WEBVTT Kind: captions Language: en-US 00:00:01.600 --> 00:00:05.120 Hello. My name is Jessie Saunders. I’m a Mendenhall postdoctoral 00:00:05.120 --> 00:00:07.600 researcher at the U.S. Geological Survey. 00:00:07.600 --> 00:00:11.040 And today I will be sharing with you some work my co-authors and I have 00:00:11.040 --> 00:00:14.160 done on comparing the real-time earthquake early warning alert 00:00:14.160 --> 00:00:19.336 performance for earthquakes that have happened in California this last year. 00:00:19.360 --> 00:00:22.560 Earthquake early warning is not earthquake prediction. 00:00:22.560 --> 00:00:26.560 Instead, earthquake early warning aims to rapidly detect that an earthquake 00:00:26.560 --> 00:00:30.000 is occurring and then issue alerts for earthquake shaking before the 00:00:30.000 --> 00:00:33.600 shaking arrives at the alerted location. There are many approaches to 00:00:33.600 --> 00:00:37.656 earthquake early warning, and today I will be talking about two of them. 00:00:37.680 --> 00:00:41.520 The first is the ShakeAlert system. ShakeAlert is the operational 00:00:41.520 --> 00:00:44.960 earthquake early warning system for the West Coast of the United States. 00:00:44.960 --> 00:00:48.320 Public alerts by ShakeAlert began in California in 2019 00:00:48.320 --> 00:00:51.256 and have now expanded to Oregon and Washington. 00:00:51.280 --> 00:00:54.400 ShakeAlert take a source-based approach to earthquake early warning, 00:00:54.400 --> 00:00:58.080 where ground motion observations at stations are used to rapidly estimate 00:00:58.080 --> 00:01:01.360 earthquake source parameters like magnitude and epicenter. 00:01:01.360 --> 00:01:04.800 These rapid source parameters are then input to ground motion models which 00:01:04.800 --> 00:01:09.576 estimate the extents of different levels of Modified Mercalli Intensity, or MMI. 00:01:09.600 --> 00:01:13.280 These estimated MMI distributions determine the alert regions used by 00:01:13.280 --> 00:01:17.280 ShakeAlert for a given earthquake. Here we will be focusing mainly on the 00:01:17.280 --> 00:01:22.160 alert regions determined using an MMI 3.5 alert threshold, which corresponds 00:01:22.160 --> 00:01:27.256 to the early warning alerts issued via the Wireless Emergency Alert system. 00:01:27.280 --> 00:01:30.400 The other earthquake early warning approach I will be discussing today 00:01:30.400 --> 00:01:34.616 is PLUM, which stands for Propagation of Local Undamped Motion. 00:01:34.640 --> 00:01:37.840 PLUM is a ground motion-based early warning algorithm originally 00:01:37.840 --> 00:01:40.480 developed in Japan. In the PLUM approach, 00:01:40.480 --> 00:01:44.320 observations of strong ground motions are used to forecast shaking directly 00:01:44.320 --> 00:01:47.176 without estimating earthquake source parameters. 00:01:47.200 --> 00:01:50.320 When PLUM detects strong ground motion models, the observed MMI 00:01:50.320 --> 00:01:53.280 is forward-predicted out to a specified radius and assigned to 00:01:53.280 --> 00:01:56.720 a pre-defined set of grid cells, where alerts are determined by the 00:01:56.720 --> 00:02:02.400 grid cells predicted to have MMI above a set threshold, in our case MMI 4. 00:02:02.400 --> 00:02:05.440 While the U.S. version of PLUM does not issue early warning alerts 00:02:05.440 --> 00:02:11.024 to the public, it has been in internal real-time testing since 2019. 00:02:11.883 --> 00:02:14.640 Before we examine individual earthquakes, let us first take 00:02:14.640 --> 00:02:16.800 a bird’s-eye look of all of the earthquakes issued 00:02:16.800 --> 00:02:20.616 early warning alerts by ShakeAlert or PLUM in 2021. 