WEBVTT Kind: captions Language: en-US 00:00:02.060 --> 00:00:04.040 [Silence] 00:00:04.040 --> 00:00:07.800 Okay. Welcome, everybody. Good morning. 00:00:07.800 --> 00:00:12.420 Thank you for joining today’s seminar. I hope everyone is staying safe and 00:00:12.420 --> 00:00:19.420 healthy today. Especially today. It’s pretty dicey out there. 00:00:19.420 --> 00:00:21.700 I wanted to make a few brief announcements. 00:00:21.700 --> 00:00:27.930 First, next week’s speaker is Wenyuan Fan from UCSD Scripps. 00:00:27.930 --> 00:00:30.520 So please join us for that. 00:00:30.520 --> 00:00:32.930 Second, make sure, during the talk, 00:00:32.930 --> 00:00:37.620 that your microphone is muted and your video is off. 00:00:37.620 --> 00:00:39.760 And that’s about it. 00:00:39.760 --> 00:00:45.040 So with that, I’ll pass it off to Annemarie to introduce today’s speaker. 00:00:46.900 --> 00:00:50.740 - All right. Thanks, Kathryn. So today it’s my pleasure to introduce 00:00:50.750 --> 00:00:55.720 Jessie Saunders for today’s seminar. Jessie earned her B.S. in applied physics 00:00:55.720 --> 00:01:01.489 and geophysics at UC-Davis and then went on to UCSD Scripps to get a Ph.D. 00:01:01.489 --> 00:01:04.509 focused on multi-sensor tsunami early warning and earthquake 00:01:04.509 --> 00:01:08.340 monitoring effects – monitoring earthquake effects on structures. 00:01:08.340 --> 00:01:12.750 So, Jessie has worked on a variety of topics with sort of a thread of 00:01:12.750 --> 00:01:17.020 incorporating multi-sensor classes and different data types. 00:01:17.020 --> 00:01:20.060 So, for example, she performed multi-sensor structural monitoring, 00:01:20.070 --> 00:01:24.750 both on the big UCSD shake table and also on a tall building – 00:01:24.750 --> 00:01:27.510 or, the tallest building, I guess – at Oklahoma State University. 00:01:27.510 --> 00:01:32.780 And, in those tests, she used both GPS, accelerometers, and gyroscopes. 00:01:32.780 --> 00:01:35.880 And she also included kinematic rupture simulations to assess building 00:01:35.880 --> 00:01:39.090 response there. So, again, pulling together a lot of different themes. 00:01:39.090 --> 00:01:42.970 She helped validate MEMS accelerometers for use at co-located 00:01:42.970 --> 00:01:47.170 GNSS stations for early warning. And she’s also worked on rapid tsunami 00:01:47.170 --> 00:01:50.740 earthquake detection and near-shore tsunami early warning in Cascadia, 00:01:50.740 --> 00:01:55.670 again, using a multiple-sensor approach – seismogeodetic data sets to 00:01:55.670 --> 00:02:00.580 improve slip characterization and also rapidly distinguish tsunami earthquakes. 00:02:00.580 --> 00:02:04.759 So, since arriving at the USGS in March of last year, Jessie’s become 00:02:04.760 --> 00:02:08.300 involved in several groups and projects. 00:02:08.300 --> 00:02:12.110 And I’ll just highlight the PLUM earthquake early warning testing group. 00:02:12.110 --> 00:02:15.550 She’s been an instrumental contributor there. 00:02:15.550 --> 00:02:19.010 And also particularly in the ShakeAlert ground motion group, where we really 00:02:19.010 --> 00:02:22.329 wouldn’t have made the progress that we have without her help. 00:02:22.329 --> 00:02:26.180 Her research has been focused on using both instrumental and Did You Feel It 00:02:26.180 --> 00:02:29.719 seismic intensity data to consider limits of network-based early warning 00:02:29.719 --> 00:02:34.000 approaches, uncertainty in early warning ground motion alerts, and limitations 00:02:34.000 --> 00:02:39.160 and considerations of low levels of MMI-based early warning alerts. 00:02:39.160 --> 00:02:43.099 So, throughout all this, she’s proven already to be an invaluable collaborator 00:02:43.099 --> 00:02:46.090 eager to contribute to ShakeAlert projects, and she has got a really great 00:02:46.090 --> 00:02:48.829 insight into dealing with and visualizing large data sets, 00:02:48.829 --> 00:02:53.680 which we will see today in her talk. And her talk is, how low should 00:02:53.680 --> 00:02:56.439 we alert? [echoing voice] Exploring earthquake early warning 00:02:56.440 --> 00:03:00.100 thresholds for a range of target intensities. 00:03:01.040 --> 00:03:04.000 All right. Take it away, Jessie. 00:03:04.000 --> 00:03:07.880 - Thank you, Annemarie, so much for the introduction. 00:03:07.880 --> 00:03:13.010 So, like Annemarie said, I will be talking about using different 00:03:13.010 --> 00:03:16.299 alerting thresholds for a range of target intensities. 00:03:16.299 --> 00:03:20.959 And this is work that I have done with Annemarie as well as 00:03:20.959 --> 00:03:24.409 Sarah Minson and Brad Aagaard. So thank you very much to them 00:03:24.409 --> 00:03:27.859 for all of their help with this work. 00:03:28.760 --> 00:03:31.280 Let me just start that. 00:03:31.290 --> 00:03:36.120 So the ShakeAlert earthquake early warning system is the early warning 00:03:36.120 --> 00:03:38.860 system for the West Coast of the United States. 00:03:38.860 --> 00:03:42.699 The main goal of the system is to rapidly detect earthquakes within its 00:03:42.699 --> 00:03:45.970 alerting boundaries, shown by the pink outlines here. 00:03:45.970 --> 00:03:51.140 And then, once these earthquakes are detected, trying to alert people in 00:03:51.140 --> 00:03:56.740 regions that are expected to experience strong shaking from these earthquakes. 00:03:56.740 --> 00:04:01.900 Public alerting is available in California since October 2019. 00:04:01.900 --> 00:04:05.370 And it’s currently under development in Oregon and Washington. 00:04:05.370 --> 00:04:10.969 And ShakeAlert is operated by the USGS, but it includes many other 00:04:10.969 --> 00:04:16.100 collaborators, including university partners at Caltech, UC-Berkeley, 00:04:16.100 --> 00:04:20.780 University of Oregon, and University of Washington, just to name a few. 00:04:20.780 --> 00:04:26.680 So, like my title said, my focus for today’s talk will be on alerting 00:04:26.680 --> 00:04:31.340 strategies, where I will ask what threshold is needed to ensure that 00:04:31.350 --> 00:04:35.850 a specific intensity target is included inside the alert region. 00:04:35.850 --> 00:04:40.040 And I’ll be using a couple of terms throughout this talk 00:04:40.040 --> 00:04:43.570 that I would just like to define upfront here. 00:04:43.570 --> 00:04:48.970 So when I say alert target or MMI target, I am referring to the minimum 00:04:48.970 --> 00:04:53.600 Modified Mercalli Intensity for which we want to issue an alert. 00:04:53.600 --> 00:04:57.880 And then, when I say MMI threshold, I’m referring to 00:04:57.880 --> 00:05:02.100 the MMI level used to determine the size of the alert region. 00:05:02.100 --> 00:05:08.520 And the MMI threshold does not necessarily equal the MMI target. 00:05:08.520 --> 00:05:14.700 So here is a rather detailed schematic of the ShakeAlert alerting process. 00:05:14.700 --> 00:05:18.030 So the USGS is in charge of generate early warning alerts through the 00:05:18.030 --> 00:05:23.280 ShakeAlert system, and their part is in the green right here. 00:05:23.280 --> 00:05:27.480 And then, over here are some examples of the different external 00:05:27.480 --> 00:05:32.280 entities that are in charge of delivering the alerts to the public. 00:05:32.840 --> 00:05:38.180 So, when an alert is published by the USGS through ShakeAlert, different 00:05:38.180 --> 00:05:41.920 alerting criteria must be met before these alerts can be delivered by different 00:05:41.920 --> 00:05:47.100 groups, and those are shown here. So, for example, for an alert to be 00:05:47.100 --> 00:05:51.880 delivered through the wireless emergency alert system up at the top, 00:05:51.880 --> 00:05:57.520 the magnitude must be at least magnitude 5, and then WEA will use 00:05:57.520 --> 00:06:04.790 the MMI 3.5 alert perimeter to send alerts to all cell phones in those areas. 00:06:04.790 --> 00:06:09.510 And then, for alert delivery through various early warning cell phone apps, 00:06:09.510 --> 00:06:14.630 such as MyShake, for an example, the magnitude must be at least 4.5, 00:06:14.630 --> 00:06:18.640 and then the alert perimeter that is – that is used through 00:06:18.640 --> 00:06:21.840 the early warning apps is MMI 2.5. 00:06:21.840 --> 00:06:26.740 And then, as my final example, this was announced just last week, 00:06:26.740 --> 00:06:30.490 which is very exciting. Alerts will now be sent through 00:06:30.490 --> 00:06:37.470 Google to Android phones. And they’re using a kind of two-tiered 00:06:37.470 --> 00:06:43.580 warning message where alerts within – where alerts for the MMI 4.5 perimeter 00:06:43.580 --> 00:06:49.160 are issued a take-action alert, and then phones within the MMI 2.5 00:06:49.170 --> 00:06:51.630 perimeter are issued a be-aware alert. 00:06:51.630 --> 00:06:57.840 And then, once again, this is for magnitude 4.5 and higher earthquakes. 00:06:59.240 --> 00:07:03.880 So, for my work here, I am focusing on the alert regions generated using 00:07:03.880 --> 00:07:07.640 a point source model, so the EPIC algorithm highlighted in pink. 00:07:07.640 --> 00:07:12.500 And then I’ll be discussing these different alerting thresholds here. 00:07:14.440 --> 00:07:18.980 And so now I’ll just step through an illustrative example of the 00:07:18.980 --> 00:07:23.280 alert region computation process using the magnitude 7.1 00:07:23.280 --> 00:07:25.800 Ridgecrest earthquake from last year. 00:07:25.800 --> 00:07:29.830 So here we have a map of southern California, 00:07:29.830 --> 00:07:33.920 where the different ShakeAlert stations are shown by the gray triangles. 