EURATOM METIS D4.1 Seismic source characterizations methodologies and applications

METIS Research and Innovation Action (RIA) This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 945121 Start date : 2020-09-01 Duration : 48 Months Seismic source characterizations methodologies and applications Authors : Dr. Thomas CHARTIER (GEM), Marco Pagani, GEM METIS - D4.1 - Issued on 2022-05-02 18:24:23 by GEM METIS - D4.1 - Issued on 2022-05-02 18:24:23 by GEM METIS - Contract Number: 945121 Project officer: Katerina PTACKOVA Document title Seismic source characterizations methodologies and applications Author(s) Dr. Thomas CHARTIER, Marco Pagani, GEM Number of pages 28 Document type Deliverable Work Package WP4 Document number D4.1 Issued by GEM Date of completion 2022-05-02 18:24:23 Dissemination level Public Summary Probabilistic seismic hazard assessment considers the occurrence of earthquakes as Poissonian, each even being independent of the others. As earthquakes do not occur as isolated events but for clusters of foreshocks, mainshock and aftershocks, it is necessary to identify and remove the earthquakes that are other than mainshocks from the catalog before calculating the annual rate of events used for the characterization of the seismic activity of the seismogenic sources of the hazard model. The aim of this study is to find a methodology to obtain a Poissonian declustered catalog while removing as few earthquakes as possible. To do so, we developed a methodology that allows to test the Poissonian nature of a declustered earthquake catalog and find the declustering algorithm that keeps the largest number of events while still performing well on the Poissonian test. Building a hazard model requires to perform statistic on the number of events per units of time and their spatial repartition, therefore, the Poissonian test was developed to reflect these needs. By comparing the inter-event spatio-temporal distances between the events of the tested catalog to the ones from simulated Poissonian catalogs sharing similar properties, we can score the statistical similarities between the tested and the simulated catalogs. We apply this methodology on the catalog for central Italy after being declustered with a range of different published algorithms and an additional algorithm that we propose in this study, and we perform a seismic hazard study for two locations in Central Italy to quantify the impact of the declustering algorithm on the seismic hazard level estimate. Approval Date By 2022-05-03 11:04:10 Dr. Marco PAGANI (GEM) 2022-05-03 11:13:52 Dr. Irmela ZENTNER (EDF) METIS - D4.1 - Issued on 2022-05-02 18:24:23 by GEM Research & Innovation Action NFRP-2019-2020 Methodology for the declustering of earthquake catalogs Deliverable D4.1 Version N°1 Authors: Thomas Chartier (GEM) This project has received funding from the Horizon 2020 programme under grant agreement n°945121. The content of this presentation reflects only the author’s view. The European Commission is not responsible for any use that may be made of the information it contains. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 2 Disclaimer The content of this deliverable reflects only the author’s view. The European Commission is not responsible for any use that may be made of the information it contains. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 3 Document Information Grant agreement 945121 Project title Methods And Tools Innovations For Seismic Risk Assessment Project acronym METIS Project coordinator Dr. Irmela Zentner, EDF Project duration 1st September 2020 – 31st August 2024 (48 months) Related work package WP 4 – Seismic hazard Related task(s) Task 4.1 – Source characterisation Lead organisation GEM Contributing partner(s) Due date 30/04/2022 Submission date 29/04/2022 Dissemination level X History Version Submitted by Reviewed by Date Comments N°1 Thomas Chartier Marco Pagani 29/04/2022 D4.1Methodology for the declustering of earthquake catalogs GA N°945121 4 Table of Contents 1. Introduction ................................................................................................ 8 2. Methodology: Testing the Poissonian nature of the earthquake catalog .......... 10 2.1. General philosophy of the methodology ................................................. 10 2.2. Summary of the testing procedure ......................................................... 10 2.3. Details of the testing procedure ............................................................. 10 2.3.1. Generation of the synthetic Poissonian catalog ................................. 10 2.3.2. Inter-earthquake spatiotemporal distances ...................................... 11 2.3.3. Comparison of the distribution of the IESD ...................................... 11 3. Example of application of the method in Central Italy ................................... 13 3.1. Presentation of the earthquake catalog .................................................. 13 3.2. Different declustering approaches .......................................................... 14 3.3. Testing the declustered catalogs ............................................................ 16 3.4. Results ................................................................................................ 17 3.5. Impact on the seismic hazard levels ....................................................... 18 4. Additional examples ................................................................................... 21 4.1. The Izmit Duzce Sequence .................................................................... 