00:02:20.640 --> 00:02:23.600 This map shows the magnitude 3 and larger earthquakes that have 00:02:23.600 --> 00:02:26.560 occurred within the ShakeAlert reporting boundary in 2021, 00:02:26.560 --> 00:02:29.120 colored by magnitude, where the stars indicate the 00:02:29.120 --> 00:02:33.520 earthquakes that were issued alerts. Of the 400 or so magnitude 3 and larger 00:02:33.520 --> 00:02:36.720 earthquakes that have occurred, 30 of these were issued early warning 00:02:36.720 --> 00:02:39.576 alerts either by ShakeAlert, PLUM, or both. 00:02:39.600 --> 00:02:42.480 The table shows more details about the alerted earthquakes, 00:02:42.480 --> 00:02:45.920 organized by latitude. Earthquakes are issued public alerts 00:02:45.920 --> 00:02:48.880 via ShakeAlert when the ShakeAlert magnitude estimate is above 00:02:48.880 --> 00:02:53.576 magnitude 4.5, which I indicate in green in the right-most column. 00:02:53.600 --> 00:02:56.480 The earthquakes issued PLUM alerts are indicated by purple in 00:02:56.480 --> 00:02:59.600 the column next to that. Of these 30 earthquakes, 00:02:59.600 --> 00:03:04.536 ShakeAlert issued public alerts for 17, and PLUM issued alerts for 23. 00:03:04.560 --> 00:03:07.840 Many of these earthquakes are widely felt, with 13 earthquakes having 00:03:07.840 --> 00:03:11.200 over 1,000 reports in the USGS Did You Feel It? system, 00:03:11.200 --> 00:03:14.536 where four of these have over 10,000 reports. 00:03:14.560 --> 00:03:17.280 The earthquakes that have larger numbers of Did You Feel It? reports 00:03:17.280 --> 00:03:20.480 seem to depend largely on the location of the earthquake. 00:03:20.480 --> 00:03:23.760 While the magnitude 6 Antelope Valley earthquake is the second-largest 00:03:23.760 --> 00:03:27.200 earthquake to have occurred in California last year, and has the highest 00:03:27.200 --> 00:03:31.440 number of Did You Feel It? reports at over 23,000, the other three earthquakes 00:03:31.440 --> 00:03:35.040 with over 10,000 Did You Feel It? reports all have magnitudes of less 00:03:35.040 --> 00:03:38.640 than 4.5 but are located in the San Francisco Bay Area 00:03:38.640 --> 00:03:43.280 and in the Los Angeles metro area. Nearly all of these 30 earthquakes have 00:03:43.280 --> 00:03:47.680 reports of significantly felt shaking, corresponding to MMI 4 and higher, 00:03:47.680 --> 00:03:50.880 and many of these earthquakes have reports of damaging shaking, 00:03:50.880 --> 00:03:55.788 which are indicated in the third column to the right, colored by MMI. 00:03:56.819 --> 00:04:00.128 For the rest of this talk, I will focus on a comparison of the alert regions 00:04:00.128 --> 00:04:04.987 by both PLUM and ShakeAlert for the five earthquakes highlighted here. 00:04:05.026 --> 00:04:07.840 For this alert performance comparison, I will be showing what the alert regions 00:04:07.840 --> 00:04:10.960 would look like using the real-time earthquake source characterizations 00:04:10.960 --> 00:04:15.193 and station observations but with alert regions using the preferred 00:04:15.193 --> 00:04:18.356 alerting strategies for each early warning approach. 00:04:18.380 --> 00:04:21.040 These alerting strategies are chosen to maximize alert performance for 00:04:21.040 --> 00:04:25.680 potentially damaging and significantly felt shaking targets of MMI 4 and larger 00:04:25.680 --> 00:04:32.134 while keeping over-alerting to locations that are likely to still feel some shaking. 00:04:32.158 --> 00:04:34.240 For ShakeAlert, our preferred alerting parameters are shown 00:04:34.240 --> 00:04:37.989 in the green box on the left. We will be using the ShakeAlert 00:04:37.989 --> 00:04:41.208 contour product with proposed updates to the ground motion models that 00:04:41.208 --> 00:04:44.599 make it more consistent with ShakeMap products, so these may be a little 00:04:44.599 --> 00:04:47.499 different than the operational alert regions that were issued 00:04:47.499 --> 00:04:51.899 at the time of the earthquakes. We analyzed over 140 ShakeMaps 00:04:51.899 --> 00:04:56.619 and demonstrated that an alert threshold of MMI 3.5 is optimal for alerts to 00:04:56.619 --> 00:05:00.139 all damaging shaking targets as well as for most MMI 4 shaking 00:05:00.139 --> 00:05:04.428 locations when ShakeAlert estimates source parameters accurately. 