1. Procedure Automatiche e Modulari di Analisi dei Dati Sismici Istituto Nazionale di Geofisica e Vulcanologia Taboo Workshop Ancona Aprile 23-24-2013

2. Data Crunching & Seismological Analysis Edward Ancona INGV National Network TABOO Seimic Network Rome SeedStore INGV Mini Seed Archive daily rsynch is a DELL Power Edge R510 RAM 32GB 2 Intel Xeon X5675 (3.06GHz, 6C) Ubuntu Linux Server 12.04 Edward Alto Tiberina Network

3. Elaborazione Edward’s Jobs Scheduling Start (1/day at fixed time) rsynch Mseed data (daily files per seismic channel) from SeedStore to Edward’s local archive Convert (N/day every M minutes) Converts Mseed 24H to SAC 24H files (msrouter, ms2sac, sacmerge) Trigger (N/day every M minutes) Launches Burst Trigger engine on 3 components 24H SAC  list of triggers per station Binder (N/day every M minutes) Launches Burst2Coinc Binding engine on selected stations  list of possible events per day Scissor (N/day every M minutes) Launches the events cut engine on the local MSeed archive  1 directory per event with 3 sac 180s waveforms files (Z, N, E) per station Picker Loc Mag (N/day every M minutes) Launches the CORE of the TABOO Automatic Detection and Picking System Scheduling is performed by the time-based Linux job scheduler CRON daemon All the jobs run based on a “Lock” e “Done” logic. the system never stops looking for something to do thus exploiting all the available resources

4. Elaborazione Edward’s Jobs Scheduling This tells Start: “on which day in the past from today should it work”. 0 means “TODAY” This tells Start: “on which day in the past from today should it work”. 0 means “TODAY” Quasi Real-TimeVSReal-Time What can we and what we can’t do We can elaborate whatever day and on a unlimited (except hardware resources limits) number of days at once (working dirs are independent). We can work on today now before midnight (with limits of the real- time but in quasi real-time) We can’t work in real-time … without intensive codes modifications Data: complete and quality checked seismic data Problems: whatever problem is possibly solved or identified Purpose: purely scientific Dense seismic sequences are better handeled Data: possibly incomplete or absent at the time of pick) Problems: whatever problem can’t be handled in real time Purpose: monitoring/civil protection/alarm notification This is why we have analysts seismologists a posteriori working on data for bulletins

5. Elaborazione BURST: triggering engine Orginally developed with Matlab  converted to F77 for opensource portability It works on single station, sac 3 components files Moving windows (1s every 0,5s) on 3componens (24h) Covariance Matrix is calculated Square Matrix Eigenvalues are calculated Square Matrix Sum over not-null diagonal Sum of the modules of the 3 components amplifies coherent signal dumps random noise (modulus sum ~= 0) Waveforms are filtered band pass Butterworth, 1-20Hz 4 poles Sta/Lta Algorithm SL search is performed on the CM Setup Trig on: 1.9 Trig off: 1.7 Duration: > 0.5s

6. Elaborazione Station AT07 Possible events/24h: 725 Event: 2010/04/15 01:47:36 Ml: 3.8 Sta/Lta Covariance Matrix Waveform Z BURST

7. Elaborazione Station ATLO Possible Events/24h: 1725 Event: 2010/04/15 01:47:36 Ml: 3.8 Sta/Lta Covariance Matrix Waveform Z BURST

8. Elaborazione Station ATLO Possible Events/24h: 867 Event: 2010/04/15 01:47:36 Ml: 3.8 Sta/Lta Covariance Matrix Waveform Z BURST

9. Elaborazione BURST2COINC: earthquake finder Orginally developed with Matlab  converted to F77 for open source portability It works on BURST triggers per station outputs Sort Triggers are merged and sorted by increasing time Time separation Based on stations inte-rdistance + S-wave expected mean a time lapse is defined over which consecutive triggers do not belong to the same event  single event is declared Minimum Triggers Only events with at least N of the selected stations are declared Subset of High Quality Stations Only triggers at selected stations are taken into account Final Check Double trigs and possible merged events are found and resolved 2010/04/15 Possible Events 1450

10. Elaborazione Edward’s Jobs Scheduling Start (1 daily at fixed time) rsynch Mseed data (daily files per seismic channel) from SeedStore to Edward’s local archive Convert (N/day every M minutes) Converts Mseed 24H to SAC 24H files (msrouter, ms2sac, sacmerge) Trigger (N/day every M minutes) Launches Burst Trigger engine on 3 components 24H SAC  list of triggers per station Binder (N/day every M minutes) Launches Burst2Coinc Binding engine on selected stations  list of possibile events per day Scissor (N/day every M minutes) Launches the events cut engine on the local MSeed archive  1 directory per event with 3 sac 180s waveforms files (Z, N, E) per station Picker Loc Mag (N/day every M minutes) Launches the CORE of the TABOO Automatic Detection and Picking System Scheduling is performed by the time-based Linux job scheduler CRON daemon All the jobs run based on a “Lock” e “Done” logic the system nevers stops looking for something to do thus exploiting all the available resources