21 4.2. The Christchurch Sequence ................................................................... 22 5. Conclusion ................................................................................................. 23 6. 6 Supplementary materials ......................................................................... 24 7. References ................................................................................................ 24 8. Reviewers’ comments ................................................................................. 26 8.1. Comments made by David Marsan ......................................................... 26 8.1.1. General notes: ............................................................................... 26 8.1.2. More specific comments ................................................................. 26 8.2. Comments made by Matt Gerstenberger ................................................ 27 8.2.1. General notes ................................................................................ 27 8.2.2. More specific comments ................................................................. 27 8.3. Evolutions due to the comments of the advisors ..................................... 27 D4.1Methodology for the declustering of earthquake catalogs GA N°945121 5 List of figures Figure 1: Schematic view of the impact of clustering on the temporal distribution of earthquakes .................................................................................................. 8 Figure 2 : Comparison of the density functions of the IESD for one event of the tested catalog (in blue) and the density functions from the synthetic catalogs (in grey). In red, the median value of the IESD density functions in the synthetic catalog, in purple, the 16th and 84th percentiles of the distribution. .................................. 11 Figure 3 : Earthquake catalog for Central Italy used in this study. The color scale represents the magnitude. ........................................................................... 13 Figure 4: Time distribution of the events in the catalog for Central Italy. The size of the symbols is proportional to the magnitude. ............................................... 14 Figure 5: Application of the different declustering approaches to the earthquake catalog. Mainshocks are represented in red (t411 is a declustering method based on IESD). .................................................................................................... 15 Figure 6 : Scaling of the number of events considered as mainshocks as a function of the critical distance. ..................................................................................... 15 Figure 7 : Example of a declustered catalog and the corresponding synthetic catalog used for the test. ......................................................................................... 16 Figure 8 : Temporal distribution of events in the tested catalog and the synthetic catalogs. ..................................................................................................... 17 Figure 9 : On the left, result of the Poissonian test as a function of the number of earthquakes left in the declustered catalog. On the right, we show the dependence of the Poissonian test from the critical distance used for the declustering. ....... 17 Figure 10 : Results of the Poissonian test for each declustering approach as well as for the undeclustered catalog and two sets of synthetic catalogs with different number of events. ................................................................................................... 18 Figure 11 : Simplified hazard model for Central Italy based on the area source of ESHM20. The declustered catalog used is in red, the site 1 is represented by the red triangle and site 2 is represented by the blue triangle. .............................. 19 Figure 12 : Hazard curves for Site 1 for each declustering approach. ..................... 20 Figure 13 : Hazard curves for Site 1 for each declustering approach. ..................... 20 Figure 14 : Application of the declustering methods to the catalog. Mainshocks are in red. ............................................................................................................ 21 Figure 15 : Poissonian test result and number of events for each declustered catalog. .................................................................................................................. 21 Figure 16: Declustering applied to the Christchurch sequence ............................... 22 D4.1Methodology for the declustering of earthquake catalogs GA N°945121 6 Figure 17 : Poissonian test result and number of events for each declustered catalog. .................................................................................................................. 22 Figure 1: Example of a figure ............................................................................. 28 List of tables Table 1 Result of the Poissonian test for each declustering approach ..................... 