00:05:04.452 --> 00:05:08.219 Here is an example of these optimization comparisons for the 1989 00:05:08.219 --> 00:05:12.733 magnitude 6.9 Loma Prieta earthquake. These maps show the ShakeMap of the 00:05:12.733 --> 00:05:17.108 earthquake colored by alert quality relative to a set target shaking threshold, 00:05:17.108 --> 00:05:20.882 where the alert region in each map uses different alert thresholds. 00:05:20.882 --> 00:05:24.336 The bar plots show alert quality totals for these alert thresholds 00:05:24.336 --> 00:05:28.186 in terms of area on the top and population on the bottom. 00:05:28.186 --> 00:05:31.856 An optimal alert strategy is one that minimizes the amount of missed 00:05:31.856 --> 00:05:35.776 alerts in red and reduces the amount of over-alerting in yellow, 00:05:35.776 --> 00:05:39.696 where our preferred MMI 3.5 alert threshold for the ShakeAlert contour 00:05:39.696 --> 00:05:43.740 product is highlighted by the green box in sub-plot C. 00:05:43.780 --> 00:05:46.810 And here are our preferred PLUM alert parameters. 00:05:46.810 --> 00:05:50.296 We used a cost-benefit analysis of the earthquakes in the ShakeAlert testing 00:05:50.296 --> 00:05:53.616 data set to determine optimal PLUM prediction radii and alert 00:05:53.616 --> 00:05:56.816 thresholds given the constraint that alert regions will be based on 00:05:56.816 --> 00:05:59.497 the grid used in the ShakeAlert grid product. 00:05:59.497 --> 00:06:02.816 The max here show an example of these comparisons using three PLUM 00:06:02.816 --> 00:06:07.488 prediction radii for the 2019 magnitude 7.1 Ridgecrest earthquake, where the 00:06:07.513 --> 00:06:10.553 ShakeMap is colored by alert quality and expected warning time 00:06:10.553 --> 00:06:14.073 for each strategy. Similar to the ShakeAlert alert 00:06:14.073 --> 00:06:18.073 strategy optimization shown in the previous slide, an optimal PLUM alert 00:06:18.073 --> 00:06:21.593 strategy is one that minimizes the amount of missed alerts in red and late 00:06:21.593 --> 00:06:25.809 alerts in black while also reducing the amount of over-alerting in yellow. 00:06:25.833 --> 00:06:29.593 We found that a 60-kilometer prediction radius and an MMI 4 alert 00:06:29.593 --> 00:06:32.393 threshold are the preferred PLUM alerting parameters 00:06:32.393 --> 00:06:36.433 for the shaking targets we are considering in this analysis. 00:06:37.433 --> 00:06:41.113 For the alert performance comparisons, I will be showing maps like this, 00:06:41.113 --> 00:06:44.073 which compare ShakeAlert and PLUM alert regions against the 00:06:44.073 --> 00:06:48.209 observed shaking intensity distributions from ShakeMap and Did You Feel It?. 00:06:48.233 --> 00:06:52.609 These are a bit complicated, so let’s walk through them a bit. 00:06:52.633 --> 00:06:55.593 In the background, I have plotted the ShakeMap from the earthquake as the 00:06:55.593 --> 00:06:58.953 contour map and intensity observations from Did You Feel It? reports in the 00:06:58.953 --> 00:07:02.233 squares, where I have highlighted the locations and reports of 00:07:02.233 --> 00:07:06.873 MMI 4-plus shaking in dark blue. The triangles are the real-time seismic 00:07:06.873 --> 00:07:10.873 station locations, the star is the epicenter, and the concentric circles 00:07:10.873 --> 00:07:15.489 show the approximate extent of the S wave at 10-second intervals. 