WEBVTT Kind: captions Language: en-US 00:00:00.343 --> 00:00:01.742 [inaudible background conversations] 00:00:01.770 --> 00:00:05.299 Okay. Well … 00:00:05.299 --> 00:00:07.805 This would be the time that we should be starting. 00:00:07.805 --> 00:00:11.621 People are slowly drifting in. We really need to figure out the 00:00:11.621 --> 00:00:15.549 viscosity of scientists to really work out this at a better pace. 00:00:16.012 --> 00:00:21.460 So, as the scientists viscously return to the room, I will turn things 00:00:21.460 --> 00:00:25.520 over to our excellent moderator, Jen Andrews from Caltech, 00:00:25.520 --> 00:00:29.917 who will start off the wonderfulness that is our final session of the day. 00:00:32.227 --> 00:00:35.490 - Thank you, Sara. And thank you, everyone, for joining us for the final 00:00:35.490 --> 00:00:39.608 session of the day, which is ShakeAlert earthquake early warning rollout. 00:00:39.608 --> 00:00:43.140 And, as several previous speakers alluded, it’s been a busy and exciting 00:00:43.140 --> 00:00:46.546 year for ShakeAlert going public. And I’m particularly excited about 00:00:46.546 --> 00:00:49.750 the presentations today because, as a humble worker bee on the 00:00:49.750 --> 00:00:53.560 algorithm side, what I’m really interested in is the end product, 00:00:53.560 --> 00:00:56.690 the result, how do we issue meaningful and useful alerts, 00:00:56.690 --> 00:01:00.500 and how do they get turned into actions. So three great presentations. 00:01:00.500 --> 00:01:05.491 We will be saving the discussion for the end following the previous 00:01:05.491 --> 00:01:08.800 sessions of the workshop. And so I’ll turn it over to Jessie 00:01:08.800 --> 00:01:12.640 Saunders, our first speaker, who is talking on examining regional EEW 00:01:12.640 --> 00:01:17.250 alerting strategies through the lens of the Ridgecrest earthquake. 00:01:20.090 --> 00:01:22.962 [silence] 00:01:22.962 --> 00:01:23.884 - [inaudible] 00:01:23.884 --> 00:01:26.850 Good afternoon, everybody. My name is Jessie Saunders. 00:01:26.850 --> 00:01:30.950 I am one of the Mendenhall postdocs here at the ESC 00:01:30.950 --> 00:01:33.710 in Moffett Field studying earthquake early warning. 00:01:33.710 --> 00:01:37.830 And, before I begin, I would just like to thank all of my co-authors 00:01:37.830 --> 00:01:41.479 listed here who have contributed to the work that I’m presenting today, 00:01:41.479 --> 00:01:45.042 and then, as well as all of the ShakeAlert team. 00:01:46.500 --> 00:01:49.780 So the ShakeAlert earthquake early warning system is the 00:01:49.780 --> 00:01:52.520 early warning system for the West Coast of the United States. 00:01:52.520 --> 00:01:57.760 It is operated through the USGS with partner state and academic 00:01:57.760 --> 00:02:01.620 institutions listed here. And the goal for ShakeAlert is 00:02:01.620 --> 00:02:07.500 to generate early warning alerts for MMI 4 shaking and higher shaking 00:02:07.500 --> 00:02:12.099 for earthquakes that occur in California, Oregon, and Washington – so the 00:02:12.099 --> 00:02:18.139 orange dotted lines on the map. And operationally, the algorithms that 00:02:18.139 --> 00:02:24.319 are in ShakeAlert use seismic data to rapidly estimate the earthquake source – 00:02:24.319 --> 00:02:27.739 so magnitude and location, and then finite fault effects. 00:02:27.739 --> 00:02:30.493 And then there are several additional algorithms that are currently 00:02:30.493 --> 00:02:34.489 in development. So there are three algorithms that use high-rate 00:02:34.489 --> 00:02:40.769 geodetic data to refine the finite source characterization. 00:02:40.769 --> 00:02:44.150 And then there’s also the PLUM ground motion-based algorithms. 00:02:44.150 --> 00:02:48.599 That uses direct observations of strong ground motions to generate alerts rather 00:02:48.599 --> 00:02:54.230 than using rapid source characterization. And, as was mentioned several times 00:02:54.230 --> 00:02:58.239 earlier today, the public rollout for ShakeAlert has already begun. 00:02:58.239 --> 00:03:03.470 So, at the end of December in 2018, public alerts began for residents of 00:03:03.470 --> 00:03:06.980 L.A. County through the ShakeAlertLA cellphone app. 00:03:06.980 --> 00:03:11.159 And then, in October of 2019, public alerting began for the entire 00:03:11.159 --> 00:03:14.689 state of California where ShakeAlerts will be issued through the wireless 00:03:14.689 --> 00:03:20.230 emergency alert system as well as other cellphone apps, including MyShake. 00:03:20.230 --> 00:03:25.439 Now, for this talk, I’m going to be using the magnitude 6.4 and 00:03:25.439 --> 00:03:29.559 magnitude 7.1 Ridgecrest earthquakes to examine different strategies for 00:03:29.559 --> 00:03:35.540 improving the magnitude – or, the MMI 4-plus early warning alert regions. 00:03:35.540 --> 00:03:42.049 And so, as a reminder, the way that MMI 4 is defined for USGS Did You 00:03:42.049 --> 00:03:49.004 Feel It? as well as ShakeMap is that MMI 4 is felt by many to all in the 00:03:49.004 --> 00:03:55.837 regions, and most observers describe this intensity of shaking as strong shaking. 00:03:55.837 --> 00:03:59.659 So, in the first part of my talk, I will be focusing on how we can 00:03:59.659 --> 00:04:04.618 improve the accuracy of MMI 4 alert regions, mainly for the point 00:04:04.618 --> 00:04:09.529 source alert distance computation. And then briefly, at the end, I will 00:04:09.529 --> 00:04:14.069 also talk about how we’ve used the Ridgecrest earthquakes to evaluate 00:04:14.069 --> 00:04:18.129 the alert accuracy and timeliness of the PLUM ground motion-based 00:04:18.129 --> 00:04:22.458 early warning algorithms with different regional alerting strategies. 00:04:24.169 --> 00:04:27.310 So the way that ShakeAlert generates alert regions can be 00:04:27.310 --> 00:04:31.260 broken up into four main steps. So, in the first step, an earthquake 00:04:31.260 --> 00:04:35.159 is detected, and the magnitude and the location are estimated. 00:04:35.159 --> 00:04:38.960 Then this magnitude is then input into a ground motion prediction equation to 00:04:38.960 --> 00:04:43.190 estimate the peak ground acceleration and peak ground velocity with distance. 00:04:43.190 --> 00:04:47.370 And then, at a given distance, the PGA and PGV are then input 00:04:47.370 --> 00:04:51.960 into a ground motion-to-intensity conversion equation, or GMICE, 00:04:51.960 --> 00:04:56.740 to estimate the MMI at that distance. And then the distance that corresponds 00:04:56.740 --> 00:05:01.190 to MMI 4.0, which is the alerting target, that becomes the radius 00:05:01.190 --> 00:05:05.289 of the alert region. And so that is shown here for the Ridgecrest earthquake. 00:05:05.289 --> 00:05:10.816 And this is using the actual magnitude estimate in this case. 00:05:10.816 --> 00:05:14.100 So, for this process, ShakeAlert currently uses the median expected 00:05:14.100 --> 00:05:19.460 ground motions. But we know that there is variability, both with PGA 00:05:19.460 --> 00:05:24.720 and PGV, at a given distance. And then, with PGA and PGV 00:05:24.720 --> 00:05:29.480 with MMI. And so, because of that, even if we get accurate source 00:05:29.480 --> 00:05:34.840 estimation, this will still produce an alert region that has missed alerts. 00:05:34.840 --> 00:05:38.389 And so, for the magnitude 7 earthquake, this is the 00:05:38.389 --> 00:05:41.259 Did You Feel It? map for that. And you can see that there are some 00:05:41.259 --> 00:05:47.060 regions that have MMI 4 and higher shaking that are outside the alert zone. 00:05:47.060 --> 00:05:51.530 So our question is – moving forward, is how do we improve the accuracy 00:05:51.530 --> 00:05:55.590 of this MMI 4-plus alert region? And for this, we are going to 00:05:55.590 --> 00:05:58.340 move forward assuming accurate source characterization – 00:05:58.340 --> 00:06:01.625 so with accurate magnitude estimation. 00:06:03.479 --> 00:06:05.860 So the way that we want to improve the accuracy is by 00:06:05.860 --> 00:06:08.340 expanding the alert region. And we want to make sure that we 00:06:08.340 --> 00:06:13.490 can do that with having the MMI 4 alert targets stay the same. 00:06:13.490 --> 00:06:16.349 And so our approach for this is to incorporate ground motion 00:06:16.349 --> 00:06:19.860 model uncertainties. So we have the uncertainty 00:06:19.860 --> 00:06:23.780 in PGA and PGV at a given distance from the GMPE. 00:06:23.780 --> 00:06:27.582 And then we also have uncertainty in MMI for a given PGA and PGV 00:06:27.582 --> 00:06:33.080 value from co-located data. And so we propagate these errors 00:06:33.080 --> 00:06:36.520 in the ground motion-to-intensity conversion equation to form 00:06:36.520 --> 00:06:40.370 a single standard deviation. And then we use that standard deviation to then 00:06:40.370 --> 00:06:46.259 expand the alert distance calculation. And so that looks something like this. 00:06:46.259 --> 00:06:51.499 So here, this – there we go. So this green dot shows what 00:06:51.499 --> 00:06:55.330 the MMI 4 alert distance is with one sigma uncertainty. 00:06:55.330 --> 00:06:58.270 And then we have the same thing with two sigma uncertainty. 00:06:58.270 --> 00:07:02.590 And so that corresponds to the alert regions we see on this map. 00:07:02.590 --> 00:07:06.349 And so now, let’s quantify what the overall quality 00:07:06.349 --> 00:07:09.417 of the alert is in terms of population. 00:07:11.199 --> 00:07:13.129 So that looks like that here. 00:07:13.129 --> 00:07:17.820 So all of these colors show the different alert quality. 00:07:17.820 --> 00:07:24.979 And so this middle map shows the total population that is inside the alert region 00:07:24.979 --> 00:07:29.439 as a function of alert distance. So this is the radius of our alert 00:07:29.439 --> 00:07:35.629 region as it goes out. And so here we quantify correct and missed alerts by 00:07:35.629 --> 00:07:39.520 the population that experiences MMI 4 and larger shaking according to 00:07:39.520 --> 00:07:43.130 Did You Feel It? And if these people are inside the alert region, 00:07:43.130 --> 00:07:46.889 that’s a correct alert in green. And, if they are outside the alert region, 00:07:46.889 --> 00:07:50.939 that means it is a missed alert because they did not get an alert in this case. 00:07:50.939 --> 00:07:54.560 And, as you can see, as we increase the alert distance, more and more of these 00:07:54.560 --> 00:07:58.100 people are now inside the alert region, which is great. 00:07:58.100 --> 00:08:01.689 However, as we increase the size of the alert region, we are also 00:08:01.689 --> 00:08:05.909 including a lot of people who do not feel MMI 4 and larger shaking. 00:08:05.909 --> 00:08:08.090 And these are false alerts. 00:08:08.090 --> 00:08:12.599 And that’s shown with all the population up here. 00:08:12.599 --> 00:08:17.340 And so I’ve separated these different false alerts into whether or not people 00:08:17.340 --> 00:08:21.447 still feel MMI 3 shaking, which means that, if they are indoors, they are still 00:08:21.447 --> 00:08:25.259 likely to feel some weak shaking. And that’s shown in the yellow. 00:08:25.259 --> 00:08:28.689 And then the population shown in orange are people who are 00:08:28.689 --> 00:08:32.820 unlikely to feel any shaking. And then, in gray, which is just 00:08:32.820 --> 00:08:38.480 a little bit up here, these are people inside ZIP codes that were not – 00:08:38.480 --> 00:08:41.730 that were not represented in Did You Feel It? 00:08:41.730 --> 00:08:44.650 And that could either be because they did not feel any shaking, 00:08:44.650 --> 00:08:48.700 and therefore were less inclined to submit a Did You Feel It? report, 00:08:48.700 --> 00:08:52.520 or because they were in a location of very sparse population, 00:08:52.520 --> 00:08:56.875 and so there were fewer people in that region to submit a report. 00:08:59.241 --> 00:09:06.165 Okay, so now the three black lines on this plot shows what the alert 00:09:06.190 --> 00:09:12.180 quality is for our three alert regions. And then also I have shown in pink 00:09:12.180 --> 00:09:15.790 what the 95% correct alert rate is in terms of population. 