18 D4.1Methodology for the declustering of earthquake catalogs GA N°945121 7 Abbreviations and Acronyms Acronym Description WP Work Package PSHA Probabilistic Seismic Hazard Assessment IESD Inter Event Spatio-temporal Distance Summary Probabilistic seismic hazard assessment considers the occurrence of earthquakes as Poissonian, each even being independent of the others. As earthquakes do not occur as isolated events but for clusters of foreshocks, mainshock and aftershocks, it is necessary to identify and remove the earthquakes that are other than mainshocks from the catalog before calculating the annual rate of events used for the characterization of the seismic activity of the seismogenic sources of the hazard model. The aim of this study is to find a methodology to obtain a Poissonian declustered catalog while removing as few earthquakes as possible. To do so, we developed a methodology that allows to test the Poissonian nature of a declustered earthquake catalog and find the declustering algorithm that keeps the largest number of events while still performing well on the Poissonian test. Building a hazard model requires to perform statistic on the number of events per units of time and their spatial repartition, therefore, the Poissonian test was developed to reflect these needs. By comparing the inter-event spatio-temporal distances between the events of the tested catalog to the ones from simulated Poissonian catalogs sharing similar properties, we can score the statistical similarities between the tested and the simulated catalogs. We apply this methodology on the catalog for central Italy after being declustered with a range of different published algorithms and an additional algorithm that we propose in this study, and we perform a seismic hazard study for two locations in Central Italy to quantify the impact of the declustering algorithm on the seismic hazard level estimate. Keywords PSHA, declustering, earthquake catalog, Poissonian D4.1Methodology for the declustering of earthquake catalogs GA N°945121 14 Figure 4: Time distribution of the events in the catalog for Central Italy. The size of the symbols is proportional to the magnitude. 3.2. Different declustering approaches We apply different published declustering algorithms on the catalog for Central Italy (Figure 5): ► Gardner and Knopoff 1972 with the following windows: o Gardner and Knopoff 1974 (GK GK) o Uhrhammer 1986 (GK Urh) o Gruental 1985 (GK Gruental) ► Reasenberg 1985 ► Zaliapin and Ben-Zion 2020 D4.1Methodology for the declustering of earthquake catalogs GA N°945121 15 Figure 5: Application of the different declustering approaches to the earthquake catalog. Mainshocks are represented in red (t411 is a declustering method based on IESD). In addition, we apply a simple declustering methodology based on the same spatiotemporal distance (IESD) used for testing the Poissonian nature of the catalog. The IESD is computed between each event and is this distance is below a given critical distance, the two events are considered to be part of the same cluster. We explore a range of critical distances and as the critical distance increases, the clusters are larger and the number of earthquakes remaining in the declustered catalog is decreasing (Figure 6). This declustering method with further on be referred as t411. In Figure 5, for t411, a critical distance value of 0.002 is used as an example, if the IESD between two earthquakes is less than 0.002, the earthquakes are considered to be part of the same cluster. Figure 6 : Scaling of the number of events considered as mainshocks as a function of the critical distance. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 16 3.3. Testing the declustered catalogs The declustered catalog obtained with the considered declustering approaches are tested using the methodology presented in the methodology section of this report. First, a set of synthetic catalogs are generated, where each catalog has the same number of events and the same time span as the tested catalog and is reproducing the same spatial pattern of the original catalog (Figure 7). Figure 7 : Example of a declustered catalog and the corresponding synthetic catalog used for the test. In Figure 8, we note that the tested catalog still contains some cluster events, creating steps in the cumulative distribution with time, while the Poissonian synthetic catalogs do not show major steps in their distribution. However, the synthetic catalogs show some random variation of the annual earthquake rate relative to the average rate. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 17 Figure 8 : Temporal distribution of events in the tested catalog and the synthetic catalogs. For the application of the t411 methodology, we explore a range of critical distances and test the Poissonian nature of each of the declustered catalogs obtained. In Figure 6, we showed that the number of events remaining in the catalog decreases as the critical distance increases. As the number of events decreases, only events that are more isolated from the other are left in the catalog, therefore, the Poissonian test result value increases (Figure 9). For the rest of this study, we will use the critical distance of 0.002, as it leaves a large number of earthquakes in the catalog (261) while having a good Poissonian test score of 0.39. Figure 9 : On the left, result of the Poissonian test as a function of the number of earthquakes left in the declustered catalog. On the right, we show the dependence of the Poissonian test from the critical distance used for the declustering. 3.4. Results Following the methodology presented in this study, the IESD distribution of the tested catalog is compared to the distribution obtained from the synthetic catalogs. The scores are presented in Table 1 and Figure 10. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 18 Table 1 Result of the Poissonian test for each declustering approach Declustering GK GK GK Gruental GK Uhr Reasenberg Zaliapin t411 Poissonian test score 0.38 0.39 0.37 0.29 0.38 0.39 Number of events 187 159 259 513 250 261 Figure 10 : Results of the Poissonian test for each declustering approach as well as for the undeclustered catalog and two sets of synthetic catalogs with different number of events. In addition to the different declustered catalogs, the Poissonian test is also applied to the full catalog and on a range of synthetic catalogs of different sizes. As expected, the full undeclustered catalog performs poorly with the overall score of 0.2. In comparison, the synthetic catalogs have the highest possible scores of around 0.5, with some small variability due to the randomness of the generation process. The Reasenberg approach leaves a too many earthquakes in the declustered catalog (459) and therefore it shows a low score for the Poissonian test (0.3). The Zaliapin and Ben Zion (2020), Gardner and Knopoff with the Gruental windows and the t411 approaches keep a similar number of earthquake and perform better on the Poissonian test (around 3.8, 3.9). Gardner and Knopoff using the other windows perform similarly on the test but keep fewer events in the catalog. 3.5. Impact on the seismic hazard levels D4.1Methodology for the declustering of earthquake catalogs GA N°945121 19 Figure 11 : Simplified hazard model for Central Italy based on the area source of ESHM20. The declustered catalog used is in red, the site 1 is represented by the red triangle and site 2 is represented by the blue triangle. In order to observe the impact on the seismic hazard, we developed a hazard model using each one of the declustered catalog. This model is using the geometry of area sources included in the European Seismic Hazard Model 2020 (ESHM20 see https://gitlab.seismo.ethz.ch/efehr/eshm20), the GMPE developed by Ameri et al. (2017) is used and the earthquake rates are calculated using a maximum likelihood approach on the declustered earthquake catalogs. The hazard is calculated at two theoretical sites. Site 1 (in red in Figure 11) located in a relatively stable area of Italy, and site 2 (in blue in Figure 11) is in a more active area, in the Apennines. The impact of the declustering approach in the hazard results is greater for the site located in an active region, closer to many earthquake clusters, than for the site in the stable region where the declustering removes only few earthquakes. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 20 Figure 12 : Hazard curves for Site 1 for each declustering approach. Figure 13 : Hazard curves for Site 1 for each declustering approach. The results of the Poissonian testing can be used to select and weight the different declustered to be explored in a logic tree. Given the values presented in Table 1, Zaliapin, GK using the Gruenthal windows and t411 approaches would be ones with the highest weights in a logic tree. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 21 4. Additional examples 4.1. The Izmit Duzce Sequence We test different declustering methodologies using the 1999 North Anatolian Fault (Turkey) earthquake sequence. The ISC-GEM catalog from 1998 to 2012 in the Izmit region is used (Figure 14). Figure 14 : Application of the declustering methods to the catalog. Mainshocks are in red. Figure 15 : Poissonian test result and number of events for each declustered catalog. Interestingly, for this earthquake sequence most of the declustering approaches (but Reasenberg) lead to a very good score, equivalent to the score of the synthetic catalogs. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 22 4.2. The Christchurch Sequence Different declustering approaches are applied to the 2010 Christchurch (NZ) earthquake sequence, using the ISC-GEM catalog (Figure 16). Figure 16: Declustering applied to the Christchurch sequence Figure 17 : Poissonian test result and number of events for each declustered catalog. For this earthquake sequence, the different declustering algorithms perform quite differently, with the Gardner and Knoppof windowing approach leaving few earthquakes in the declustered catalog than the other approaches but performing very well on the test for two of the windows used. D4.1Methodology for the declustering of earthquake catalogs GA N°945121 23 5. Conclusion In this study, we have developed an approach to test the Poissonian nature of a declustered earthquake catalog. This allows to find the declustering method that keeps the largest number of events in a catalog of events that can be considered to a high degree independent. This is the best compromise available when we preparing seismicity information used for the computation of the magnitude frequency distribution of the seismogenic sources of the seismic hazard model. This methodology relies on the statistical comparison of the declustered catalog with a set of synthetic Poissonian catalogs. Interearthquake spatiotemporal distances are computed and the distribution of the distance values in the tested catalog is compared to analogous distributions obtained from synthetic Poissonian catalogs. The closer the distributions are, the higher the is score measuring the level of a catalog of being Poissonian. The proposed methodology is tested using a catalog for central Italy and, compared against several published declustering methodology. The performance of each declustering method computed using the proposed methodology allows to propose a weighting for these methodologies in a logic tree. The impact of the declustering methodology on the variability of hazard results is evaluated for two sites in Italy. Our results show a larger impact for the one located in the most seismically active region.