Observatories and Research Facilities for EUropean Seismology
Volume 5, no 1 March 2003 Orfeus Newsletter


The Kandilli Observatory Real-Time Automated Seismic Data Processing System

Dean Childs and Ahu Kömeç

Kandilli Observatory Earthquake Research Institute (KOERI), Geophysics Dept., Bogaziçi University, Istanbul, Turkey

Brief History to present
Real Time Automated Seismic Data Processing System Overview
Earthworm Real Time Data Processing Subsystem Performance and Problems
Conclusions
Personnel


View of the Bosphorus from Kandilli Observatory


Brief history to present

Kandilli Observatory Earthquake Research Institute (KOERI) is located in Istanbul, Turkey and has been part of Bogaziçi (Bosphorus) University since 1987. KOERI began making single instrument earthquake observations in the 1930s. In the early 1980s a 9-station radio telemetered centrally digitized network was established in the Marmara Sea Region near Istanbul. In 1992 this same analog network was expanded into greater Turkey using telephone lines but central digitizing was abandoned. In 1997 the first digital broadband station was installed at Kandilli followed in 1998 by the installation of a 5-station digital radio telemetered network encompassing the Marmara Sea. Full scale conversion to digital recording began within KOERI after the tragic August 17, 1999, Mw 7.4 Izmit earthquake. At least 17,118 people were killed, nearly 50,000 were injured, thousands missing, about 500,000 people homeless and an estimated 3 to 6.5 billion U.S. dollars damage in Istanbul, Kocaeli and Sakarya Provinces (U.S. National Earthquake Information Center). At present, KOERI, responsible for monitoring seismicity within Turkey, has online: 52 short period analog (1 component), 6 broadband digital (3 component), 1 short period digital (3 component) and 4 semi-broadband digital (3 component) stations for a total of 52 analog (centrally digitized) and 11 digital stations. Expansion and modernization of the network is still in progress. Most notably, installation of up to 30 three component digital broadband stations is planned over the next 3 years. Conversion to the digital realm naturally led to the development of the Kandilli Observatory Real Time Automated Seismic Data Processing System (KORTASDPS) whose idea was initiated by Mustafa Aktar and Dean Childs in April of 2001. The KORTASDPS located its first earthquakes in November of the same year.

Internet Information Access:

  1. KOERI station list
  2. KOERI station network map
  3. KOERI recent earthquakes
  4. KOERI home page
  5. Official Earthworm web page

Real Time Automated Seismic Data Processing System Overview

The KORTASDPS is a collection of PCs on a closed data network (Figure 1). The core of this automated system uses the Earthworm software initially developed by the United States Geological Survey in Menlo Park, California. Earthworm is a suite of programs called modules which each work interactively and/or independently and are designed to be flexible enough to adapt to a wide variety of seismic network configurations. In addition, KORTASDPS utilizes commercial server software provided by the different seismic instrumentation manufacturers. The primary function of each PC connected to the KORTASDPS closed data network usually falls into 1 or 2 of the following five operational subsystems shown in Figure 1.
  1. Field
  2. Data Acquisition
  3. Earthworm Real Time Data Processing
  4. Data Archiving and Public Distribution
  5. Public Information Access (web server)


Figure 1. Overall system architecture. The 5 subsystems are enclosed in dashed lined boxes, solid line boxes represent computers with the following information listed top to bottom: computer name, operating system, primary software packages and main functions. Heavy arrowed lines represent direction of real time data stream flow. Heavy dashed arrowed lines represent direction of data file flow.

Field PCs are used to configure digital seismic instrumentation and backup data at the station in the event of failures.

Data Acquisition PCs interact directly with the seismic stations, record continuous data streams to the HDD using commercial software and broadcast the same data stream onto the closed data network.

Earthworm Real Time Data Processing PCs primarily run Earthworm modules. They receive the broadcasted data stream from the data acquisition PCs and convert this to Earthworm format where it is stored, served by request and re-broadcasted onto the data network. Waveforms from this Earthworm formatted data stream are fed to a picker, picks are associated into events, events are located, and finally waveforms are automatically assembled into a SAC formatted event files and written to disk. In addition other pseudo real-time tasks are carried out such as producing 24-hour heliplots in GIF image format for each data channel. These images are useful for quick review of station and telemetry health. Because these PCs are core to the real-time component of the overall system and because these PCs are relatively difficult to configure a duplicate or mirror set of computers are run in parallel as a backup.