00:09:15.790 --> 00:09:19.350 And then, in blue, shows the same thing except in terms of area. 00:09:19.350 --> 00:09:22.980 So we can see that there is a pretty big difference if we look at 00:09:22.980 --> 00:09:26.890 population versus area. And so that could affect what 00:09:26.890 --> 00:09:29.550 our overall target should be – so much uncertainty should we 00:09:29.550 --> 00:09:33.600 actually include moving forward. And this also looks pretty different 00:09:33.600 --> 00:09:38.780 for the magnitude 6.4 Ridgecrest earthquake, which is shown here. 00:09:38.780 --> 00:09:42.630 So the magnitude 6.4 Ridgecrest earthquake – obviously, the MMI 4 00:09:42.630 --> 00:09:50.020 shaking didn’t extend as far. So there’s fewer people who need to be alerted. 00:09:50.020 --> 00:09:53.180 But, because some of those people who need to be alerted extend into 00:09:53.180 --> 00:09:56.870 the densely populated L.A. area, we can see that there is a pretty 00:09:56.870 --> 00:10:03.923 big spike in false alerts when we hit that densely populated region. 00:10:03.923 --> 00:10:07.965 So now, once again, our question becomes, how much uncertainty should 00:10:07.965 --> 00:10:15.280 we include in our alerting strategy? And so these two tables show the totals 00:10:15.280 --> 00:10:19.480 in population for the correct, missed, and false alerts for both of these 00:10:19.480 --> 00:10:24.200 earthquakes for our three example alerting strategies here. 00:10:24.200 --> 00:10:27.690 And so they have actually pretty similar characteristics. 00:10:27.690 --> 00:10:33.152 So if we were just to use the median expected alert distance for MMI 4 00:10:33.152 --> 00:10:38.440 shaking, we only get 7 and 6% correct alerts. 00:10:38.440 --> 00:10:42.120 And then, if we include one sigma of uncertainty, we end up getting 00:10:42.120 --> 00:10:46.303 that increased to over 80%. And then, if we get two sigma 00:10:46.303 --> 00:10:52.230 uncertainty, that gets us to over 99%, which is great, except getting to that 00:10:52.230 --> 00:10:59.350 99% causes us to have to send false alerts to tens of millions of people. 00:10:59.350 --> 00:11:03.910 And, for the magnitude 7 Ridgecrest earthquake, that ends up being about 00:11:03.910 --> 00:11:07.540 twice as many people over-alerted who actually require alerts – 00:11:07.540 --> 00:11:12.670 or, than people who require alerts. And then, for the magnitude 6.4 00:11:12.670 --> 00:11:17.190 Ridgecrest earthquake, that ends up being about 15 times as many people 00:11:17.190 --> 00:11:23.630 who receive unnecessary alerts compare to the people who actually need alerts. 00:11:23.630 --> 00:11:28.130 And so, when we move forward and look at more earthquakes to try to 00:11:28.130 --> 00:11:32.130 pick out how much uncertainty should we include and what our 00:11:32.130 --> 00:11:37.820 alerting strategy should be, it’s really important to take into account this 00:11:37.820 --> 00:11:41.990 tradeoff between trying to minimize the number of false alerts and then also 00:11:41.990 --> 00:11:47.690 trying to minimize the amount of over- alerting that we might expect to happen. 00:11:48.720 --> 00:11:55.760 So then, for the second half of my talk, I will now kind of go away from 00:11:55.760 --> 00:11:58.640 source-based early warning and then talk a bit about 00:11:58.640 --> 00:12:01.180 ground motion-based early warning. 00:12:01.180 --> 00:12:05.660 So the ground motion-based early warning algorithm that currently is 00:12:05.660 --> 00:12:11.270 in use in Japan and is under consideration for incorporation to 00:12:11.270 --> 00:12:14.500 the West Coast is called PLUM. It’s down – it stands for 00:12:14.500 --> 00:12:17.520 the Propagation of Local Undamped Motion. 00:12:17.520 --> 00:12:21.630 And so, instead of determining the earthquake source characteristics 00:12:21.630 --> 00:12:26.580 in order to issue an alert, what it does is that it monitors 00:12:26.580 --> 00:12:32.990 a station, and when a station observes MMI 4 or larger shaking, it issues 00:12:32.990 --> 00:12:38.030 an alert to regions within 30 kilometers of that station. 00:12:38.030 --> 00:12:42.370 However, in order for this to work, we need to have a set of pre-defined 00:12:42.370 --> 00:12:47.754 alerting regions. And we do not have that yet for the West Coast. 00:12:47.754 --> 00:12:53.430 And, fortuitously, PLUM was actually running on an internal development 00:12:53.430 --> 00:12:57.770 server with real-time data streams during the Ridgecrest earthquakes. 00:12:57.770 --> 00:13:02.450 And so that has allowed us to perform a postmortem analysis of PLUM using 00:13:02.450 --> 00:13:08.140 real-time data availability and latencies. And so, using that, we were able to 00:13:08.140 --> 00:13:12.999 test two – or, three candidate regional alerting strategies. 00:13:12.999 --> 00:13:17.760 So we have county boundaries here in this middle map, which represent 00:13:17.760 --> 00:13:19.720 kind of the largest alerting regions. 00:13:19.720 --> 00:13:24.180 You can see, particularly in southern California, these are very large. 00:13:24.180 --> 00:13:28.050 And then, second are the National Weather Service public 00:13:28.050 --> 00:13:32.190 forecast zones, which are shown here. And these are essentially county 00:13:32.190 --> 00:13:36.517 boundaries that are broken up into regions of similar weather. 00:13:36.517 --> 00:13:41.900 And one of the ways that that’s broken up is in terms of topography. 00:13:41.900 --> 00:13:46.970 So we actually happen to end up seeing a really nice correlation between the 00:13:46.970 --> 00:13:50.550 National Weather Service forecast zones as well as topographically 00:13:50.550 --> 00:13:55.092 derived Vs30 estimates, which is what the map is colored by. 00:13:55.092 --> 00:13:57.587 And then, for our third alerting strategy, 00:13:57.587 --> 00:14:02.936 we do a very simple 50-kilometer-by-50-kilometer grid. 00:14:02.936 --> 00:14:06.790 So let’s jump right into the results. 00:14:06.790 --> 00:14:12.610 So here, these maps show the results for PLUM with the magnitude 6.4 00:14:12.610 --> 00:14:15.900 Ridgecrest earthquake on top and the magnitude 7 Ridgecrest earthquake 00:14:15.900 --> 00:14:20.540 on the bottom for our three different candidate regional alerting strategies. 00:14:20.540 --> 00:14:24.450 And so each of these regions are colored by the alert quality. 00:14:24.450 --> 00:14:27.870 And so, sorry, these colors are a little bit different than earlier, 00:14:27.870 --> 00:14:33.070 but they’re basically the same thing. So, in blue, those are – those indicate 00:14:33.070 --> 00:14:37.710 a correct alert. So that would be a region that is alerted by PLUM, 00:14:37.710 --> 00:14:41.050 and then it also experiences MMI 4-plus shaking somewhere – 00:14:41.050 --> 00:14:43.830 excuse me – somewhere within that boundary. 00:14:43.830 --> 00:14:47.060 And, in the red, we have the missed alerts. 00:14:47.060 --> 00:14:51.660 So these are regions that are not alerted, but then there is some MMI 4 or larger 00:14:51.660 --> 00:14:56.410 shaking inside of that. And then yellow is for false alerts. 00:14:56.410 --> 00:15:00.990 And then, in gray, these are regions that are not alerted. 00:15:00.990 --> 00:15:07.520 And so, for all three alert strategies, they all do a really well – 00:15:07.520 --> 00:15:11.700 a really good job. There are some regions of missed alerts, notably 00:15:11.700 --> 00:15:15.740 up here, where there’s a tiny bit of MMI 4 and larger shaking 00:15:15.740 --> 00:15:18.800 in some of these regions. And then there’s also some missed 00:15:18.800 --> 00:15:23.800 alerts down here by the Salton Sea. And this was because the way that 00:15:23.800 --> 00:15:28.340 PLUM computes MMI is very slightly different than the way that ShakeMap 00:15:28.340 --> 00:15:32.010 does it, which is what we’re using for this comparison. 00:15:32.010 --> 00:15:36.510 And so that also explains some of the false alerts over in here where some of the 00:15:36.510 --> 00:15:40.610 stations that participate in the alert – so that measure MMI 4 and larger 00:15:40.610 --> 00:15:43.980 shaking are shown. But that wasn’t included – 00:15:43.980 --> 00:15:48.690 or that wasn’t the case for ShakeMap for some reason. 00:15:48.690 --> 00:15:54.684 And so, if we look at this in terms of the total population that was alerted, 00:15:54.684 --> 00:15:59.950 all three strategies actually achieved 99% correct alerts, which is great. 00:15:59.950 --> 00:16:05.570 And then, if we look at the smaller alerting regions, those actually reduce 00:16:05.570 --> 00:16:11.020 the total area that is falsely alerted. So that could indicate that the smaller 00:16:11.020 --> 00:16:15.930 50-kilometer grid might actually be the strategy to select. 00:16:15.930 --> 00:16:20.110 However, how does alert timeliness factor in? 00:16:21.009 --> 00:16:27.560 So, because we were able to use the real-time data streams for this, we’re 00:16:27.560 --> 00:16:31.720 actually able to estimate the warning times for these different regions. 00:16:31.720 --> 00:16:34.760 So what we do for this is that we assume that strong shaking 00:16:34.760 --> 00:16:39.170 arrives with the S wave. So then, for any location, 00:16:39.170 --> 00:16:42.970 the warning time for that location is the S wave arrival minus the time 00:16:42.970 --> 00:16:46.410 that that region is alerted. And so, once again, we are able to 00:16:46.410 --> 00:16:51.290 include the data telemetry latencies in our alerting time. 00:16:51.290 --> 00:16:58.100 However, we do not include any alert delivery – alert delivery latencies. 00:16:58.100 --> 00:17:03.670 So these would be the time it takes from when a ShakeAlert is issued to when it 00:17:03.670 --> 00:17:06.699 goes through, say, the WEA system and then arrives on your phone. 00:17:06.699 --> 00:17:09.420 So that’s not included in this. 00:17:09.420 --> 00:17:12.850 Okay. And so the amount of warning time for these 00:17:12.850 --> 00:17:15.366 different cases are shown here. 00:17:15.366 --> 00:17:20.630 We got the longest warning times for the largest alerting regions. 00:17:20.630 --> 00:17:24.301 And so you can see that really clearly with the counties over here. 00:17:24.301 --> 00:17:28.520 However, most of the warning times – or, the longest warning times are 00:17:28.520 --> 00:17:32.170 in regions that do not experience MMI 4 shaking. 00:17:32.170 --> 00:17:38.262 And so the boundary for MMI 4 is the black lines on these maps. 00:17:38.262 --> 00:17:41.220 So one other thing that’s pretty noticeable is that 00:17:41.220 --> 00:17:44.910 there are some regions in gray. And regions in gray means that there 00:17:44.910 --> 00:17:50.440 is no useful warnings for these regions. And so there will always be a late alert 00:17:50.440 --> 00:17:54.490 zone near the epicentral region. And this has to do with the amount of 00:17:54.490 --> 00:17:58.560 time that it takes to issue the first alert, for the most part. 00:17:58.560 --> 00:18:03.300 And then we also see that, when we have smaller alerting regions, there ends 00:18:03.300 --> 00:18:08.100 up being additional late alert zones that are beyond the epicentral region. 00:18:08.100 --> 00:18:11.556 And so, because of this, if we were to look at these two earthquakes, 00:18:11.556 --> 00:18:17.490 even though the 50-kilometer alerting grid does a better job with reducing 00:18:17.490 --> 00:18:21.740 the overall area of false alerts, because there are more regions 00:18:21.740 --> 00:18:27.470 that have late-alert zones, even without the additional late – 00:18:27.470 --> 00:18:33.840 or, additional delivery latencies, we might say that this style of 00:18:33.840 --> 00:18:39.220 regional alerting strategy might not be the optimal one to use. 00:18:39.220 --> 00:18:42.625 We might want to use a slightly bigger one. 00:18:45.613 --> 00:18:50.870 Okay, so in conclusion, in the first part, we developed a method to incorporate 00:18:50.870 --> 00:18:54.620 ground motion uncertainties into the alert distance computation. 00:18:54.620 --> 00:18:58.660 And so that actually is able to improve the accuracy of the 00:18:58.660 --> 00:19:04.868 MMI 4-plus alert regions for source-based alerting regions. 00:19:04.868 --> 00:19:10.410 And then, in Part 2, we examined potential regional alerting strategies 00:19:10.410 --> 00:19:13.040 for the ground motion-based early warning algorithms. 00:19:13.040 --> 00:19:16.220 And we were able to use real-time data streams from the Ridgecrest 00:19:16.220 --> 00:19:21.496 earthquakes in order to do that. For both alerting styles in Part 1 and 00:19:21.496 --> 00:19:28.290 Part 2, we were able to achieve 99% correct alerts for MMI 4-plus shaking. 