00:07:33.920 --> 00:07:39.060 And first, an earthquake is detected when the P waves arrive 00:07:39.060 --> 00:07:42.280 at four stations, which are shown in red. 00:07:42.280 --> 00:07:46.250 This P wave data from these stations, as well as later-arriving stations, 00:07:46.250 --> 00:07:50.180 are then used to estimate the source parameters, which for here, we will be 00:07:50.180 --> 00:07:55.830 using the magnitude and the location. And note here, for this work, we are 00:07:55.830 --> 00:08:01.260 assuming that the magnitude and the epicenter are estimated accurately. 00:08:02.740 --> 00:08:09.690 So now, the magnitude and a constant Vs30 are then input into ground motion 00:08:09.690 --> 00:08:14.010 prediction equations to estimate the peak ground acceleration and peak 00:08:14.010 --> 00:08:17.900 ground velocity with distance. And then, for each distance, 00:08:17.900 --> 00:08:22.200 the PGA and PGV are then converted into MMI using a ground 00:08:22.200 --> 00:08:27.320 motion-to-intensity conversion equation. So now we can then find the epicentral 00:08:27.320 --> 00:08:31.180 distances for a range of possible alerting thresholds, which are 00:08:31.180 --> 00:08:39.229 shown here by the circles. And the distance at which – 00:08:39.229 --> 00:08:44.370 has an MMI that corresponds to the threshold is now the alert distance, 00:08:44.370 --> 00:08:49.430 which, in this case, we’ll use MMI 4.0 as our example. 00:08:49.430 --> 00:08:52.320 And this arrow here represents the alert distance here. 00:08:52.320 --> 00:08:58.040 And a circular region with that radius is then issued as the alert 00:08:58.040 --> 00:09:03.400 perimeter, which is shown here. So now we have our alert region for this 00:09:03.400 --> 00:09:09.520 earthquake using this MMI threshold. However, we do have an issue. 00:09:09.520 --> 00:09:15.100 And that issue is that ground motion variabilities are not accounted for 00:09:15.100 --> 00:09:19.430 very well with this alert region computation approach. 00:09:19.430 --> 00:09:23.670 So we know that variations in the source and site conditions can produce 00:09:23.670 --> 00:09:29.430 a variation in the peak ground motions relative to the median expected PGA 00:09:29.430 --> 00:09:34.740 and PGV with distance from the GMPE, and that’s shown as an example here 00:09:34.740 --> 00:09:39.069 with the figure with the blue dots. And then, as well, we also know that 00:09:39.069 --> 00:09:44.249 there’s variation between how these peak ground motions relate to 00:09:44.249 --> 00:09:48.420 the MMI that people experience. And so that’s shown here in the purple. 00:09:48.420 --> 00:09:56.110 And, when we combine these two uncertainties, that ends up with – 00:09:56.110 --> 00:10:00.290 for our alert region, where we have a wide range of uncertainty in the 00:10:00.290 --> 00:10:06.740 expected extent for a given MMI. So this is the median expected alert 00:10:06.740 --> 00:10:10.040 distance for MMI 4 shaking as a function of magnitude shown 00:10:10.040 --> 00:10:13.779 in the solid line. And then the dashed lines are the plus and minus 00:10:13.779 --> 00:10:17.149 one standard deviation when we use the uncertainties 00:10:17.149 --> 00:10:20.540 from the ground motion models that we use. 00:10:23.509 --> 00:10:28.100 And so, how does this affect who gets alerted – who needs alerts? 00:10:28.110 --> 00:10:33.040 So here again is our map view for our example earthquake with the 00:10:33.040 --> 00:10:37.829 same alert distance-with-magnitude plot of MMI 4 that we just saw. 00:10:37.829 --> 00:10:40.850 And so, if we choose the median expected alert distance, 00:10:40.850 --> 00:10:45.089 that gives you the alert region here. And so that corresponds to this dot. 00:10:45.089 --> 00:10:49.059 And then, taking into account one standard deviation 00:10:49.059 --> 00:10:52.800 of the combined uncertainties in the ground motion models, 00:10:52.800 --> 00:10:56.589 those are the regions in the dashed lines here. 00:10:56.589 --> 00:11:02.980 And so, while people within the median expected alert region in black will 00:11:02.980 --> 00:11:07.070 still get an alert, there will be people who experience 00:11:07.070 --> 00:11:10.899 higher-than-expected shaking that will not receive an alert in this case. 00:11:10.899 --> 00:11:15.749 And, if we compare these regions to the ShakeMap for the earthquake, 00:11:15.749 --> 00:11:22.700 which is shown here in the color, we see that, to the south, over here, 00:11:22.700 --> 00:11:28.059 there’s very clearly some MMI 5-plus regions that are not included inside 00:11:28.059 --> 00:11:35.920 the black alert region for this example. And it’s a little bit hard to distinguish 00:11:35.920 --> 00:11:40.369 the MMI 4 region with the ShakeMap colors, 00:11:40.369 --> 00:11:43.430 so let’s color this by alert quality. 00:11:43.430 --> 00:11:49.029 And so that’s shown here. So here the red and the green colors 00:11:49.029 --> 00:11:53.360 on the map, those are all of the regions in the ShakeMap that 00:11:53.360 --> 00:11:59.730 experience MMI 4 and larger. And so, if they are in the alert region 00:11:59.730 --> 00:12:04.410 here, that is a correct alert, so it’s green. And if they are outside of the alert 00:12:04.410 --> 00:12:07.740 region, that is a missed alert, and that’s red. So all of the people in these 00:12:07.740 --> 00:12:14.480 regions would not receive an alert that they would need for this example. 00:12:14.480 --> 00:12:18.339 And then also, there are some regions shown in the yellow here. 00:12:18.339 --> 00:12:22.050 And these are regions that are incorrect alerts. 00:12:22.050 --> 00:12:24.910 So these are locations that end up experiencing shaking 00:12:24.910 --> 00:12:27.980 less than MMI 4 that are alerted. 00:12:29.040 --> 00:12:32.520 So, for this example, this, I think, clearly demonstrates that using 00:12:32.520 --> 00:12:38.680 an MMI threshold that is the same as the MMI alert target, 00:12:38.680 --> 00:12:41.119 we won’t be able to include all of the regions that 00:12:41.120 --> 00:12:45.880 ultimately experience shaking at the alert target level. 00:12:46.540 --> 00:12:50.320 So what do we do here? So our goal is to find an alerting 00:12:50.320 --> 00:12:55.300 strategy such that the locations that experience MMI at or above our 00:12:55.310 --> 00:13:00.220 alert target can be alerted reliably. And, for the alert perimeter approach 00:13:00.220 --> 00:13:04.459 that we’re considering here, this means that we want to 00:13:04.459 --> 00:13:09.209 expand the size of the alert region. And there are a couple of options 00:13:09.209 --> 00:13:14.829 that we can choose here. Both of them accomplish the same thing. 00:13:14.829 --> 00:13:19.420 So Option 1 is to choose a different alerting threshold. 00:13:19.420 --> 00:13:24.200 And that would be one that would be lower than the alert target. 00:13:24.209 --> 00:13:30.180 So, instead of alerting at, say, this circle, we alert at this one right here. 00:13:30.180 --> 00:13:36.179 And then, Option 2 is to incorporate ground motion model uncertainty. 00:13:36.179 --> 00:13:39.519 So that, for a given alert target, instead of choosing the median 00:13:39.519 --> 00:13:46.100 expected alert distance, we end up choosing the median expected plus 00:13:46.100 --> 00:13:50.500 one or two sigmas. So that will increase the alert region to around here. 00:13:50.500 --> 00:13:55.980 And we explored the framework in Option 2 using the 2019 Ridgecrest 00:13:55.980 --> 00:14:01.379 earthquakes for an alert target at MMI 3.5, and that was just published 00:14:01.379 --> 00:14:05.850 in the BSSA Ridgecrest special issue, and I definitely encourage you to 00:14:05.850 --> 00:14:13.380 check that out. But today, we will be looking at the Option 1, so … 00:14:16.940 --> 00:14:21.800 So let’s go back to our Ridgecrest example and take a look at what 00:14:21.800 --> 00:14:25.500 the alerting threshold strategy would look like. 00:14:25.500 --> 00:14:30.580 So, once again, here is our map showing the alert quality for 00:14:30.580 --> 00:14:35.959 our MMI 4 alert target using an MMI threshold of MMI 4. 00:14:35.959 --> 00:14:42.040 And so these bar plots here show the total alert quality in terms of area 00:14:42.040 --> 00:14:47.589 and then in population for this particular alert region. 00:14:47.589 --> 00:14:52.139 And so, as we can see here, the green numbers indicate 00:14:52.140 --> 00:14:56.400 the correct alert percentage for this strategy. 00:14:56.400 --> 00:15:03.520 So, for this alert strategy, while we can get 42% correct alerts in terms of area, 00:15:03.520 --> 00:15:08.300 we actually only capture 1% correct alerts in terms of population. 00:15:09.880 --> 00:15:15.580 And so now, if we decrease our MMI threshold to 3.5, you can 00:15:15.589 --> 00:15:20.060 see this expands the alert region. And we end up capturing a lot 00:15:20.060 --> 00:15:25.290 more correct alerts. So we get 92% correct alerts in terms of area 00:15:25.290 --> 00:15:29.439 and then 45% in terms of population. 00:15:29.439 --> 00:15:34.249 And you can now see, as well, that there’s a lot of more yellow here. 00:15:34.249 --> 00:15:40.050 So we are incorporating more incorrect alerts using this strategy. 