generating a catalog of detected events by stacking detections from all templates and removing duplicates. Magnitudes are also estimated. 4. Creating a simple figure comparing a detected event waveform to the template. 5. Measuring differential travel times between templates to refine the origin time estimation of detected events.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    36 views4 pages

    TheMatchedFilterTutorial Oct 2017

    Uploaded by

    Ade Surya Putra
    The document provides instructions for using matched filter detection to detect earthquakes in continuous seismic data. It involves 5 main steps: 1. Preparing the continuous waveform data and template data, which includes filtering, normalizing headers, and computing SNR for the templates. 2. Running the detection on all continuous data using each template to obtain correlation traces and detect potential events. 3. Generating a catalog of detected events by stacking detections from all templates and removing duplicates. Magnitudes are also estimated. 4. Creating a simple figure comparing a detected event waveform to the template. 5. Measuring differential travel times between templates to refine the origin time estimation of detected events.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 4

    The Matched Filter (CPU version) Tutorial

    Xiaofeng Meng

    Red: Parameters, directories Blue: Command lines Purple: File names

    1. Data preparation
    1.1. Waveforms naming rule
    $network.$station.$channel.SAC
    For example, BP.FROB.BP1.SAC

    1.2. Continuous data: ContWaveform/


    Name each directory by the beginning time of the data (yyyymmddhhmmss). For example,
    20031221000000/.
    The time window of one day for continuous data is usually good enough. The detection results might be
    very unstable if the time window is too short. If the number of stations is too large that one day’s data can
    not fit into memory, one should use shorter time window.
    Create a file that lists all directories cont.list.

    1.2.1. Headers
    Set one common kztime for all channels for one day and set the ‘o’ header to 0.
    All waveforms should have the same ‘b’ header (i.e., 0) and ‘e’ header (i.e., 86400). If one channel is much
    shorter than others, it should be excluded.
    Use gsact command to shift time (this step is included in check_cont_waveforms.bash).
    All channels should have the same sampling rate delta. If not, use SAC command interpolate to
    correct (this step is in check_cont_waveforms.bash).
    Command: check_cont_waveforms.bash
    Output: 'sampling_rate.list' lists directories that have different sampling rate among channels.
    Output: 'length.list' lists directories that have significantly different length (default value is 3600s, which
    one can change in check_cont_waveforms.bash by themselves).
    Then, one should decide if those channels need to be removed.

    1.2.2. Band-pass filter


    To enhance local earthquakes' signal to noise ratio, one need to apply a band-pass or high-pass filter.
    For local/regional (~100 km) scale detection, one usually use filters like 2-8 Hz, 5-15 Hz etc. However, due
    to the variations in noise level, signal strength and source/receiver distance, the final filter should be
    decided by users.
    Command: bp.bash
    Output: BP.FROB.BP1.SAC.bp

    1.2.3. Sampling rate


    The computing time grows linearly with sampling rate. Therefore, it is recommended to desample the data
    if possible.
    The filter's frequency range must below Nyquist frequency (i.e., sampling rate/2).
    Modify 'bp.bash' to allow or disallow desampling.

    1.3. Template data TempWaveform/


    Name template id as the origin time of the earthquake (yyyymmddhhmmss). For example,
    20031221211539.
    Create a file that lists all directories temp.list.
    Prepare the catalog of all templates template.catalog, whose format should be:
    [template_id] [year] [month] [day] [hour] [minute] [second] [lat] [lon] [depth] [magnitude] [epoch_ti
    me]
    1.3.1. Length
    Half minute before to two minutes after the origin time of the template is good enough.

    1.3.2. Headers
    Set the kztime to the origin time of the template event and set the ‘o’ header to 0.
    Command: set_otime.bash
    Set desired arrival times for detection as ‘t1’ header. How to get arrival times?
    1) Phase information listed in the local catalog;
    2) Hand-pick;
    3) Predict the arrival time based on 1D velocity model.

    1.3.3. Band-pass filter and sampling rate


    Template waveforms must have the same band-pass filter and sampling rate as the continuous data.
    Command: bp.bash

    1.3.4. Signal to noise ratio (SNR)


    Compute the signal to noise ratio of each channel. Only channels with SNR larger than 5 will be used for
    detection.
    Command: compute_SNR.bash
    Output: wf_SNR.dat in each template’s directory, lists SNR of each channel, whose format is:
    [wf name] [SNR] [energy of signal window] [energy of noise window]
    Output: dir_event_SNR.dat in parent directory, lists SNR information of each template, whose format is:
    [template ID] [median SNR] [Total number of channels] [Channels with SNR>5]

    2. Detection WorkDir/
    2.1. Running detection on single channel
    sliding_wfcc_fix_v5 [-f template_sac_file] [-s continuous_sac_file] [-b template_start_time] [-a
    template_end_time] [-B continuous_start_time] [-A continuous_end_time] [-S time_step] [-O
    shift _time] [-t taper] [-D debug] [-F output] [-o sacfile] -h help
    sliding_wfcc_fix_v5 -
    f /home/xmeng/Tutorial/CPU_WFCC/TempWaveform/20031221211539/BP.VCAB.BP1.SAC.bp -
    s /home/xmeng/Tutorial/CPU_WFCC/ContWaveform/20031221000000/BP.VCAB.BP1.SAC.bp -b 4.7216
    -a 10.7216 -B 1 -A 86448 -S 0.05 -O 4.7216 -F 1 -o
    VCAB_BP1_20031221191400_t2_bp_20031221191400_wfcc.sac
    Output: VCAB_BP1_20031221191400_t2_bp_20031221191400_wfcc.sac