00:19:28.290 --> 00:19:33.080 However, this does require falsely alerting tens of millions of people. 00:19:33.080 --> 00:19:37.990 And, for the magnitude 6.4 earthquake, once again, that’s about 15 to 20 times 00:19:37.990 --> 00:19:41.690 more false alerts than necessary alerts. And, for the magnitude 7.1, 00:19:41.690 --> 00:19:46.030 that’s about 2 to 2-1/2 more false alerts than necessary alerts. 00:19:46.030 --> 00:19:50.230 And, of those people who are falsely alerted, about 20 to 40% 00:19:50.230 --> 00:19:54.228 of them may not even feel shaking. 00:19:54.228 --> 00:19:58.860 So, in this case, there is this tradeoff between minimizing the amount of 00:19:58.860 --> 00:20:02.960 missed alerts and then also minimizing the amount of over-alerting. 00:20:02.960 --> 00:20:06.240 And it’s – and we would need to examine more earthquakes in order to 00:20:06.240 --> 00:20:13.100 better pick out what type of alerting strategy would be the most optimal. 00:20:13.100 --> 00:20:18.360 But we can’t just look at the overall alerts. 00:20:18.360 --> 00:20:22.320 With Part 2, with looking at the alert timeliness, we have found that it’s 00:20:22.320 --> 00:20:28.180 really important to also factor in what our expected alert timeliness is in order 00:20:28.180 --> 00:20:32.940 to find the correct alerting strategy. And then finally, we really need to 00:20:32.940 --> 00:20:36.147 understand what the public’s expectations are for early warning, 00:20:36.147 --> 00:20:38.630 particularly their false alert tolerance. 00:20:38.630 --> 00:20:43.790 If we expect – if people are really false alert-tolerant, then we would be able to 00:20:43.790 --> 00:20:48.160 tailor our alerting strategy in order to better minimize the missed alerts 00:20:48.160 --> 00:20:51.754 without having to worry about as much over-alerting. 00:20:51.754 --> 00:20:54.680 But, once again, we need to make sure that that’s something that 00:20:54.680 --> 00:20:57.600 everybody is okay with. And so we would need lots of 00:20:57.600 --> 00:21:01.417 social science in order to do that, and hopefully Sara will 00:21:01.417 --> 00:21:04.160 talk about that in the next talk. 00:21:04.160 --> 00:21:05.642 Thank you. 00:21:05.642 --> 00:21:10.708 [applause] 00:21:12.882 --> 00:21:16.240 - Great. Thank you, Jessie. Our next speaker is the social 00:21:16.240 --> 00:21:19.300 scientist Sara McBride on failure to alert – exploring 00:21:19.300 --> 00:21:23.792 perceptions of ShakeAlert during the 2019 Ridgecrest earthquake sequence. 00:21:26.827 --> 00:21:32.046 [silence] 00:21:32.046 --> 00:21:34.517 - Good afternoon, everyone. I’m Sara McBride. 00:21:34.517 --> 00:21:38.275 I’m the social science coordinator for ShakeAlert. 00:21:38.275 --> 00:21:43.833 Oh. I need to get the presentation going first. That would be helpful. 00:21:46.647 --> 00:21:51.045 [silence] 00:21:51.083 --> 00:21:53.342 [inaudible whispers] 00:21:53.367 --> 00:21:58.982 [silence] 00:21:59.007 --> 00:22:03.290 All right. Sorry for that false start. What a great way to start this 00:22:03.290 --> 00:22:09.309 failure to alert talk. A failure to start, really. 00:22:09.309 --> 00:22:11.730 So I am the social science coordinator for ShakeAlert, 00:22:11.730 --> 00:22:15.800 so I’m sort of an expert on human perception around ShakeAlert. 00:22:15.800 --> 00:22:18.170 And I actually know a lot about failure, so I think that 00:22:18.170 --> 00:22:24.210 this is a good topic for me [chuckles] to talk about today. 00:22:24.210 --> 00:22:28.690 So I just want to thank my legion of co-authors on this particular one. 00:22:28.690 --> 00:22:33.110 And it is quite a legion there who helped get together 00:22:33.110 --> 00:22:35.811 all of this information and help with the analysis. 00:22:35.811 --> 00:22:40.430 Particularly Jason Baumgartner from Pushshift who scraped Twitter for me, 00:22:40.430 --> 00:22:42.780 and we got 600,000 tweets regarding Ridgecrest. 00:22:42.780 --> 00:22:47.650 We had a huge data set that I’m going to talk about today. 00:22:47.650 --> 00:22:53.848 I just quickly want to talk – give a little plug for social science and ShakeAlert 00:22:53.848 --> 00:22:58.430 and talk about the Joint Committee for Communication, Education, 00:22:58.430 --> 00:23:02.880 and Outreach and our national committee so that you know that we’re 00:23:02.880 --> 00:23:08.270 embedding social science within the ShakeAlert endeavor at the USGS. 00:23:08.270 --> 00:23:12.020 And we have a number of other activities going on, including 00:23:12.020 --> 00:23:14.540 working with all the geological surveys on state levels. 00:23:14.540 --> 00:23:18.580 All the emergency management groups – our wonderful friends, 00:23:18.580 --> 00:23:21.480 who are our emergency managers. 00:23:21.480 --> 00:23:25.890 And then our technical implementation and engagement group, which Dr. Jen 00:23:25.890 --> 00:23:29.240 Strauss is in the room here, who is one of the heads of the committee. 00:23:29.240 --> 00:23:32.190 And Jen is actually a vice chair of this initiative. 00:23:32.190 --> 00:23:34.760 And we have colleagues working on this in British Columbia 00:23:34.760 --> 00:23:38.937 as well as Oregon, Washington, and California. 00:23:38.962 --> 00:23:43.370 And, just quickly on the ShakeAlert CEO Plan – that’s communication, 00:23:43.370 --> 00:23:47.540 education, and outreach – we really have five focus areas, which are public 00:23:47.540 --> 00:23:51.500 safety, preparedness, and resilience. Technical implementation and user 00:23:51.500 --> 00:23:55.179 engagement, of which Jen Strauss is one of the leaders on that. 00:23:55.179 --> 00:23:56.790 Consistent messaging and communication. 00:23:56.790 --> 00:24:00.750 Integration with other federal and state earthquake hazard products. 00:24:00.750 --> 00:24:03.750 And education resource development and dissemination. 00:24:03.750 --> 00:24:08.670 And these are our five strains. And really just talking about the social 00:24:08.670 --> 00:24:11.900 science initiative, we have seven projects that are currently ongoing 00:24:11.900 --> 00:24:17.400 right now from 2019 and 2020, of which this project is a part of it. 00:24:17.400 --> 00:24:19.690 And we really want to develop an understanding of where our 00:24:19.715 --> 00:24:23.831 people are currently in terms of risk perception and protective action, 00:24:23.831 --> 00:24:27.116 knowledge, and basic earthquake preparedness in Washington, 00:24:27.116 --> 00:24:30.610 Oregon, and California. And we also want to establish a long-term 00:24:30.610 --> 00:24:36.080 monitoring and evaluation plan for ShakeAlert to see how ShakeAlert 00:24:36.080 --> 00:24:40.790 changes our communities and builds in more community resilience. 00:24:40.790 --> 00:24:42.929 And so we do have these seven active projects working with 00:24:42.929 --> 00:24:45.170 nine universities and research agencies from 00:24:45.170 --> 00:24:48.860 five different states and two different countries. 00:24:48.860 --> 00:24:52.678 So I just want to talk about this project now and about 00:24:52.678 --> 00:24:55.298 our alerting platforms that were available 00:24:55.298 --> 00:25:00.140 to the public on July 4th and July 6th of 2019 for Ridgecrest. 00:25:00.140 --> 00:25:05.150 There was only one platform that was really publicly available at the time, 00:25:05.150 --> 00:25:08.840 which was ShakeAlertLA, and that was developed in conjunction with 00:25:08.840 --> 00:25:12.010 the city of Los Angeles. And the alerting thresholds 00:25:12.010 --> 00:25:15.690 there were set at a magnitude 5 with an MMI shaking of 4. 00:25:15.690 --> 00:25:21.990 And, due to the public response and perception that people wanted an alert 00:25:21.990 --> 00:25:28.960 for lesser magnitudes and lesser MMI shakings, the thresholds were lowered. 00:25:28.960 --> 00:25:32.390 Also just a note to Caltech. They have developed UserDisplay, 00:25:32.390 --> 00:25:37.520 which is not – it’s not available to the public but to select ShakeAlert users. 00:25:37.520 --> 00:25:41.050 But it was filmed live during a media conference, and I want to talk 00:25:41.050 --> 00:25:45.850 a little bit about that and what that did in the media environment. 00:25:45.850 --> 00:25:49.480 Now just quickly about the Ridgecrest earthquakes and ShakeAlertLA. 00:25:49.480 --> 00:25:53.900 Alerts were not sent out via the ShakeAlert app for either the magnitude 00:25:53.900 --> 00:25:59.240 6.4 or the magnitude 7.1 due to neither really meeting the 00:25:59.240 --> 00:26:03.050 estimated shaking threshold. And that’s – the “estimated” 00:26:03.050 --> 00:26:06.010 is a real important term here, and we’ll get to that in a minute. 00:26:06.010 --> 00:26:09.809 And while the thresholds might have been met for the magnitude 7.1, 00:26:09.809 --> 00:26:15.191 the ShakeAlert system estimated the earthquake at a 6.3 rather than an M 7.1. 00:26:15.191 --> 00:26:19.030 And the app did not provide information on the earthquakes on their 00:26:19.030 --> 00:26:23.000 map or in their recent earthquake list after the earthquake occurred. 00:26:23.000 --> 00:26:26.150 So, when people looked for more information about shaking 00:26:26.150 --> 00:26:31.360 on their ShakeAlertLA app, they just couldn’t find anything. 00:26:31.360 --> 00:26:33.350 And so I just quickly want to go through – so that’s really 00:26:33.350 --> 00:26:37.546 what happened, and that was for both kind of earthquakes. 00:26:37.546 --> 00:26:41.929 So, for the data collection analysis, I analyzed 69 media stories between 00:26:41.929 --> 00:26:45.720 July and November of 2019 about the Ridgecrest earthquakes. 00:26:45.720 --> 00:26:49.230 I went through Google News. And I usually wait about six months. 00:26:49.230 --> 00:26:52.300 I collect – I go through and collect new stories every month because of 00:26:52.300 --> 00:26:55.700 the way that Google archives differently at different times using pretty 00:26:55.700 --> 00:26:59.179 consistent search terms. The searches – my searches to 00:26:59.179 --> 00:27:04.630 gather media started 36 hours post the first event on the July 4th and 00:27:04.630 --> 00:27:08.908 then continued monthly and ongoing until November 2019. 00:27:08.908 --> 00:27:11.600 And all the stories were coded in Nvivo, which is a content 00:27:11.600 --> 00:27:16.728 analysis software program, using interpretive thematic analysis. 00:27:16.728 --> 00:27:21.420 Again, there were 600,000 tweets with the data set that I was provided 00:27:21.420 --> 00:27:24.549 by Pushshift by Jason Baumgartner. 00:27:24.549 --> 00:27:29.540 The collection began – I think it was about six hours post the M 6.4 quake. 00:27:29.540 --> 00:27:34.050 So it was in time to collect real rich, real-time data from the magnitude 7.1. 00:27:34.050 --> 00:27:36.860 The problem is, with social media, is that people often delete their 00:27:36.860 --> 00:27:39.830 tweets after a period of time or they delete their accounts. 00:27:39.830 --> 00:27:43.740 And so you don’t always get really good, rich data that is representative 00:27:43.740 --> 00:27:47.790 of the conversation that was going on if you, say, search Twitter and look, 00:27:47.790 --> 00:27:50.960 you know, months or years afterwards. 00:27:50.960 --> 00:27:53.900 And so what we did with the social media analysis is we analyzed 00:27:53.900 --> 00:28:00.870 2,100 tweets, and we coded 1,600 because, in the hashtags and in the – 00:28:00.870 --> 00:28:06.510 using the ShakeAlert terms #shakealertla, #shakealert – 00:28:06.510 --> 00:28:09.780 sometimes you just get, like, junk tweets that have absolutely 00:28:09.780 --> 00:28:12.429 nothing to do with the event. Someone just wants to tag along 00:28:12.429 --> 00:28:16.290 on the conversation and then add their own whatever to the – 00:28:16.290 --> 00:28:20.140 [chuckles] to the discourse on that hashtag. Again, all the tweets were 00:28:20.140 --> 00:28:24.290 coded in NVivo, and I used interpretive thematic analysis there. 00:28:24.290 --> 00:28:27.180 So the types of media I analyzed were print – so newspapers, television, 00:28:27.180 --> 00:28:32.490 radio, electronic media – which is only available online, magazines, blogs. 00:28:32.490 --> 00:28:35.100 And I looked at local, regional, national, and international news, 00:28:35.100 --> 00:28:39.983 particularly the AP and Reuters wire services. 