Data Archiving and Public Distribution PCs automatically pull SAC format continuous and event based data files from the data processing subsystem and create duplicate copies on their local HDDs. This data is open via FTP access to the public. Each month the data is archived to DDS-3 DAT tape. Because of its importance this subsystem is also mirrored.

The Public Information Access PC is the KORTASDPS web server. Plans are to make available to the public real-time earthquake location, magnitude and waveform images, produced by the KORTASDPS as well as general information regarding the overall system architecture.

Earthworm Real Time Data Processing Subsystem Performance and Problems

Table 1 shows the results of a quantitative analysis of the Earthworm Real Time Data Processing subsystem performance over a 3.5-month period. This subsystem was compared to the existing Manual Method, which relies upon visual detection, by a seismic analyst on-site 24 hours/day.

Number of Earthquakes Detected March 13th through June 26, 2002
(A) 467 Only by the Earthworm Subsystem
(B) 403 Only by the Manual Method
(C)
311
Both the Manual Method & Earthworm Subsystem
  1181 Total Earthquakes Detected

Table 1. Subset A is the number of events detected by the Earthworm subsystem and missed by the Manual Method. Conversely subset B is events detected by the Manual Method but missed by the Earthworm subsystem. Subset C is events detected jointly by both methods. Note that significant numbers of events would be missed if either method were used alone. Naturally the question is why.

Figure 2 is a plot of all the earthquakes located by the Earthworm System shown in subsets A and C from Table 1. Overlying the event locations are the stations used by the Earthworm subsystem. Figure 2 shows that earthquake location density is highest in Northwest and North central Turkey where station density is high. Central Turkey has few events, which one might expect, as it is relatively less tectonically active. However there is a clear lack of events located by Earthworm in Eastern and South central Turkey where seismicity levels are known to be high. Eastern Turkey station density is clearly low and station health can be below average, especially in winter, due to difficult accessibility and poor data communication infrastructure. When seismic network coverage is sparse the Earthworm subsystem misses events because it requires a minimum of 4 associated station picks in order for an event is declared. An analyst often can only find 2 or 3 stations with recorded seismic energy for events less than ~3.0 Md in Eastern Turkey.


Figure 2. Map plot of 549 Earthquakes (yellow dots) detected by the Earthworm system from groups A and C in Table 1. Analog and digital stations that were connected to KORTASDPS are represented by black triangles and white squares respectively. The ellipse surrounds aftershocks from the February 3, 2002 Mw 6.2 Sultandagi earthquake. Note the concentration of events located in the northwest and north central Turkey. In these areas station density is high which provides the Earthworm system with a lower magnitude threshold detection capability. In contrast eastern and south central Turkey, which are also seismically very active, are nearly empty of events due to very sparse station coverage. The 467 events located by the Earthworm system but missed by the analyst using just the Manual Method strongly points out the advantage of an automated system. Auto-located events shown in the southwest Black Sea are in error. (note: 229 events from group A in Table 1 were excluded from this plot due to unreliable locations)

In Figure 3 403 events located only by the Manual Method during the same 3.5 month time period are plotted. In contrast to Figure 2 we see many more earthquakes in South central and Eastern Turkey. As discussed earlier an analyst can attempt to locate these small earthquakes with poor data quality and a sparse seismic network whereas an automated system will reject them for insufficient data. Still, in Western Turkey many events have been rejected or have escaped the Earthworm subsystem. Some of these events are in areas of low station density such as the 74 aftershocks enclosed within the ellipse from the February 3, 2002 Mw 6.2 Sultandagi earthquake. However earthquakes within the Marmara Sea area occur in a region of high station density. As before these earthquakes may have been rejected because they are not detected by at least 4 stations. Also, the inclusion of one or more noise picks may have corrupted the location solution made by the associater also leading to rejection. The latter, yet untested, hypothesis is based on the fact that very noisy radio data transmission used by most of the stations located about the Maramara Sea results in some of the highest rates of false picks ranging 2,000 to 27,000 per month. The highest rates usually, but not always, occur during storms. Needless to say such poor quality data creates havoc for an automated system, particularly the event associater.

Station health is clearly an influencing factor with regards to earthquake detectability for any system. The pie chart in Figure 4 shows the condition of the seismic station state of health during the June 22 2002 Western Iran Mw 6.5 earthquake. In June, when weather conditions are mild throughout Turkey, only 66% of the stations are healthy. The effect of poor station performance on earthquake detectability in areas of high station density is not as noticeable, but, in areas where station density is poor, it can be greatly amplified.


