How Low Should We Alert? Quantifying Intensity Threshold Alerting Strategies for Earthquake Early Warning in the United States

Abstract

Alerting the public of potential dangers during rapid-onset natural hazard events always carries some uncertainty due to the limited real-time information available to develop accurate warnings as well as the natural randomness of the system. For extreme weather phenomena such as tornadoes during supercell thunderstorms, flash floods during atmospheric river events, and lahars during volcanic eruptions, warning centers do not know exactly where these events will occur nor the exact path they will take when they do. As such, public alerts for these phenomena are issued to wider areas than may ultimately be impacted to account for these uncertainties (e.g.,

Incorporating Intensity Distance Attenuation Into PLUM Ground‐Motion‐Based Earthquake Early Warning in the United States: The APPLES Configuration

“…EEW alerting strategies are typically a trade‐off between alert accuracy and alert timeliness. While matching predicted ground motions with observations does not generally yield the best EEW performance because such strategies tend to increase missed alerts and reduce available warning times (Cochran et al., 2022; Minson et al., 2018, 2019, 2022; Saunders, Minson, & Baltay, 2022), one of our goals for the APPLES configuration is to produce ground‐motion distributions that are more consistent with the median‐expected predictions used in current ShakeAlert operations. For this analysis, we consider only onshore MMI ≥ 2.5 locations within the ShakeAlert states of California, Oregon, and Washington.…”

“…ShakeAlert outputs the median‐expected shaking distributions on a grid or contour as calculated using ground‐motion models (Thakoor et al., 2019). Alert thresholds used for public alerting (e.g., MMI 3.5) are then selected to better ensure timely alerts are issued to locations that experience higher, potentially damaging shaking intensities (Kohler et al., 2020; Saunders, Minson, & Baltay, 2022).…”

“…In the current PLUM approach, the predicted ground motion at a given location corresponds to the maximum‐observed ground motion at seismic stations within a set distance of the location. By forward‐predicting ground‐motion distributions directly from observed ground motions, PLUM avoids potential sources of errors present in source‐characterization‐based EEW approaches, as errors in rapid earthquake location and magnitude estimates propagate into errors in the resulting alert regions computed from ground‐motion models using these source estimates (Chung et al., 2020; Cochran, Aagaard, et al., 2018; Saunders, Minson, & Baltay, 2022; Williamson et al., 2023). PLUM was added to the Japan Meteorological Agency (JMA) EEW system because source‐parameter‐based algorithm performance was found to decline significantly due to widely‐spaced, but near coincident events during the 2011 M 9.0 Tohoku aftershock sequence (Kodera et al., 2018).…”

“…Thirdly, the ground‐motion predictions produced by PLUM, being outward predictions of maximum shaking, are different from the median‐expected ground‐motion predictions produced by ShakeAlert's source‐parameter‐based EEW algorithms (Böse, Andrews, O’Rourke, et al., 2023). This results in these approaches having different preferred alert thresholds for achieving similar alert performance when considering a given target intensity threshold for EEW alerts (Cochran et al., 2022; Saunders, Minson, & Baltay, 2022). These differences prevent a straightforward combination of PLUM ground‐motion predictions with those from ShakeAlert, which makes it difficult to integrate PLUM into the ShakeAlert system.…”

Detecting the Unseen: Understanding the Mechanisms and Working Principles of Earthquake Sensors

“…The results of physical experiments show that using these wireless sensors, the monitoring center can display the monitoring image of the monitoring area in real time and visualize the collected sensor data [9]. The ongoing research have been using intelligent monitoring algorithms (such as object recognition or intrusion detection) on monitoring nodes to achieve better monitoring performance [49]. Other advancements include optimization of the mechanical design of the monitoring nodes (e. g., miniaturization or lightweight) and the positioning algorithms for the sensor nodes.…”