00:28:39.983 --> 00:28:43.760 And just quickly, on the media analysis, so what I found really 00:28:43.760 --> 00:28:47.640 interesting was that there was a pretty significant spike – you know, 00:28:47.640 --> 00:28:52.460 obviously, after July the 4th. But what I found was interesting was 00:28:52.460 --> 00:28:57.840 I couldn’t find as many stories on the 6th, even though that was 00:28:57.840 --> 00:29:03.040 a larger earthquake and ostensibly felt by more people and more intense. 00:29:03.040 --> 00:29:06.250 So I’m not entirely sure why that is. I probably need to rerun some of 00:29:06.250 --> 00:29:09.190 my searches and just double-check my analysis. 00:29:09.190 --> 00:29:13.950 But the stories I analyzed were the – I analyzed 12 stories in that – 00:29:13.950 --> 00:29:20.320 from that first area in that first couple of days and then nine from the – you 00:29:20.320 --> 00:29:25.150 know, day three. And then, as time went on, obviously, the story sort of died. 00:29:25.150 --> 00:29:29.450 And what’s interesting is you can see a long tail, right, on these stories. 00:29:29.450 --> 00:29:33.160 You can start seeing August and October is when these really 00:29:33.160 --> 00:29:35.680 rich science stories are coming out in the media. 00:29:35.680 --> 00:29:39.800 And they can see the scientists, you know, were publishing in 00:29:39.800 --> 00:29:43.290 Science and in Nature, and our journal articles are coming out. 00:29:43.290 --> 00:29:45.830 And those were – those are what those stories are in the 00:29:45.830 --> 00:29:51.270 longer tail of the analysis. So just quickly – so 25 out of the 00:29:51.270 --> 00:29:55.210 69 articles I analyzed mentioned ShakeAlert specifically. 00:29:55.210 --> 00:29:59.670 It was the third most common theme in all the media stories – the first, 00:29:59.670 --> 00:30:03.220 of course, was aftershocks, which was 52 stories – other than 00:30:03.220 --> 00:30:08.170 the original earthquake itself. And the top three themes underneath 00:30:08.170 --> 00:30:12.740 that about ShakeAlert were ShakeAlertLA, which had 19 articles 00:30:12.740 --> 00:30:15.790 mentioning it; threshold, which had 19 articles mentioned it; 00:30:15.790 --> 00:30:18.760 and failure, which had 10 articles mentioning it. 00:30:18.760 --> 00:30:22.020 And I thought that was really interesting that failure was kind of 00:30:22.020 --> 00:30:24.460 a discourse that was mentioned in social media, and I’ll talk about 00:30:24.460 --> 00:30:29.030 that in a second. But that the media really did focus on talking about thresholds. 00:30:29.030 --> 00:30:32.670 So when they – I noted that, when they talked about failure, 00:30:32.670 --> 00:30:34.561 they also talked about the thresholds. 00:30:34.561 --> 00:30:38.320 The media did a pretty good job at explaining what had occurred 00:30:38.320 --> 00:30:43.370 with ShakeAlertLA and why people didn’t receive the alert. 00:30:43.370 --> 00:30:46.490 So I want to talk about failure. And I just want to say, before I start 00:30:46.490 --> 00:30:52.330 this, what I’m going to talk about now is not how the USGS itself or the 00:30:52.330 --> 00:30:56.950 scientists in this room necessarily perceive the performance of ShakeAlert. 00:30:56.950 --> 00:31:00.610 This is the perception and the discourse and the commentary 00:31:00.610 --> 00:31:03.580 of people on Twitter. So I’m going to go through the dark 00:31:03.580 --> 00:31:09.970 belly of [chuckles] Twitter right now. And this is not me saying this is 00:31:09.970 --> 00:31:14.480 what I think. This is me just reflecting on discourse that was 00:31:14.480 --> 00:31:16.970 going on using these terms, okay? 00:31:16.970 --> 00:31:20.810 So I’m just going to – that’s my caveat here. 00:31:20.810 --> 00:31:27.473 So failure was one of the biggest themes mentioned – 446 tweets out of 00:31:27.473 --> 00:31:32.332 1,668 mentioned that people were disappointed in the app. 00:31:32.332 --> 00:31:36.169 But, while they mentioned failure, they also just mentioned general 00:31:36.169 --> 00:31:42.340 disappointment in going to the app and not receiving any more information. 00:31:42.340 --> 00:31:45.402 And then the poor performance was also mentioned about the app with 00:31:45.402 --> 00:31:49.090 battery drains and locality issues – so location issues. 00:31:49.090 --> 00:31:54.910 And so that was another issue as well with 38 tweets just exploring 00:31:54.910 --> 00:31:59.860 sort of issues with the app there. And the media stories were 00:31:59.860 --> 00:32:03.530 highly re-tweeted. But after sort of 72 hours after 00:32:03.530 --> 00:32:08.020 the largest earthquake, on Twitter, I noted that really 00:32:08.020 --> 00:32:10.740 it was just media stories being shared. People weren’t really commenting 00:32:10.740 --> 00:32:13.290 on it anymore. They were just – these media stories were just 00:32:13.290 --> 00:32:19.480 being shared along to each other. So what I found interesting was that 00:32:19.480 --> 00:32:22.340 a lot of the discourse wasn’t just people being frustrated that 00:32:22.340 --> 00:32:25.450 they didn’t receive an alert. It was that they didn’t receive 00:32:25.450 --> 00:32:31.840 an alert despite perceived intense shaking by people that was felt widely 00:32:31.840 --> 00:32:35.810 throughout southern California. But, on top of that, when they searched 00:32:35.810 --> 00:32:39.410 for authoritative information from the app, and within the app, 00:32:39.410 --> 00:32:41.940 they couldn’t find any. And then, when they went to 00:32:41.940 --> 00:32:46.591 the USGS website, they couldn’t find the USGS – the USGS website 00:32:46.591 --> 00:32:50.370 experienced some technical issues as well and was unavailable for 00:32:50.370 --> 00:32:54.470 a period of time after both earthquakes. Now, I just want to say, I have 00:32:54.470 --> 00:32:57.840 personally talked with people who work on the USGS on the website, 00:32:57.840 --> 00:33:00.200 and we had a long conversation about me mentioning it, and they really 00:33:00.200 --> 00:33:04.350 wanted to say, yep, this happened. We want to be transparent about it. 00:33:04.350 --> 00:33:05.920 We did have connectivity issues. 00:33:05.920 --> 00:33:08.910 But we fixed it, and we fixed it right away. 00:33:08.910 --> 00:33:13.130 So they’re pretty confident that we’re not going to have those issues again. 00:33:13.130 --> 00:33:15.549 But they were comfortable with me talking about that 00:33:15.549 --> 00:33:19.040 being mentioned on Twitter. 00:33:19.040 --> 00:33:22.340 So there was also some – so just – sorry, backing up. 00:33:22.340 --> 00:33:27.470 I just – failure wasn’t just the lack of an alert. 00:33:27.470 --> 00:33:31.410 Failure in people’s minds was the lack of authoritative 00:33:31.410 --> 00:33:34.725 information that they could find from their app or from the USGS. 00:33:34.725 --> 00:33:37.950 So I just want to get that in people’s minds a little bit. 00:33:37.950 --> 00:33:40.929 And I found that personally really interesting. 00:33:40.929 --> 00:33:42.900 And then people had some confusion points, right? 00:33:42.900 --> 00:33:44.947 They were confused between MMI and magnitude. 00:33:44.947 --> 00:33:48.382 And I don’t think that’s necessarily surprising to anyone in the room. 00:33:48.382 --> 00:33:52.470 But people were most confused about why the magnitude – you know, 00:33:52.470 --> 00:33:56.480 what is MMI, what is magnitude, why are they different thresholds. 00:33:56.480 --> 00:33:59.540 And then we had also conducted wireless emergency alert tests 00:33:59.540 --> 00:34:03.490 in San Diego 10 days prior. And some people in San Diego 00:34:03.490 --> 00:34:08.379 thought, well, we have ShakeAlert and wireless emergency alerts 00:34:08.379 --> 00:34:12.000 because you just did a test on it. And so they were confused that 00:34:12.000 --> 00:34:14.529 they didn’t actually even have ShakeAlert yet because 00:34:14.529 --> 00:34:17.879 they thought they had it because we had had those tests. 00:34:17.879 --> 00:34:21.489 And then ShakeAlert and ShakeAlertLA and UserDisplay 00:34:21.489 --> 00:34:25.759 were sort of used interchangeably. People didn’t – people confused – 00:34:25.759 --> 00:34:28.729 the UserDisplay they were seeing on the – on the – on the media 00:34:28.729 --> 00:34:32.659 interviews that went off during the press conferences, they thought 00:34:32.659 --> 00:34:34.889 that’s what they were going to receive in ShakeAlertLA. 00:34:34.889 --> 00:34:37.899 And then they thought ShakeAlert was the same as ShakeAlertLA, which is, 00:34:37.899 --> 00:34:41.200 you know, kind of understandable because of the name sharing. 00:34:41.200 --> 00:34:44.079 [chuckles] And then misinformation on Twitter. 00:34:44.079 --> 00:34:46.839 Of course, we always have the delightful earthquake prediction 00:34:46.839 --> 00:34:51.589 people come out in force claiming that, you know, their cat told them this 00:34:51.589 --> 00:34:58.109 was going to happen and that – you know, 21 tweets on that. 00:34:58.109 --> 00:35:00.160 So just some questions and comments about ShakeAlert. 00:35:00.160 --> 00:35:05.140 So 57 tweets were dedicated to asking about having it in their county, 00:35:05.140 --> 00:35:09.215 which I found really interesting. Even though the discourse was 00:35:09.215 --> 00:35:13.920 around failure to alert, people were, like, hey, your app didn’t work, 00:35:13.920 --> 00:35:16.690 but we still want it. [laughs] 00:35:16.690 --> 00:35:21.160 And that was in San Bernardino, Kern County, and San Diego County. 00:35:21.160 --> 00:35:25.400 Eleven tweets out of the 1,600 claimed to uninstall the app, and five explored 00:35:25.400 --> 00:35:29.359 issues of loss of trust in the ability for the app to deliver messages. 00:35:29.359 --> 00:35:31.728 And four tweets talked about testing the system. 00:35:31.728 --> 00:35:34.720 And I found those tweets really interesting. 00:35:34.720 --> 00:35:38.630 They came from, like, software engineers who were, like, oh, well, 00:35:38.630 --> 00:35:41.410 yeah. You’re just testing the product and seeing if it works. 00:35:41.410 --> 00:35:43.109 And, hey, you guys learned some things. 00:35:43.109 --> 00:35:47.150 I thought that was, you know, kind of a very charitable approach from people. 00:35:47.150 --> 00:35:50.940 And then 85 tweets suggested for improvements of the app. 00:35:50.940 --> 00:35:55.119 Mostly around people being allowed to choose their own alerting thresholds. 00:35:55.119 --> 00:35:58.599 So one thing – if you know anything about me personally, 00:35:58.599 --> 00:36:03.310 you know that I do love humor. And I coded humor 60 times. 00:36:03.310 --> 00:36:06.690 And humor is something that is well-known of a way of coping 00:36:06.690 --> 00:36:09.099 around stressful situations. So I just want to read these out 00:36:09.099 --> 00:36:11.610 because they’re kind of small. 00:36:11.610 --> 00:36:15.900 But my favorite one is all, like – is the second one, which is, true, 00:36:15.900 --> 00:36:21.470 the ShakeAlert app failed to notify L.A. residents of two major earthquakes 00:36:21.470 --> 00:36:24.989 this weekend, but animals are supposed to sense quakes, too, 00:36:24.989 --> 00:36:27.270 and my dog didn’t say a damn word. 00:36:27.270 --> 00:36:31.790 So I’m feeling all kinds of betrayed right now. [laughter] 00:36:31.790 --> 00:36:34.440 Which is by far my favorite one. 00:36:34.440 --> 00:36:36.609 And then one of my other – the other good ones was, 00:36:36.609 --> 00:36:40.637 did ShakeAlertLA work? Asking for 2 million friends. 00:36:40.637 --> 00:36:46.809 And, I hope ShakeAlert works better than Shake Weight. 00:36:46.809 --> 00:36:50.579 And there were some that I can’t share in polite company. 00:36:50.579 --> 00:36:54.989 But this is one of the delightful parts of the job. 00:36:54.989 --> 00:36:58.799 So just quickly on who had the most popular tweets about ShakeAlert. 00:36:58.799 --> 00:37:04.569 Well, Representative Adam Schiff had the most re-tweeted ShakeAlert 00:37:04.569 --> 00:37:08.680 tweet with, then, L.A. city and then some personal accounts. 00:37:08.680 --> 00:37:12.229 L.A. Times – Ron Lin, congratulations. You made the top 10. 