00:15:40.050 --> 00:15:46.049 And then, finally, if we go down to a threshold of MMI 3.0, 00:15:46.049 --> 00:15:49.860 we can see that we can alert … 00:15:49.860 --> 00:15:53.740 [audio cuts out] 00:15:53.740 --> 00:15:58.620 … requires alerts for MMI 4, but this does come with sending 00:15:58.629 --> 00:16:01.559 incorrect alerts to nearly twice the amount of people who 00:16:01.560 --> 00:16:06.640 actually need alerts for this earthquake using this alert target. 00:16:08.920 --> 00:16:14.360 So, as we can see with this example, if we choose a lower MMI threshold 00:16:14.360 --> 00:16:17.949 compared to the alert target, we are able to reduce the amount 00:16:17.949 --> 00:16:20.879 of missed alerts, but this does come at the cost of increasing 00:16:20.879 --> 00:16:24.679 the number of correct alerts. And, depending on the location 00:16:24.680 --> 00:16:28.000 of the earthquake relative to large population centers, 00:16:28.000 --> 00:16:31.589 the majority of the population that requires alerts may not be 00:16:31.589 --> 00:16:35.279 near the [audio cuts out] for the Ridgecrest earthquakes. 00:16:35.279 --> 00:16:38.921 And so, because of that, that has implications on what MMI threshold 00:16:38.921 --> 00:16:46.220 may be the preferred alerting strategy for a given earthquake. 00:16:46.220 --> 00:16:51.680 And, as such, we shouldn’t make decisions about the preferred 00:16:51.689 --> 00:16:54.920 alerting strategy using just one or two earthquakes. 00:16:54.920 --> 00:17:00.589 So here, we will use this MMI threshold framework to then perform 00:17:00.589 --> 00:17:04.900 a systematic assessment of the alerting strategies for different alert 00:17:04.900 --> 00:17:08.040 targets using a large catalog of earthquakes. 00:17:10.360 --> 00:17:13.580 And so, for this assessment, we will consider alert targets 00:17:13.589 --> 00:17:19.089 between MMI 4 through MMI 6. So here is a table that shows the 00:17:19.089 --> 00:17:24.589 descriptions used to define these different MMI levels for USGS 00:17:24.589 --> 00:17:29.430 Did You Feel It reports. And so, MMI 4 shaking is categorized 00:17:29.430 --> 00:17:34.420 as light shaking, but despite not causing damage or falling objects, 00:17:34.420 --> 00:17:39.250 most people still describe the shaking as strong shaking. 00:17:39.250 --> 00:17:43.460 And most people in these regions do feel shaking from the earthquake. 00:17:44.220 --> 00:17:50.140 MMI 4.5 and 5.0 shaking fall within the category of moderate shaking. 00:17:50.140 --> 00:17:53.800 And here, we see objects starting to fall from shelves as well as 00:17:53.800 --> 00:17:56.900 minor damage to walls. And then people can also have 00:17:56.900 --> 00:18:00.700 difficulty standing or walking when experiencing this level of shaking. 00:18:00.700 --> 00:18:04.880 So it’s important that people take protective actions at this intensity, 00:18:04.880 --> 00:18:09.900 such as drop, cover, hold on, which is illustrated down here. 00:18:09.900 --> 00:18:14.400 And then finally, MMI 5.5 and 6.0 shaking are within the category of 00:18:14.400 --> 00:18:19.820 strong shaking, where we see most objects falling from shelves, windows 00:18:19.820 --> 00:18:25.720 can start to break, and then damage to unreinforced masonry can occur. 00:18:28.580 --> 00:18:32.120 And so, with our alerting targets, we will also be considering 00:18:32.120 --> 00:18:37.690 a range of MMI thresholds. We will be considering MMI thresholds 00:18:37.690 --> 00:18:44.190 that vary between zero to 2 MMI units below the MMI target. 00:18:44.190 --> 00:18:47.950 And so, with these different alerting strategies and our given alert targets, 00:18:47.950 --> 00:18:52.980 this table here shows the specific MMI thresholds that we will be testing. 00:18:53.720 --> 00:18:59.000 And, because we are able to look at these specific MMI thresholds, 00:18:59.010 --> 00:19:03.250 this approach allows us to examine the effectiveness of different 00:19:03.250 --> 00:19:05.770 alert strategies used by the ShakeAlert system. 00:19:05.770 --> 00:19:09.290 So, once again, alerts sent through the WEA system 00:19:09.290 --> 00:19:15.740 use the MMI 3.5 alert region. Cell phone apps use MMI 2.5, 00:19:15.740 --> 00:19:21.580 and then Google and Android use a combination of MMI 2.5 and MMI 4.5. 00:19:23.420 --> 00:19:26.360 [Silence] 00:19:26.360 --> 00:19:30.440 And so, other assumptions that we are using for our analysis. 00:19:30.440 --> 00:19:36.150 So, once again, as I outlined in the example from earlier, we are assuming 00:19:36.150 --> 00:19:41.550 that the magnitude and the epicenter are estimated accurately by ShakeAlert, and 00:19:41.550 --> 00:19:46.890 we only considered the point-source- based alert regions for this analysis. 00:19:46.890 --> 00:19:50.490 The ground motion models that we use here are a little bit different from what 00:19:50.490 --> 00:19:55.040 ShakeAlert uses operationally right now, but these are similar to the ground 00:19:55.040 --> 00:19:59.700 motion models used in the USGS NEIC ShakeMap procedures. 00:19:59.700 --> 00:20:04.880 So we used the NGA-West2 GMPEs to find the average PGA 00:20:04.880 --> 00:20:08.130 and PGV with distance. And then we used the 00:20:08.130 --> 00:20:13.890 Worden et al. 2012 GMICE to then estimate MMI from 00:20:13.890 --> 00:20:16.500 these average peak ground motions. 00:20:16.500 --> 00:20:21.210 So, once we have an alert region, we will not approximate the 00:20:21.210 --> 00:20:24.220 alert region as an octagon. This is done operationally 00:20:24.220 --> 00:20:29.380 by ShakeAlert to help improve the speed that an alert can go out. 00:20:29.380 --> 00:20:31.220 So we will keep it as a circular region, 00:20:31.220 --> 00:20:34.860 but we will truncate the alerts at the state boundaries. 00:20:36.360 --> 00:20:41.560 So, because of that, the alerts will not extend beyond California, Oregon, 00:20:41.560 --> 00:20:46.100 and Washington for our analysis. And then finally, we are only 00:20:46.100 --> 00:20:49.820 considering the spatial distribution of shaking in our alert quality analysis 00:20:49.820 --> 00:20:55.040 for right now. So we are not considering alert timeliness at this moment. 00:20:56.780 --> 00:20:59.020 [Silence] 00:20:59.020 --> 00:21:05.280 So here is some figures showing the data set – the data set that we are using. 00:21:05.280 --> 00:21:10.400 We are using ShakeMaps from 143 magnitude 5 to 7.3 earthquakes 00:21:10.400 --> 00:21:15.480 that have occurred between 1980 and, I believe, June of this year. 00:21:15.480 --> 00:21:21.060 So here in our map we have all of the earthquakes that are sized and 00:21:21.060 --> 00:21:25.430 colored according to magnitude. And then these middle plots, 00:21:25.430 --> 00:21:30.700 the top row is the earthquake timeline, and then the earthquake magnitude 00:21:30.700 --> 00:21:37.000 distribution, and then finally, the cumulative distribution 00:21:37.010 --> 00:21:41.980 of MMI along the ShakeMap boundaries for this case. 00:21:41.980 --> 00:21:47.540 So all of the earthquakes in our catalog have produced MMI 4.0 00:21:47.540 --> 00:21:51.480 and larger shaking within California, Oregon, and/or Washington. 00:21:51.480 --> 00:21:57.180 And all of these ShakeMaps have been processed using the 00:21:57.180 --> 00:22:02.110 current ShakeMap software. And, where available, finite fault 00:22:02.110 --> 00:22:05.680 slip information, ground motion observations from seismic sensors, 00:22:05.680 --> 00:22:07.830 and Did You Feel It reports are incorporated into the 00:22:07.830 --> 00:22:11.260 ShakeMap for a given earthquake. 00:22:11.260 --> 00:22:16.220 And so, looking at the bottom plot that shows the cumulative distribution of 00:22:16.220 --> 00:22:21.520 MMI along the edges of the ShakeMap, we can see that nearly all ShakeMaps 00:22:21.520 --> 00:22:28.360 have edges with MMI less than 4.0. And so this means that the MMI 00:22:28.360 --> 00:22:32.620 distribution for the alert targets that we are analyzing will be 00:22:32.620 --> 00:22:37.430 complete for this catalog. And so this is important because, 00:22:37.430 --> 00:22:43.860 that way, we won’t be underestimating the extent of shaking at our targets. 00:22:43.860 --> 00:22:49.100 And so we also assume that any location outside of the ShakeMap 00:22:49.100 --> 00:22:53.200 boundary for a given earthquake is unlikely to have felt shaking. 00:22:55.640 --> 00:22:59.020 And so here is a slide that details our methods. 00:22:59.020 --> 00:23:03.390 So, for each earthquake in our catalog, we will use its ShakeMap to compute 00:23:03.390 --> 00:23:08.180 the cumulative MMI distribution as a function of the epicentral distance, 00:23:08.180 --> 00:23:13.110 which is also the alert distance, in terms of both area and population. 00:23:13.110 --> 00:23:15.980 And then we can use these MMI distributions to then compute the 00:23:15.980 --> 00:23:20.930 alert quality as a function of alert distance for a given alert target. 00:23:20.930 --> 00:23:25.780 And so, this table here shows our definitions for alert quality. 00:23:25.780 --> 00:23:32.360 So correct and missed alerts correspond to locations that experience shaking at 00:23:32.360 --> 00:23:35.960 or above the alert target that are either inside the alert region for a 00:23:35.960 --> 00:23:40.140 correct alert or outside of the alert region, which is a missed alert. 