Earthquake Moment Tensor Analysis Using Broadband Seismic Waveforms 
Automatic moment tensor analysis is triggered by an email of a quick earthquake hypocenter
report from JMA (Japan Meteorological Agency).
It includes origin time, hypocenter, and magnitude of earthquake.
This analysis uses broadband waveforms of about 70 Fnet online stations whose data are
telemetried by leased telephone lines.
It provides moment tensor solution of earthquake automatically.
Usually, JMA email arrives 10 minutes after the earthquake occurrence and this system analyze in
1  2 minutes and append the result to the WEB automatically. So within 15 minutes,
everybody can see the result (written by blue characters) from the world.
We have finished examining all the automatically estimated focal mechanisms that occurred after
January 1, 1997. This inspection is usually done during the business hours of the institute
(Mon through Feb, from 00:30 to 08:00, UT).
And if we find possible to improve the solution by manual operation of the data,
we reanalyse by using humanmade dataset, which improves the reliability of
earthquake mechanism solutions especially its depth.
They were written by green characters on the earthquake information pages.
In this analysis, we use a longperiod fullwave inversion code developed by
Professor Douglas S. Dreger at UC Berkeley.
The main feature is that since we use long period surface waves (10100 second),
we do not have to pick up seismic phases, which usually requires verification by human eyes.
So this code is suitable for automatic processing.
Filters used here are 2050 seconds for between 3.5 and 5.0,
20100 seconds for between 5.0 and 7.5, and 50200 seconds for greater than 7.5.
Initial magnitude is referred as a JMA magnitude "Mj" in the email.
You will find a table with origin times (Universal time), locations, magnitudes, and region names,
which are provided by JMA. You will also find qualities in the table.
Quality is a parameter to qualify the result
(100% is best, more than 80% is good 50% is farely well, less than 20% is not good).
We only show reliable solutions which satisfy the criteria that magnitude is greater
than 3.5 and quality (variance reduction) is greater than 50%.
If you click a origin time of preferred earthquake, you will find two tables;
"Hypocenter Information Provided by JMA" and "Focal Mechanism Solution Determined Automatically".
Upper table is the information provided by JMA and the lower table is the result of the analysis.
Strike is the direction of the fault measured from clockwise to the north.
Dip is the dipping angle of the fault measured from horizontal plane.
Rake is the slip direction on the fault measured from anticlockwise to the horizontal plane.
The focal mechanism gives you two possible fault planes and either of them is an actual fault surface.
Mo is moment, whose unit is Nm. Mw is the magnitude measured from moment release
based on the relation Mw=log(Mo*10^{7})/1.510.7.
VarRed is variance reduction, which is the same as quality mentioned before.
The variance reduction is defined as an index of a waveform fit between observation and synthetic waveforms.
Variance reduction is defined as the sum of squares of the difference in amplitude
normalized by the observed waveforms.
Depth is determined by assuming several point sources.
We chose the solution whose variance reduction becomes the maximum.
Point source interval is 3km between 5 km and 125 km, 5 km between 120 km and 200 km,
10 km between 200 km and 300 km, and 20 km at a depth of 300 km or more.
You will find the figure showing the result at the bottom.
Left hand side is the comparison between filtered observed waveforms
(solid line) and calculated waveforms (broken line).
Right hand side is the moment tensor solution. Moment tensor is shown in equal area
lower hemisphere projection.
White area is dilatational region at the source and black is compressional region.
Either of two boundaries shows the fault surface.
If you click the station code, you will see original waveforms.
The unit of horizontal axis is second and vertical axis is count.
You can roughly convert from count to m/s by multiplying 10^9 for BH (high gain)
component and 10^7 for BL (low gain) components.
Among these waveforms, we used at most three station whose hypocentral distance is
between 50km and 400km and whose data quality is good.
[Caution]
When you use automatically determined moment tensor solutions,
due to unexpected noise the result might happen to be completely wrong.
I would recommend you to use the result mainly for the urgent information.
If you want to use the result for other than the above purpose,
you had better examine the original waveforms we used for the analysis.
All the moment tensor solutions are registered in the database so that everybody
search all the moment tensor solutions that the user required through
our Web.
We will renew the database, as required.
We have published "NIED Seismic Moment Tensor Catalogue" in 1997, 1998, 1999 and 2000 as"Technical Note of the National Research Institute for Earth Science and Disaster Resilience".
We would appreciate if you quote this result as "NIED CMT solutions".
We do not have any responsibilities for your any damages from the usage of the results presented here.
If you have any questions, comments, please click
here. 