11. Risultati Scissors: Based on Burst2Coinc output extracts 180s waveforms per channell per station directly from local Mseed converts to SAC format Place the set in one directory per event (YYYY/JDAY/YYYYMMDDHHMISS) Writes the trigger time per station in each triggered waveform as a reference for PickerLockMag) Picker Loc Mag checks and prepares waveforms for the automatic picking) Refine Burst triggers to better match P and to trig untriggered stations (we use a basic picking engine for this) Launches the well tested “Mpx Shell”, a set of programs which core is MannekenPix (by Dr. Freddy Aldersons) Picker Loc Mag P- S-wave pick, P-onset polarity, 1D Location, Magnitude Calculation

12. Risultati Picker Loc Mag Filtering is performed with adaptive Wiener Filter based on Noise and Signal+Noise Windows separated by Safety Gaps around the reference picking/traveltime/trigger Identification Algorithm (IA) based on the C5 algorithm (Quinlan, 1993) classifies P, S, N samples MPX is guided toward the proper event by a trigger in the waveform. Guiding theoretical travel times are calculated for un-triggered additional waveforms after Run1 MPX can be run in Highest Hit Rate (HHR) mode or in Highest Accuracy (HA) mode. Run1/1b: HHR (Takanami-Kitagawa AIC algorithm + Baer-Kradolfer); Run2: HA (Baer-Kradolfer). Picking of S-waves. Implemented SEDSL impulsive onset: Identification limit is S (qA) emergent onset: lower limit + takanami (qB)

13. Risultati Picker Loc Mag Location 1d Location is performed at each MPX Run1 We use Hypoellipse correctly setup for the network size, data type, 1d velocity model Magnitude Calculation MPX determines maximum amplitude (using sensor and digitizer information) Ml per channel with Hutton & Boore (1987) ML = log10(amp) + 1.110 log10(hd) + 0.00189 hd + 3.591 M-estimator for robust averages + huber estimation of weights to avoid strong outliers  Ml + Weighted  Ml is also calculated with simple Mean, Mean without outliers, Median We essentially use the same formulas as INGV bulletin to be able to compare the catalogues Primary Output of Edward: TABOO seismic catalogue A line per event, with Time/space coordinates, quality control, Magnitude Ml A set of P-wave onset times and polarities, S-waves onsets, weights

14. Risultati Primary Output of Edward TABOO seismic catalogue A line per event, with Time/space coordinates, quality control, Magnitude Ml A set of P-wave onset times and polarities, S-waves onsets, weights Locations Map Vertical Sections e.g. 2010/04/15 From 1450 triggers to 510 high quality earthquakes out of 801 located events Picker Loc Mag

15. MPX compared to Manual Analysis January 2013

16. MPX compared to Manual Analysis January 2013 30 days picked in 10 days

17. MPX compared to Manual Analysis January 2013

18. MPX compared to Manual Analysis January 2013

19. MPX compared to Manual Analysis January 2013

20. MPX compared to Manual Analysis January 2013

21. MPX compared to Manual Analysis January 2013

22. Full Datasets Taboo Target Region Datasets Selection Comparison between P and S Phases MPX compared to Manual Analysis January 2013

23. Saturday, 2013/04/20 – Ml 3.6 h 07:58 Mail delle 12:15 Work on 2013.110.b2c.out OVER. Scissor at 2013-04-22_13:31 Last Pick at 2013-04-23_01:19: to be picked = 1669, picked = 1068, picking = 9 Located Events: all=862 and Selected=315 2013 04 20 07 57 41.82 43.4395 12.2935 0.52 059 051 0.29 0.03 0.05 1 092 3.49 3.58 3.55 3.56 0.32

24. Database Appendix what do I do with all this?

25. Database Appendix Edward will be Database Oriented Everything will be stored in tables of an interrogable DataBase Stations’ parameters Earthquakes’ basic parameters (location, quality, magnitude, phases onsets and polarities) Elaborations Non linear 1d and 3d locations parameters Linearized 1d and 3d locations parameters Relative 1D and 3D locations + waveforms crosscorrelation focal mechanisms parameters seismic velocities at nodes of a 3d grid b-values in space and time Vp/Vs ratio in space and time … and so on … Back to the Future: we plan to stop using files at all. Only “space-time” info are stored. This step will come after a confrontation with other groups and project working with the same data at INGV.