00:07:15.513 --> 00:07:19.193 For ShakeAlert, I show the ShakeAlert alert parameters by the octagons 00:07:19.193 --> 00:07:24.233 colored by the alert threshold MMI. Our preferred MMI 3.5 alert contour 00:07:24.233 --> 00:07:27.993 is outlined in black, but I also show the other ShakeAlert contours 00:07:27.993 --> 00:07:31.042 used for public alerts outlined in gray. 00:07:31.042 --> 00:07:34.313 The green circle shows the extent of the S wave at the time the first alert 00:07:34.313 --> 00:07:37.513 was issued by ShakeAlert. This represents the approximate 00:07:37.513 --> 00:07:41.433 minimum size of the initial late alert zone where locations outside 00:07:41.433 --> 00:07:44.073 of the circle may get some warning while locations 00:07:44.073 --> 00:07:47.969 inside the circle will not get a timely alert. 00:07:47.993 --> 00:07:51.913 For PLUM, the PLUM alert region perimeter is highlighted in pink, 00:07:51.913 --> 00:07:56.929 and the purple circle shows the estimated initial late alert zone. 00:07:56.953 --> 00:07:59.513 Combining them all together gives us this map. 00:07:59.513 --> 00:08:02.953 The magnitude 5.3 Calipatria earthquake near the Salton Sea 00:08:02.953 --> 00:08:06.313 shown here is a nice example of good alert performance behavior 00:08:06.313 --> 00:08:08.449 by both ShakeAlert and PLUM. 00:08:08.473 --> 00:08:12.153 Both approaches issue alerts at approximately the same time, and both 00:08:12.153 --> 00:08:17.249 have alert regions that encompass the locations that experience target shaking. 00:08:17.273 --> 00:08:20.153 Another example of good alert performance for both ShakeAlert 00:08:20.153 --> 00:08:22.713 and PLUM is the magnitude 6.2 Petrolia 00:08:22.713 --> 00:08:25.249 earthquake sequence that occurred last month. 00:08:25.273 --> 00:08:28.233 This event is complicated, as Kathryn Materna will discuss 00:08:28.233 --> 00:08:32.793 in her talk later in this session. Analysis of this event is still ongoing, 00:08:32.793 --> 00:08:35.753 but recent results suggest that there were two earthquakes, 00:08:35.753 --> 00:08:40.713 the magnitude 5.7 foreshock offshore, followed by the magnitude 6.2 00:08:40.713 --> 00:08:44.793 main shock 11 seconds later. During this earthquake sequence, 00:08:44.793 --> 00:08:47.673 both ShakeAlert and PLUM issued alerts with the detection of the 00:08:47.673 --> 00:08:50.553 foreshock, where these alerts were issued several seconds 00:08:50.553 --> 00:08:53.809 before the S wave from the main shock reached the surface. 00:08:53.833 --> 00:08:57.113 For ShakeAlert, there was some epicenter mislocation, 00:08:57.113 --> 00:09:01.089 and the magnitude estimate did not get above magnitude 5.6. 00:09:01.113 --> 00:09:06.313 However, the MMI 3.5 alert contour is able to compensate for these errors 00:09:06.313 --> 00:09:09.913 in most locations that experience MMI 4-plus shaking by the 00:09:09.913 --> 00:09:13.249 main shock are issued an alert by ShakeAlert. 00:09:13.273 --> 00:09:16.953 For PLUM, PLUM’s first alert was issued at about 2 seconds after 00:09:16.953 --> 00:09:20.473 ShakeAlert’s first alert once detection thresholds were exceeded by the 00:09:20.473 --> 00:09:24.049 foreshock’s S wave arrival at enough onshore stations. 00:09:24.073 --> 00:09:27.673 PLUM’s alert region for this sequence is able to encompass all onshore 00:09:27.673 --> 00:09:32.369 locations of MMI 4-plus shaking from the magnitude 6.2 main shock. 00:09:32.393 --> 00:09:35.833 Alongside reducing missed alerts, the additional alerts by PLUM 00:09:35.833 --> 00:09:38.929 would have provided timely warning to these locations. 00:09:38.953 --> 00:09:42.873 Next, let’s take a look at the magnitude 6 Antelope Valley earthquake. 