00:37:12.229 --> 00:37:16.059 High-five to you. Keep doing your great work. 00:37:16.059 --> 00:37:20.319 USGS ShakeAlert – yeah, L.A. Times was three and then Ron, so I just count 00:37:20.319 --> 00:37:23.891 you guys as one, so you guys got four of the top 10 tweets there. 00:37:23.891 --> 00:37:26.450 So, well done there on social media. 00:37:26.450 --> 00:37:32.230 And then personal accounts. Four out of the top 10 personal 00:37:32.230 --> 00:37:36.989 accounts were linked to media stories. So their re-tweets were the most 00:37:36.989 --> 00:37:39.289 re-tweeted, but they were just re-tweeting media stories. 00:37:39.289 --> 00:37:44.349 Only one of the top re-tweeted tweets came from a top 10 personal account. 00:37:44.349 --> 00:37:47.339 And four out of the top 12 tweets came from personal accounts 00:37:47.339 --> 00:37:50.029 rather than organizational, media, and political accounts. 00:37:50.029 --> 00:37:53.999 Which means that, while organizational, media, and political accounts are 00:37:53.999 --> 00:37:57.819 really important, we can’t dismiss influencers and 00:37:57.819 --> 00:38:01.579 social media influencers in the discourse here. 00:38:01.579 --> 00:38:04.569 So what did we learn? So ShakeAlert is now live and in testing 00:38:04.569 --> 00:38:07.559 mode in California with MyShake and QuakeAlert also available. 00:38:07.559 --> 00:38:10.335 So we have different apps now than ShakeAlertLA. 00:38:10.335 --> 00:38:14.410 ShakeAlerts are also available via wireless emergency alert. 00:38:14.410 --> 00:38:16.450 And the ShakeAlert website has been updated and has held up 00:38:16.450 --> 00:38:18.930 in a recent earthquake in Berkeley. 00:38:18.930 --> 00:38:23.640 And ShakeAlertLA – and we have lowered the thresholds for public apps. 00:38:23.640 --> 00:38:27.030 So I think that we learned the lessons really quickly. 00:38:27.030 --> 00:38:31.849 And we were able to assess what the important lessons were and 00:38:31.849 --> 00:38:35.180 make those changes internally. And I think that that’s a really 00:38:35.180 --> 00:38:39.349 laudable thing for an organization to do so quickly. 00:38:39.349 --> 00:38:42.849 So just conclusions. The communication of failure is 00:38:42.849 --> 00:38:46.809 not always straightforward. And people see failure in many different 00:38:46.809 --> 00:38:51.099 dimensions, and it’s not just about necessarily not receiving an alert, 00:38:51.099 --> 00:38:55.769 but it’s really important about what information they receive afterwards. 00:38:55.769 --> 00:39:01.710 And that they can find authoritative information to confirm their experience. 00:39:01.710 --> 00:39:05.079 Media stories did talk about failure but also talked about thresholds. 00:39:05.079 --> 00:39:08.509 So the media was very curious and wanted to report correctly 00:39:08.509 --> 00:39:10.650 what had occurred and why. 00:39:10.650 --> 00:39:16.040 And, you know, I think it’s important that failure is perceived more as 00:39:16.040 --> 00:39:20.410 a lack of information in our communication ecosystem. 00:39:20.410 --> 00:39:22.830 So diversity in our communication channels could improve this 00:39:22.830 --> 00:39:27.039 outcome for future earthquakes. So, if one channel fails, people can 00:39:27.039 --> 00:39:31.150 go to another channel to find the information that they’re looking for. 00:39:31.150 --> 00:39:35.400 And, while – I found this really interesting, which I didn’t mention here, 00:39:35.400 --> 00:39:38.710 but I thought was intriguing that lowering the thresholds was popular 00:39:38.710 --> 00:39:41.130 on Twitter, but there was definitely dissenting opinions. 00:39:41.130 --> 00:39:43.420 I found about 40 tweets that said, no, we shouldn’t 00:39:43.420 --> 00:39:48.249 lower the thresholds. And they wanted to keep the thresholds. 00:39:48.249 --> 00:39:51.069 And then media continues to have a really critical role in the 00:39:51.069 --> 00:39:55.319 communication of ShakeAlert. You know, as they shift their platforms 00:39:55.319 --> 00:40:01.640 from broadcast to social media, media is a main social media influencer. 00:40:01.640 --> 00:40:04.190 And so getting messages out quickly to media, I think, 00:40:04.190 --> 00:40:07.999 is a really critical thing to do. 00:40:07.999 --> 00:40:12.720 So that’s the end of my presentation, so thank you, everyone, for listening. 00:40:12.720 --> 00:40:17.667 [applause] 00:40:21.522 --> 00:40:24.359 - Thanks, Sara, for exploring our failures there. 00:40:24.359 --> 00:40:28.500 And finally, our last talk of the day, Jill Barnes on ShakeAlert 00:40:28.500 --> 00:40:31.833 and disaster readiness in LAUSD. 00:40:31.833 --> 00:40:35.667 - I’m not sure I’ll be able to switch this. - Oh, [inaudible]. 00:40:35.667 --> 00:40:38.250 - Because I’m so not a Mac person. 00:40:41.000 --> 00:40:46.491 [silence] 00:40:46.491 --> 00:40:49.583 - [inaudible] Yeah, that one. 00:40:49.583 --> 00:40:57.708 [silence] 00:40:57.708 --> 00:41:00.720 There we are. 00:41:00.720 --> 00:41:04.488 So I’m not sure if I should be insulted by the fact that 00:41:04.488 --> 00:41:08.099 I was preceded by failure. [laughter] 00:41:08.099 --> 00:41:11.670 So hopefully I won’t give you that. [chuckles] 00:41:11.670 --> 00:41:15.335 Just to give you a little bit of context on the LAUSD. 00:41:15.335 --> 00:41:17.390 The Los Angeles Unified School District is the 00:41:17.390 --> 00:41:20.229 second-largest school district in the country. 00:41:20.229 --> 00:41:29.349 It’s the largest with an elected board. We have about 650,000 K-12 students. 00:41:29.349 --> 00:41:32.529 We have 100,000 adult school students. 00:41:32.529 --> 00:41:35.579 We have almost 100,000 preschool students. 00:41:35.579 --> 00:41:37.465 We have 10 infant programs. 00:41:37.465 --> 00:41:40.082 We have a couple of overnight programs. 00:41:40.082 --> 00:41:43.233 And we’ve got about 80,000 employees. 00:41:43.233 --> 00:41:49.039 So, all told, we have direct contact with about a million people within 00:41:49.039 --> 00:41:55.109 Los Angeles – the Los Angeles area. And we cover multiple cities 00:41:55.109 --> 00:41:58.519 in Los Angeles. I don’t know that there’s a more complicated 00:41:58.519 --> 00:42:06.210 setup than L.A. We are one of 83 school districts in Los Angeles County. 00:42:06.210 --> 00:42:09.059 So the L.A. Unified is not L.A. County schools, but we 00:42:09.059 --> 00:42:12.579 cover more than Los Angeles city. We do have L.A. city, and we have 00:42:12.579 --> 00:42:18.730 parts or all of 23 different other cities and unincorporated areas of the county. 00:42:18.730 --> 00:42:20.829 We don’t do anything simply. 00:42:20.829 --> 00:42:25.039 So we cover 710 square miles, and the number of schools – you’ll get 00:42:25.039 --> 00:42:28.420 a different number depending on who you ask and what your criteria are. 00:42:28.420 --> 00:42:35.125 It’s between 900 and 1,400, depending on how you’re asking. But many. 00:42:36.253 --> 00:42:39.619 In general, when we’re thinking about disasters in school, we’re really 00:42:39.619 --> 00:42:42.600 putting things into four buckets. And, within those buckets, 00:42:42.600 --> 00:42:46.109 each bucket has two. So you’re looking at either – 00:42:46.109 --> 00:42:49.740 you’re getting outside – you’re either getting outside to get away from 00:42:49.740 --> 00:42:53.489 an emergency that’s inside, rather like we did today earlier. 00:42:53.489 --> 00:42:58.609 Or we’re coming inside because the danger is outside. 00:42:58.609 --> 00:43:03.230 We have 20 different types of emergencies in our plan. 00:43:03.230 --> 00:43:06.799 I’m not going to go over all of this in detail. 00:43:06.799 --> 00:43:11.339 There are several different laws that cover some of the disaster planning 00:43:11.339 --> 00:43:18.479 and preparedness that we do have in schools, including earthquake drills. 00:43:20.246 --> 00:43:26.519 And, for us, we plan for – we have three days of supplies at every 00:43:26.519 --> 00:43:29.739 single one of our schools. Because we know that, 00:43:29.739 --> 00:43:35.089 for K-12 schools, we care for those kids until each one of them 00:43:35.089 --> 00:43:37.780 is reunited with a custodial parent. 00:43:37.780 --> 00:43:42.690 So we know we’ll be in it for a while there with them. 00:43:42.690 --> 00:43:46.430 So, in terms of ShakeAlert, we started with ShakeAlert with 00:43:46.430 --> 00:43:51.210 a desktop beta test – beta testing of the desktop program at 00:43:51.210 --> 00:43:54.730 Eagle Rock High School, which is – they’ve been a really great partner. 00:43:54.730 --> 00:44:00.279 They have a very vibrant science program at the school and a 00:44:00.279 --> 00:44:04.369 fantastically motivated principal. They’ve done a lot there. 00:44:04.369 --> 00:44:10.710 So we started by putting the desktop version of the program into all of 00:44:10.710 --> 00:44:14.279 the science classrooms and on the principal’s laptop as well. 00:44:14.279 --> 00:44:17.410 And they used it in science lessons, and they would actually kind of 00:44:17.410 --> 00:44:22.512 manually trigger the system so they could use it in drills as well. 00:44:22.512 --> 00:44:24.799 And so we used that for a couple of years there. 00:44:24.799 --> 00:44:28.535 We started that in 2015. 00:44:28.535 --> 00:44:32.760 And then we actually were able to get them the device. 00:44:32.760 --> 00:44:37.069 So most of our schools – we’re planning on doing it 00:44:37.069 --> 00:44:40.660 through the PA systems because that’s a system we already have. 00:44:40.660 --> 00:44:44.980 And actually, that’s generally where – if there’s an announcement made 00:44:44.980 --> 00:44:47.799 about an earthquake, which is often generally when there’s – 00:44:47.799 --> 00:44:50.779 the earthquake is actually occurring, it’s going to be the principal 00:44:50.779 --> 00:44:54.289 on the PA system telling everyone to drop, cover, and hold on. 00:44:54.289 --> 00:44:57.699 And that’s generally how we run our drills as well for earthquakes. 00:44:57.699 --> 00:45:00.670 So the PA system is already being used in that way. 00:45:00.670 --> 00:45:04.839 It’s also used – that’s how we announce lockdowns and shelter-in-place events. 00:45:04.839 --> 00:45:08.969 So it’s already a system that we’re using to announce emergencies. 00:45:08.969 --> 00:45:14.108 So it makes sense for us to do that. I know that there may be – as we move 00:45:14.108 --> 00:45:17.630 further into technology, apparently some of the new fire alarm systems 00:45:17.630 --> 00:45:21.579 have some vocal capabilities as well, and we’re just barely beginning 00:45:21.579 --> 00:45:24.952 to explore whether or not we can do anything with 00:45:24.952 --> 00:45:26.989 early warning with that as well. 00:45:26.989 --> 00:45:29.880 So right now, what we’ve done is, we’ve taken the device – we’re working 00:45:29.880 --> 00:45:36.379 with Early Warning Labs, and we have the device actually put into the PA 00:45:36.379 --> 00:45:40.960 system at Eagle Rock High School. And it’s actually giving out the alerts. 00:45:40.960 --> 00:45:43.339 And it can also be used for tests, which is wonderful because 00:45:43.339 --> 00:45:47.543 that way the school uses it for all of its drills. 00:45:47.543 --> 00:45:54.450 We have two more schools that we – that we will be putting the devices in. 00:45:54.450 --> 00:45:57.163 Hopefully sooner rather than later. 00:45:57.163 --> 00:46:01.099 And the other piece of this is we have about 50 schools that 00:46:01.099 --> 00:46:04.630 have upgraded PA systems. And, because of those upgraded 00:46:04.630 --> 00:46:09.819 PA systems, we would actually be able to do all of the ShakeAlert 00:46:09.819 --> 00:46:13.829 software kind of over the air without even going to those school sites, 00:46:13.829 --> 00:46:16.789 which makes it a lot simpler. Unfortunately, that’s only going to 00:46:16.789 --> 00:46:20.299 cover about 50 of our schools. But we’ve lumped those 50 together 00:46:20.299 --> 00:46:25.400 and put in an application with Cal OES for hazard mitigation grant funding. 00:46:25.400 --> 00:46:27.579 But we haven’t yet received that funding. 00:46:27.579 --> 00:46:30.729 So we’re kind of waiting on that one. 00:46:30.729 --> 00:46:38.540 So one thing that I learned through this is that it helps to get – 00:46:38.540 --> 00:46:43.269 I like to say yes to things. And I like to forge ahead. 00:46:43.269 --> 00:46:45.569 There’s a lot on this that I can’t do myself. 