WEBVTT Kind: captions Language: en 00:00:03.429 --> 00:00:09.700 Good morning, and welcome to the August 8th – August 9th version 00:00:09.700 --> 00:00:15.639 of the Earthquake Science seminar. And we’re pleased to have you. 00:00:15.639 --> 00:00:18.770 Next week, we have a very special speaker. 00:00:18.770 --> 00:00:25.910 Our own Jack Norbeck will be talking about hazard of induced earthquakes. 00:00:25.910 --> 00:00:33.180 And before I introduce the speaker, I’d just like to let you know that we’re 00:00:33.180 --> 00:00:41.350 planning to have lunch with Erol after his talk – probably about quarter to noon. 00:00:41.350 --> 00:00:45.160 So most of you know Erol, looking around in the room. 00:00:45.160 --> 00:00:49.620 But for those of you who may not know Erol, 00:00:49.620 --> 00:00:53.660 Erol has both his bachelor’s of science and his master’s 00:00:53.660 --> 00:00:58.280 in civil and structural engineering and engineering seismology from the 00:00:58.280 --> 00:01:04.290 Middle East Technological University and Bosphorus University. 00:01:04.920 --> 00:01:09.800 His Ph.D. in structural engineering is from UC-Davis, 00:01:09.810 --> 00:01:13.290 and he has a minor in geotechnical engineering. 00:01:13.290 --> 00:01:18.549 He did a postdoc at UC-Berkeley and was a seismologist 00:01:18.549 --> 00:01:22.770 at the Geological – sorry, the California Geological Survey 00:01:22.770 --> 00:01:29.220 for about four years before he joined the USGS in 2008. 00:01:29.220 --> 00:01:34.970 He’s been a member since that time in the National Strong Motion Project, 00:01:34.970 --> 00:01:38.310 and he’s worked on such things as instrumenting 00:01:38.310 --> 00:01:44.110 the Veteran Affairs hospitals for alerting that system 00:01:44.110 --> 00:01:47.170 for earthquake damage and earthquake early warning. 00:01:47.170 --> 00:01:52.090 And he’s played many roles in the American Society of Civil Engineers. 00:01:52.090 --> 00:01:53.460 So, Erol? 00:01:53.980 --> 00:01:56.460 - Thank you so much for the introduction. 00:01:57.000 --> 00:02:00.960 I hope you all hear me well? Okay. 00:02:01.760 --> 00:02:05.399 So my name is Erol Kalkan. I am a structural engineer 00:02:05.399 --> 00:02:08.979 with the Earthquake Science Center. Before, you know, starting, 00:02:08.979 --> 00:02:15.150 I would like to, you know, acknowledge my co-authors and my colleagues. 00:02:15.150 --> 00:02:18.609 Jeanne and Peter – they are sitting right on the back. 00:02:18.609 --> 00:02:22.379 They work for the Western Geographic Center. 00:02:22.379 --> 00:02:24.840 So they are our very talented – you know, 00:02:24.840 --> 00:02:28.420 the programmers that basically make this software happen. 00:02:29.200 --> 00:02:35.019 And Chris is the data manager for NSMP – National Strong 00:02:35.019 --> 00:02:39.909 Motion Project. And today, talk is going to be on software. 00:02:39.909 --> 00:02:43.629 I find myself a little bit difficult to actually present the software, 00:02:43.629 --> 00:02:49.019 but I’ll do my best. So this software is – we develop 00:02:49.019 --> 00:02:53.529 for processing strong ground motions, you know, automatically and 00:02:53.529 --> 00:02:58.230 some really neat features and somewhat interesting – 00:02:58.230 --> 00:03:04.139 the modules that I will try to, you know, show you guys. 00:03:04.139 --> 00:03:08.249 So, you know, the first thing I would really want to say – 00:03:08.249 --> 00:03:11.379 thank you for the members of the NSMP working group, 00:03:11.379 --> 00:03:15.620 especially David Oppenheimer who initiated this project 00:03:15.620 --> 00:03:18.209 and the name – you know, the PRISM. 00:03:18.209 --> 00:03:23.099 And Lind Gee has been very supportive for – you know, 00:03:23.100 --> 00:03:28.300 for this project basically all the way through here and, 00:03:28.300 --> 00:03:31.480 you know, the providing it – you know, enough funding. 00:03:31.489 --> 00:03:34.609 And like to thank Dave Boore for sharing his – 00:03:34.609 --> 00:03:38.880 you know, the Fortran codes for filtering. 00:03:39.560 --> 00:03:42.080 Joe Fletcher and Vladimir Graizer for many – 00:03:42.080 --> 00:03:44.300 you know, the discussions on signal processing. 00:03:45.020 --> 00:03:49.560 Jamie, Brad – you know, Bob Darragh and Tadahiro Kishida. 00:03:49.569 --> 00:03:52.639 They were really helpful for reviewing our – you know, 00:03:52.639 --> 00:03:55.309 the publications and providing, you know, the comments and – 00:03:55.309 --> 00:03:58.739 very constructive comments and suggestions. 00:03:58.739 --> 00:04:01.609 And finally, I would like to thank to NSMP technicians 00:04:01.609 --> 00:04:05.010 who install and maintain the strong motion network 00:04:05.010 --> 00:04:09.349 and those involved in collecting and vetting the strong motion data. 00:04:09.349 --> 00:04:13.860 So with that said, here’s the outline of my talk. 00:04:14.680 --> 00:04:20.040 Just want to first explain you why we started this project. 00:04:20.050 --> 00:04:25.380 And then give a snapshot of PRISM. And then I want to compare 00:04:25.380 --> 00:04:30.550 the PRISM with other softwares available in the market. 00:04:30.550 --> 00:04:34.250 And then the integration of the PRISM with the earthquake – 00:04:34.250 --> 00:04:39.060 you know, Advanced Quake Management System of the USGS. 00:04:39.060 --> 00:04:42.370 Like to introduce you to data format that we used. 00:04:42.370 --> 00:04:44.860 And then we’re going to look under the hood 00:04:44.860 --> 00:04:48.699 about actually PRISM processing engine works. 00:04:48.699 --> 00:04:54.189 We did very comprehensive testing that I want to mention about that. 00:04:54.189 --> 00:04:56.770 And then I’m going to introduce the graphic user interface 00:04:56.770 --> 00:04:59.260 of PRISM software. 00:04:59.260 --> 00:05:05.550 And then finally, I’ll show you our documentation, and in a couple of slides, 00:05:05.550 --> 00:05:08.699 show how we can run the PRISM processing engine. 00:05:08.700 --> 00:05:12.420 And finally, updates and my concluding remarks. 00:05:13.870 --> 00:05:18.640 So the primary motivation is that – you know, NSMP has been 00:05:18.650 --> 00:05:21.930 deploying more and more stations every year. 00:05:21.930 --> 00:05:28.080 And the – processing ground motion has been a manual process, 00:05:28.080 --> 00:05:31.120 you know, by our – you know, the expert people here. 00:05:31.120 --> 00:05:35.090 You know, as we get more records in-house, so automatic 00:05:35.090 --> 00:05:37.370 strong motion record processing becomes a – you know, 00:05:37.370 --> 00:05:42.840 the essential part of our – you know, the daily routine. 00:05:43.500 --> 00:05:47.380 With that saying, so we like to have the expert review 00:05:47.389 --> 00:05:51.449 only limited to selected very significant events or the records that, 00:05:51.449 --> 00:05:53.370 you know, automatically process but have some – 00:05:53.370 --> 00:05:56.800 you know, the issues that needs to be manually addressed. 00:05:57.820 --> 00:06:05.720 So, so far, you know, since 1992, USGS had been using venerable processing 00:06:05.720 --> 00:06:09.610 software, BAP, which is the Basic Strong-Motion Accelerogram Processing. 00:06:09.610 --> 00:06:13.280 It’s, again, written in-house at the USGS. 00:06:13.280 --> 00:06:17.060 But this – you know, the Fortran codes become outmoded. 00:06:17.060 --> 00:06:20.900 So it doesn’t actually allow for automated processing. 00:06:22.289 --> 00:06:28.280 So in a nutshell, so PRISM is – basically we got the raw acceleration 00:06:28.280 --> 00:06:33.080 data into the system, and we have the – you know, the product distribution. 00:06:33.080 --> 00:06:37.060 We primarily provide acceleration, velocity, displacement – 00:06:37.069 --> 00:06:40.289 those are all corrected waveforms, and we provide, you know, 00:06:40.289 --> 00:06:42.900 pseudo acceleration, response spectrum, velocity, and displacement. 