No abstract

The 2016 Kumamoto earthquake (Kumamoto earthquake sequence) is an extremely high-seismicity event that has been occurring across Kumamoto and Oita Prefectures in Japan since April 14, 2016 (JST). The earthquake early warning system of the Japan Meteorological Agency (JMA) issued warnings for 19 events in the Kumamoto earthquake sequence from April 14 to 19, under some of the heaviest loading conditions since the system began operating in 2007. We analyzed the system performance for cases where a warning was issued and/or strong motion was actually observed. The results indicated that the system exhibited remarkable performance, especially for the most destructive earthquakes in the Kumamoto earthquake sequence. In addition, the system did not miss or seriously under-predict strong motion of any large earthquake from April 14 to 30. However, in four cases, the system issued over-predicted warnings due to the simultaneous occurrence of small earthquakes within a short distance, which implies a fundamental obstacle in trigger-data classifications based solely on arrival time. We also performed simulations using the integrated particle filter (IPF) and propagation of local undamped motion (PLUM) methods, which JMA plans to implement to address over-prediction for multiple simultaneous earthquakes and under-prediction for massive earthquakes with large rupture zones. The simulation results of the IPF method indicated that the IPF method is highly effective at minimizing over-prediction even for multiple simultaneous earthquakes within a short distance, since it adopts a trigger-data classification using velocity amplitude and hypocenter determinations using not-yet-arrived data. The simulation results of the PLUM method demonstrated that the PLUM method is capable of issuing warnings for destructive inland earthquakes more rapidly than the current system owing to the use of additional seismometers that can only be incorporated by this method.

Although numerous Earthquake Early Warning (EEW) algorithms have been developed to date, we lack a detailed understanding of how often and under what circumstances useful ground motion alerts can be provided to end users. In particular, it is unclear how often EEW systems can successfully alert sites with high ground motion intensities. These are the sites that arguably need EEW alerts the most, but they are also the most challenging ones to alert because they tend to be located close to the epicenter where the seismic waves arrive first. Here we analyze the alerting performance of the Propagation of Local Undamped Motion (PLUM), Earthquake Point‐Source Integrated Code (EPIC), and Finite‐Fault Rupture Detector (FinDer) algorithms by running them retrospectively on the seismic strong‐motion data of the 219 earthquakes in Japan since 1996 that exceeded Modified Mercalli Intensity (MMI) of 4.5 on at least 10 sites (Mw 4.5–9.1). Our analysis suggests that, irrespective of the algorithm, EEW end users should expect that EEW can often but not always provide useful alerts. Using a conservative warning time (tw) definition, we find that 40–60% of sites with strong to extreme shaking levels receive alerts with tw > 5 s. If high‐intensity shaking is caused by shallow crustal events, around 50% of sites with strong (MMI~6) and <20% of sites with severe and violent (MMI ≥ 8) shaking receive alerts with tw > 5 s. Our results provide detailed quantitative insight into the expected alerting performance for EEW algorithms under realistic conditions. We also discuss how operational systems can achieve longer warning times with more precautionary alerting strategies.

We test the Japanese ground‐motion‐based earthquake early warning (EEW) algorithm, propagation of local undamped motion (PLUM), in southern California with application to the U.S. ShakeAlert system. In late 2018, ShakeAlert began limited public alerting in Los Angeles to areas of expected modified Mercalli intensity (IMMI) 4.0+ for magnitude 5.0+ earthquakes. Most EEW systems, including ShakeAlert, use source‐based methods: they estimate the location, magnitude, and origin time of an earthquake from P waves and use a ground‐motion prediction equation to identify regions of expected strong shaking. The PLUM algorithm uses observed ground motions directly to define alert areas and was developed to address deficiencies in the Japan Meteorological Agency source‐based EEW system during the 2011 Mw 9.0 Tohoku earthquake sequence. We assess PLUM using (a) a dataset of 193 magnitude 3.5+ earthquakes that occurred in southern California between 2012 and 2017 and (b) the ShakeAlert testing and certification suite of 49 earthquakes and other seismic signals. The latter suite includes events that challenge the current ShakeAlert algorithms. We provide a first‐order performance assessment using event‐based metrics similar to those used by ShakeAlert. We find that PLUM can be configured to successfully issue alerts using IMMI trigger thresholds that are lower than those implemented in Japan. Using two stations, a trigger threshold of IMMI 4.0 for the first station and a threshold of IMMI 2.5 for the second station PLUM successfully detect 12 of 13 magnitude 5.0+ earthquakes and issue no false alerts. PLUM alert latencies were similar to and in some cases faster than source‐based algorithms, reducing area that receives no warning near the source that generally have the highest ground motions. PLUM is a simple, independent seismic method that may complement existing source‐based algorithms in EEW systems, including the ShakeAlert system, even when alerting to light (IMMI 4.0) or higher ground‐motion levels.

ABSTRACT We evaluate the timeliness and accuracy of ground-motion-based earthquake early warning (EEW) during the July 2019 M 6.4 and 7.1 Ridgecrest earthquakes. In 2018, we began retrospective and internal real-time testing of the propagation of local undamped motion (PLUM) method for earthquake warning in California, Oregon, and Washington, with the potential that PLUM might one day be included in the ShakeAlert EEW system. A real-time version of PLUM was running on one of the ShakeAlert EEW system’s development servers at the time of the 2019 Ridgecrest sequence, allowing us to evaluate the timeliness and accuracy of PLUM’s warnings for the M 6.4 and 7.1 mainshocks in real time with the actual data availability and latencies of the operational ShakeAlert EEW system. The latter is especially important because high-data latencies during the M 7.1 earthquake degraded ShakeAlert’s performance. PLUM proved to be largely immune to these latencies. In this article, we present a retrospective analysis of PLUM performance and explore three potential regional alerting strategies ranging from spatially large regions (counties), to moderate-size regions (National Weather Service public forecast zones), to high-spatial specificity (50 km regular geographic grid). PLUM generated initial shaking forecasts for the two mainshocks 5 and 6 s after their respective origin times, and faster than the ShakeAlert system’s first alerts. PLUM was also able to accurately forecast shaking across southern California for all three alerting strategies studied. As would be expected, a cost-benefit analysis of each approach illustrates trade-offs between increasing warning time and minimizing the area receiving unneeded alerts. Choosing an optimal alerting strategy requires knowledge of users’ false alarm tolerance and minimum required warning time for taking protective action, as well as the time required to distribute alerts to users.