00:09:42.873 --> 00:09:46.073 This was the second-largest earthquake to occur in the ShakeAlert reporting 00:09:46.073 --> 00:09:50.529 boundary this year, and it unfortunately caused some problems for ShakeAlert. 00:09:50.553 --> 00:09:54.153 The sparse station coverage, as well as triggers from a small foreshock on the 00:09:54.153 --> 00:09:57.833 closest stations caused ShakeAlert to severely mislocate the epicenter 00:09:57.833 --> 00:10:01.433 of the main shock over 70 kilometers to the southeast, which led to 00:10:01.433 --> 00:10:05.673 magnitude underestimation. This mislocation then prevented 00:10:05.673 --> 00:10:09.353 triggers from later stations from being properly associated with this event, 00:10:09.353 --> 00:10:13.593 which caused ShakeAlert to create split events, including a magnitude 4.8 00:10:13.593 --> 00:10:17.913 event near Stockton to the southwest. Because of these large source estimation 00:10:17.913 --> 00:10:23.113 errors, neither of the ShakeAlert events that created public alerts have MMI 3.5 00:10:23.113 --> 00:10:28.275 alert contours that include the areas that experienced significant shaking. 00:10:28.300 --> 00:10:31.820 For PLUM, the ground motions from the small foreshock did not exceed 00:10:31.820 --> 00:10:35.660 detection thresholds, so PLUM issued an alert for the magnitude 6 00:10:35.660 --> 00:10:38.836 main shock about 10 seconds faster than ShakeAlert. 00:10:38.860 --> 00:10:42.540 Due to the sparse station coverage, the first stations to detect strong 00:10:42.540 --> 00:10:44.940 shaking weren’t immediate neighbors of each other. 00:10:44.940 --> 00:10:48.436 So PLUM also had multiple detections for this earthquake. 00:10:48.460 --> 00:10:51.900 However, unlike source-based early warning algorithms, 00:10:51.900 --> 00:10:54.780 multiple PLUM detections for the same event do not affect 00:10:54.780 --> 00:10:58.220 the PLUM alert region. This is because PLUM alert regions, 00:10:58.220 --> 00:11:00.700 by definition, surround the stations observed to have 00:11:00.700 --> 00:11:04.140 strong ground motions. So multiple PLUM event detections 00:11:04.140 --> 00:11:08.660 would not cause source estimation errors that would impact alert regions. 00:11:09.580 --> 00:11:13.796 Now let’s take a look at the magnitude 4.7 Truckee earthquake. 00:11:13.820 --> 00:11:17.740 Sparse station spacing also caused ShakeAlert to mislocate the earthquake, 00:11:17.740 --> 00:11:21.820 however for this events, ShakeAlert’s initial magnitude estimate was too high 00:11:21.820 --> 00:11:25.580 at magnitude 6, and this magnitude overestimate was used in the 00:11:25.580 --> 00:11:30.220 first alert issued to the public. While the preferred MMI 3.5 alert 00:11:30.220 --> 00:11:33.660 threshold would have kept alerts to locations that still felt shaking, 00:11:33.660 --> 00:11:38.940 the operational MMI 2.5 alert region used to issue public alerts sent alerts all 00:11:38.940 --> 00:11:44.196 the way to the San Francisco Bay Area, which did not feel the earthquake. 00:11:44.220 --> 00:11:47.740 PLUM detected the earthquake about 10 seconds faster than ShakeAlert. 00:11:47.740 --> 00:11:51.020 But the sparse station spacing still prevented PLUM alerts from being 00:11:51.020 --> 00:11:55.220 timely in the locations of target shaking. And while PLUM alerts will have 00:11:55.220 --> 00:11:58.755 [inaudible] at over-alerting surrounding the locations that experienced target 00:11:58.780 --> 00:12:03.020 shaking, over-alerting is limited to the extent of the 60-kilometer PLUM 00:12:03.020 --> 00:12:08.848 prediction radius. So large-scale over- alerting is not possible in PLUM. 00:12:09.740 --> 00:12:13.660 Finally, let’s go back to the Mendocino area and examine the alert performance 00:12:13.660 --> 00:12:18.676 for the offshore magnitude 5.1 Petrolia earthquake that occurred last July. 00:12:18.700 --> 00:12:22.540 This earthquake was a deeper event with 31-kilometer source depth. 