00:06:42.900 --> 00:06:45.650 Response spectrum, we provide Fourier spectrum 00:06:45.650 --> 00:06:49.260 and a bunch of earthquake intensity measures. 00:06:51.200 --> 00:06:56.009 So it’s fully – it’s a modern software. It’s completely Java-powered. 00:06:56.009 --> 00:06:59.250 Its design is based on API, which is an application 00:06:59.250 --> 00:07:03.010 programming interface, which allows actually PRISM processing 00:07:03.010 --> 00:07:07.300 engine works, you know, seamlessly with the graphic user interface. 00:07:07.960 --> 00:07:11.719 It has two modules. One of them is the batch processing 00:07:11.719 --> 00:07:15.539 engine where we do, you know, the automatic processing. 00:07:15.539 --> 00:07:18.949 We also have a review tool, which is the graphic user interface, 00:07:18.949 --> 00:07:24.249 where the user can, you know, grab it – a bunch of records and, you know, do – 00:07:24.249 --> 00:07:28.080 manipulate them – you know, do processing manually. 00:07:29.860 --> 00:07:32.900 So PRISM both is processing engine and the review tool 00:07:32.900 --> 00:07:37.499 is very easy to install and run as a standalone system. 00:07:37.499 --> 00:07:40.620 So because it’s Java-based, it works on, you know, 00:07:40.620 --> 00:07:45.760 the Mac operating system, Windows, or, you know, the Linux computers. 00:07:46.830 --> 00:07:49.300 So here I’d like to compare PRISM with other, you know, 00:07:49.319 --> 00:07:52.960 available software in the market. 00:07:52.960 --> 00:07:56.120 PRISM definitely joins, you know, a suite of tools also used with 00:07:56.120 --> 00:08:01.110 other processing codes. But we certainly have some really unique 00:08:01.110 --> 00:08:06.699 new features that doesn’t present in the – you know, in others. 00:08:06.700 --> 00:08:11.760 So in terms of – you know, so our software is completely free, you know, 00:08:11.760 --> 00:08:17.700 and open-source, you know, similar to the older – the USGS software BAP. 00:08:17.700 --> 00:08:19.460 There are two commercial softwares – 00:08:19.469 --> 00:08:23.490 SeismoSignal and Seismogram 2K from Europe. 00:08:23.490 --> 00:08:27.400 So SeismoSignal is – well, it’s not free. It’s not open-source. 00:08:27.400 --> 00:08:30.300 It doesn’t allow for batch processing. 00:08:30.300 --> 00:08:35.840 Well, Seismogram 2K is free, but the source is not open. 00:08:35.840 --> 00:08:38.659 So in terms of graphic user interface, 00:08:38.659 --> 00:08:41.969 except for BAP, you know, all other software has that. 00:08:41.969 --> 00:08:45.240 So we use phase picking – picking through, you know, 00:08:45.240 --> 00:08:49.340 the P wave arrival time, we use that for ground motion processing. 00:08:49.340 --> 00:08:53.620 We thought it was very useful to have a – 00:08:53.620 --> 00:08:57.520 you know, the better or more reliable processing. 00:08:57.529 --> 00:09:01.760 So only PRISM and, you know, the Seismogram 2K does that. 00:09:01.760 --> 00:09:07.180 So in terms of the platforms, so we are platform-independent. 00:09:07.180 --> 00:09:11.180 You know, the BAP works on, you know, the Windows operating system. 00:09:11.180 --> 00:09:15.570 SeismoSignal, again, works only on Windows operating system. 00:09:15.570 --> 00:09:19.690 Seismogram 2K is Java-based, so it can work on three operating systems. 00:09:19.690 --> 00:09:23.920 So in terms of updates and support, you know, there’s no updates for BAP, 00:09:23.920 --> 00:09:26.899 but the other – you know, the three softwares, they have the 00:09:26.899 --> 00:09:32.480 updates and support. So that’s just basically how we compare with others. 00:09:34.400 --> 00:09:37.490 So the whole idea to PRISM is making all the strong motion data 00:09:37.490 --> 00:09:41.149 coming to USGS process automatically. 00:09:41.149 --> 00:09:46.940 So the USGS is using Advanced Quake – the management system. 00:09:46.940 --> 00:09:49.980 And so this system detect the earthquake, you know, 00:09:49.980 --> 00:09:53.589 collect all the waveforms. And then, you know, well, 00:09:53.589 --> 00:09:58.450 thanks to Tim MacDonald, so the AQMS is fully integrated with PRISM. 00:09:58.450 --> 00:10:02.110 So when the data comes, so the PRISM – you know, 00:10:02.110 --> 00:10:04.839 kicks in, process all the – all the records automatically 00:10:04.839 --> 00:10:07.810 without any, you know, interference. 00:10:07.810 --> 00:10:11.390 So that was the whole idea why – you know, why we started this project. 00:10:11.390 --> 00:10:15.300 And I’m actually really glad to see that it’s – you know, the final is, 00:10:15.300 --> 00:10:19.580 you know, up and running and fully integrated with the AQMS. 00:10:20.930 --> 00:10:25.040 So PRISM uses standard COSMOS data format. 00:10:25.050 --> 00:10:28.610 So these are the Volume 1, or what we call it, V1. 00:10:28.610 --> 00:10:31.540 These are uncorrected acceleration. 00:10:31.540 --> 00:10:34.430 So these are all our – you know, the products. 00:10:34.430 --> 00:10:37.089 And we have the Volume 2, where we have the 00:10:37.089 --> 00:10:41.250 corrected acceleration, velocity, and displacement time series. 00:10:41.250 --> 00:10:45.240 And then the Volume 3 is the response spectrum, Fourier amplitude spectrum, 00:10:45.240 --> 00:10:50.710 and engineering intensity parameters, which I’m going to explain them shortly. 00:10:50.710 --> 00:10:56.440 So if you look at the – you know, how actually the processing occurs, 00:10:56.440 --> 00:11:00.100 so the PRISM processing engine – first I would like to give all the – 00:11:00.100 --> 00:11:01.720 you know, the key features. 00:11:01.720 --> 00:11:07.120 And I want to show all the – you know, the – how the processing flow works. 00:11:07.120 --> 00:11:09.000 It’s fully automated, as I mentioned. 00:11:09.000 --> 00:11:13.780 It’s platform-independent, modular, extensible, and open-source. 00:11:13.780 --> 00:11:18.890 It doesn’t depend on any closed-source or any licensed software. 00:11:18.890 --> 00:11:22.740 So processing engine implements the – you know, the API to allow, 00:11:22.740 --> 00:11:26.050 you know, incorporation of alternative implementation steps. 00:11:26.050 --> 00:11:32.200 That’s – you know, that’s very typical with all the modern software today. 00:11:32.200 --> 00:11:37.600 So we have a customizable configuration file that the user can, you know, go and, 00:11:37.600 --> 00:11:40.759 you know, change some parameters for the automated processing. 00:11:40.759 --> 00:11:46.949 So we collapse everything into this – you know, the configuration file. 00:11:46.949 --> 00:11:48.699 So it’s really flexible. 00:11:48.699 --> 00:11:52.380 And it does P wave arrival time picking. 00:11:52.380 --> 00:11:56.350 So currently we have two pickers implemented, and the user can, 00:11:56.350 --> 00:12:01.040 you know, decide which one they’re going to use in this configuration file. 00:12:01.040 --> 00:12:05.569 So we have the PphasePicker and the AIC picker. 00:12:05.569 --> 00:12:09.910 So PRISM also do a maximum amplitude picking. 00:12:09.910 --> 00:12:13.740 We do, you know, integration and differentiation, obviously. 00:12:13.740 --> 00:12:16.860 Right now everything is in the time domain. 00:12:18.410 --> 00:12:23.140 In terms of digital signal processing, we use frequency domain 00:12:23.150 --> 00:12:29.740 zero-padding for resampling. We use acausal bandpass filtering. 00:12:31.260 --> 00:12:33.120 And then filtering is, you know, performed on 00:12:33.120 --> 00:12:36.330 acceleration – you know, the time domain. 00:12:36.330 --> 00:12:40.139 So our products, you know, includes compatible acceleration, 00:12:40.139 --> 00:12:43.540 velocity, and displacement time series. So that means, you know, 00:12:43.540 --> 00:12:46.420 one can get the acceleration records and, you know, integrate to get the 00:12:46.420 --> 00:12:49.980 velocity and integrate, you know, the displacement and, you know, 00:12:49.980 --> 00:12:54.860 vice-versa, go back by differentiating, and then get the same acceleration. 