00:46:45.569 --> 00:46:49.989 I can’t go out there and do the technical components of it. 00:46:49.989 --> 00:46:54.150 In terms of – this has a lot of different pieces in the district that it touches. 00:46:54.150 --> 00:46:58.585 It’s not just our IT personnel. And it’s not just our maintenance 00:46:58.585 --> 00:47:02.049 personnel. It’s a combination of those two departments, who generally 00:47:02.049 --> 00:47:07.390 don’t do a lot of things collaboratively. Plus the emergency warning piece. 00:47:07.390 --> 00:47:11.630 So it helps, when you’re pursuing this, to make sure that you’ve got all of 00:47:11.630 --> 00:47:15.239 that support lined up ahead of time before you 00:47:15.239 --> 00:47:18.540 start doing anything, which is great. 00:47:18.540 --> 00:47:23.069 Which we do have, but it still is managing to proceed rather slowly. 00:47:23.069 --> 00:47:26.299 So the thing that I really like about ShakeAlert is there 00:47:26.299 --> 00:47:31.390 really is no downside. It’s really great for the life safety angle. 00:47:31.390 --> 00:47:35.349 It gives us extra time. And that extra time is really important 00:47:35.349 --> 00:47:40.329 for us with a couple of different groups. It’s very important for the kids who 00:47:40.329 --> 00:47:44.920 use wheelchairs or have other devices or difficulties getting – 00:47:44.920 --> 00:47:48.950 some of them need extra time. They can get under something to drop, 00:47:48.950 --> 00:47:51.989 cover, and hold on, and it just takes them a little extra time. 00:47:51.989 --> 00:47:54.430 Some of them cannot. The ones in earthquakes, then you 00:47:54.430 --> 00:47:56.779 need to have – or, the ones in wheelchairs that needed to 00:47:56.779 --> 00:47:59.640 have somebody roll them into a protective corner of the classroom. 00:47:59.640 --> 00:48:02.290 And then that person needs to be able to drop, cover, and hold on. 00:48:02.290 --> 00:48:04.890 So having that extra time comes in really handy. 00:48:04.890 --> 00:48:08.630 There are also some kids with processing disabilities, where – 00:48:08.630 --> 00:48:11.979 these are the kids who kind of freeze when you tell them something. 00:48:11.979 --> 00:48:14.569 And they need a minute to kind of get over that before they 00:48:14.569 --> 00:48:19.380 actually start moving. And ShakeAlert actually provides them that time so that 00:48:19.380 --> 00:48:24.730 they actually get a chance to process – do their freeze, process, and then 00:48:24.730 --> 00:48:27.829 they’re still able to drop, cover, and hold on before the shaking starts. 00:48:27.829 --> 00:48:32.212 It really does have a really fantastic protective capability for those kids. 00:48:32.212 --> 00:48:35.150 It’s also going to be really nice with our – with our younger ones, 00:48:35.150 --> 00:48:38.289 with our preschoolers, with our infant programs because it gives us 00:48:38.289 --> 00:48:41.670 a chance to do something for those kids and then have the adults that 00:48:41.670 --> 00:48:46.029 are taking care of them also be able to have an adequate time to drop, 00:48:46.029 --> 00:48:50.290 cover, and hold on for themselves. I think it’s a really great way 00:48:50.290 --> 00:48:55.170 to look at resilience. I’m all about the resilience from the 00:48:55.170 --> 00:49:00.739 social aspect of it, that community disaster resilience aspect of it. 00:49:00.739 --> 00:49:05.089 I think that, in putting ShakeAlert in the schools – as we know, 00:49:05.089 --> 00:49:08.849 kids tends to take the lessons that they’ve learned home with them. 00:49:08.849 --> 00:49:12.739 If the warning tones are standardized, if the messaging is standardized, 00:49:12.739 --> 00:49:16.200 if they hear that out in the community, they will then communicate 00:49:16.200 --> 00:49:18.709 to their parents. If they know there’s an app, they may ask 00:49:18.709 --> 00:49:21.999 their mom or dad to download it. If there’s translation that needs 00:49:21.999 --> 00:49:24.680 to take place, the kids can kind of get on board with that. 00:49:24.680 --> 00:49:27.660 By the time these kids grow up, then everybody will be used to it, 00:49:27.660 --> 00:49:34.103 and it’ll be wonderful. So I think that it will do a lot to help us, 00:49:34.103 --> 00:49:38.319 as a region, as the West Coast, to move forward by getting it into 00:49:38.319 --> 00:49:43.559 more of our schools, hopefully. And it – the nice thing for me – 00:49:43.559 --> 00:49:46.519 a lot of people have asked, well, you know, what are you 00:49:46.519 --> 00:49:48.299 going to have to redo? You know, you’re going to have to do 00:49:48.299 --> 00:49:49.890 all this training. You’re going to have to do all these things. 00:49:49.890 --> 00:49:52.960 You’re going to have to really go out there and teach people different things. 00:49:52.960 --> 00:49:55.309 Well, I don’t. Because it doesn’t change what we do. 00:49:55.309 --> 00:49:58.349 What did we do when we heard the alert? We drop, cover, and hold on. 00:49:58.349 --> 00:50:01.849 We’re doing the exact same action. We’re just doing it earlier. 00:50:01.849 --> 00:50:05.950 Which is really, really nice because it makes – the learning is – 00:50:05.950 --> 00:50:11.233 we’re that much further down the road with learning. 00:50:11.233 --> 00:50:15.269 All I have to do, then, is update the procedures to make sure that it accounts 00:50:15.269 --> 00:50:21.119 for having alerts or tones given through ShakeAlert, as well as, if you don’t – 00:50:21.119 --> 00:50:26.069 if you don’t get those alerts, then if you, of course, then feel the shaking, then 00:50:26.069 --> 00:50:28.514 you’ll still drop, cover, and hold on. 00:50:28.514 --> 00:50:31.739 And then it’s really nice that we can use it for the drills as well. 00:50:31.739 --> 00:50:35.469 Because they do drill with the drop, cover, and hold on every month. 00:50:35.469 --> 00:50:38.861 So that will be a lot of practice with the – 00:50:38.861 --> 00:50:42.980 with it that we’ll be experiencing. 00:50:42.980 --> 00:50:46.970 So I spend a lot of my time making sure that our kids with disabilities 00:50:46.970 --> 00:50:49.910 have the support that they need. And I’ve already mentioned a little bit 00:50:49.910 --> 00:50:54.730 about that in terms of ShakeAlert. But we really do get a nice microcosm 00:50:54.730 --> 00:50:57.059 of society going through our public schools. 00:50:57.059 --> 00:51:00.827 And you’ll get kids with mobility issues, visual issues, the deaf and 00:51:00.827 --> 00:51:07.460 hard of hearing, cognitive, autism spectrum, the medically fragile, 00:51:07.460 --> 00:51:10.480 kids with emotional disabilities, medication-dependent, 00:51:10.480 --> 00:51:14.496 and multiple disabilities. Because nobody ever says you get just one. 00:51:14.496 --> 00:51:17.819 And, for most of these populations, 00:51:17.819 --> 00:51:22.190 ShakeAlert does give them a little something extra in it. 00:51:22.190 --> 00:51:25.969 One other thing. We do have a school for the deaf. 00:51:25.969 --> 00:51:33.009 And it is a – it’s all taught in ASL, which means that the majority of the 00:51:33.009 --> 00:51:38.170 staff that works at that school is also deaf because who’s fluent in ASL? 00:51:38.170 --> 00:51:40.819 People who are deaf. So the majority of the staff is deaf. 00:51:40.819 --> 00:51:44.009 All the students are deaf. They start their programs there 00:51:44.009 --> 00:51:47.310 at 3 years old, and it goes through the age of 20. 00:51:47.310 --> 00:51:52.599 So it’s a comprehensive K-12 program, and they have a very specialized – 00:51:52.599 --> 00:51:55.650 they have a one-of-a-kind – I don’t even think it’s produced 00:51:55.650 --> 00:51:58.839 anywhere else in the country from what I’ve heard – they have a very 00:51:58.839 --> 00:52:04.660 unique – communication devices that are – that are throughout 00:52:04.660 --> 00:52:09.517 the school that were developed specifically for them. 00:52:09.517 --> 00:52:13.950 And I think it will be interesting to see how we can incorporate 00:52:13.950 --> 00:52:18.930 ShakeAlert for that population as well. Because then we’ll be able to start 00:52:18.930 --> 00:52:22.819 seeing the benefits of ShakeAlert in some of those specialized 00:52:22.819 --> 00:52:28.085 communities as well. And that is short and sweet. 00:52:28.085 --> 00:52:30.250 And I think I’m done with that. 00:52:30.250 --> 00:52:32.194 [applause] 00:52:32.194 --> 00:52:34.708 [inaudible] 00:52:37.200 --> 00:52:45.283 [silence] 00:52:45.283 --> 00:52:50.958 - Okay, great. All right. Do we have any questions for Jessie, Sara, and Jill? 00:52:54.566 --> 00:53:01.239 - So a question for Jessie. What – did you ever look at what 00:53:01.239 --> 00:53:06.035 the difference in – the improvement would be in all of the aspects if you 00:53:06.035 --> 00:53:12.910 simply took the MMI 3 level for alerting instead of MMI 4 plus 00:53:12.910 --> 00:53:16.540 sigma or plus two sigma? Because if you just say MMI 3-plus, 00:53:16.540 --> 00:53:21.512 then you don’t have all of that stuff that’s MMI 3 00:53:21.512 --> 00:53:25.160 and not quite MMI 4, but you still alerted it. 00:53:25.160 --> 00:53:28.440 - Okay, so I just want to try to understand your question correctly. 00:53:28.440 --> 00:53:32.869 So you’re saying that, instead of trying to expand the alert region 00:53:32.869 --> 00:53:37.260 by incorporating uncertainty, we just lower the MMI threshold … 00:53:37.260 --> 00:53:38.816 - Yeah. - … and expand it that way? 00:53:38.816 --> 00:53:45.059 - Yeah. - Okay. So, yes, it does help. 00:53:45.059 --> 00:53:51.920 For the – so incorporating one sigma of uncertainty ends up expanding the 00:53:51.920 --> 00:53:58.589 alert region by close to 100 kilometers. And, with lowering the threshold 00:53:58.589 --> 00:54:04.600 by one MMI unit, that also does approximately the same thing. 00:54:04.600 --> 00:54:08.630 But our goal with what we were trying to do is we were trying 00:54:08.630 --> 00:54:15.200 to improve the alert accuracy while keeping the MMI target the same. 00:54:15.200 --> 00:54:18.083 So you can do both methods. 00:54:20.958 --> 00:54:41.280 [silence] 00:54:41.280 --> 00:54:44.637 - This question is for Sara. 00:54:44.637 --> 00:54:48.979 I’m wondering – well, I know why I sometimes see people posting 00:54:48.979 --> 00:54:53.729 on social media is to complain. And I’m wondering if you came across 00:54:53.729 --> 00:54:58.160 a lot of tweets, or any tweets, of people saying that they were 00:54:58.160 --> 00:55:01.880 happy with ShakeAlert? Or if it was all 100% negative. 00:55:01.880 --> 00:55:03.207 [laughter] 00:55:03.207 --> 00:55:05.009 Not necessarily 100% negative, but just … 00:55:05.009 --> 00:55:07.119 - It is Twitter, right? - Exactly. 00:55:07.119 --> 00:55:09.349 - Like, you know – you know what I’m working with. 00:55:09.349 --> 00:55:11.049 - Right. - Right? 00:55:12.049 --> 00:55:17.109 Yeah. Like, what I found really interesting was the discourse for 00:55:17.109 --> 00:55:19.979 people who didn’t want to see the threshold lowered. 00:55:19.979 --> 00:55:23.150 That liked only being alerted when they thought it was really serious. 00:55:23.150 --> 00:55:26.309 And that was – you know, I found 30, 40 tweets that were, 00:55:26.309 --> 00:55:30.859 like, hey, I don’t want to be – I don’t want to be aware of every little shake. 00:55:30.859 --> 00:55:34.880 And I’m fine with that, so don’t lower the threshold right away. 00:55:34.880 --> 00:55:38.150 And I found that really interesting. That wasn’t really a complaint. 00:55:38.150 --> 00:55:40.469 And it wasn’t really, I really like the app the way that it is. 00:55:40.469 --> 00:55:43.999 It was more, I don’t want to see this threshold change. 00:55:43.999 --> 00:55:50.369 So I wouldn’t say that that wasn’t necessarily – it wasn’t necessarily 00:55:50.369 --> 00:55:52.140 positive or negative. It was just people being happy 00:55:52.140 --> 00:55:55.089 with what – you know, they liked the status quo 00:55:55.089 --> 00:55:59.083 of only being alerted at certain levels of shaking. 00:56:02.899 --> 00:56:08.709 - So, sort of on that note – this is kind of a question for both Jessie and Sara. 00:56:08.709 --> 00:56:12.199 Other than people opining about that sort of thing, what kind of research 00:56:12.199 --> 00:56:17.529 is going into whether or not there are benefits of the threshold 00:56:17.529 --> 00:56:20.759 being higher or lower? You know, does it – 00:56:20.759 --> 00:56:25.165 does it hurt people to hear about just minor shaking? 