00:12:22.540 --> 00:12:25.900 So ShakeAlert likely detected the earthquake before the S waves reached 00:12:25.900 --> 00:12:30.596 the surface, which is why you don’t see a green late alert zone on the map. 00:12:30.620 --> 00:12:34.380 However, ShakeAlert underestimated the magnitude for this event, 00:12:34.380 --> 00:12:37.476 preventing a public alert from being issued. 00:12:37.500 --> 00:12:41.260 PLUM detected and issued and alert for this earthquake, and despite 00:12:41.260 --> 00:12:44.460 PLUM detecting the event much later than ShakeAlert once the S waves 00:12:44.460 --> 00:12:47.980 reached the surface, a PLUM alert would likely have been timely for 00:12:47.980 --> 00:12:50.996 the locations that experienced target shaking. 00:12:51.020 --> 00:12:53.900 Offshore and edge-of-network earthquakes can be problematic 00:12:53.900 --> 00:12:57.740 for ShakeAlert source characterization due to the one-sided station azimuthal 00:12:57.740 --> 00:13:01.820 coverage. For events like the one here, PLUM would help confirm that 00:13:01.820 --> 00:13:07.820 significant shaking is occurring onshore and that an alert is needed. 00:13:07.820 --> 00:13:11.500 In summary, we discussed the real-time alerting behavior of ShakeAlert and 00:13:11.500 --> 00:13:16.860 PLUM for earthquakes in 2021. For ShakeAlert, the MMI 3.5 contour 00:13:16.860 --> 00:13:19.900 product using proposed updates to the ground motion models used 00:13:19.900 --> 00:13:23.820 in the alerting process yields improved alert performance for significant 00:13:23.820 --> 00:13:27.636 shaking targets compared to other operational contour products. 00:13:27.660 --> 00:13:32.220 For PLUM, once significant shaking is detected, PLUM alerts areas that 00:13:32.220 --> 00:13:35.956 surround the stations with observed significant ground motions. 00:13:35.980 --> 00:13:39.420 For most earthquakes, PLUM and ShakeAlert have comparable alert 00:13:39.420 --> 00:13:43.180 performance, where first alert times are usually within one or two seconds 00:13:43.180 --> 00:13:46.460 of each other, and the alert region coverage is similar, 00:13:46.460 --> 00:13:50.196 like for the magnitude 5.3 Calipatria earthquake. 00:13:50.220 --> 00:13:54.860 However, for cases where ShakeAlert has large errors in its source estimations, 00:13:54.860 --> 00:13:58.276 these source errors contribute to errors in the alert region. 00:13:58.300 --> 00:14:01.980 For cases where the ShakeAlert epicenter is off by a large distance, 00:14:01.980 --> 00:14:05.820 and/or if the ShakeAlert magnitude is significantly underestimated, 00:14:05.820 --> 00:14:09.180 the errors in the resulting alert contours may prevent locations 00:14:09.180 --> 00:14:12.276 that experience target shaking from being alerted. 00:14:12.300 --> 00:14:15.820 Because PLUM alerts surround the stations with significant shaking, 00:14:15.820 --> 00:14:19.660 PLUM alerts are not affected by source estimation errors, so it can 00:14:19.660 --> 00:14:23.396 help provide a backup to source-based alerts when there are problems. 00:14:23.420 --> 00:14:26.620 We also observed that there were no negative impacts to the alert 00:14:26.620 --> 00:14:30.869 performance due to PLUM, such as extreme over-alerting. 00:14:31.740 --> 00:14:35.260 PLUM is a simple and robust approach to earthquake early warning, 00:14:35.260 --> 00:14:38.540 and I hope this presentation demonstrated how integrating PLUM 00:14:38.540 --> 00:14:42.060 into ShakeAlert in the future could improve early warning performance 00:14:42.060 --> 00:14:46.996 in the United States as it has already done for early warning in Japan. 00:14:47.020 --> 00:14:49.420 Thank you very much for listening to my talk.