00:12:54.860 --> 00:12:59.569 So I’m going to explain that, you know, in a little, but not every – you know, 00:12:59.569 --> 00:13:06.440 the data produces, their products are not – they’re compatible. 00:13:06.440 --> 00:13:10.160 So compatible data products basically includes initial values required, 00:13:10.160 --> 00:13:16.600 which we save them in V2 – you know, the Volume 2 file in header information. 00:13:16.600 --> 00:13:18.560 So we can produce – we can go back 00:13:18.560 --> 00:13:21.620 between acceleration, velocity, and displacement. 00:13:22.810 --> 00:13:26.220 So PRISM creates really nice, you know, the list of log files. 00:13:26.220 --> 00:13:30.980 So this way, like, every step of the processing can be traced. 00:13:31.800 --> 00:13:34.580 So, as I mentioned, for input, we are currently using, you know, 00:13:34.580 --> 00:13:37.060 the COSMOS V-zero input format. 00:13:37.070 --> 00:13:42.579 So V-zero is – the raw data comes to PRISM is basically in counts – 00:13:42.579 --> 00:13:46.610 not even – not even in, you know, the physical units. 00:13:46.610 --> 00:13:49.279 So as I mentioned, the products are in COSMOS data format, 00:13:49.279 --> 00:13:52.709 Volume 1, Volume 2, and Volume 3. 00:13:52.709 --> 00:13:57.240 So the Volume 1 processing is the simplest, you know, start. 00:13:57.240 --> 00:13:59.709 We got the raw data coming to PRISM. 00:13:59.709 --> 00:14:02.470 So we convert the counts into physical units, 00:14:02.470 --> 00:14:07.899 and then we remove the mean from the entire time series. 00:14:07.899 --> 00:14:11.300 We don’t do instrument response correction, you know, 00:14:11.300 --> 00:14:15.009 for there isn’t that – you know, the cutoff frequency of the 00:14:15.009 --> 00:14:18.110 low-pass filter that we use is lower than that of the natural 00:14:18.110 --> 00:14:22.310 frequency of the accelerometer. With that saying, you know, 00:14:22.310 --> 00:14:25.750 PRISM doesn’t process any records if there’s something 00:14:25.750 --> 00:14:29.340 where it is less than 50 sample per second. 00:14:31.149 --> 00:14:33.800 So Volume 2 processing is actually where all the – 00:14:33.810 --> 00:14:39.129 you know, the magic happens. So this is much more complicated. 00:14:39.129 --> 00:14:41.970 So the first, once we get the data, we check the sampling rate. 00:14:41.970 --> 00:14:44.750 If anything – you know, less than 200 sample per second, 00:14:44.750 --> 00:14:48.810 we resample it to 200 sample per second. 00:14:48.810 --> 00:14:52.520 And then we do event onset detection to find where the – 00:14:52.520 --> 00:14:58.620 you know, the P wave arrival time is. And then we remove the pre-event mean 00:14:58.620 --> 00:15:02.560 from the – you know, from the rest of the record. 00:15:02.560 --> 00:15:06.840 And then we apply, you know, the baseline correction, 00:15:06.840 --> 00:15:11.820 first- or second-order by fitting a first- or second-order polynomial to velocity. 00:15:11.820 --> 00:15:16.569 And then we take the derivative of the baseline correction from acceleration. 00:15:16.569 --> 00:15:21.750 And then we go back, and then we do some QC to make sure that, you know, 00:15:21.750 --> 00:15:26.279 the velocity – it oscillates in the beginning. And at the end of the 00:15:26.280 --> 00:15:30.900 velocity oscillate around zero. So those are our physical constraints. 00:15:32.149 --> 00:15:38.780 So we do bandpass – once the QC pass, we do bandpass filter, the acceleration. 00:15:38.790 --> 00:15:41.910 And then we do – you know, if everything goes well, 00:15:41.910 --> 00:15:45.000 we integrate to get the velocity and displacement. 00:15:45.000 --> 00:15:49.019 You know, sometimes these initial baseline correlation is not sufficient. 00:15:49.019 --> 00:15:53.690 But in that case, our QC fails, and then, you know, the system 00:15:53.690 --> 00:15:56.750 starts a more complicated baseline correction algorithm 00:15:56.750 --> 00:16:00.240 which we call it adaptive baseline. 00:16:02.730 --> 00:16:06.080 So I’m going to go back to adaptive baseline in the next slide, but this is – 00:16:06.089 --> 00:16:10.649 let’s assume that, you know, we finished the Volume 2 processing successfully. 00:16:10.649 --> 00:16:16.490 And then that product goes back to the Volume 2 – Volume 3 processing. 00:16:16.490 --> 00:16:19.779 So here we have the corrected acceleration, 00:16:19.779 --> 00:16:22.350 and then we calculate the pseudo acceleration, velocity, 00:16:22.350 --> 00:16:28.410 and displacement response spectra for 2, 5, 10, and 20% damping levels. 00:16:28.410 --> 00:16:31.589 And then we compute the Fourier spectra. 00:16:31.589 --> 00:16:35.040 And then we compute a series of 00:16:35.040 --> 00:16:37.160 earthquake engineering intensity measures. 00:16:37.160 --> 00:16:42.350 These are the default intensity measures in COSMOS Volume 3 data format. 00:16:42.350 --> 00:16:46.220 So we calculate the Arias intensity. We calculate the, you know, 00:16:46.220 --> 00:16:48.520 strong motion duration, the bracket duration, 00:16:48.520 --> 00:16:52.440 duration interval, response spectrum intensity, 00:16:52.440 --> 00:16:56.120 root mean square acceleration, and also cumulative absolute velocity. 00:16:56.120 --> 00:17:03.980 So all this – the outcomes goes back to the Volume 3 sent as ASCII file. 00:17:05.600 --> 00:17:08.700 So want to come back to adaptive baseline correction. 00:17:08.700 --> 00:17:11.680 You know, sometimes records may have, you know, step-like offsets – 00:17:11.680 --> 00:17:17.310 you know, the spikes in the baseline, or they may be contaminated by, 00:17:17.310 --> 00:17:21.700 you know, some instrumental noise or some rotational, gravitational, you know, 00:17:21.700 --> 00:17:25.950 effects or sometimes, you know, the static – fault displacement. 00:17:25.950 --> 00:17:29.640 So this case, you know, the PRISM regular processing 00:17:29.640 --> 00:17:34.270 may not be sufficient to correct for such distortions. 00:17:34.270 --> 00:17:37.120 So here is – I would like to give one example. 00:17:37.120 --> 00:17:41.680 So we have the acceleration record, you know, after pre-event mean removal. 00:17:41.680 --> 00:17:46.260 So we do – we apply our initial baseline correction. 00:17:46.260 --> 00:17:48.130 So as you see – I’m sorry. 00:17:48.130 --> 00:17:52.660 We just integrated the pre-event mean removed acceleration to velocity, 00:17:52.660 --> 00:17:59.750 and you can see, you know, the trend at the – at the tail of the record. 00:17:59.750 --> 00:18:02.030 So when we apply, you know, initial baseline correction, 00:18:02.030 --> 00:18:06.340 you know, we fix that distortion. But in this case, the beginning of the 00:18:06.340 --> 00:18:08.260 record has, you know, some distortions. 00:18:08.260 --> 00:18:14.320 So basically, the record failed our QC because the velocity is – the mean 00:18:14.320 --> 00:18:19.270 average velocity in the beginning of the record is not close to zero. 00:18:19.270 --> 00:18:22.210 So once the record – you know, fails in QC, 00:18:22.210 --> 00:18:25.450 it goes back to our, you know, adaptive routine. 00:18:25.450 --> 00:18:31.860 So the adaptive baseline correction is based on the idea that, you know, 00:18:31.860 --> 00:18:36.070 we can divide the waveform into multiple segments 00:18:36.070 --> 00:18:39.470 and do some fitting of those multiple segments. 00:18:39.470 --> 00:18:44.470 So the idea has come, you know, way back in 1979 from Vladimir Graizer 00:18:44.470 --> 00:18:48.530 published a – you know, the paper, and then – and then the folks in 00:18:48.530 --> 00:18:52.600 Caltech and then, you know, Boore has published another paper in 2001. 00:18:52.600 --> 00:18:56.060 The whole idea is, you know, get the record and divide into 00:18:56.060 --> 00:19:00.120 multiple segments and, you know, fit, you know, different polynomials 00:19:00.120 --> 00:19:04.310 to have – to have a better baseline correction instead of, you know, like, 00:19:04.310 --> 00:19:10.600 a first-order or second-order simple polynomial for the entire record. 