    2.2. Running detection on all continuous days with one template


    Prepare a config file ‘config’. [wfcc_code] is full path to code; [data_dir] is full path to data and working
    directories; [output_option]: 0 outputs ascii file, 1 outputs sac file, 2 only shows script; [slide_win] is
    temporal step (usually sampling rate); [phase] is header for arrival time; [t1] is starting time of detecting
    time window (1s before arrival time); [t2] is end time of detecting time window (4s after arrival time);
    [minimum_wf_cutoff] is the minimum number of channels per template; [min_SNR] is the minimum SNR
    for detection; [allow_shift] is the number of data points allowed to shift while stacking.

    Command: Run_detection.bash 20031221211539, where 20031221211539 is template id.

    2.3. Stack correlation traces and output positive detections


    XmengStackShift [list_of_correlating_traces] [number_of_correlating_traces] [number_of_points_allo
    w_to_shift] [output_sac_file] > [output_detected_events]
    Default detection threshold is {median value+9*MAD}. You may change it at line 123 in the source code
    and recompile it.
    XmengStackShift sac_file_SNR_20031221191400.id 27 0 shift_20031221191400.sac >
    9times_20031221191400.dat
    Output: shift_20031221191400.sac, stacked CC trace.
    Output: 9times_20031221191400.dat, list of detected events. Format is following:
    First line: mad:median_value:MAD
    Other lines: detection_time CC (CC-median_value)/MAD
    This step is included in 'Run_detection.bash'.

    2.4. Running detection on all continuous days with all templates


    In the sample dataset, we have 6 templates and 1-day continuous data. Since we usually have multiple
    CPUs on one computer, we can utilize them at once.
    We can divide the template library into three equal parts and submit a job for each one of them. In this way,
    we can shorten the computing time to one third.
    Command: Run_Whole_Dataset.bash 3, where 3 is the number of CPUs you want to use.

    3. Prepare catalog Catalog/


    Prepare a config file ‘config’. [min_chan_number] is the minimum number of channels per template;
    [threshold] is detection threshold (median+9*MAD); [time_step] is the time window for removing
    duplicate detections (2 s); [data_dir] is full path to data and working directories; [min_SNR] is the
    minimum SNR for detection; [phase] is header for arrival time.
    Command: Generate_catalog.bash 20031221000000, where 20031221000000 is continuous data ID.

    3.1. Combine detections from all templates


    3.2. Remove duplicate detections
    Many close templates may detect the same earthquake, with subtle differences in origin time. Therefore, we
    only keep the one with the highest CC every a few seconds (default is 2s).
    3.3. Compute the magnitude of the detected events
    Obtain the peak amplitude of the detected event and the corresponding template, respectively.
    magnitude_detected_event = magnitude_template + log10(peak_amp_ratio).
    3.4. Generate final catalog
    Output: catalog_20031221000000.dat. Final catalog format:
    [1. detected event date] [2. detected event time] [3. detected event epoch time] [4. detected event time
    since midnight] [5. detected event magnitude] [6. CC] [7. (CC-median value)/MAD] [8. template id] [9.
    template year] [10. template month] [11. template day] [12. template hour] [13. template minute] [14.
    template second] [15. template latitude] [16. template longitude] [17. template depth] [18. template
    magnitude] [19. template epoch time]

    4. Figure Figures/
    Generate the a simple figure for waveform comparison between the detected and template event.
    Command: waveforms_comparison.bash catalog_20031221000000.dat 36,
    where catalog_20031221000000.dat is the detected catalog and 36 is the number of detected events.
    Users can change the absolute path and parameters in waveforms_comparison.bash.
    Output: 75448.00.eps

    5. Measure differential travel time Detected/


    Prepare a config file ‘config’. [search_range] assume different templates detect one event within the time
    window; [delta] is sampling rate; [fit_points] is number of data points around peak CC for quadratic
    interpolation; [wfcc_code] is full path to code; [data_dir] is full path to data and working directories;
    [phase] is header for arrival time; [threshold] is detection threshold (median+9*MAD); [threshold2] is
    threshold on single channel for measuring differential travel time; [min_SNR] is the minimum SNR for
    detection.
    Command: Correlate.bash catalog_20031221000000.dat 11

    5.1. Correlate around the origin time of the detected event


    Perform correlation 60s before and 60 after the origin time of the detected event from all templates that
    detected the particular event. Make sure to differentiate P- and S-arrivals.
    5.2. Output the differential travel time and weight
    5.2.1. The origin time of the detected event is the detection time from the template that has the highest CC;
    5.2.2. Find the highest CC (CCpeak) on each channel, as well as the second highest CC (CCsecond), within
    search_range of the origin time, and determine if CCpeak is above threshold2;
    5.2.3. Perform a spline interpolation to the three data points around CCpeak;
    5.2.4. Weight is CCpeak*CCpeak*(0.1+3*(CCpeak-CCsecond));
    5.2.5. Write the differential travel time and weight to the output file ‘hypoDD.$template’.

    You might also like