00:23:40.140 --> 00:23:46.200 For an incorrect alert, this has commonly been referred to as a false alert in the 00:23:46.200 --> 00:23:53.320 past, but I am using terminology that’s consistent with a really nice paper from 00:23:53.330 --> 00:23:59.860 Sara McBride and others who came up with a lot of post-alert follow-up 00:23:59.860 --> 00:24:05.660 messages specifically for ShakeAlert. And so they use incorrect alerts when 00:24:05.660 --> 00:24:12.300 an earthquake has occurred, but an alert is sent to an area where 00:24:12.300 --> 00:24:18.150 shaking at that location is actually below the alert threshold for this case. 00:24:18.150 --> 00:24:23.070 And then, for these incorrect alerts, we can additionally categorize these 00:24:23.070 --> 00:24:26.890 in terms of whether or not they are likely to still feel some shaking from 00:24:26.890 --> 00:24:29.670 the earthquake and therefore have immediate confirmation that an 00:24:29.670 --> 00:24:32.430 earthquake has occurred. And so, for our analysis, 00:24:32.430 --> 00:24:39.200 we will use MMI 3.0 as our threshold for whether or not people may 00:24:39.200 --> 00:24:43.160 still feel some shaking or may be unlikely to feel shaking. 00:24:43.160 --> 00:24:47.470 And then finally, a correct no alert is a location that experiences shaking 00:24:47.470 --> 00:24:51.120 below the alert target that is outside of the alert region. 00:24:52.080 --> 00:24:56.380 So, for each of our different alerting strategies, we will then evaluate 00:24:56.390 --> 00:25:00.900 the alert quality at the alert distance generated by that strategy. 00:25:00.900 --> 00:25:05.340 And so we will be comparing the percent of missed versus correct alerts, 00:25:05.340 --> 00:25:09.030 the total number of incorrect alerts relative to the amount of necessary 00:25:09.030 --> 00:25:12.280 alerts for that earthquake, and then the percentage of 00:25:12.280 --> 00:25:17.480 the alert region that is incorrect alerts versus correct alerts. 00:25:20.120 --> 00:25:23.620 And so, once again, here is an illustrative example of our 00:25:23.630 --> 00:25:28.690 methods using the magnitude 7.1 Ridgecrest earthquake. 00:25:28.690 --> 00:25:31.750 And so, for each earthquake in our catalog, we will take the 00:25:31.750 --> 00:25:35.460 ShakeMap shown here, and then we will compute the cumulative MMI 00:25:35.460 --> 00:25:39.920 distribution as a function of the alert distance. So that is shown here. 00:25:39.920 --> 00:25:45.120 And we have it both in terms of area and then population. 00:25:45.120 --> 00:25:49.040 And so, for a given alert distance – say 300 kilometers, right here where 00:25:49.040 --> 00:25:58.140 my mouse is – this shows the total amount of different MMI values 00:25:58.150 --> 00:26:02.940 within the alert region for that radius. And then the dotted lines here on 00:26:02.940 --> 00:26:08.260 the map shows the maximum radius considered in these plots. 00:26:08.260 --> 00:26:14.980 And we see that, as we increase the alert distance, the alert region expands. 00:26:14.980 --> 00:26:21.140 And we actually see a pretty smooth increase in area, but for population, 00:26:21.140 --> 00:26:25.850 we see, for the Ridgecrest earthquake, that there’s a big increase here 00:26:25.850 --> 00:26:29.880 when we start to include the Los Angeles area. 00:26:32.500 --> 00:26:36.880 And so we can then take this ShakeMap and then compute the 00:26:36.880 --> 00:26:40.470 alert quality for a given alert target. So, once again, here we have 00:26:40.470 --> 00:26:44.710 MMI 4.0 for our example. And so, all of those regions 00:26:44.710 --> 00:26:48.700 that experience MMI 4 and higher are shown in the green. 00:26:48.700 --> 00:26:52.930 And then we have our cumulative MMI distribution maps that are now 00:26:52.930 --> 00:27:00.260 colored by the different alert quality. So we can see the tradeoff between 00:27:00.260 --> 00:27:04.790 missed and correct alerts and that, as we increase our alert distance, 00:27:04.790 --> 00:27:08.440 we are expanding our alert region and including more 00:27:08.440 --> 00:27:12.710 correct alerts, which is great. But we also see that we are also 00:27:12.710 --> 00:27:17.840 including more incorrect alerts, which are shown in the yellow. 00:27:17.840 --> 00:27:22.280 And so, with this, we can then take our different alert regions for our 00:27:22.280 --> 00:27:27.030 different alert thresholds, and then we can evaluate the alert quality 00:27:27.030 --> 00:27:29.480 at those different locations. So, on the map, these are 00:27:29.480 --> 00:27:34.600 shown by the circles. And then their different alert distances 00:27:34.600 --> 00:27:40.090 for these thresholds are shown by the vertical lines on the alert quality plots. 00:27:40.090 --> 00:27:47.270 And then finally, we can also look at the incorrect alerts in terms of likely 00:27:47.270 --> 00:27:53.890 still felt versus likely not felt shaking. And so the likely not felt shaking 00:27:53.890 --> 00:27:58.260 is shown in the orange here. And then, when I do this analysis, 00:27:58.260 --> 00:28:03.120 I am assuming that the regions that are outside of the ShakeMap, 00:28:03.120 --> 00:28:06.990 which are shown in gray on these distribution plots – 00:28:06.990 --> 00:28:10.900 I am assuming that these are also unlikely to feel shaking. 00:28:10.900 --> 00:28:15.940 So, for this analysis, the gray would be included with the orange. 00:28:18.000 --> 00:28:20.840 [Silence] 00:28:20.840 --> 00:28:27.500 So, before I show the combined analysis for all of the earthquakes in our catalog, 00:28:27.500 --> 00:28:31.300 I’ll just quickly show you some examples from some 00:28:31.310 --> 00:28:35.070 individual earthquakes. And so these are all listed here. 00:28:35.070 --> 00:28:41.720 And so I will show the ShakeMap for this earthquake plus plots like this. 00:28:41.720 --> 00:28:50.540 So this is the alert quality total plots for the magnitude 7 Ridgecrest earthquake. 00:28:50.540 --> 00:28:55.690 So the column here shows the correct alert percentages in terms of area at the 00:28:55.690 --> 00:28:59.630 top and then population at the bottom. And then this one here shows the 00:28:59.630 --> 00:29:05.470 amount of over-alerting for these different alert strategies. 00:29:05.470 --> 00:29:10.070 And so we have these in terms of the alert total – so the total area 00:29:10.070 --> 00:29:13.190 that is over-alerted and then the total population 00:29:13.190 --> 00:29:16.040 that is over-alerted, and then – whoop. 00:29:16.720 --> 00:29:22.360 And then the dashed line here shows the total amount of necessary 00:29:22.370 --> 00:29:27.730 alerts for a given alert target. So if – so, for example, here, 00:29:27.730 --> 00:29:30.990 for the Ridgecrest earthquakes, in this case, this blue dot – 00:29:30.990 --> 00:29:34.890 this alerting strategy is below the dashed line. 00:29:34.890 --> 00:29:39.570 So the amount of over-alerts is comparatively less than the 00:29:39.570 --> 00:29:44.020 amount of necessary alerts. But then, if we go to the next 00:29:44.020 --> 00:29:49.520 alerting strategy, we can see that we are alerting more people 00:29:49.520 --> 00:29:53.180 incorrectly compared to correctly for this case. 00:29:54.740 --> 00:29:59.140 And so, first off is the magnitude 6.4 Ridgecrest earthquake. 00:29:59.140 --> 00:30:04.700 This has pretty similar behavior to the magnitude 7.1 earthquake, but we can 00:30:04.700 --> 00:30:14.350 see that, because it doesn’t hit the L.A. area as much with higher intensity 00:30:14.350 --> 00:30:17.920 shaking, we can see that the amount of incorrect alerts is actually 00:30:17.920 --> 00:30:21.900 really significant compared to the amount of necessary alerts. 00:30:23.620 --> 00:30:28.840 And then here is the magnitude 5.4 Brawley earthquake. 00:30:28.840 --> 00:30:33.140 This was from the 2012 swarm in this area. 00:30:33.140 --> 00:30:38.720 So this is also pretty similar characteristics as the Ridgecrest 00:30:38.720 --> 00:30:45.080 earthquakes because this was also in an area that wasn’t as densely populated. 00:30:45.080 --> 00:30:54.320 So we can see that – and so now, here is another magnitude 5.4 00:30:54.320 --> 00:30:57.870 earthquake, but this happened in the Los Angeles area. 00:30:57.870 --> 00:31:01.430 So we can see that, for this case, because the population distribution 00:31:01.430 --> 00:31:05.600 is fairly even, though it’s dense, the behavior of the correct alert 00:31:05.600 --> 00:31:10.330 percentage is actually pretty similar between area and population. 00:31:10.330 --> 00:31:14.530 And so now let’s look at a couple of examples in the Bay Area. 00:31:14.530 --> 00:31:16.760 So this is for the Loma Prieta earthquake. 00:31:16.760 --> 00:31:23.320 And this is a case where the shaking was really amplified in some regions. 00:31:23.320 --> 00:31:26.500 So the alert thresholds that we consider here are actually unable to 00:31:26.500 --> 00:31:33.240 get a high percentage of correct alerts for our MMI 6 alert target. 00:31:33.240 --> 00:31:38.300 And this is also a case where the relative amount of over-alerting 00:31:38.300 --> 00:31:42.460 is actually fairly low compared to the amount of necessary alerts. 00:31:43.840 --> 00:31:50.620 And finally, for our example, here is the 2014 South Napa earthquake. 