00:19:10.600 --> 00:19:14.090 So we developed adaptive baseline correction, which we call it ABC, 00:19:14.090 --> 00:19:17.640 you know, following the – Iwan’s method. 00:19:17.640 --> 00:19:23.650 So in ABC processing, we fit a nth-order polynomial to the 00:19:23.650 --> 00:19:29.770 beginning of the record and also the end of the record, where we know that, 00:19:29.770 --> 00:19:32.810 you know, the beginning shows, you know, physically at zero, 00:19:32.810 --> 00:19:34.150 and then the record should go back to – 00:19:34.150 --> 00:19:38.270 you know, oscillate around zero. Those are our physical constraints. 00:19:38.270 --> 00:19:43.770 And then these two polynomials, we join them with a cubic-spline 00:19:43.770 --> 00:19:48.440 in the beginning part – in the middle part. 00:19:48.440 --> 00:19:51.430 So here’s the flow chart of our adaptive baseline correction. 00:19:51.430 --> 00:19:56.320 It’s a little bit complicated. It’s an integrated procedure. 00:19:56.320 --> 00:20:01.630 The challenge here is that, to find out, you know, what part of the waveform is 00:20:01.630 --> 00:20:06.420 going to be at the beginning, the end and, you know, the middle sections. 00:20:06.420 --> 00:20:10.750 So the beginning part is – you know, it start from the zero, and then we 00:20:10.750 --> 00:20:16.990 choose the pre-event – end of the pre-event duration, or the onset time 00:20:16.990 --> 00:20:20.140 of P wave arrival, as the first segment. 00:20:20.140 --> 00:20:24.120 So that’s where we are actually fitting the first polynomial. 00:20:24.120 --> 00:20:30.540 And then the challenge is deciding on the end segment of the waveform. 00:20:30.540 --> 00:20:32.720 So here is – we have the, you know, t-sub-2. 00:20:32.720 --> 00:20:35.630 That’s the beginning of this end segment. 00:20:35.630 --> 00:20:42.430 And in ABC processing, so we basically are iteratively changing this t-sub-2 00:20:42.430 --> 00:20:48.430 to have a – you know, to come up with the optimal baseline correction fit. 00:20:48.430 --> 00:20:52.720 And in between those two polynomials, we have, you know, the cubic-spline. 00:20:52.720 --> 00:20:58.650 So not only the ABC is iteratively changing the beginning of the 00:20:58.650 --> 00:21:02.510 end segment that we’re going to fit a polynomial, but also changing 00:21:02.510 --> 00:21:06.610 the polynomial orders to come up with, you know, the best solution. 00:21:06.610 --> 00:21:10.390 So this is an iterative procedure all the way – you know, it looks like it’s 00:21:10.390 --> 00:21:15.870 going to take time, but it’s like a – you know, it works extremely fast. 00:21:15.870 --> 00:21:19.160 So at the end of this adaptive baseline correction, we come up with the 00:21:19.160 --> 00:21:24.020 best solution for the baseline and we – in the velocity domain, and then we take 00:21:24.020 --> 00:21:29.350 the derivative, go back to acceleration, and remove it from the acceleration. 00:21:29.350 --> 00:21:33.540 So once we do that, we do all these – you know the quality check again. 00:21:33.540 --> 00:21:37.780 And if everything pass, you know, so the record – 00:21:37.780 --> 00:21:40.090 the PRISM goes back to the next record. 00:21:40.090 --> 00:21:43.320 So if something fails, it went with the adaptive baseline correction. 00:21:43.320 --> 00:21:47.270 So PRISM flag that record as a trouble and put it in the trouble folder. 00:21:47.270 --> 00:21:51.560 So that’s where the manual processing is needed. 00:21:53.090 --> 00:21:57.100 So here is – I would like to show you one – you know, the weak motion 00:21:57.110 --> 00:22:02.440 that we process using PRISM from 2014 South Napa earthquake. 00:22:02.440 --> 00:22:05.890 The red lines are the PRISM products, and the black is the uncorrected 00:22:05.890 --> 00:22:09.750 acceleration at the top panel and the uncorrected – 00:22:09.750 --> 00:22:16.280 basically the velocity in the lower – I just, you know, integrate the 00:22:16.280 --> 00:22:20.680 uncorrected acceleration just to show the trend in the velocity. 00:22:20.680 --> 00:22:22.970 So once we, you know, apply the PRISM, so this record 00:22:22.970 --> 00:22:27.850 didn’t go to adaptive baseline correction, so basically, our initial baseline 00:22:27.850 --> 00:22:31.230 correction was sufficient, so the – you know, once we done 00:22:31.230 --> 00:22:38.660 with this record, so we have a – you know, the nice signal and – 00:22:38.660 --> 00:22:42.200 yeah, so that’s what it is for this slide. 00:22:44.000 --> 00:22:47.500 So we have many, many examples in our – you know, the open file reports 00:22:47.500 --> 00:22:49.880 show how basically the PRISM generates results. 00:22:49.880 --> 00:22:51.940 That was just one example. 00:22:51.940 --> 00:22:55.660 Here I’d like to show you how fast actually the PRISM works. 00:22:55.660 --> 00:22:58.410 So we did a runtime performance test using, 00:22:58.410 --> 00:23:00.770 you know, the [inaudible] desktop computer. 00:23:00.770 --> 00:23:06.500 So it’s a 2.6 gigahertz CPU with 16-gigabyte RAM. 00:23:06.500 --> 00:23:10.430 So we processed the South Napa records, which has about – 00:23:10.430 --> 00:23:14.250 more than 700 channels. And we also picked two – 00:23:14.250 --> 00:23:19.220 you know, the small-magnitude events around 100 channels each. 00:23:19.220 --> 00:23:24.150 So if you look at those small-magnitude events, so the passing grade is 100%. 00:23:24.150 --> 00:23:29.130 So PRISM processed them – all the records without any problem. 00:23:29.130 --> 00:23:32.440 When you look at the processing time, you know, for each channel, 00:23:32.440 --> 00:23:36.600 it’s less than half a second. So the data comes extremely fast. 00:23:36.600 --> 00:23:38.780 You know, the process – we have all these data products 00:23:38.780 --> 00:23:42.130 right in the – you know, the output folders. 00:23:42.130 --> 00:23:45.420 So in the South Napa case, you know, the processing time is, 00:23:45.420 --> 00:23:48.740 again, less than a second. It’s basically 0.8 seconds 00:23:48.740 --> 00:23:51.960 because some of the records required, you know, the ABC processing, 00:23:51.960 --> 00:23:54.660 which may require a little bit more time. 00:23:54.660 --> 00:23:59.660 But overall, you know, for 700 records, 00:23:59.660 --> 00:24:04.280 we are practically done, you know, less than 10 minutes, everything is ready. 00:24:05.780 --> 00:24:09.900 So we compare PRISM first with the BAP as well as 00:24:09.900 --> 00:24:14.280 the California Strong Motion Instrumentation Processing. 00:24:14.280 --> 00:24:19.350 So it’s like – because every – you know, the BAP has its own – 00:24:19.350 --> 00:24:21.350 you know, the flavors. The same thing with the 00:24:21.350 --> 00:24:25.620 CSMIP processing. So it’s not each software is exactly the same. 00:24:25.620 --> 00:24:30.620 So it’s kind of like comparing, a little bit, apples to, you know, the oranges. 00:24:30.620 --> 00:24:35.220 You know, nonetheless, we did a series of benchmark tests. 00:24:35.220 --> 00:24:38.970 So we have a dedicated open file report showing all the – 00:24:38.970 --> 00:24:41.560 you know, the comparison. 00:24:42.680 --> 00:24:46.900 And we tried to match the processing parameters between, you know, 00:24:46.900 --> 00:24:52.060 the BAP, the CSMIP, and the PRISM as closely as possible. 00:24:52.060 --> 00:24:55.430 Although, you know, we were expecting to see some difference because there are 00:24:55.430 --> 00:24:59.460 some, you know, fundamental differences in the techniques that we used. 00:25:01.090 --> 00:25:06.020 So one of the main differences here, you know, among many – 00:25:06.020 --> 00:25:10.330 so in BAP processing, low-frequency filter is applied in time domain. 00:25:10.330 --> 00:25:15.530 And at high frequencies, the cosine taper is used in frequency domain. 