00:56:25.165 --> 00:56:29.169 You know, at a lower – you know, intensity 3 instead of 4? 00:56:29.790 --> 00:56:35.219 - Is this one for me? I mean, we are looking at the 00:56:35.219 --> 00:56:38.440 different thresholds and why people like them the way that they do. 00:56:38.440 --> 00:56:44.319 And a lot of that, we’re still kind of – we have a project of doing surveying 00:56:44.319 --> 00:56:49.349 in Ridgecrest and L.A. about this very question and really unpacking 00:56:49.349 --> 00:56:54.130 more what people think about the thresholds. 00:56:54.130 --> 00:56:59.390 And what motivates personal desires or personal preferences 00:56:59.390 --> 00:57:02.279 in terms of thresholds. I would only be speculating 00:57:02.279 --> 00:57:05.428 at this point, which I love to do, and we can do later. 00:57:05.428 --> 00:57:09.460 But maybe not in front of these people. [laughs] 00:57:09.460 --> 00:57:12.940 - So I have maybe a follow-on question there. Thanks, Austin, for the lead-in. 00:57:12.940 --> 00:57:16.618 To Jill, so you mentioned that ShakeAlert is only positive, and it will 00:57:16.618 --> 00:57:20.493 give your students, and particularly maybe students who have a disability 00:57:20.493 --> 00:57:22.829 or need a few extra moments to drop, cover, and hold on. 00:57:22.829 --> 00:57:27.789 But, based on what Jessie presented and what Sara sort of just said, 00:57:27.789 --> 00:57:31.371 you can expect some false alerts. So that you’ll get a ShakeAlert, 00:57:31.371 --> 00:57:34.740 and the students maybe drop, cover, and hold on, and will not feel shaking. 00:57:34.740 --> 00:57:39.103 So do you – can you speculate, I guess, on – or, from your experience, 00:57:39.103 --> 00:57:41.380 particularly for students with disabilities for whom drop, covering, 00:57:41.380 --> 00:57:45.459 and hold on might not be trivial, how do you – will that be a negative 00:57:45.459 --> 00:57:48.769 experience, or what would you expect? Or maybe it’s just more practice. 00:57:48.769 --> 00:57:51.670 They already do drills. - Yeah. I don’t anticipate it being 00:57:51.670 --> 00:57:54.269 too much of a – of a big deal for the schools. 00:57:54.269 --> 00:57:57.450 Because they do have to – they are required to do an earthquake drill 00:57:57.450 --> 00:58:00.339 every month anyway. And, if they – if they do that drill 00:58:00.339 --> 00:58:04.949 through an alert that doesn’t turn out to result in shaking, then, hey, 00:58:04.949 --> 00:58:07.332 they can check that box that they’ve done that drill. 00:58:07.332 --> 00:58:10.980 A little – a little extra practice isn’t a bad thing. 00:58:10.980 --> 00:58:13.759 And it generally doesn’t take more than a couple of minutes, especially if we’re 00:58:13.759 --> 00:58:16.539 just doing drop, cover, and hold on, and we’re not evacuating buildings. 00:58:16.539 --> 00:58:18.849 It really only takes a couple of minutes in the classroom. 00:58:18.849 --> 00:58:23.119 So it doesn’t have a large impact on instructional time, which is 00:58:23.119 --> 00:58:28.000 what teachers and administrators are usually concerned about. 00:58:30.032 --> 00:58:31.032 - [inaudible] 00:58:31.057 --> 00:58:32.057 [laughter] 00:58:32.082 --> 00:58:35.269 - [inaudible] the other side of the room. 00:58:37.035 --> 00:58:40.249 - So, for Sara, you mentioned that there were some earthquake 00:58:40.249 --> 00:58:44.240 prediction tweets. And it only recently came to my 00:58:44.240 --> 00:58:49.289 attention, but there was actually, I gather, quite a well-known example 00:58:49.289 --> 00:58:55.049 of Kern County Fire tweeting – and not only that, but Jennifer Lazo 00:58:55.049 --> 00:58:58.619 pointed out that they also did a Nixle alert. 00:58:58.619 --> 00:59:03.147 So that was an official agency tweet, and it was based on some wacky 00:59:03.147 --> 00:59:06.240 something, I don’t know where, from California City Police. 00:59:06.240 --> 00:59:07.855 You look like you know what this is about. 00:59:07.855 --> 00:59:09.849 I just wanted to cue you up on that. - Yeah. Yeah. I actually do know 00:59:09.849 --> 00:59:13.029 a little bit about what this – what you’re asking. 00:59:13.029 --> 00:59:17.150 Yeah. Essentially, someone called – from what we can gather, because we 00:59:17.150 --> 00:59:21.089 haven’t been able to really assess from the people who actually sent the tweet. 00:59:21.089 --> 00:59:24.671 There’s been numerous attempts to talk to them back channel and find out 00:59:24.671 --> 00:59:28.960 what’s going on – not from myself in particular, but from other – from other 00:59:28.960 --> 00:59:33.019 emergency managers that do social media and researchers as well. 00:59:33.019 --> 00:59:40.849 It looks like someone called in from a – pretending to be from the military base 00:59:40.849 --> 00:59:46.779 saying that they had a prototype of some description that was – 00:59:46.779 --> 00:59:49.760 equipment of some description that was telling them that 00:59:49.760 --> 00:59:52.967 an earthquake was going to occur in 15 minutes. 00:59:52.967 --> 00:59:59.249 And I can kind of emphasize with the Kern County officials because they – 00:59:59.249 --> 01:00:02.589 you know, a new situation had arisen. There was a lot of stress. 01:00:02.589 --> 01:00:05.700 They couldn’t – I mean, I don’t know all the details of the decision-making 01:00:05.700 --> 01:00:09.339 process, but someone sounded authoritative saying that they had 01:00:09.339 --> 01:00:12.779 special information, and they made the best decision they could at the time. 01:00:12.779 --> 01:00:15.634 And I’m not going to stand up here and criticize any other agency 01:00:15.634 --> 01:00:18.366 in terms of their decision-making process. 01:00:18.366 --> 01:00:22.039 But I think what’s interesting is that we also had a very similar case 01:00:22.039 --> 01:00:26.849 occur in Anchorage as well. So – and fortunately, that got stopped 01:00:26.849 --> 01:00:31.890 before it went to sending out a message. But what was interesting and – 01:00:31.890 --> 01:00:35.829 you know, I’m not a conspiracy theorist, either, but what was interesting was, 01:00:35.829 --> 01:00:38.880 is that the messages were quite similar and claimed to come from 01:00:38.880 --> 01:00:42.390 official sources or military institutions. 01:00:42.390 --> 01:00:47.359 So there is – there is a – there is more to unpack here. 01:00:47.359 --> 01:00:50.479 And I know a couple of researchers who are looking at this piece 01:00:50.479 --> 01:00:53.270 of misinformation. And hopefully they will 01:00:53.270 --> 01:00:57.000 have their articles and research published soon. 01:00:59.467 --> 01:01:01.973 [silence] 01:01:01.998 --> 01:01:05.572 - So this is kind of jointly for all three of you all, I guess, 01:01:05.572 --> 01:01:08.743 back to the haters-gonna-hate question. 01:01:08.743 --> 01:01:15.359 So, short of having to launch these massive social science surveys, 01:01:15.359 --> 01:01:18.920 where you go, post-earthquake, and have a specific question about 01:01:18.920 --> 01:01:23.617 thresholding or asking a school district, what is your specific, you know, 01:01:23.617 --> 01:01:30.239 response to a false alert, because social media and a lot of the traditional ways 01:01:30.239 --> 01:01:35.829 that we can information from the public is very biased towards haters gonna 01:01:35.829 --> 01:01:39.699 hate, you know, and you don’t get a whole lot of people taking the time 01:01:39.699 --> 01:01:43.960 while they’ve, you know, experienced something, and they’re, like, yay, you 01:01:43.960 --> 01:01:50.219 know, we’re happy with the status quo. How can we, as ShakeAlert or, 01:01:50.219 --> 01:01:56.279 you know, users of the system actually probe what’s happening? 01:01:56.279 --> 01:02:00.735 How do we, you know, not get bogged down in the Google Play, you know, 01:02:00.770 --> 01:02:07.619 responses or tweetstorms or a random article in the small village gazette? 01:02:07.619 --> 01:02:10.900 You know, how do we actually probe what people want? 01:02:10.900 --> 01:02:14.092 Because individual surveys, while important and gives you 01:02:14.092 --> 01:02:20.458 empirical evidence, is not something we can do on a monthly basis. 01:02:23.061 --> 01:02:26.417 - Crickets. [laughter] 01:02:29.300 --> 01:02:33.764 - I mean, I don’t really have an answer to that question, but perhaps one way 01:02:33.764 --> 01:02:39.199 that we could pursue that that’s not sending out separate surveys, why not 01:02:39.199 --> 01:02:44.079 just add a couple of extra questions to, say, the Did You Feel It? survey? 01:02:44.079 --> 01:02:48.299 And have that be some ShakeAlert- specific questions that don’t necessarily 01:02:48.299 --> 01:02:55.109 go into say, the estimations for MMI, but then perhaps can give us some of 01:02:55.109 --> 01:03:02.989 the answers that we might want in terms of, did you feel strong shaking and, 01:03:02.989 --> 01:03:06.119 like, did you get a ShakeAlert, or would you want to. 01:03:06.119 --> 01:03:10.630 And maybe that could be one way that we get those answers that 01:03:10.630 --> 01:03:15.239 don’t necessarily require mobilizing a bunch of people to survey 01:03:15.239 --> 01:03:18.708 everybody every time that there’s an earthquake. 01:03:23.558 --> 01:03:25.941 - So, yeah, we’re actually already doing that. 01:03:25.941 --> 01:03:29.619 So we’re in the process of developing questions for Did You Feel It? with 01:03:29.619 --> 01:03:32.632 Dave Wald. He’s kindly said, hey, we would like to add a few questions. 01:03:32.632 --> 01:03:35.509 So we’re working with him on what the most appropriate questions are. 01:03:35.509 --> 01:03:39.729 Because we agree. Like, collecting data on a regular basis and 01:03:39.729 --> 01:03:44.279 making it a part of a process the public is already – some people in the public are 01:03:44.279 --> 01:03:49.535 already familiar with, I think, is a really good thing to do. 01:03:49.567 --> 01:03:52.160 I mean, I’m an advocate for doing social science research. 01:03:52.160 --> 01:03:55.499 The other thing that we’re looking at doing is analyzing human behavior 01:03:55.499 --> 01:03:58.510 via CCTV footage. So we watch what people actually do. 01:03:58.510 --> 01:04:01.099 Because I think understanding human behavior, versus what 01:04:01.099 --> 01:04:05.339 they tell us they did, which is – which is fantastic to learn. 01:04:05.339 --> 01:04:07.509 People are, like, yeah, of course, I dropped, covered, and held on, 01:04:07.509 --> 01:04:11.119 and I grabbed my whole family. And actually, they, you know, 01:04:11.119 --> 01:04:12.789 ran somewhere else. 01:04:12.789 --> 01:04:16.180 So analyzing people’s actually behavior is another way that 01:04:16.180 --> 01:04:20.689 we’re collecting sort of more latent data that doesn’t require asking people 01:04:20.689 --> 01:04:23.660 necessarily their perceptions. Because I love that question, 01:04:23.660 --> 01:04:26.789 Jen, about haters gonna hate. Of course, in Twitter, you’re going to 01:04:26.789 --> 01:04:30.019 get people who are going to be wanting to be funny, wanting to 01:04:30.019 --> 01:04:32.650 make a statement. I think that that asking for 01:04:32.650 --> 01:04:35.640 2 million friends tweet was, like, re-tweeted 500 times, 01:04:35.640 --> 01:04:39.178 and that person was really happy and then posted their sound cloud. 01:04:39.178 --> 01:04:41.459 [laughter] 01:04:41.459 --> 01:04:43.949 You know, so there is that whole thing of, if I say something quite funny 01:04:43.949 --> 01:04:47.559 and complain-y on Twitter, I will get recognition. 01:04:47.559 --> 01:04:50.479 And so you have to bear that in mind when you are doing social media 01:04:50.479 --> 01:04:54.109 analysis is that people are making a comment because they want to be seen, 01:04:54.109 --> 01:04:57.749 not necessarily because this is actually what they feel or what they perceive. 01:04:57.749 --> 01:05:00.769 And surveys tell you a certain piece of information, but they tell you 01:05:00.769 --> 01:05:04.789 what people want you to think they think they think. 01:05:04.789 --> 01:05:08.640 So I’ll just let you guys think about that. [laughter] 01:05:08.640 --> 01:05:10.010 And then, you know, we also have focus groups. 01:05:10.010 --> 01:05:12.630 And I could go through all the methods of – the many methods of social science 01:05:12.630 --> 01:05:15.630 if we want to, but I’m sure we have other things to talk about today. 