NFR/NORSAR, Kjeller, Norway

The UN Conference on Disarmament's Group of Scientific Experts (GSE) was established in 1976 to consider international co operative measures to detect and identify seismic events. Over the years, the GSE has developed and tested several concepts for an International Seismic Monitoring System (ISMS) for the purpose of assisting in the verification of a potential comprehensive test ban treaty. The GSE is now planning its third global technical test. (GSETT 3) in order to test new and revisled concepts for an ISMS. GSETT 3 wili be an unprecedented global effort to conduct an operationally realistic test of rapid collection, distribution and processing of seismie data. A global network of seismograph stations will provide data to an International Data Center, where the data will be processed an results made available to participants. The full scaIe phase of GSETT 3 is scheduled to begin in January 1995.

The Intelligent Monitoring Systern (IMS) currently provides for joint processing of data from six arrays located in Northern and Central Europe. From experience with analyst review of events automatically defined by the IMS, we bave realized that the quality of the automatic event locations can be significantly improved if the event intervals are reprocessed with signal processing pararneters tuned to phases from events in the given region. The tuned processing parameters are obtained from off line analysis of events located in the region of interest. The primary goal of such intelligent post processing is to provide event definitions of a quality that minimizes the need for subsequent manual analysis. The first step in this post processing is to subdivide the arca to be monitored in order to identify sites of interest. Clearly, calibration will be the easiest and potential savings in manpower are the largest for areas of high, recurring seismicity. We bave identified 8 mining sites in Fennoscandia/NW Russia and noted that 65.6% of the events of ML > 2.0 in this region can be associated with one of these sites. This result is based on 1 year and a half of data. The second step is to refine the phase arrival and azimuth estimates using frequency filters and processing parameters that are tuned to the initial event location provided by the IMS. In this study, we have analyzed a set of 52 mining explosions from the Khibiny Massif mining area in the Kola peninsula of Russia. Very accurate locations of these events bave been provided by the seismologists from the Kola Regional Seismology Centre. Using an autoregressive likelihood technique we have been able to estimate onset times to an accuracy (standard deviation) of about 0.05 s for P phases and 0.15 0.20 s for S phases. Using fixed frequency bands, azimuth can be estimated to an accuracy (one standard deviation) of 0.9 degrees for the ARCESS array and 3 4 degrees for the small array recently established near Apatity on the Kola peninsula. The third step in the post processing is a relocation of the event, using refined arrivai times and recomputed azimuths from broad band flk analysis. By introducing region specific travel time corrections, a median error of 1.4 km from the reported location has been obtained. This should be compared to the median error of 10.8 km for the automatie IMS processing for these events. This improvement in location accuracy clearly demonstrates the usefulness of the intelligent post processing approach.

We have investigated the potential automatic use of an onset picker based on autoregressive likelihood estimation. Both a single component version and a three component version of this method have been tested on data from events located in the Khibiny Massif of the Kola peninsula, recorded at the Apatity array, the Apatity three component station and the ARCESS array. Using this method, we have been able to estimate onset times to an accuracy (standard deviation) of about 0.05 s for P-phases and 0.15 0.20 s for S phases. These accuracies are as good as for analyst picks, and are considerably better than the accuracies of the current onset procedure used for processing of regional array data at NORSAR. In another application, we have developed a generic procedure to reestimate the onsets of all types of first arriving P phases. By again applying the autoregressive likelihood technique, we have obtained automatic onset times of a quality such that 70% of the automatic picks are within 0.1 s of the best manual pick. For the onset time procedure currently used at NORSAR, the corresponding number is 28%. Clearly, automatic reestimation of first arriving P onsets using the autoregressive likelihood technique has the potential of significantly reducing the retiming efforts of the analyst.