00:25:15.530 --> 00:25:20.560 So in CSMIP processing, so initial long-period filter is applied to 00:25:20.560 --> 00:25:25.930 the instrument-corrected acceleration. And then, you know, the velocity 00:25:25.930 --> 00:25:29.210 and displacement are subsequently computed by integrating the 00:25:29.210 --> 00:25:34.000 acceleration and then filtered again by using the same long-period filter. 00:25:35.290 --> 00:25:37.600 So in contrast, you know, the PRISM applies filtering 00:25:37.610 --> 00:25:41.660 to corrected acceleration only. And the velocity and displacement 00:25:41.660 --> 00:25:45.260 are obtained by, you know, integrating the filtered acceleration. 00:25:45.260 --> 00:25:49.740 So this basically give us a – this process leads to a 00:25:49.740 --> 00:25:53.840 compatible acceleration, velocity, and displacement time series. 00:25:55.140 --> 00:26:00.200 So, you know, we did all this testing – the first – you know, we got a – 00:26:00.200 --> 00:26:02.200 around, you know, the [inaudible] records. 00:26:02.200 --> 00:26:04.620 And we look at very detailed – like, how, you know, 00:26:04.630 --> 00:26:09.660 the PRISM processing compare with the BAP and CSMIP processing. 00:26:09.660 --> 00:26:13.400 So for that purpose, we first compare the peak values. 00:26:13.400 --> 00:26:15.920 And then we computed the, you know, coherence – you know, 00:26:15.920 --> 00:26:19.510 the cross-spectrum phase, cross-correlation of the time series. 00:26:19.510 --> 00:26:22.850 We also compared the Fourier amplitude spectrum between, 00:26:22.850 --> 00:26:26.870 you know, the different ground motion processing routines. 00:26:26.870 --> 00:26:29.990 And then finally we computed and compared the spectrograms 00:26:29.990 --> 00:26:35.340 of power spectra of acceleration, velocity, and displacement time series. 00:26:35.340 --> 00:26:38.460 So I’m not showing all the results here, but everything is, 00:26:38.470 --> 00:26:42.600 you know, in our open file report. But here is just, you know, one example. 00:26:42.600 --> 00:26:45.040 Here is the PRISM versus BAP processing. 00:26:45.049 --> 00:26:51.500 So we picked this 2005 magnitude 5 earthquake from Samoa. 00:26:51.500 --> 00:26:54.827 So this record is 100 sample per second. 00:26:54.827 --> 00:26:57.990 And the PGA occurs at very high frequencies. 00:26:57.990 --> 00:27:01.970 So this is one of our – you know, the test record. 00:27:01.970 --> 00:27:05.920 So when we do the PRISM versus BAP comparison – so the first 00:27:05.920 --> 00:27:10.210 panel show the acceleration time series. These are all processed. 00:27:10.210 --> 00:27:13.870 And the following panels show the difference between the acceleration. 00:27:13.870 --> 00:27:17.520 And then, you know, the same thing for the velocity and displacement. 00:27:17.520 --> 00:27:23.570 So overall, you know, except in the final few seconds of the process record, 00:27:23.570 --> 00:27:28.490 you know, the BAP and PRISM is basically producing the same waveforms. 00:27:28.490 --> 00:27:30.850 And the difference between the peak values are – 00:27:30.850 --> 00:27:34.630 well, we haven’t seen any difference in the PGA and PGV, and peak 00:27:34.630 --> 00:27:39.760 ground displacement difference is less than 0.1 second – 1%. 00:27:40.610 --> 00:27:43.720 So we did, you know, similar comparisons with many other records, 00:27:43.730 --> 00:27:47.970 but here’s one example from South Napa earthquake – Napa College. 00:27:47.970 --> 00:27:52.400 So we compare our processing with the CSMIP processing. 00:27:52.400 --> 00:27:55.450 You know, peak values, again, you know, very close to each other. 00:27:55.450 --> 00:27:58.970 So the difference is around, you know, the 1%. 00:27:58.970 --> 00:28:03.220 But we – you know, we see some, you know, small variations in the acceleration 00:28:03.220 --> 00:28:07.190 time series and, you know, the much less on velocity and displacement. 00:28:07.190 --> 00:28:10.070 So these are – we thought these are all, you know, the possible – 00:28:10.070 --> 00:28:12.760 because we are using, you know, different techniques for processing than, 00:28:12.760 --> 00:28:15.960 you know, what – you know, CSMIP does. 00:28:16.960 --> 00:28:20.000 So after looking, you know, at those records individually, 00:28:20.000 --> 00:28:24.140 much details, you know, with the feedback from 00:28:24.140 --> 00:28:28.800 the NSMP working group, we conducted a statistical evaluation. 00:28:28.800 --> 00:28:33.540 So this time we collected more than 1,800 V1 records 00:28:33.540 --> 00:28:38.040 from the Center for Engineering Strong Motion Data website. 00:28:38.040 --> 00:28:42.800 So that’s the website operated by CGS and USGS. 00:28:42.800 --> 00:28:49.750 So we pick eight earthquakes that occur in California between 2005 and 2006. 00:28:49.750 --> 00:28:53.100 So this is the distribution of our – you know, the events with the 00:28:53.100 --> 00:28:56.900 moment magnitude and epicenter distance. 00:28:56.900 --> 00:28:59.910 So we have a bunch of records, you know, in the near field as well as, 00:28:59.910 --> 00:29:02.240 you know, the far field where we have the lower, 00:29:02.240 --> 00:29:05.400 you know, signal-to-noise ratio. 00:29:05.400 --> 00:29:08.880 And then we compare – we process them with the PRISM 00:29:08.880 --> 00:29:15.760 and compare how our products compare with the CSMIP processing. 00:29:15.760 --> 00:29:19.610 So we computed misfit between two times series by comparing, again, 00:29:19.610 --> 00:29:23.920 the peak values, as well as normalized Euclidian distance 00:29:23.920 --> 00:29:27.520 between acceleration, velocity, and displacement time series. 00:29:27.520 --> 00:29:32.690 We also did moving window root mean square level of acceleration, 00:29:32.690 --> 00:29:35.310 velocity, displacement time series. It’s actually – it’s really difficult 00:29:35.310 --> 00:29:39.060 to compare two time series, when you look at – I mean, it’s easy to compare the 00:29:39.060 --> 00:29:42.260 peak values, but how about the whole – you know, the time series. 00:29:42.260 --> 00:29:48.410 That’s why we used these L2 norm as well as the moving window RMS. 00:29:48.410 --> 00:29:51.380 So among those eight records, here is, you know, 00:29:51.380 --> 00:29:56.870 one example from, again, the South Napa earthquake. 00:29:56.870 --> 00:29:59.750 So what we are looking, on the Y axis, is the peak values – 00:29:59.750 --> 00:30:03.580 peak ground acceleration, velocity, and displacement. 00:30:03.580 --> 00:30:06.040 And on the X axis, I show the difference between 00:30:06.040 --> 00:30:09.530 the PRISM processing as well as CSMIP processing. 00:30:09.530 --> 00:30:15.100 So I color-coded – so everything blue is the difference in the PGA, PGV, 00:30:15.100 --> 00:30:18.060 and PGD is within 1%. 00:30:18.060 --> 00:30:21.690 And, you know, similarly, orange is between 1 and 2%. 00:30:21.690 --> 00:30:27.380 Yellow is between 2 and 3%, and then the red is, you know, 3 to 4%. 00:30:27.380 --> 00:30:31.740 And the vertical line showing the median value, and then 00:30:31.740 --> 00:30:35.360 the standard deviation is shown by the – you know, the horizontal. 00:30:35.360 --> 00:30:38.660 So overall, you know, most of the data is collected 00:30:38.660 --> 00:30:43.390 between minus 1 and, you know, the 1% difference. 00:30:43.390 --> 00:30:46.110 So for the measure of two of the records, you know, 00:30:46.110 --> 00:30:50.410 the difference overall between 1 and 2. And then the other seven earthquakes, 00:30:50.410 --> 00:30:54.660 basically, you know, provide the same kind of results. 00:30:56.140 --> 00:31:01.210 And now, when we look at the overall – you know, more than 1,800 records, 00:31:01.210 --> 00:31:04.270 for 80% of the records, PRISM provides peak values, 00:31:04.270 --> 00:31:08.010 you know, equal or less than 1% of the CSMIP values. 