Survey Suggest Directions for Development

The ISC conducted a survey of users last year, looking to find which services benefit most seismologists. The ISC accepted completed survey forms from anybody who responded. While each respondent indicated that they use ISC data regularly, selection bias is difficult to quantify in this sort of survey.

The most common uses of ISC data among respondents are seismic hazard analysis (45% of respondents), seismotectonics (41%) and arrival-time tomography (31%). The total exceeds 100% because multiple selections were allowed. Half of all respondents report using ISC data monthly, usually from the web site (55%) or CD (35%).

Two-thirds of respondents use ISC epicentres and magnitudes, while arrival times, epicentres and magnitudes reported by other agencies are each used by about half of all respondents, and macroseismic reports by about 40%. One-quarter of respondents reported using ISC data to select waveforms for further study.

Given that phase data are printed for only some earthquakes, 73% favoured the present policy of including them for large earthquakes, and most said that this was the most important criterion.

Only 39% of respondents felt that a monthly version of the annual CD could satisfy needs now met by the printed Bulletin. Respondents were not enthusiastic about a printed summary to accompany a monthly CD, but 63% of them said that a CD with a graphical interface for selecting events and making maps might be useful in place of the printed Bulletin. Of respondents seeking a graphical interface, 68% indicated that Microsoft Windows was their preferred platform, while 19% preferred X-windows and 6% Macintosh.

The predominance of ISC epicentres and magnitudes among data used shows that high quality analysis remains the ISC's key strength. But use of the Bulletin for selecting waveforms and the demand for a more capable CD show the importance of continuing to develop user services.

1998 Bulletin CD Available

The 1998 Bulletin CD comprises 66,031 earthquakes and explosions, including 47,078 hypocentres computed by the ISC and 82,525 hypocentres contributed by other agencies. The disk includes complete data for 1,187,006 phase readings used to compute the hypocentres, which incorporate 620,074 secondary phases.

All of the 1998 data, plus 1964-1997 hypocentres, magnitudes and macro-seismic data are in fixed-format plain-text files that are read by the event selection program included on the disk, and readily parsed by user programs. The CD also includes PDF documents of the 1998 Regional Catalogue and the "long form" 1998 Bulletin.

The new CD is available at discount to academic institutions and individuals, and further discounts are offered on complete sets of CD's from earlier years. For details contact the ISC or fill-in and print the order form at http://www.gxjydl.com/CD.html then fax or post it to the ISC.

Earthquake Parameters and Standardized Information (EPSI)

With funding from the European Commission, the ISC is participating in the Earthquake Parameters and Standardized Information (EPSI) project, which is led by the European-Mediterranean Seismological Center. Eleven institutes from across Europe participating in EPSI aim to develop standard procedures for computing a number of fundamental earthquake parameters, as well as producing a European-Mediterranean bulletin whose quality would benefit from recent research.

This region-wide bulletin will result from the automatic fusion of tens of bulletins provided by European institutes using newly developed software. The results ultimately will be incorporated into the ISC Bulletin, and lessons learned from automatic fusion on a regional scale may be useful in improving ISC procedures.

Eliminating Data Collection Thresholds

The ISC has always set aside some data from contributions. Early on, the main limitation was time required to re-key data from printed bulletins. Nearly all data are contributed over the Internet now, so re-keying is no longer an issue. But concern remains about time spent reviewing small events, which adds little to regional analysis. Nearly 20% of our 1994 primary hypocentres were simply adopted from the JMA, for example, but only after ISC seismologists tried to re-compute their locations.

The data system used at the ISC through last year had no means for analysing just some of the data. Thus the only alternative to manually reviewing every event was to exclude some of them from the ISC's data collection file. Data were excluded very reluctantly, of course, but by 1998 we routinely put aside some data from each of 14 agencies.

The ISC's new data management system does allow analysis of selected data, so for earthquakes from 1999 onwards we have eliminated collection thresholds. By waiting to select events for analysis until collection is complete, we can take advantage of more data. If several agencies contribute readings for the same event, for example, we can use of all of the data in our analysis if any reported magnitude is sufficiently large.

We prefer analysis thresholds that are uniform within each section of the ISC Catalogue, which conform to the Flinn-Endgahl (FE) seismic regions. As shown in the map below, we use no thresholds at all in sparsely monitored FE regions. In most other FE regions, coverage by local and regional agencies is sufficient that ISC analysis of smaller events serves no purpose so we do not re-analyse events with a maximum reported magnitude smaller than 3.

In some regions coverage is good enough that too many small events are reported for us to analyse all of them, even though coverage gaps within the same FE regions are so common that we must be especially careful. We address these competing concerns by using a lower threshold in these cases, and by using spatially variable thresholds within just a few FE regions. To minimise the possibility of missing out important data, our re-analysis includes both events with no reported magnitude and phase readings that are reported to us without an associated event.

Previous evaluations have shown that the Bulletin is incomplete below about magnitude 4 in almost every region, i.e. that some earthquakes smaller than that are missing. Our analysis thresholds are at least a full magnitude unit below the completeness level, and so do not effect most uses of the Bulletin. As shown in the table below, however, even these very low thresholds achieve a great saving in the number of events that ISC seismologists must re-analyse.

While only analysed events are included in the printed Bulletin, we are confident that the new system provides a better selection than previously. What's more, we plan to offer the complete set of collected data on the web site and AutoDRM server within a few months, and on our CD of 1999 events.

Many users will be surprised that the ISC now collects data for 10,000 to 12,000 events each month, which is more than twice the monthly average in the 1998 Bulletin. The doubling is only partly due to elimination of collection thresholds. Several agencies previously were reluctant to contribute data for regional events, since the ISC made no use of them. But our plan to include all contributed data in a comprehensive summary has helped to motivate more contributions.

The majority of phases are included in our analysis, since so many of them are associated with large events or are reported without an association. It is notable, however, that nearly a quarter of the secondary phases reported to the ISC are associated with events too small to warrant re-analysis. These readings, mostly of regional phases, number well over 100,000 after just a few months using the new system. We're working hard to make them available soon.

Evaluating Residuals

In a paper at the European Geophysical Society meeting, Ray Willemann asked how to tell if a new travel time model significantly improves the fit between observed and expected arrival times. A pitfall for some schemes is that an unknown fraction of arrivals are outliers. Schemes to compute hypocentres by adaptively excluding outliers are widely used, but with a new model the set of purported outliers may change so much that comparisons between two sets of residuals are difficult to interpret.

Instead, Ray advocated trimming a fixed proportion of residuals from each event, without regard for the number of outliers. He used this technique to test for statistical significance of regional variations in results from Chen Qi-fu's study last year relocating ISC events with 3-D mantle models. He concluded that residuals with respect to the 3-D models are significantly better than the Jeffreys-Bullen residuals in many regions. But in regions with many local stations to constrain locations it is more difficult to achieve improvement. In one region, Pamir, Ray reckons that residuals became significantly worse. Details are posted at http://www.gxjydl.com/doc/analysis/2001p02/.