00:31:50.620 --> 00:31:55.200 And so this is somewhat opposite to that of the Loma Prieta earthquake, 00:31:55.200 --> 00:32:00.710 where we actually see high correct alert percentages are able to be achieved 00:32:00.710 --> 00:32:06.340 at MMI thresholds that are actually closer to the alert target. 00:32:07.420 --> 00:32:12.560 But, however, the number of incorrect alerts ended up being about twice that as 00:32:12.560 --> 00:32:15.620 compared to the Loma Prieta earthquake for this case. 00:32:15.620 --> 00:32:22.600 So, for all of these different individual earthquakes, I hope that you were able 00:32:22.610 --> 00:32:27.260 to see fairly quickly that there is a lot of variation between them. 00:32:27.260 --> 00:32:31.440 So here are the alert quality comparisons for all of the earthquakes 00:32:31.440 --> 00:32:38.080 in our ShakeMap catalog. And so the dots here show the median 00:32:38.080 --> 00:32:42.960 values, where the shaded lines represent plus-and-minus 1 standard deviation 00:32:42.960 --> 00:32:50.160 and spread for our catalog. And, as for the individual earthquake 00:32:50.160 --> 00:32:57.110 examples, we see that the correct alerts increase as the MMI threshold decreases 00:32:57.110 --> 00:33:01.640 relative to the alert target, where generally, higher correct 00:33:01.640 --> 00:33:06.460 alert percentages are reached more quickly for the lower alert targets 00:33:06.460 --> 00:33:10.540 compared to the higher MMI alert targets. 00:33:10.540 --> 00:33:15.220 And then, for quantifying the amount of over-alerting – so these figures 00:33:15.220 --> 00:33:20.780 here show the relative amount of over-alerting on a log scale. 00:33:20.780 --> 00:33:25.330 So the value of 1 here shown in the dashed lines, this reflects 00:33:25.330 --> 00:33:31.800 the amount of necessary alerts. So anything below that represents 00:33:31.800 --> 00:33:35.790 over-alerting that is comparatively less than 1 times the amount of 00:33:35.790 --> 00:33:39.941 necessary alerts, while anything higher than this dashed line 00:33:39.941 --> 00:33:44.080 represents over-alerting that is – can be many times more 00:33:44.080 --> 00:33:45.930 than the amount of necessary alerts. 00:33:45.930 --> 00:33:52.610 And we can see here, especially in terms of population, that there’s a very large 00:33:52.610 --> 00:33:57.090 spread in the amount of over-alerting for a given alert strategy. 00:33:57.090 --> 00:34:01.350 And this can actually be at two or three orders of magnitude 00:34:01.350 --> 00:34:06.260 for the lowest MMI thresholds here. 00:34:08.780 --> 00:34:13.060 So another way that we can examine the alert quality 00:34:13.060 --> 00:34:18.280 is by finding the total alert quality for our catalog. 00:34:20.320 --> 00:34:26.460 So, for this, we sum the individual alerting regions for all earthquakes for a 00:34:26.460 --> 00:34:30.889 given MMI target and MMI threshold. And this helps us find the total alert 00:34:30.889 --> 00:34:37.629 quality for that alert strategy. So, for this, I’m going to show some 00:34:37.629 --> 00:34:41.340 slides with a bunch of these plots. So, for a given alert strategy, 00:34:41.340 --> 00:34:46.460 shown here, the X axis will show the total amount of missed alerts 00:34:46.460 --> 00:34:52.771 for a given alert target, while the Y axis will show the percentage 00:34:52.771 --> 00:34:57.410 of alerts that are incorrect alerts. So, of the total region alerted, 00:34:57.410 --> 00:35:03.590 what percentage of this region were to incorrect alerts? 00:35:03.590 --> 00:35:08.369 And so, in this scheme, an ideal alerting strategy will plot close to 00:35:08.369 --> 00:35:12.330 the origins over here, which would indicate a low percentage of missed 00:35:12.330 --> 00:35:16.460 alerts and then also a low amount of incorrect alerts. 00:35:16.460 --> 00:35:21.720 So here is the total alert quality plots for each alerting strategy. 00:35:21.730 --> 00:35:28.740 So, on the left is for MMI thresholds where the threshold is the same as the 00:35:28.740 --> 00:35:33.500 alert target. And then we are decreasing our MMI threshold going to the right. 00:35:33.500 --> 00:35:39.250 So the alert region sizes increase as we go over here. 00:35:39.250 --> 00:35:44.010 And, like before, we see that, as we’ve lowered the MMI threshold 00:35:44.010 --> 00:35:49.400 and expand the alert region, that decreases the missed alerts, 00:35:49.400 --> 00:35:51.880 and we also increase the amount of over-alerting. 00:35:51.880 --> 00:35:56.740 And, in some cases, the percentage of alerts issued that are incorrect 00:35:56.740 --> 00:36:02.510 alerts can be quite high – so over 75% of the alerted region 00:36:02.510 --> 00:36:04.500 for some of these strategies. 00:36:04.500 --> 00:36:09.000 However, if we look at this same percentage in terms of the 00:36:09.000 --> 00:36:13.020 incorrect alerts to regions that are unlikely to feel shaking – 00:36:13.020 --> 00:36:17.880 so these are now shown by the diamonds, we can see that, 00:36:17.880 --> 00:36:22.930 for the alert targets that we are considering here, most alert strategies 00:36:22.930 --> 00:36:27.020 will issue alerts to people who will still feel some shaking. 00:36:27.020 --> 00:36:35.200 And this is especially true for the higher alert targets shown in the warmer colors. 00:36:35.200 --> 00:36:41.880 And we see that, for the lower alert targets in the cooler colors, 00:36:41.880 --> 00:36:48.800 this ends up still being less than 20% of the alerted region. 00:36:51.220 --> 00:36:56.380 Okay, so let’s summarize these results using our tables that list out the specific 00:36:56.380 --> 00:37:01.140 MMI thresholds that we consider for these different alert targets. 00:37:01.140 --> 00:37:06.140 So, in the top row here, I will be filling out the tables with colors 00:37:06.140 --> 00:37:10.940 that correspond to the correct alert percentages from the total alert quality. 00:37:10.940 --> 00:37:13.930 And then in terms of area and then population. 00:37:13.930 --> 00:37:17.140 So these will be using this color scale here. 00:37:17.140 --> 00:37:22.020 And then, in the bottom row, I will be doing the same thing, 00:37:22.020 --> 00:37:28.580 where I will fill in the colors according to the amount of incorrect alerts. 00:37:28.580 --> 00:37:33.869 So I’ll be using this color scale here. And so, with these two different 00:37:33.869 --> 00:37:39.500 color scales, blue is good, while red is not that good. 00:37:40.400 --> 00:37:44.300 And so here are results for the correct alert percentage. 00:37:44.300 --> 00:37:49.460 And, as expected from our example for the magnitude 7.1 Ridgecrest 00:37:49.460 --> 00:37:55.040 earthquake, when the MMI thresholds are the same as the alert target, 00:37:55.040 --> 00:37:59.870 not all regions that need alerts will be included inside the alert region. 00:37:59.870 --> 00:38:04.230 And so, for all of the alert targets that we consider here, 00:38:04.230 --> 00:38:07.570 this alert strategy actually appears to get only up to 00:38:07.570 --> 00:38:13.640 about 25% of correct alerts in terms of both area and population. 00:38:13.640 --> 00:38:19.810 And, while there is some variation between area and population, 00:38:19.810 --> 00:38:30.910 the patterns are pretty similar. So we see that, for MMI thresholds, 00:38:30.910 --> 00:38:35.840 we would need to have at least one MMI unit threshold below 00:38:35.840 --> 00:38:39.600 the alert target to achieve high correct alert percentages. 00:38:39.600 --> 00:38:44.210 And then, as we increase our alerting strategies and alert targets – 00:38:44.210 --> 00:38:51.480 so once again, at the MMI thresholds that are similar to the alert target, 00:38:51.480 --> 00:38:55.050 we see that there’s not a lot of over-alerting. 00:38:55.050 --> 00:38:59.000 And that’s because these regions aren’t often large enough to include 00:38:59.000 --> 00:39:03.020 most of the regions that need alerts. So, because of that, there’s not a lot 00:39:03.020 --> 00:39:08.440 of regions that don’t feel shaking at the level that we’re interested in. 00:39:09.080 --> 00:39:16.500 But, as you decrease your MMI thresholds relative to your alert target, 00:39:16.500 --> 00:39:20.080 you see that you do increase the amount of over-alerting. 00:39:20.080 --> 00:39:25.100 And lower MMI thresholds can have very high amounts of 00:39:25.100 --> 00:39:29.220 incorrect alerts compared to the amount of necessary alerts. 00:39:29.220 --> 00:39:32.500 However, if we do look at that in terms of the people who are 00:39:32.500 --> 00:39:37.780 unlikely to feel shaking, our table now looks like this. 00:39:38.660 --> 00:39:43.440 And so we see that most alerting strategies send alerts 00:39:43.440 --> 00:39:45.880 to people who are likely to feel some shaking. 00:39:45.880 --> 00:39:52.460 And I think this is just fine in terms of the amount of over-alerting. 00:39:52.460 --> 00:39:59.070 Because one of the main comments that we saw after the Ridgecrest earthquakes 00:39:59.070 --> 00:40:05.220 was that people in the Los Angeles area who did not experience damaging levels 00:40:05.220 --> 00:40:11.060 of shaking, they said that they still would have appreciated receiving 00:40:11.060 --> 00:40:15.670 an early warning alert for the levels of shaking that they did feel. 00:40:15.670 --> 00:40:21.040 So I think this is just fine, personally. 00:40:21.740 --> 00:40:23.920 And so, with these results, we can pick our preferred 00:40:23.930 --> 00:40:26.730 alerting strategies, which are shown here in the table. 