00:31:08.010 --> 00:31:11.470 So this – we did this testing to give a confidence to the 00:31:11.470 --> 00:31:14.860 engineering community that, you know, what we are processing here is, 00:31:14.860 --> 00:31:19.340 you know, not much difference than what the CSMIP processing does. 00:31:21.240 --> 00:31:26.910 So this part so far – done all the – you know, the automated processing part. 00:31:26.910 --> 00:31:30.520 So as I mentioned, so PRISM has also graphic user interface 00:31:30.520 --> 00:31:34.580 for manually editing the records if needed. 00:31:35.570 --> 00:31:38.320 So PRISM review tool is – these are the features. 00:31:38.330 --> 00:31:41.420 It’s a desktop application. It’s a standalone application. 00:31:41.420 --> 00:31:44.720 You don’t need to install the – you know, the processing engine. 00:31:45.500 --> 00:31:48.540 Again, using the COSMOS data sets. 00:31:49.700 --> 00:31:53.400 But it – because it’s based on the API, design is 00:31:53.400 --> 00:31:57.080 basically using all the tools that – what we have in the processing engine. 00:31:58.100 --> 00:32:01.640 It has multiple viewers and editors for handling seismic and, 00:32:01.640 --> 00:32:04.520 you know, the spectral – you know, the trace data. 00:32:04.520 --> 00:32:07.920 It provides template processing, so if you – if you can create your 00:32:07.920 --> 00:32:12.480 template and apply the same template for, you know, the multiple records 00:32:12.480 --> 00:32:16.220 without each time going and changing the parameters. 00:32:16.220 --> 00:32:20.060 Has a customizable, you know, graphic user interface. 00:32:20.060 --> 00:32:22.460 And it’s platform-independent. 00:32:23.690 --> 00:32:28.220 So here’s the main – you know, the interface of the review tool. 00:32:28.230 --> 00:32:32.660 So on the left side, you have your – you know, the navigation pane. 00:32:32.660 --> 00:32:37.730 You can see all the – you know, the – all the records. You can move them around. 00:32:37.730 --> 00:32:41.410 In the middle, you can see the – you know, the wave traces. 00:32:41.410 --> 00:32:47.160 And what is really neat is that Pete put a really nice feature on the 00:32:47.160 --> 00:32:51.370 last panel – the new channel – you can see all your metadata there – 00:32:51.370 --> 00:32:54.740 you know, the earthquake information, you know, the sampling rate, 00:32:54.740 --> 00:32:59.200 your station number, you know, so on and so forth. 00:33:00.890 --> 00:33:05.300 So when you look at the seismic editor, so you have your acceleration, 00:33:05.310 --> 00:33:08.680 velocity, displacement. You know, you can choose your – 00:33:08.680 --> 00:33:11.070 you know, the baseline correction parameters here. 00:33:11.070 --> 00:33:15.940 You know, like, the application range, and then, once you click the Commit, 00:33:15.940 --> 00:33:18.620 so it’s basically [inaudible] baseline correction – 00:33:18.620 --> 00:33:22.370 basically baseline from the – you know, the record. 00:33:22.370 --> 00:33:26.510 Again, what’s really nice is that, on the right panel, 00:33:26.510 --> 00:33:28.240 this is the history of your processing. 00:33:28.240 --> 00:33:31.910 So you can – by clicking, you can go back to your – you know, 00:33:31.910 --> 00:33:34.970 the previous step. So you can go back to all the way up 00:33:34.970 --> 00:33:42.320 and then – you know, to start over. So it’s quite flexible, the interface. 00:33:42.320 --> 00:33:48.110 And finally has the seismic editor for your spectrum where you manually – 00:33:48.110 --> 00:33:50.810 you can click on the – on the screen and, you know, 00:33:50.810 --> 00:33:56.330 come up with your filter range and apply your filtering. 00:33:56.330 --> 00:34:01.450 So this is all about, you know, how we do the manual processing. 00:34:01.450 --> 00:34:07.640 So we published, so far in 2017, a very comprehensive report 00:34:07.640 --> 00:34:13.519 on the batch processing part. And then another open file report 00:34:13.519 --> 00:34:17.230 fully explaining, you know, our testing with – comparing with 00:34:17.230 --> 00:34:22.319 the BAP as well as the CSMIP ground motion processing. 00:34:22.319 --> 00:34:26.359 We published a summary paper in Seismological Research Letters, 00:34:26.359 --> 00:34:32.340 you know, a few months ago. And PRISM, right now, is at GitHub. 00:34:32.340 --> 00:34:39.270 The whole source code is available under USGS/PRISM. 00:34:39.270 --> 00:34:44.790 We also have a dedicated website under USGS research, where we 00:34:44.790 --> 00:34:50.569 tried to put, you know, the more information for the – for the user. 00:34:50.569 --> 00:34:53.389 We have links to our – you know, the documentation, as well as – 00:34:53.389 --> 00:34:55.589 as well as link to source code. 00:34:55.589 --> 00:35:00.500 So if you look at that website, we provided very simple four-step 00:35:00.500 --> 00:35:06.420 procedure to how to run PRISM processing engine if you want to test it. 00:35:06.420 --> 00:35:09.991 So if you click on the second one – you know, Download PRISM Processing 00:35:09.991 --> 00:35:16.569 Engine Executable File, which is a JAR file, and then we provided examples. 00:35:16.569 --> 00:35:22.190 So there’s a PDF showing – it’s actually – it doesn’t even take, like, 00:35:22.190 --> 00:35:26.740 a half a minute to, you know, read it and implement it. 00:35:26.740 --> 00:35:32.670 So when you download our – the folder – the ZIP folder, 00:35:32.670 --> 00:35:36.130 so you have – the JAR file is available here, and then our – 00:35:36.130 --> 00:35:39.490 you know, the four-step procedure as a PDF. 00:35:39.490 --> 00:35:43.799 We have the configuration file folder where the XML files stays. 00:35:43.799 --> 00:35:45.789 So you don’t need to change that unless, you know, 00:35:45.789 --> 00:35:49.880 there is a need to change some of the – you know, the filtering or some of – 00:35:49.880 --> 00:35:55.010 you know, P phase picking parameters or some other parameters. 00:35:55.010 --> 00:35:57.289 And then we have the computed parameters. 00:35:57.289 --> 00:36:02.180 So in the “in” folder, we have the input motion as V-zero files. 00:36:02.180 --> 00:36:10.160 And when you, you know, run it on the terminal – so the folder structure is 00:36:10.160 --> 00:36:15.619 going to have all these outputs for each station, you know, nicely organized. 00:36:15.619 --> 00:36:18.490 And then, when you look at the – each station, you have your 00:36:18.490 --> 00:36:23.640 V-zero folder, V1 folder, V2, and V3 folder with all the products. 00:36:25.280 --> 00:36:32.180 So, so far, that’s, you know, what we have done with PRISM version 1. 00:36:32.180 --> 00:36:34.869 So we would like to continue with PRISM 2. 00:36:34.869 --> 00:36:37.000 So we have some – you know, the proposed updates 00:36:37.000 --> 00:36:40.410 for FY – fiscal year ’18. 00:36:41.780 --> 00:36:47.520 We would like to add a new feature for pre-screening V-zero files. 00:36:47.530 --> 00:36:53.240 And we would like to reject any files when the signal-to-noise ratio less than 3. 00:36:53.240 --> 00:36:57.680 So we don’t want to actually – you know, to process the really noisy records. 00:36:57.680 --> 00:37:02.840 And another feature that we want to implement is that right now, 00:37:02.840 --> 00:37:10.890 the corner filter – corner filter – the filter corners that we use for [inaudible] filter, 00:37:10.890 --> 00:37:14.529 we have a lookup table based on magnitude of the earthquake. 00:37:14.529 --> 00:37:19.290 So they are really conservative values. So we would like to get rid of that. 00:37:19.290 --> 00:37:23.740 Instead, we want to look at the frequency content of the record 00:37:23.740 --> 00:37:29.880 to come up with, you know, what corner filters that – 00:37:29.880 --> 00:37:34.240 the corner filter frequencies that we need to use for filtering. 00:37:34.240 --> 00:37:39.630 So this is – we thought these are – you know, our high-priority updates. 00:37:39.630 --> 00:37:43.349 So in addition, we’d like to implement frequency domain decimation algorithm. 