Electronic Supplement to
The SCEC Geodetic Transient-Detection Validation Exercise

doi: 10.1785/0220130041

by Rowena B. Lohman and Jessica R. Murray

Synthetic Timeseries, Input Models and Sample Scripts

The files below include synthetic data that can be used as input to test codes developed for the detection of anomalous transient signals in geodetic data, the parameters of the noise and underlying signals used to generate each test data set, and a set of example MATLAB^® scripts that can read data provided through the testing center hosted at the Southern California Earthquake Center.

---

Test Data

The data provided to the participants for Phase I are time series for a network of GPS sites that will mirror the continuous GPS station distribution of southern California, extending east to the Colorado River, spanning 1999-2006. The synthetic data are may include data gaps, noise with a character unknown to the participants, as well as a deformation signal.

The files listed below each uncompress to folders containing a set of files, e.g.,

2009.016.045624.tar.gz

2009.016.045727.tar.gz

2009.016.045829.tar.gz

2009.016.050004.tar.gz

each of which is a separate simulation. These files in turn unpack to a set of .csv files (warning, in the same directory), which have names like

pin1.2009.016.005407.csv

The 2009.... is simply the creation time used as a unique ID. The .csv files look like:

Station pin1,    33.6122,   -116.4582,   1256.17
2000-01-01,    -0.276,    -0.059,     2.562, 51545.0000
2000-01-02,     0.347,     0.077,    -0.626, 51546.0000
2000-01-03,     0.360,     0.633,    -0.664, 51547.0000
2000-01-04,     0.163,     0.401,    -4.012, 51548.0000
2000-01-05,     0.546,     0.870,     1.427, 51549.0000
2000-01-06,     0.122,     0.055,     1.210, 51550.0000
2000-01-07,     0.198,     0.320,     2.132, 51551.0000

which is very similar to the PBO csv file, with the MJD included in the last column. The displacements are in order East, North, Up.

Each simulation uses a different random number sequence to generate the noise, and the noise parameters for the stations are randomly altered between simulations. So there should be no common elements.

simulations.1a [Zip Archive; ~43.2 MB]

simulations.2a [Zip Archive; ~77.7 MB]

simulations.2b [Zip Archive; ~77.7 MB]

simulations.2c [Zip Archive; ~77.7 MB]

Phase3 [Zip Archive; ~487 MB]

---

Description of True Sources for Synthetic Data

Phase 1 slip models [Zip Archive; ~432 KB]. This file contains the secular rate, noise characteristics, and true source-time functions used in Phase 1 and contains:

2009.016.045624corr.log

2009.016.045727corr.log

2009.016.045829corr.log

2009.016.050004corr.log

Explan

File "Explan" provides an explanation of the parameters.

Phase 2,3 Full models of fault slip for the datasets can be found below. Note that there is no slip model for Phase 3B, since this contained real data.

Phase 2A [Zip Archive; ~63 KB]

Phase 2B [Zip Archive; ~31 KB]

Phase 3A [ASCII Text; ~935 KB]

Phase 3C [ASCII Text; ~8.2 MB]

Phase 3D [ASCII Text; ~656 KB]

Phase 3E [ASCII Text; ~1.4 MB]

Phase 3F [ASCII Text; ~1 MB]

Phase 3G [ASCII Text; ~1.3 MB]

Once untarred and gunzipped, these files contain a text file for each of the datasets that had an actual signal added to it - i.e., if you don't see a number matching one of the existing data sets, then this was a "control", with no signal. Set PhaseIIIB was real data, hence no solution. File names should look like the following: slip.2009.167.001253

Lines beginning with a "g" describe a particular fault plane: for example

g 34.0300 -118.4348 34.0646 -118.4525 t t 5.000 18.600 16.0

describes a plane whose top edge runs from the first location to the second, at a depth of 5 km. The dip is 16 degrees, down from the horizontal, and to the right looking from the first point to the second; the fault width down-dip is 18.6 km (which gives a bottom depth of 10 km). The two letters are depth codes used in some of my other software; here they are always set to t.

Each line that describes a fault plane is followed by lines giving the slip history on that plane; for example

s 0.0109 90.0 2001 2 12 0 0

s 0.0106 90.0 2001 3 12 0 0

is a slip of 0.0109 m, rake 90 degrees (thrust), at time 2001:002:12:00:00, followed a day later by one of 0.0106 m. The slips are taken as a series of steps, but of course the displacements are computed quasistatically.

---

Set of MATLAB^® Scripts for Running Transient Detection

Lohman_scripts [Zip Archive; ~1.7 MB]. Set of example scripts that accept inputs and outputs according to the requirements of the SCEC testing center.

Usage:

Step 1: Download file and uncompress in the area on your personal machine where you plan on working

Step 2: Fix paths in two spots

• in Lohman/CommonScripts, set the GPSDataDir in get_data_UNAVCO.pl to be the location where you want your GPS data to go.

• in setup_paths.m, adjust GPSDataDir (same as above), CMTDataDir, CommonScriptsDir, DetectionsDir (must exist). Change RunDate to "now" if you want it to be the same day as you run the code. RunDate allows SCEC to repeat the test on a subsequent day as if it were performed on the same day, for validation purposes.

Step 3: Update GPS data: Run get_data_UNAVCO.pl, no arguments. Can run from CommonScripts dir if paths are set.

Step 4: open MATLAB and run fit_timeseries.m (may have to change your MATLAB path if you are not running from the "Lohman" directory).

Step 5: If desired, adjust the parameters in "general_params.m":

• AC/product: Leave as is
• Cflag: set to 1 to have a lot of comments about data quality come out
• Pflag: 0 = no plts. 1 = plot map of all stations with info about their quality, detections, etc, 2 = plot the time series and fit for each detection, 3 = plot time series and fit for all time series that meet the basic length and neighbor criteria
• Minlat/maxlat/minlon/maxlon = area of interest
• Distneighb: distance around each point where other points are considered "neighbors"
• Numneighb: Number of neighbors that must also move for a detection to be considered "real". Sites with fewer than that number of neighbors existing will not be considered at all.
• Nsig: Number of standard deviations above the noise, after fitting model, where time series are considered "moving" and will flag as a detection.
• Velnum: minimum # of dates that must exist in order for the site to be considered at all.
• Testcutoff: time period in days (from RunDate) used to assess detections. I.e., the algorithm looks at only those points after a model is removed, and assess whether they are Nsig above the noise.
• Testnum: How many dates within the testcutoff time that must be above Nsig for there to be a detection. I.e., if it is = testcutoff, then ALL recent dates must be above Nsig for a site to be flagged
• Minint: minimum time between earthquake to allow the fitting program to assess postseismic signal.
• EQstart: Only consider earthquakes after this date

[ Back ]