00:40:26.730 --> 00:40:33.881 So, for MMI 4 and 4.5, having an alert threshold that is one MMI unit 00:40:33.881 --> 00:40:40.380 below this seems be pretty good. And then, as we increase our alert 00:40:40.380 --> 00:40:49.080 target, the MMI threshold also decreases to around 1.5 MMI units 00:40:49.080 --> 00:40:53.630 below the target for MMI 5, and then around 2 MMI units 00:40:53.630 --> 00:40:56.870 below the target for MMI 5.5 and 6. 00:40:56.870 --> 00:41:02.340 And these alerting strategies correspond with the specific 00:41:02.340 --> 00:41:06.040 MMI thresholds shown here on this column. 00:41:06.040 --> 00:41:14.840 And lower MMI thresholds than these values will have around the same, 00:41:14.850 --> 00:41:18.820 if not slightly better, amounts of correct alerting. 00:41:18.820 --> 00:41:23.490 But they will have more over-alerting. While higher MMI thresholds than these 00:41:23.490 --> 00:41:28.720 values will have more missed alerts for these specific alert targets here. 00:41:28.720 --> 00:41:34.770 And it’s important to keep in mind that, in this case, most missed alerts will 00:41:34.770 --> 00:41:40.050 tend to be for locations that experience shaking near the alert target. 00:41:40.050 --> 00:41:45.510 So the near-source, or near-epicentral regions that experience the highest 00:41:45.510 --> 00:41:52.230 MMI levels, these will still be included in the smaller alert regions, 00:41:52.230 --> 00:41:54.810 with the caveat that some of these locations might be 00:41:54.810 --> 00:42:00.780 in the late-alert zone when alert timeliness is considered. 00:42:00.780 --> 00:42:08.430 And then incorporation of finite fault information from ShakeAlert’s 00:42:08.430 --> 00:42:14.700 FinDer algorithm will likely improve the alert quality for MMI 4.5 and higher 00:42:14.700 --> 00:42:18.680 thresholds, especially for the larger magnitude events 00:42:18.690 --> 00:42:23.050 where the alert distances will then be in terms of the 00:42:23.050 --> 00:42:26.390 rupture distance instead of just the epicentral distance. 00:42:26.390 --> 00:42:29.840 And then, once again, we are assuming here that the magnitude 00:42:29.840 --> 00:42:35.040 is estimated accurately and it’s consistent with the catalog magnitude. 00:42:35.040 --> 00:42:39.420 And, while underestimating the magnitude will underestimate the size 00:42:39.420 --> 00:42:43.990 of the alert region, it’s possible that ShakeAlert may overestimate 00:42:43.990 --> 00:42:48.740 the magnitude, and so the alert regions will be larger. 00:42:48.740 --> 00:42:52.280 But there will be variation in what happens with that. 00:42:54.700 --> 00:42:58.840 So this analysis only covered part of the picture. 00:42:58.840 --> 00:43:04.800 And, for better determination of the original alerting strategies – or, of the 00:43:04.800 --> 00:43:09.660 optimal alerting strategies, we do need to consider the timing of the alerts. 00:43:09.660 --> 00:43:14.620 And there are several factors that affect the alert timeliness, 00:43:14.630 --> 00:43:17.750 and ultimately warning time, for a given location. 00:43:17.750 --> 00:43:20.800 This would be the time it takes to estimate the magnitude 00:43:20.800 --> 00:43:25.480 of the earthquake, which is illustrated here in this figure 00:43:25.480 --> 00:43:30.090 by Daniel Trugman and others from their 2019 paper. 00:43:30.090 --> 00:43:35.541 Other factors include the station distribution relative to the earthquake 00:43:35.541 --> 00:43:41.140 location and then data telemetry latencies, alert delivery latencies, and 00:43:41.140 --> 00:43:45.900 when shaking at the MMI target level ultimately arrives at that location. 00:43:47.160 --> 00:43:52.070 So right now, I’m currently in the process of estimating alert times 00:43:52.070 --> 00:43:55.580 for the different alerting strategies for the earthquakes in our catalog. 00:43:55.580 --> 00:44:03.440 And, like how using a lower MMI threshold increases the number of 00:44:03.440 --> 00:44:09.700 correct alerts for a given alert target, studies like Minson et al. in 2018 have 00:44:09.700 --> 00:44:14.760 shown that, for cases where a location will experience high MMI during a 00:44:14.770 --> 00:44:18.580 large-magnitude earthquake, those locations may get alerted 00:44:18.580 --> 00:44:23.580 faster if the MMI threshold is lower. Because they will be included 00:44:23.580 --> 00:44:27.600 inside the alert region at a lower magnitude estimate. 00:44:28.600 --> 00:44:34.680 And I think conducting a similar analysis using our catalog will help 00:44:34.690 --> 00:44:38.700 determine if the preferred alerting strategies, in terms of the spatial 00:44:38.700 --> 00:44:41.390 distribution, will also be sufficient for providing 00:44:41.390 --> 00:44:46.300 timely alerts for these different alert targets. 00:44:48.020 --> 00:44:53.080 And so, to summarize what I have covered today, I hope I have 00:44:53.080 --> 00:44:56.990 demonstrated that the median expected MMI estimates will 00:44:56.990 --> 00:45:01.670 not be able to produce alert regions that include all of the locations that 00:45:01.670 --> 00:45:05.920 experience shaking at the MMI threshold level used to generate the alert. 00:45:05.920 --> 00:45:11.340 And, to remedy this, we want to make sure that either the MMI threshold 00:45:11.350 --> 00:45:15.011 is lower than the alert target or that the ground motion model 00:45:15.011 --> 00:45:18.990 uncertainty is incorporated into the alert region computation. 00:45:18.990 --> 00:45:22.310 And so, for this latter strategy, once again, I encourage you to 00:45:22.310 --> 00:45:27.340 check out our paper that looks at the Ridgecrest earthquakes. 00:45:27.340 --> 00:45:34.340 And so here we used a large ShakeMap catalog of magnitude 5 to 7 earthquakes 00:45:34.350 --> 00:45:38.900 to examine different MMI threshold alerting strategies for a range of 00:45:38.900 --> 00:45:43.610 alert targets between MMI 4 and 6. And, with this analysis, we found 00:45:43.610 --> 00:45:49.030 that a threshold of MMI 3.5, which is the current threshold 00:45:49.030 --> 00:45:54.500 that defines where ShakeAlert messages are delivered through the WEA system – 00:45:54.500 --> 00:45:59.340 we found that this threshold can include most regions that experience potentially 00:45:59.340 --> 00:46:03.580 damaging levels of shaking – so MMI 4.5 and larger – 00:46:03.580 --> 00:46:07.820 without significant over-alerting to people who are unlikely to feel shaking. 00:46:07.820 --> 00:46:10.290 And then, once again, this assumes accurate magnitude 00:46:10.290 --> 00:46:14.440 and epicenter estimation for this analysis. 00:46:14.440 --> 00:46:17.320 And thank you. I’m happy to take any questions. 00:46:17.940 --> 00:46:22.600 - Thank you very much, Jessie. It’s been an excellent talk. 00:46:23.340 --> 00:46:25.180 Let’s take questions. 00:46:25.180 --> 00:46:29.670 So we have Tom Hanks. You raised – he raised his hand. 00:46:29.670 --> 00:46:32.350 Tom, do you have a question? Please unmute your microphone 00:46:32.350 --> 00:46:36.560 and ask or – yeah, just unmute your microphone and ask, please. 00:46:36.560 --> 00:46:40.600 - Okay. I’m – thank you, Noha. I’m not quite sure how I raised 00:46:40.600 --> 00:46:49.220 my hand, but I do have a question. And that is, 45 minutes ago, Jess, 00:46:49.220 --> 00:46:56.820 you showed a plot of MMI versus PGV or PGA. 00:46:56.820 --> 00:46:59.520 - Mm-hmm. - Early in the talk. 00:46:59.520 --> 00:47:03.520 - Yes. - And there’s a break in slope, 00:47:03.520 --> 00:47:07.579 what, about MMI 4? - Mm-hmm. 00:47:07.579 --> 00:47:11.490 - And what causes that break in slip? Why … 00:47:11.490 --> 00:47:14.340 - That is a good question. 00:47:16.640 --> 00:47:22.940 So I know that when – let me find the slide in question. 00:47:22.950 --> 00:47:27.099 Okay, so this is the figure that you’re referring to, Tom. 00:47:27.099 --> 00:47:35.770 And so this shows the – so this shows – in the purple, these are – I believe these 00:47:35.770 --> 00:47:42.630 are data from Worden et al. 2012 where they used co-located 00:47:42.630 --> 00:47:48.450 Did You Feel It MMI data with PGV and PGA observations 00:47:48.450 --> 00:47:54.630 from seismic sensors. And so, when they developed their 00:47:54.630 --> 00:48:03.380 GMICE, they used a similar piecewise function that was 00:48:03.380 --> 00:48:14.750 used by Wald et al. 1999. And so I don’t remember if there 00:48:14.750 --> 00:48:20.910 was additional reasoning behind choosing that piecewise function, 00:48:20.910 --> 00:48:27.140 other than looking at the data and seeing that there was a change in slope. 00:48:28.660 --> 00:48:30.560 - Okay. - Tom? 00:48:31.520 --> 00:48:32.960 Hey, Tom, let me add to that. I mean … 00:48:32.960 --> 00:48:35.860 - Oh, yes, thank you. [laughs] - The data is … 00:48:35.860 --> 00:48:39.480 - Who are you? - [laughs] This is Dave Wald. 00:48:39.480 --> 00:48:42.840 Hi, Tom. The data show this, 00:48:42.840 --> 00:48:45.790 and that’s why we did the regression with the break. 00:48:45.790 --> 00:48:49.700 But really, fundamentally, there’s no reason to think that 00:48:49.700 --> 00:48:53.390 equal steps of intensity – integer intensity units – 00:48:53.390 --> 00:48:56.740 would have equal steps in ground motion thresholds. 00:48:56.740 --> 00:49:00.340 There’s no reason that intensity was really defined based on 00:49:00.340 --> 00:49:04.490 a continuous increases, like log 2 of ground motion. 