00:37:43.349 --> 00:37:46.921 You know, as I mentioned before, everything less than 200 sample 00:37:46.921 --> 00:37:51.549 per second, the PRISM first resample the record to 200 sample per second 00:37:51.549 --> 00:37:57.640 to eliminate, you know, numerical errors with the differentiation and integration. 00:37:57.640 --> 00:38:02.370 So – and that keeps the waveform like that, you know, until – and it ends. 00:38:02.370 --> 00:38:07.600 So we – you know, based on feedback from NSMP working group, again, 00:38:07.600 --> 00:38:14.480 we want to – so there is a demand to bring back the original sampling rates. 00:38:14.480 --> 00:38:15.960 So to do that, we would like to 00:38:15.960 --> 00:38:19.700 implement a frequency domain decimation algorithm. 00:38:20.720 --> 00:38:25.060 And then, as I mentioned, our integration and differentiation 00:38:25.079 --> 00:38:26.980 is currently in time domain. 00:38:26.980 --> 00:38:29.440 We would like to move it to the frequency domain, again, 00:38:29.440 --> 00:38:34.240 to further reduce the numerical issues – numerical – you know, the errors. 00:38:34.240 --> 00:38:38.680 And finally, you know, PRISM only accept V-zero COSMOS data format. 00:38:38.680 --> 00:38:41.980 That’s kind of like limitation. We would like to have some 00:38:41.980 --> 00:38:45.750 conversion tools for different input formats, you know, 00:38:45.750 --> 00:38:50.131 similar to COSMOS – you know, the converter – so the user doesn’t have to 00:38:50.131 --> 00:38:56.069 have, you know, the V-zero – COSMOS V-zero data format to use PRISM. 00:38:56.069 --> 00:39:01.619 So our dream is, bring the PRISM 3 to the cloud, 00:39:01.619 --> 00:39:05.390 and hopefully in the, you know, coming years. 00:39:05.390 --> 00:39:09.890 And these are my concluding remarks. So we believe the – you know, 00:39:09.890 --> 00:39:12.380 that it will offer a very robust automated processing – 00:39:12.380 --> 00:39:17.769 ground motion processing software, you know, for use by, you know, 00:39:17.769 --> 00:39:22.280 engineering applications and, you know, seismological studies. 00:39:22.280 --> 00:39:26.400 So we realize that, you know, there is no unique way to process ground motion. 00:39:26.400 --> 00:39:29.000 So we cannot that, you know, we are doing the best. 00:39:29.000 --> 00:39:33.540 But we use, you know, mostly accepted – you know, the procedures, you know, 00:39:33.540 --> 00:39:39.079 developed over the years by, you know, the Tony Shakal and his group. 00:39:39.079 --> 00:39:44.559 You know, and then, you know, USGS folks, as well as, 00:39:44.560 --> 00:39:48.480 you know, the other – you know, the COSMOS partners. 00:39:50.200 --> 00:39:54.279 So, you know, all of the essential steps apply in PRISM, you know, 00:39:54.279 --> 00:39:59.519 whether automatic or manual, are documented as, you know, 00:39:59.519 --> 00:40:01.789 ASCII file headers of the processed time series. 00:40:01.789 --> 00:40:06.660 We know this is very important. So we don’t want to give a black box. 00:40:06.660 --> 00:40:09.980 So actually, once the processing is done, 00:40:09.980 --> 00:40:13.549 so user have all the information – you know, what parameters are used. 00:40:13.549 --> 00:40:16.880 So we try to provide as many information as possible 00:40:16.880 --> 00:40:21.309 in the header of our products. 00:40:21.309 --> 00:40:25.660 So these are the references cited during my presentation. 00:40:27.640 --> 00:40:32.760 And, you know, the credits goes for the – for the logos that I used. 00:40:33.900 --> 00:40:36.720 So that’s basically – wrap up my presentation, and I would like to 00:40:36.720 --> 00:40:41.650 thank you for your time and for coming here and as well as for your attention. 00:40:41.650 --> 00:40:46.960 [ Applause ] 00:40:46.960 --> 00:40:49.960 - Thank you, Erol. It’s really nice to hear that – 00:40:49.960 --> 00:40:53.369 although BAP was a very useful product, that it’s now 00:40:53.369 --> 00:40:57.950 been superseded by something that will automate the process. 00:40:57.950 --> 00:41:04.269 I remember waiting quite a while to get records using BAP, and so it’s really nice 00:41:04.269 --> 00:41:10.430 to hear about new software that will do that very quickly and automatically. 00:41:10.430 --> 00:41:13.260 Are there some questions? 00:41:18.780 --> 00:41:22.540 - Hey, Erol. Thanks. That was a really nice presentation. 00:41:22.549 --> 00:41:25.190 So just a couple things. One, I think it would be great 00:41:25.190 --> 00:41:30.880 for your 2.0 to intake other data formats, which I think you … 00:41:30.880 --> 00:41:32.500 - Right. 00:41:32.500 --> 00:41:36.840 - And so then, to use this, you would download the V-zero separately, 00:41:36.840 --> 00:41:39.900 and then process it all locally. Is that correct? 00:41:39.900 --> 00:41:43.220 - Well, the processing engine is implemented to AQMS. 00:41:43.220 --> 00:41:45.960 So AQMS is providing all the V-zero files automatically. 00:41:45.960 --> 00:41:47.190 There is no … - Oh, okay. 00:41:47.190 --> 00:41:50.990 - There is no intervention on that part. - So you don’t have to download 00:41:50.990 --> 00:41:54.340 the files independently. It’s … - For manual processing, yes. 00:41:54.340 --> 00:41:57.220 You know, you need to download the V-zero – you need the V-zero files. 00:41:57.229 --> 00:42:01.660 - So you also showed a couple of your test data sets or examples. 00:42:01.660 --> 00:42:05.779 Do you – would you have any of those available just to use? 00:42:05.780 --> 00:42:08.580 Or you would advocate that you should do the – 00:42:08.580 --> 00:42:11.640 run PRISM ourselves? [chuckles] 00:42:11.640 --> 00:42:15.340 - For testing one? - No. Just for – I mean, you’re 00:42:15.340 --> 00:42:21.940 producing a lot of really nice data sets. Do you have those personally available? 00:42:22.040 --> 00:42:24.700 - Oh, yeah, yeah. We have … - Like, the South Napa. 00:42:24.700 --> 00:42:27.860 All the processed data that you’ve done. - Right. Yeah. Yeah. Everything is – 00:42:27.860 --> 00:42:32.220 you know, we can definitely provide it. Well, actually it’s a good point. 00:42:32.220 --> 00:42:35.369 Maybe we can put all onto the website that, you know, a user can 00:42:35.369 --> 00:42:37.680 get the whole Napa waveforms. - Right. 00:42:37.680 --> 00:42:40.000 - Like V-zero files, and then they can do their own … 00:42:40.000 --> 00:42:41.640 - Yeah, yeah. Okay. - … you know, processing. Yeah. 00:42:41.640 --> 00:42:43.460 - Great. - That’s very good point. Thank you. 00:42:43.460 --> 00:42:44.780 - Thank you. 00:42:49.160 --> 00:42:52.420 - Erol, you mentioned that you’ve done a lot of testing to convince 00:42:52.430 --> 00:42:55.240 the engineering community that PRISM is working. 00:42:55.240 --> 00:42:57.240 Have you heard back from them? 00:42:57.240 --> 00:43:00.900 Do they accept PRISM, and do they plan to use it? 00:43:00.910 --> 00:43:03.900 - Yeah, that – well, that’s an excellent point. So that’s – actually, that’s why 00:43:03.900 --> 00:43:08.499 we put together about this 150-pages-long open file report. 00:43:08.499 --> 00:43:14.390 I presented PRISM during AGU. I haven’t got a chance to 00:43:14.390 --> 00:43:17.869 go to engineering community yet, you know, to make presentations, 00:43:17.869 --> 00:43:22.839 but that’s the plan. We put the PRISM, you know, 00:43:22.840 --> 00:43:26.800 up in ResearchGate, you know, a couple of, you know, the websites. 00:43:26.800 --> 00:43:29.756 We are getting, you know, a lot of hits and then, you know, the questions. 00:43:29.756 --> 00:43:32.369 I believe people are, you know, using it. 00:43:32.369 --> 00:43:36.660 But one of the downside is that we are really limited to the 00:43:36.660 --> 00:43:42.080 COSMOS V-zero format, which has quite a bit of header information, 00:43:42.080 --> 00:43:46.640 you know, including, you know, earthquake magnitude used to be there, 00:43:46.640 --> 00:43:48.779 you know, among many, many others. 