Figure 3. Map plot of 403 earthquakes (yellow dots) from group B in Table 1 detected by the Manual Method but missed by the Earthworm System. Analog and digital stations that were connected to KORTASDPS are represented by black triangles and white squares respectively. Compared to Figure 2 there is a clear increase of events located in eastern, central and southwestern Turkey. These events were missed by the Earthworm System because many of them are small and have been located by less than 4 stations (Earthworm must have a minimum of 4 stations) and/or were located using poor 3 and 4 weight signals not detected or poorly timed by the Earthworm auto picker. This figure points to the importance of retaining the human vigilance element until the seismic network improves to a point where the automated system can detect nearly all the events. Nearly 20% of the total 403 events plotted (shown within the ellipse) are from the February 3, 2002 Mw 6.2 earthquake. These aftershock events were missed by the Earthworm system because station coverage is not sufficient and one critical near station, ISP, is not yet part of the Earthworm automated system.


Figure 4. Overall seismic network state of health may greatly influence the performance of an automated system's earthquake detection ability therefore it is important to monitor its status. This figure shows the state of health of the seismic stations and/or data communication links which were connected to the KORTASDPS during the June 22, 2002 western Iran Mw 6.5 earthquake. Dead stations (in red) are defined as those which have no data or data with no seismic signal. Sick stations (in green) have an unhealthy seismic signal but the P wave onset can still be read by an analyst.

Conclusions

Since the August 17, 1999, Mw 7.4 Izmit earthquake Kandilli Observatory has undergone, and is still in the process of making, dramatic changes to its seismic network and real-time data handling procedures.

The existing Manual System of locating earthquakes at Kandilli, though still an important aspect of the detection procedure, is missing significant numbers of earthquakes that are now being detected by the Kandilli Observatory Real Time Automated Seismic Data Processing System (KORTASDPS). With the coming improvements to the seismic network, such as higher quality digital stations and increased station density, visual human vigilance of the real-time data stream will not only be unnecessary but overwhelming. At that time the seismic analyst job will shift to system oversight and location quality control as is done in successful, Earthworm based, earthquake detection networks such as the United States Geological Survey's Northern California Seismic Network located in Menlo Park, California. However until these seismic network improvements are made the Manual Method should be utilized in tandem with the automated system in order to optimize the maximum earthquake detection capability.

Currently, the Earthworm system misses or rejects events due to insufficient station coverage, poor data quality and the high numbers of false or noise picks caused primarily by noisy phone or radio telemetry. The planned addition of 30 high quality digital broadband stations within Turkey will greatly improve problems with insufficient station coverage and eliminate telemetry noise (digital data is inherently free of telemetry noise). Future reliance upon digital stations will improve the automated system's earthquake detection ability as well as provide high quality full waveform data to the scientific community. However, during this transition period the analog stations will still be providing the majority of online data and therefore should not be neglected.

The Earthworm software suite is adaptable to almost all current day seismic network configurations. It is stable and runs on inexpensive PCs. Its modularity allows relatively easy, step-by-step system development and step-by-step learning. Any intermediate to advanced computer operator can self-learn the basics of the system within a month. Also, its modularity and verbose log files aid greatly in isolating system problems to the module level. Once the initial learning pains have been overcome the Earthworm system is easy to expand or install at another location. Though documentation is clearly not complete there exists good documentation on most of the core modules. Best of all Earthworm is free making it accessible to institutions with limited financial resources.

  • Mustafa Aktar
    Bogaziçi University, Geophysics Dept.
    Project initiator, general oversight, promoter and idea development
  • Dean Childs
    Bogaziçi University, Geophysics Dept.
    Project initiator, project and system manager, coordinator, promoter and developer
  • Ahu Kömeç
    Kandilli Observatory Seismology Laboratory
    System manager and Seismic Networm State of Health Monitoring project developer
  • Gonca Örgülü
    Bogaziçi University, Geophysics Dept.
    Archiving subsystem manager and developer
  • Mehmet Yilmazer
    Kandilli Observatory Seismology Laboratory
    Seismic analyst software developer
  • Lynn Dietz
    United States Geological Survey, Northern California Seismic Network
    Technical support
  • David Oppenheimer
    United States Geological Survey, Northern California Seismic Network
    Technical support
Table 2 lists personnel directly involved with the kandilli Observatory Real Time Automated Seismic Data Processing System.

 

page 4
Copyright © 2003. Orfeus. All rights reserved.