00:49:04.490 --> 00:49:10.660 So it’s not surprising that, at some place, that linearity just breaks down. 00:49:11.640 --> 00:49:13.940 There’s no physical reason for it, though. 00:49:13.940 --> 00:49:19.240 - Yeah. I guess I’m still interested. That’s a – that’s a very sharp break. 00:49:19.240 --> 00:49:26.900 And why is it there? Why isn’t it 3 or 4 or 5? 00:49:26.900 --> 00:49:29.220 I mean, I understand that you did it – 00:49:29.220 --> 00:49:33.670 I understand what’s going on empirically, I just … 00:49:33.670 --> 00:49:36.280 - It’s when things start falling off shelves. 00:49:36.280 --> 00:49:40.680 And that has a different threshold than felt. 00:49:40.680 --> 00:49:44.770 And so those are the nuances. It’s a continuous observations, 00:49:44.770 --> 00:49:47.910 but then you start adding new features of observations, 00:49:47.910 --> 00:49:50.460 like things falling off shelves. 00:49:52.720 --> 00:49:54.820 [Silence] 00:49:54.820 --> 00:49:58.720 - Yeah. I kind of feel I’m falling off the shelf myself. 00:49:58.720 --> 00:50:01.580 [laughter] 00:50:01.580 --> 00:50:06.940 All right. Well, let’s have a little bit more thought about that break, Dave. 00:50:06.940 --> 00:50:09.200 You know, you just can’t do something like 00:50:09.200 --> 00:50:13.400 drop that on us and not explain it, really. 00:50:14.840 --> 00:50:18.480 This is not your fault, Jessie. [laughter] 00:50:19.880 --> 00:50:22.900 - It’s my fault. - Yeah. Yeah. It’s all his fault. 00:50:22.900 --> 00:50:29.520 Okay. Thanks. And – okay. That was my question. 00:50:29.520 --> 00:50:33.600 - Thanks, Tom, and thanks, Dave. - Jessie, you have a question 00:50:33.609 --> 00:50:38.090 from Alex Grant. Great talk. Do you mind [audio cuts out] 00:50:38.090 --> 00:50:41.450 differences you found between these two alerting strategies – 00:50:41.450 --> 00:50:46.109 lower MMI versus adding uncertainty? He said, I’ll do the homework 00:50:46.109 --> 00:50:50.100 reading after. - Mm-hmm. So … 00:50:51.520 --> 00:50:56.360 I would say they’re – let me find … 00:50:58.640 --> 00:51:02.580 They’re basically functionally equivalent. 00:51:02.580 --> 00:51:11.440 It’s just, for the Option 1 here – so, for the choosing a different MMI threshold, 00:51:11.440 --> 00:51:17.720 that would just be stepping through, in this case, MMI increments of 0.5. 00:51:17.720 --> 00:51:23.490 And then, for the different ground motion model uncertainties, that would 00:51:23.490 --> 00:51:33.760 require knowing what the combined standard deviation is for the different 00:51:33.760 --> 00:51:42.520 ground motion models that we’re using for these alert region calculations. 00:51:42.520 --> 00:51:49.950 And the different ground motion models that I’m using right now, 00:51:49.950 --> 00:51:53.980 they’re a little bit different than what we used for our Ridgecrest paper. 00:51:53.980 --> 00:52:02.160 And I haven’t been able to have the time to calculate what the specific standard 00:52:02.160 --> 00:52:07.100 deviation is for these specific ground motion models that I use here. 00:52:07.100 --> 00:52:14.160 But, for our Ridgecrest paper, we found that this was around, 00:52:14.160 --> 00:52:25.120 I believe, 0.7 or 0.8 MMI units. So one standard deviation would be 00:52:25.120 --> 00:52:31.640 kind of a little bit in between 1 to 2 of these steps here. 00:52:35.800 --> 00:52:41.420 - Jessie, you have another question from [inaudible]. 00:52:41.420 --> 00:52:44.540 Jessie, great presentation. Kudos to you for developing these 00:52:44.540 --> 00:52:49.740 tools to systematically evaluate the alerting accuracy moving forward. 00:52:49.740 --> 00:52:51.780 I just have a minor clarification question. 00:52:51.780 --> 00:52:58.400 In all those – in your example, did you use final ShakeMap – 00:52:58.400 --> 00:53:01.790 fully [inaudible] ShakeMap strong motion data? 00:53:01.790 --> 00:53:07.310 If so, will there be additional inaccuracies introduced with 00:53:07.310 --> 00:53:11.880 first or early versions of ShakeMap without any 00:53:11.880 --> 00:53:14.849 strong motion constraints [inaudible] earthquake? 00:53:14.849 --> 00:53:19.270 It would be interesting to test this on Version 1 unconstrained ShakeMap. 00:53:19.270 --> 00:53:26.880 - Mm-hmm. Yeah. So these would be using the final ShakeMap estimates. 00:53:26.880 --> 00:53:33.210 So, once all of the data comes in for a particular earthquake using all of that 00:53:33.210 --> 00:53:39.540 data to help constrain the ShakeMap for these different cases in our catalog. 00:53:39.540 --> 00:53:46.240 And I agree. It would be interesting to look and see how this compares 00:53:46.240 --> 00:53:50.940 with the other iterations of the ShakeMaps. 00:53:53.060 --> 00:53:55.600 [Silence] 00:53:56.080 --> 00:54:00.660 - Okay, there is another question from Austin – Austin Elliott. 00:54:00.670 --> 00:54:04.630 This is – this is a really clever and interesting analysis. 00:54:04.630 --> 00:54:05.860 Great work, Jessie. 00:54:05.860 --> 00:54:09.290 You expressed that the incorrect alerting might not be as much of 00:54:09.290 --> 00:54:13.359 a problem given that the majority of the area’s population 00:54:13.360 --> 00:54:17.520 correctly alerted still do feel shaking at some level. 00:54:17.520 --> 00:54:21.100 Do you anticipate that the level of tolerance would change with, 00:54:21.119 --> 00:54:23.520 for example, the Android two-mode alerts? 00:54:23.520 --> 00:54:28.630 Do you think the difference between the take action and near alert introduces 00:54:28.630 --> 00:54:35.440 some genuine sensitivity in whether – in whether incorrectness is tolerable? 00:54:35.440 --> 00:54:41.200 - That’s a great question, Austin. And this is something that I’ll have to 00:54:41.200 --> 00:54:45.560 think about a little bit more. But I know … 00:54:47.620 --> 00:54:52.060 Just my personal experience with looking at, say, Twitter when a big 00:54:52.060 --> 00:54:56.560 earthquake happens, or an earthquake happens where an alert is issued, 00:54:56.560 --> 00:55:00.390 I do know that there have been cases – and then Sara McBride has looked at 00:55:00.390 --> 00:55:07.310 this as well – where incorrect alerts are issued to cell phones, 00:55:07.310 --> 00:55:12.400 and then people aren’t entirely sure about what to do yet. 00:55:12.400 --> 00:55:18.060 And improving public education about what to do when receiving 00:55:18.060 --> 00:55:20.839 a ShakeAlert message will definitely help with this. 00:55:20.839 --> 00:55:30.190 But I think, with the two-level alerting strategy that Google Android will use, 00:55:30.190 --> 00:55:36.040 I think that might be quite helpful to people receiving alerts to know that, 00:55:36.040 --> 00:55:39.020 okay, I’m actually very close to this earthquake. 00:55:39.020 --> 00:55:41.570 I definitely need to take action right now. 00:55:41.570 --> 00:55:47.140 Versus, oh, I should take action. But later, if they end up not feeling 00:55:47.140 --> 00:55:52.260 shaking, they know that, from that comparatively lower alert level 00:55:52.260 --> 00:55:55.320 sent to their phone that, okay, maybe I wasn’t close enough 00:55:55.320 --> 00:56:00.500 to this earthquake to feel shaking. And I think another thing that 00:56:00.500 --> 00:56:02.620 would be very helpful, just in general, 00:56:02.620 --> 00:56:08.180 would be trying to add some additional follow-up alert messages. 00:56:08.180 --> 00:56:11.330 And so, once again, Sara McBride has done a great job with 00:56:11.330 --> 00:56:15.090 developing specific messages to send out for different cases. 00:56:15.090 --> 00:56:18.960 So hopefully those will be incorporated in the future. 00:56:21.500 --> 00:56:24.860 - Dave Wald, did you have your hand raised? 00:56:24.860 --> 00:56:26.780 - I had my hand up. Thank you. 00:56:26.790 --> 00:56:29.460 Yeah. This is – Jessie, this really is great stuff. 00:56:29.460 --> 00:56:32.130 I’m impressed, and it really works out well. 00:56:32.130 --> 00:56:34.730 Some questions, though. One is just a clarification. 00:56:34.730 --> 00:56:40.200 I think going from false to incorrect is a nice step forward. 00:56:40.200 --> 00:56:42.910 But I really like the over-alerting as a statement. 00:56:42.910 --> 00:56:46.430 Because I don’t consider that – I don’t think many would 00:56:46.430 --> 00:56:50.589 consider that incorrect. I mean, I think that over-alerting 00:56:50.589 --> 00:56:55.160 terminology should stick because, especially for a large earthquake, 00:56:55.160 --> 00:56:59.040 if there’s a large event, and people were alerted, and they didn’t feel it, 00:56:59.040 --> 00:57:01.400 I still think they’ll declare success. 00:57:01.400 --> 00:57:05.720 So that’s just terminology – you know, semantics, I think. 