00:43:48.779 --> 00:43:52.390 So we are hoping that, with the PRISM 2, once we double up those 00:43:52.390 --> 00:43:58.180 conversion tools, then the engineers will be much more interested in using it. 00:43:59.549 --> 00:44:03.860 And our graphic user interface, we are doing a second update on that, 00:44:03.869 --> 00:44:06.420 so we haven’t released it yet. 00:44:06.420 --> 00:44:09.619 But Pete and Chris, they’ve been working on that very hard. 00:44:09.619 --> 00:44:13.670 So it should be done, you know, in a couple of weeks. 00:44:13.670 --> 00:44:17.890 So once it’s done, we’re going to put that on GitHub and our – you know, 00:44:17.890 --> 00:44:23.319 the USGS website as well. I think that will bring more interest 00:44:23.319 --> 00:44:26.249 from the engineering community that they can see and, you know, 00:44:26.249 --> 00:44:28.580 do a manual processing. 00:44:31.460 --> 00:44:34.420 - Could you tell us … - Oh, sorry. 00:44:39.720 --> 00:44:44.860 - Perhaps you mentioned it but I – and I missed it, but suppose a magnitude 6 00:44:44.869 --> 00:44:51.090 earthquake occurs in the Bay Area. How quickly can the data be processed? 00:44:51.090 --> 00:44:57.339 Is it limited by the delivery of the accelerogram data itself? 00:44:57.339 --> 00:45:02.360 And what kind of lag time is there? - Yeah, that’s an excellent question. 00:45:02.360 --> 00:45:08.000 Maybe Tim can tell us how quickly the AQMS can pull the data. 00:45:08.000 --> 00:45:11.619 Because the PRISM does – you know, the processing is very easy for those 00:45:11.619 --> 00:45:16.580 700-channel Napa. Every channel is, you know, less than a second. 00:45:16.580 --> 00:45:20.799 So it all depends on how quickly AQMS is going to decide on the 00:45:20.799 --> 00:45:25.700 magnitudes information, right, and then pull together the waveforms. 00:45:25.700 --> 00:45:30.289 So as soon as those waveforms – you know, they’re compiled, 00:45:30.289 --> 00:45:32.269 the PRISM is going to – kicks in and then, you know, 00:45:32.269 --> 00:45:36.390 just generate all the – all the outcomes pretty fast. 00:45:36.390 --> 00:45:41.029 And I believe that Chris is working closely with the CSMIP folks 00:45:41.029 --> 00:45:45.170 to push the data automatically to their – to their website as well. 00:45:45.170 --> 00:45:49.690 So hopefully, you know, in a couple of months, 00:45:49.690 --> 00:45:53.720 everything will done automatically, and everything will be available on the, 00:45:53.720 --> 00:45:56.620 you know, engineering website. 00:45:59.100 --> 00:46:04.039 - Erol, the original networks are acquiring a lot and a great deal of strong motion 00:46:04.040 --> 00:46:09.970 data these days as well. Are those data being processed through PRISM as well? 00:46:11.420 --> 00:46:16.180 Well, whenever the data comes to NSMP, that’s where the PRISM, 00:46:16.180 --> 00:46:21.380 you know, works. But if you are talking about the 00:46:21.380 --> 00:46:27.780 other networks, yeah, they’re very welcome to use, you know, the PRISM. 00:46:27.780 --> 00:46:30.440 I think it will nice to have some – you know, the outreach to those – 00:46:30.440 --> 00:46:36.249 you know, the local networks to show what we have. Yeah. 00:46:36.249 --> 00:46:40.920 - I think it would be nice for whatever products the USGS issues 00:46:40.920 --> 00:46:44.380 to be standard and uniform. - Right. 00:46:46.700 --> 00:46:53.840 [ Silence ] 00:46:54.780 --> 00:47:00.620 - It’s interesting to me to hear these things going on. 00:47:00.620 --> 00:47:07.040 Every five or 10 years or so, a group connected with strong 00:47:07.040 --> 00:47:11.320 ground motion comes forward with bright guys and computer displays 00:47:11.320 --> 00:47:21.940 to say – and to advertise the fact that their processing is going very well, 00:47:21.940 --> 00:47:38.769 and people should use it. These emerging systems come from 00:47:38.769 --> 00:47:48.900 Caltech long ago, Sacramento, USC, Ralph Archuleta at Santa Barbara. 00:47:48.900 --> 00:47:57.540 And each has a period where the user is really intrigued 00:47:57.540 --> 00:48:04.360 by the great steps that are – that are being taken – yours too. 00:48:05.860 --> 00:48:18.519 In five or 10 years, when your interests change, and your followers choose 00:48:18.519 --> 00:48:24.549 other things to do, this will fade away, and another group will emerge. 00:48:24.549 --> 00:48:29.880 And will all of us – will all of us being paying attention to which group 00:48:29.880 --> 00:48:39.600 that is going to be? So that’s what we’re all going to do as users of data. 00:48:39.600 --> 00:48:46.260 I have one clarification point to make, and that is that, one of your early slides 00:48:46.260 --> 00:48:53.320 says that the BAP procedures was written up by Converse and Brady. 00:48:54.560 --> 00:49:02.740 It is now out of date or about to be – on the slide – I forget the phrase you used. 00:49:02.740 --> 00:49:05.020 - Oh, we don’t say outdated. Outmoded. 00:49:05.020 --> 00:49:06.680 - Hm? - Outmoded. 00:49:06.680 --> 00:49:09.640 - Outmoded, yes. This is very … 00:49:09.640 --> 00:49:12.410 - I didn’t mean to – you know, to offend. 00:49:12.410 --> 00:49:16.980 We – actually, we used BAP a lot to start the PRISM project. 00:49:16.980 --> 00:49:22.119 And, you know, the overall layout and then the structure – how the flow is. 00:49:22.119 --> 00:49:27.009 So we still use, you know, a lot of things that, you know, the BAP does. 00:49:27.009 --> 00:49:31.550 But for every software, you know, as time goes, you know, like, 00:49:31.550 --> 00:49:34.749 the newer one comes. So right now, nobody is using 00:49:34.749 --> 00:49:40.730 Windows 3 or 3.1, right, or the Vista or – that’s my point. 00:49:40.730 --> 00:49:43.660 So there is this – you know, the flow of information, flow of, 00:49:43.660 --> 00:49:48.960 you know, the new stuff. So by no means that we try to, like, 00:49:48.960 --> 00:49:53.539 undermine the efforts that, you know, put then. 00:49:53.539 --> 00:49:58.559 And BAP has been successfully used, you know, for the last 15 years. 00:49:58.559 --> 00:50:05.109 - I was actually going to say that, indeed, BAP has become outmoded 00:50:05.109 --> 00:50:08.460 with all the work that is going on and all the different inputs 00:50:08.460 --> 00:50:17.109 that you have to attend to. But the – “outmoded” is not quite right. 00:50:17.109 --> 00:50:23.900 The group here at the Geological Survey who was – who was preparing 00:50:23.900 --> 00:50:28.859 data for the use by the engineering community were all offered, 00:50:28.859 --> 00:50:38.640 back in the early ’90s, the option to get an early out from the Survey – 00:50:38.640 --> 00:50:44.300 that is, to take – to retire gracefully. 00:50:45.880 --> 00:50:55.240 This led to a situation where the leader of this group was left with no staff. 00:50:55.310 --> 00:51:05.180 And for – and, I mean, I exaggerate a little, but the – 00:51:05.180 --> 00:51:13.759 this early-out offering was followed quite closely by the 00:51:13.760 --> 00:51:21.440 well-known rift within the Survey in 1994 and ’95. 00:51:21.440 --> 00:51:30.340 And through some – through some procedure and for – 00:51:30.349 --> 00:51:39.779 probably through some intrigue as well, I, as the leader, was rift. 00:51:39.780 --> 00:51:44.820 So the Survey was left pretty much without 00:51:44.820 --> 00:51:50.759 hands-on capability of fulfilling the BAP system. 00:51:50.760 --> 00:51:57.280 I think that the Survey has recovered remarkably well since that time. 00:52:02.860 --> 00:52:05.440 - Any further comments or questions? 00:52:06.160 --> 00:52:08.820 If not, let’s thank Erol again for a nice presentation. 00:52:08.820 --> 00:52:10.020 - Thank you. Appreciate it. 00:52:10.020 --> 00:52:13.080 [ Applause ] 00:52:13.089 --> 00:52:15.880 - And we’ll meet over at Tectonic – anyone who wants to join us 00:52:15.880 --> 00:52:19.680 over in Tectonic Grill, we’ll be there in about 15 minutes.