01:05:15.630 --> 01:05:18.359 But, yes, so we are looking at doing that, and we’ve got other 01:05:18.359 --> 01:05:20.749 social science projects on the way to explore more 01:05:20.749 --> 01:05:23.799 about people’s perceptions on ShakeAlert. 01:05:26.490 --> 01:05:29.059 - All very fascinating. Are the thresholds based on 01:05:29.059 --> 01:05:30.760 what people feel outside? 01:05:30.760 --> 01:05:36.368 And do you know when you get false readings if people are inside or outside? 01:05:36.368 --> 01:05:40.250 It makes quite a difference if you’re in a multi-story building. 01:05:43.819 --> 01:05:49.881 - So, right now, ShakeAlert does not tailor the alerts based on the type of 01:05:49.881 --> 01:05:56.029 building that someone might be in. So right now, it’s just a pretty simple – 01:05:56.029 --> 01:06:00.589 we just use the magnitude and epicenter estimate of the earthquake and 01:06:00.589 --> 01:06:06.950 then also some finite fault information. and then it’s just one map – like, one 01:06:06.950 --> 01:06:12.199 very simple ShakeMap, essentially. And then we use that to define the 01:06:12.199 --> 01:06:18.438 alert region. And, for MMI 4 shaking, which is our minimum threshold target 01:06:18.438 --> 01:06:21.985 that we’re trying to alert for, that’s felt by pretty much everybody, both 01:06:21.985 --> 01:06:27.777 inside buildings and outside as well. But it isn’t a level of damaging shaking. 01:06:27.777 --> 01:06:29.917 - I appreciate that. - Yeah. 01:06:29.917 --> 01:06:32.917 - Just interested, if somebody reports that they feel shaking, and they’re in a 01:06:32.917 --> 01:06:38.720 four-story building, the people on the ground will not feel it [inaudible]. 01:06:38.720 --> 01:06:42.349 - Yeah. So in – I know in the Did You Feel It? surveys, they ask about whether 01:06:42.349 --> 01:06:46.618 or not you were inside a building or outside, but I don’t think they ask about 01:06:46.650 --> 01:06:53.339 the story of building that you were on. So that information is not something 01:06:53.339 --> 01:06:56.941 that we use to tease out the MMI. - Might be something to keep in mind. 01:06:56.941 --> 01:06:59.103 - Yeah. - It is radically different. 01:06:59.103 --> 01:07:02.167 - Yeah. For sure. 01:07:05.092 --> 01:07:09.020 [silence] 01:07:09.045 --> 01:07:14.180 - Question is for Jessie. I haven’t really used ShakeAlert, 01:07:14.180 --> 01:07:18.975 but is there any follow-up at all? An alert goes out, and then does 01:07:18.975 --> 01:07:22.900 anything else go out immediately? Say within a minute or so 01:07:22.900 --> 01:07:27.289 to let people know? And also, given that we have 01:07:27.289 --> 01:07:31.855 instruments everywhere, is there any feedback from the system 01:07:31.855 --> 01:07:36.369 whereby the strong ground motion instruments can then be queried 01:07:36.369 --> 01:07:40.657 to see where things are really shaking? 01:07:40.657 --> 01:07:48.270 - So, to answer the last part of your question first, for the PLUM earthquake 01:07:48.270 --> 01:07:51.049 early warning algorithm that’s in development – so this is not 01:07:51.049 --> 01:07:55.809 in operation for ShakeAlert, it’s essentially real-time monitoring of 01:07:55.809 --> 01:08:01.102 the ground shaking that’s happening. So PLUM would potentially capture 01:08:01.102 --> 01:08:06.749 some of the non-uniform behavior that we would expect to see. 01:08:06.749 --> 01:08:12.140 Whereas, for the ShakeAlert algorithms that are in operation, their main 01:08:12.140 --> 01:08:15.719 objective is to rapidly characterize the earthquake. 01:08:15.719 --> 01:08:22.759 So they’re not – so those algorithms aren’t necessarily adding additional 01:08:22.759 --> 01:08:26.657 complexity, such as, for example, basin amplification in L.A., 01:08:26.657 --> 01:08:33.079 were that to come to play. And then, when an alert is issued – 01:08:33.079 --> 01:08:40.904 Sara might know more about this – I don’t believe there’s any follow-up 01:08:40.904 --> 01:08:43.209 messaging that’s happening yet. 01:08:43.209 --> 01:08:47.000 So, Sara, do you want to answer that? [laughs] 01:08:50.691 --> 01:08:53.003 - Yeah, we’ve actually been developing post-alert messaging for about 01:08:53.003 --> 01:08:57.179 18 months. Because we know that it’s critical that people receive information 01:08:57.179 --> 01:09:01.680 very quickly after they receive an alert. And so we have templated WEA 01:09:01.680 --> 01:09:08.549 messages and templated media releases for successful alerts and false alerts 01:09:08.549 --> 01:09:12.589 and missed alerts at the moment. And then we can make modifications 01:09:12.589 --> 01:09:15.279 as we learn more about the earthquake behavior, particularly 01:09:15.279 --> 01:09:19.619 in the media template. Because, yeah, as we saw in the failure 01:09:19.619 --> 01:09:22.199 to alert, it wasn’t just that people didn’t receive an alert. 01:09:22.199 --> 01:09:24.549 It was that they didn’t receive any information afterwards. 01:09:24.549 --> 01:09:27.659 So post-alert messaging is a really critical part of the project, and we’re 01:09:27.659 --> 01:09:31.190 in the process of finalizing that development. 01:09:31.190 --> 01:09:35.389 - So it only goes to the media? It didn’t – why not just send out the 01:09:35.389 --> 01:09:39.299 same message again with [inaudible] … - No, we do – I said that we had WEA – 01:09:39.299 --> 01:09:43.310 wireless emergency alert messages that do explain a bit of that. 01:09:43.310 --> 01:09:46.159 The complexity with that is that we’re trying to work in 01:09:46.159 --> 01:09:50.630 a 90-character message. Now, we have – we do know that 01:09:50.630 --> 01:09:55.860 the FCC has expanded some of the character allowances, but it is quite 01:09:55.860 --> 01:09:59.570 complex to explain why people might not have received an alert or why 01:09:59.570 --> 01:10:04.738 they did receive an alert in a 90 – in a 90-character limit. 01:10:04.738 --> 01:10:07.798 And that’s been part of it, but we do have some in draft, so when 01:10:07.798 --> 01:10:12.159 people do receive a WEA, and it’s quite a big area, we’re 01:10:12.159 --> 01:10:16.899 prepared to say something via wireless emergency alerts, if required. 01:10:19.423 --> 01:10:22.980 - [laughs] Hi. I just had a question about whether or not there’s been 01:10:22.980 --> 01:10:26.520 a comparison with the National Weather Service’s, for example, 01:10:26.520 --> 01:10:30.100 tornado alerts and similar weather alerts. 01:10:30.100 --> 01:10:33.329 Obviously, that’s probably a lot higher probability, but people are probably 01:10:33.329 --> 01:10:37.230 going to tweet about those, too, and, you know, have similar responses 01:10:37.230 --> 01:10:39.890 about whether or not – well, probably want to know about a tornado just 01:10:39.890 --> 01:10:44.599 about – like a big earthquake. - Yeah, so to develop the post-alert 01:10:44.599 --> 01:10:49.820 messaging, we took a lot of – we took a real look at rapid-onset 01:10:49.820 --> 01:10:54.270 hazard warning systems. And tornadoes were definitely one of them. 01:10:54.270 --> 01:10:56.000 And so we looked at the messaging on there. 01:10:56.000 --> 01:10:58.980 And there is – there are differences. There are fundamental differences 01:10:58.980 --> 01:11:02.477 in behaviors of the systems and timeframes. 01:11:02.477 --> 01:11:07.570 But we did use some of the messaging, particularly around the shortness and 01:11:07.570 --> 01:11:12.574 the directiveness around the messaging to develop these messages. 01:11:12.609 --> 01:11:16.196 We looked at lahar warning systems. We looked at tornadoes. 01:11:16.196 --> 01:11:20.980 We looked at flash flooding messages. So we looked at all of these hazards. 01:11:20.980 --> 01:11:23.400 We also looked at other earthquake early warning messages, you know, 01:11:23.400 --> 01:11:28.070 in Japan and Mexico – well, Mexico City has the sirens, among other things. 01:11:28.070 --> 01:11:31.074 So, yeah, we did take a pretty broad look at the 01:11:31.074 --> 01:11:34.451 different hazards and their messaging. 01:11:34.483 --> 01:11:38.243 - Do we have one more quick question from anyone? 01:11:44.811 --> 01:11:48.929 - Am I doing this right? This is a little bit of a follow-up 01:11:48.929 --> 01:11:55.699 on Rufus’ question because I think I didn’t quite understand the answer. 01:11:55.699 --> 01:12:00.574 If you have – you talked about the follow-up messaging being – 01:12:00.574 --> 01:12:05.221 it sounded to me, with respect to, did the alert work well for you, 01:12:05.255 --> 01:12:08.450 and that’s what you follow up. Like, oh, you got a false alert. 01:12:08.450 --> 01:12:11.121 Oh, we forgot to give you an alert. 01:12:11.121 --> 01:12:15.090 But, in one of your slides, you said something about – 01:12:15.090 --> 01:12:18.909 you showed that they wanted follow-up information. 01:12:18.909 --> 01:12:25.280 And is there a – does this happen, or is it in the works to have a follow-up 01:12:25.280 --> 01:12:30.000 tweet to the actual alert with information about the event – 01:12:30.000 --> 01:12:34.480 where was it, actually, and the kinds of things that people – 01:12:34.480 --> 01:12:37.847 that, well, for example, I immediately go to my computer and look up 01:12:37.847 --> 01:12:41.699 all kinds of things. But is that kind of follow-up stuff? 01:12:41.699 --> 01:12:46.128 - Yeah. So we actually have our own Twitter account – USGS_ShakeAlert. 01:12:46.128 --> 01:12:49.719 I recommend everyone follow it. Because it makes Bob de Groot really 01:12:49.719 --> 01:12:53.640 happy when he sees numbers go up. And he’s, like, hey, Sara, we got 01:12:53.640 --> 01:12:56.596 95,000 – we got 9,500 followers today. 01:12:56.596 --> 01:12:59.347 And that makes me happy when he’s happy. 01:12:59.347 --> 01:13:03.280 And that’s actually the account to follow that we’ll be updating 01:13:03.280 --> 01:13:06.659 when people receive an alert. And we do post follow-up information 01:13:06.659 --> 01:13:09.750 about what occurred and where to go to get more information. 01:13:09.750 --> 01:13:12.469 And you can also look at the USGS websites as well. 01:13:12.469 --> 01:13:16.840 It’s unlikely we’ll post that in a WEA because a WEA is a very specific 01:13:16.840 --> 01:13:22.780 type of public safety message, and there are very specific constraints regarding 01:13:22.780 --> 01:13:25.699 wireless emergency alerts. And you can’t just send out wireless 01:13:25.699 --> 01:13:28.899 emergency alerts for public information and follow-up information. 01:13:28.899 --> 01:13:31.530 You need to – it needs to be very directive and very specific 01:13:31.530 --> 01:13:35.199 to people’s situations. So there is limitations in terms of the 01:13:35.199 --> 01:13:38.929 purpose of wireless emergency alerts. But, in terms of social media and 01:13:38.929 --> 01:13:41.659 Twitter, yeah, we’ve been definitely developing that 01:13:41.659 --> 01:13:44.292 and in the media templates as well. 01:13:46.249 --> 01:13:50.628 - I just wanted to say something too. Since I didn’t help develop it, 01:13:50.628 --> 01:13:53.428 I can say this. The MyShake cellphone app does provide 01:13:53.428 --> 01:13:57.391 follow-up information. So if you get an alert on MyShake, 01:13:57.391 --> 01:14:01.280 there’s – [inaudible] to have it right now, you can follow the link to learn 01:14:01.280 --> 01:14:05.380 more about the earthquake, go to the USGS website and get the definitive 01:14:05.380 --> 01:14:08.746 solution and ShakeMaps and all those sort of things for that event. 01:14:08.746 --> 01:14:13.650 So there is – depends on the app, but at least that one does have follow-up. 01:14:13.650 --> 01:14:17.200 - Thanks. This is great discussion, but I’m mindful that we’ve reached 4:30. 01:14:17.200 --> 01:14:21.589 Sarah, did you want to say – I’m happy for the discussion to go on, but did you 01:14:21.589 --> 01:14:27.139 have any closing messages [laughs] if people do want to sneak out? 01:14:29.756 --> 01:14:31.340 - Good night, everyone. Drive safe. See you in the morning? 01:14:31.340 --> 01:14:33.058 [laughter] 01:14:33.058 --> 01:14:35.074 - Great. [laughs] 01:14:35.074 --> 01:14:37.062 [applause] 01:14:37.089 --> 01:14:39.065 - Thanks to our speakers again. Thank you. 01:14:41.574 --> 01:14:54.945 [silence]