00:57:07.280 --> 00:57:12.460 The main – the main thing that I wanted to mention is that you have a really 00:57:12.470 --> 00:57:16.210 big challenge here because the lower intensities are just fundamentally 00:57:16.210 --> 00:57:20.400 more ambiguous than higher intensities. At intensity 6, nobody will – 00:57:20.400 --> 00:57:23.560 there’s no such thing as over-alerting. Everyone will experience it. 00:57:23.560 --> 00:57:28.660 They’ll experience it differently. But at lower thresholds – intensity 3, 00:57:28.660 --> 00:57:31.940 especially with intensity 2, and even intensity 4, you know, 00:57:31.940 --> 00:57:34.870 only a fraction of the population experiences it. 00:57:34.870 --> 00:57:38.170 And it’s very situational. People in a high-rise will 00:57:38.170 --> 00:57:42.089 feel something, and people in low-rises won’t. 00:57:42.089 --> 00:57:46.140 And so, you know, there’s fundamental ambiguity on whether they should have 00:57:46.140 --> 00:57:49.090 experienced it or not. [chuckles] So how you declare success there 00:57:49.090 --> 00:57:51.290 becomes fairly ambiguous. - Yeah. 00:57:51.290 --> 00:57:55.310 - And that’s – you know, that’s all a manifestation of wanting to 00:57:55.310 --> 00:57:59.430 go to lower alert thresholds. I mean, the Japanese use intensity 6 00:57:59.430 --> 00:58:03.780 with – 6 to 7, effectively, intensity. And we’re looking at a much lower 00:58:03.780 --> 00:58:08.440 intensity where defining success is going to be really difficult. 00:58:08.440 --> 00:58:11.860 And you showed that it – you know, from event to event, even with this – 00:58:11.860 --> 00:58:15.940 you know, an optimized threshold, you have different successes – 00:58:15.940 --> 00:58:20.720 quote, unquote, successes. So that’s why this is so challenging is 00:58:20.720 --> 00:58:25.240 because we’ve gone to a lower intensity rather than unambiguous intensities. 00:58:25.240 --> 00:58:28.600 - Yeah. And that’s actually the – one of the main reasons why I made 00:58:28.600 --> 00:58:35.130 sure that, when I was calculating alert quality, I was making sure to do that 00:58:35.130 --> 00:58:41.430 in terms of the specific alert target. Because, if you calculate alert quality 00:58:41.430 --> 00:58:46.440 in terms of just your alert thresholds, then you’ll just continue to get the same 00:58:46.440 --> 00:58:54.000 thing that we saw in the beginning of my talk, which was, as you lower your 00:58:54.000 --> 00:58:58.560 alert target by lowering the threshold in the same way, you’re still going 00:58:58.560 --> 00:59:03.369 to be missing a lot of alerts for people who experience that intensity 00:59:03.369 --> 00:59:09.369 shaking at the MMI threshold. So that’s why I made sure that the 00:59:09.369 --> 00:59:13.420 alert target, and then therefore the alert quality, was separate 00:59:13.420 --> 00:59:19.940 from the actual MMI that we were using to create the alert. 00:59:19.940 --> 00:59:25.400 And, yeah, I agree. The shaking intensities that 00:59:25.400 --> 00:59:29.800 we are currently looking at for ShakeAlert for the different targets, 00:59:29.800 --> 00:59:37.410 these are lower than in Japan. But I think, if I remember correctly, 00:59:37.410 --> 00:59:42.810 there have been people who commented in Japan that, because they have so 00:59:42.810 --> 00:59:46.830 many more large earthquakes compared to here, if they were to 00:59:46.830 --> 00:59:51.580 lower their alert target, people would be getting alerts all the time. 00:59:51.580 --> 00:59:55.980 And so they decided that, for their early warning system, 00:59:55.980 --> 01:00:02.900 they do prefer having that higher alert target and then higher thresholds. 01:00:04.760 --> 01:00:08.820 - Yeah. Well, they have a very clear break where they’re alerting where 01:00:08.820 --> 01:00:14.490 there’s potential for life safety issues. And when we say 4-1/2 is the beginning, 01:00:14.490 --> 01:00:20.130 or potential for damage, really, it’s 5-1/2 where you get any significant damage. 01:00:20.130 --> 01:00:23.640 And it’s not until 6-1/2 before you get any injuries. 01:00:23.640 --> 01:00:27.670 So, you know, we’ve lowered the bar on what people get alerts for. 01:00:27.670 --> 01:00:31.570 And it’s not about life safety anymore. It’s about communication. 01:00:31.570 --> 01:00:34.620 - Yeah. - Just, last comment, is that, 01:00:34.620 --> 01:00:36.730 you know, when – you mentioned going and using timing. 01:00:36.730 --> 01:00:39.190 And I think that’s going to be very important because there’s going to 01:00:39.190 --> 01:00:43.650 be a lot of people that will be not in the successful alert because 01:00:43.650 --> 01:00:49.820 they’ll be in the no-alert zone. And that calculation, in addition 01:00:49.820 --> 01:00:52.720 to using uncertainty in the magnitude and location, 01:00:52.720 --> 01:00:55.420 is going to change things considerably. 01:00:55.420 --> 01:00:58.240 You know, you’ve assumed the magnitude and location are correct. 01:00:58.240 --> 01:01:01.460 But that’s a game-changer, especially for Ridgecrest. 01:01:01.460 --> 01:01:02.460 - Yeah. 01:01:04.040 --> 01:01:10.140 Mm-hmm. I agree. Depending on the situation, especially with the earthquake 01:01:10.140 --> 01:01:14.320 source relative to the stations that ShakeAlert is using, the magnitude – 01:01:14.320 --> 01:01:18.760 or, the magnitude and especially the location might not be correct, 01:01:18.760 --> 01:01:20.360 especially right away for the initial alert. 01:01:20.360 --> 01:01:24.680 So, yes, I agree that, once we include these uncertainties, 01:01:24.680 --> 01:01:31.060 there might be significant changes. But when more observations are 01:01:31.060 --> 01:01:35.580 included, these estimations do tend to improve. 01:01:35.580 --> 01:01:42.970 So – but quantifying the initial alert zone will be important for this analysis. 01:01:42.970 --> 01:01:46.560 So hopefully, I’ll be able to get that soon. [laughs] 01:01:47.320 --> 01:01:50.800 - Great. Yeah. You’re asking all the right questions. This is great. Thanks. 01:01:50.800 --> 01:01:52.540 - Thank you. 01:01:52.540 --> 01:01:57.140 - Thanks, Dave. Tom Hanks, is your hand – is your hand up? 01:01:58.040 --> 01:02:01.740 - Yes, it is. Yeah, to kind of follow up on what 01:02:01.740 --> 01:02:07.380 Dave said, and going back to that slide that I was concerned about 01:02:07.390 --> 01:02:18.450 with the break in slope, so if you kind of shut off at Modified Mercalli about 4, 01:02:18.450 --> 01:02:25.240 then you would only be dealing with the upper limb of data. 01:02:25.240 --> 01:02:33.740 And that might save a lot of uncertainty when turning that corner and trying to 01:02:33.740 --> 01:02:38.600 get the correct alerts versus missed alerts and something like that. 01:02:38.600 --> 01:02:46.000 So it would be an interesting thing just to try this again at the magnitude, 01:02:46.000 --> 01:02:50.329 you know, 4 – level 4 Modified Mercalli. 01:02:50.329 --> 01:02:53.810 That would be the lower – the lower limit, and so that 01:02:53.810 --> 01:02:58.780 would exclude a lot of data, but it might improve the results 01:02:58.780 --> 01:03:03.290 in terms of what’s correct and what’s incorrect. 01:03:05.000 --> 01:03:09.620 But I’ll let you calculate that and tell the answer. 01:03:09.620 --> 01:03:12.460 - [laughs] I’ll definitely have to think about it. 01:03:12.460 --> 01:03:16.000 - Yeah. [laughs] Okay. Thanks again. It was a cool talk. 01:03:16.000 --> 01:03:17.940 - Mm-hmm. Thank you. 01:03:17.940 --> 01:03:22.650 - Thank you. We have another question. Thanks for the nice talk. 01:03:22.650 --> 01:03:26.200 Evaluating the variables – and this is important to try to get 01:03:26.200 --> 01:03:28.990 all the experiences correct for the technical and geek – from the 01:03:28.990 --> 01:03:33.420 technical and geek perspective. Sorry. My son is crying. 01:03:33.420 --> 01:03:37.160 From the perspective of those who experienced shaking, like my sister in 01:03:37.160 --> 01:03:44.119 L.A. for the Ridgecrest earthquake – for the Ridgecrest earthquake, 01:03:44.119 --> 01:03:47.230 how does one include what is most effective for them? 01:03:47.230 --> 01:03:51.360 That is, there will be some shaking, and maybe more than you expect? 01:03:51.360 --> 01:03:55.750 Or you will experience this amount of shaking and damage but no more? 01:03:55.750 --> 01:03:58.950 How does one develop confidence in the delivered message? 01:03:58.950 --> 01:04:08.280 - Mm-hmm. So the ShakeAlert message that is currently delivered through 01:04:08.280 --> 01:04:13.060 WEA and then through the early warning apps are – it’s a very simple 01:04:13.060 --> 01:04:16.240 message, and it’s just, earthquake, earthquake, drop, cover, hold on. 01:04:16.240 --> 01:04:23.800 So, for the most part, we do not send out information about the earthquake 01:04:23.800 --> 01:04:27.980 magnitude and the distance just because this does evolve in time. 01:04:27.980 --> 01:04:34.040 And our main goal for sending these messages is so that people are aware 01:04:34.040 --> 01:04:37.250 that an earthquake is happening and that they do take protective action. 01:04:37.250 --> 01:04:42.150 So mainly, drop, cover, hold on for this case. 01:04:42.150 --> 01:04:46.640 So that would be the main thing I would say. 01:04:46.640 --> 01:04:51.580 Just be prepared to drop, cover, hold on when you either feel shaking 01:04:51.580 --> 01:04:54.450 or you get a ShakeAlert message. 01:04:56.960 --> 01:05:01.160 [Silence] 01:05:01.160 --> 01:05:06.140 - All right. Are there any other questions? Any last questions for Jessie? 01:05:06.140 --> 01:05:12.200 Either unmute your microphone or type in the meeting chat. 01:05:14.280 --> 01:05:18.120 All right. So thank you very much for joining us today. 01:05:18.120 --> 01:05:21.180 Great talk, Jessie. People are thanking you. 01:05:21.190 --> 01:05:25.559 And thank you, Jessie, for this excellent talk. And see you all next week. 01:05:25.560 --> 01:05:26.600 - Thank you so much.