Earthquake early warning system
An earthquake early warning (EEW) system is a technological network that detects the onset of an earthquake through seismic sensors, processes initial data to estimate its magnitude and location, and rapidly disseminates alerts to provide seconds to minutes of advance notice before strong ground shaking reaches populated areas.[1] These systems exploit the physics of seismic wave propagation, where faster but less damaging primary (P) waves arrive first, allowing time for warnings to be issued via electronic communication— which travels near the speed of light—before the slower, more destructive secondary (S) waves and surface waves impact structures and people.[1] Core components of EEW systems include dense arrays of seismometers and accelerometers for real-time monitoring, centralized processing hubs that employ algorithms to analyze wave data and predict shaking intensity, and dissemination infrastructure such as cell broadcasts, mobile apps, and automated controls for critical systems like transportation and utilities.[1] The United States' ShakeAlert system, managed by the U.S. Geological Survey (USGS), exemplifies this with over 1,500 sensors across the West Coast, delivering alerts to millions through partnerships with tech companies and public agencies since achieving public operational status in 2019.[2] Benefits include enabling life-saving actions—such as "drop, cover, and hold on"—halting high-speed trains, safeguarding semiconductor fabrication, and minimizing economic disruptions, with studies showing potential reductions in fatalities by up to 50% in regions with effective systems.[1] The concept of EEW originated in the 19th century with early proposals for telegraph-based alerts following the 1868 Hayward earthquake, but systematic research accelerated in the 1980s through efforts in Japan and the U.S.[3] The first public operational system launched in Mexico in 1991, using coastal sensors to warn Mexico City of distant subduction zone events, followed by Japan's nationwide rollout in 2007 after extensive testing. As of 2025, EEW systems are operational in several countries including Japan, Mexico, the United States, Taiwan, China, South Korea, Canada, Turkey, Romania, and Italy, with public alerts available in many of these.[4][5][6][7] Despite these advances, limitations persist, including no warnings for sites within 10-20 kilometers of the epicenter due to rapid wave arrival, potential false alarms from aftershocks or non-earthquake events, and challenges in regions with sparse sensor coverage or complex geology.[1]Fundamentals
Seismic Wave Basics
Earthquakes generate seismic waves that propagate through the Earth, with the primary distinction in early warning systems arising from the differing speeds and properties of these waves. Primary waves, or P-waves, are compressional waves that cause particles in the medium to vibrate parallel to the direction of wave propagation, similar to sound waves. These waves travel through solids, liquids, and gases at speeds typically ranging from 6 to 8 km/s in the Earth's crust, depending on the rock type and depth.[8][9] Secondary waves, known as S-waves, are shear waves that cause particles to oscillate perpendicular to the direction of propagation, resulting in a shearing motion that cannot pass through liquids. S-waves travel more slowly than P-waves, at approximately 3 to 4 km/s in crustal rocks, arriving after the P-waves and contributing significantly to ground shaking due to their transverse motion.[10][11] Surface waves, which include Love and Rayleigh waves, propagate along the Earth's surface and generally arrive after the body waves (P and S). These waves travel at speeds of about 2 to 4.5 km/s and have larger amplitudes, making them responsible for much of the structural damage during earthquakes, as their motion causes prolonged rolling or rocking of the ground.[12] The feasibility of earthquake early warning systems hinges on the speed difference between P-waves and S-waves: detecting the faster P-waves at distant sensors allows estimation of the impending arrival of the more destructive S-waves and surface waves. By identifying the P-wave onset, systems can project the location and magnitude of the earthquake in real-time, providing seconds to tens of seconds of warning before strong shaking begins, enabling protective actions.[1] This lead time is calculated using the basic wave travel time equation, d = v \times t, where d is the distance from the epicenter to the location, v is the wave speed, and t is the travel time. The difference in travel times for P- and S-waves over the same distance yields the available warning period, with longer distances generally providing more lead time due to the velocity disparity.[1]Detection and Lead Time
Earthquake early warning systems detect the onset of an earthquake by monitoring the arrival of primary (P) waves at seismic stations, triggering alerts when specific ground motion parameters exceed predefined thresholds. These thresholds are typically based on P-wave acceleration or velocity to ensure rapid identification of potentially significant events while minimizing false alarms. For example, systems like MyShake use a P-wave acceleration threshold of greater than 0.01 g (where g is the acceleration due to gravity) to initiate processing, as this level indicates sufficient energy release for warning issuance.[13] Similarly, research on threshold-based methods employs parameters such as peak displacement (Pd) or a characteristic period (τc) in the initial P-wave window to confirm triggering.[14] These thresholds are calibrated to balance sensitivity and reliability, drawing from empirical data on small-magnitude events that evolve into larger ones.[15] The lead time provided by these systems—the interval between alert issuance and the arrival of stronger secondary (S) waves—is fundamentally governed by the physics of seismic wave propagation. It is calculated as the difference in travel times: lead time = (d / v_S) - (d / v_P), where d is the hypocentral distance, v_S is the S-wave velocity (typically around 3.5 km/s in the crust), and v_P is the P-wave velocity (approximately 6 km/s).[16] For near-field locations (e.g., d ≈ 10 km), this yields lead times of just a few seconds (about 1-3 s), sufficient for basic automated responses like halting trains. In far-field scenarios (e.g., d ≈ 100 km), lead times extend to tens of seconds (around 10-20 s), allowing for evacuations or infrastructure protections.[1] These values assume idealized crustal velocities and do not account for processing delays, which can reduce effective lead time by 5-10 seconds in operational systems.[17] Several factors influence the achievable lead time beyond basic wave speeds. Hypocenter depth plays a key role: shallower events (e.g., <10 km) produce shorter wave paths and thus minimal lead times, while deeper hypocenters (e.g., 20-30 km) can slightly increase them due to longer travel distances.[18] A critical limitation is the "blind zone," an area surrounding the epicenter where lead time is zero or effectively negative, meaning strong shaking arrives before or simultaneously with any alert. This zone typically extends 10-20 km from the rupture initiation point, depending on network density and processing speed, as the short S-P wave travel time difference (often <5 s) cannot be overcome by current technology.[19] Within this radius, residents experience immediate shaking without prior warning, underscoring the need for resilient building codes in high-seismic areas.[20]Historical Development
Early Concepts
The earliest conceptual foundation for an earthquake early warning system emerged in 1868, when J.D. Cooper, a physician in San Francisco, proposed using telegraph cables radiating from the city to detect initial ground tremors at distant points and transmit alarms via electrical signals before stronger shaking arrived in urban areas.[21] This idea, inspired by the 1868 Hayward earthquake, envisioned a network of sensors to provide seconds to minutes of lead time for protective actions, though it relied on rudimentary technology and was never implemented.[22] Building on such notions, foundational observations of seismic wave propagation in the early 20th century provided the scientific groundwork for distinguishing fast-arriving primary (P) waves from slower secondary (S) waves, essential for early detection. Andrija Mohorovičić's analysis of the 1909 Kulpa Valley earthquake in Croatia identified distinct wave arrivals, revealing velocity differences that later enabled the timing critical to warning systems—P waves traveling at about 6 km/s versus S waves at 3.5 km/s in the crust. These insights, formalized in his 1910 paper, shifted focus from post-event analysis to real-time wave monitoring, influencing subsequent prototype designs. In the 1980s, Japan advanced these concepts through practical experiments using off-the-shelf seismometers to detect P waves and issue rapid alerts. The Urgent Earthquake Detection and Alarm System (UrEDAS), developed by Japanese National Railways, employed strong-motion accelerometers along the Shinkansen high-speed rail lines to trigger automatic train braking within 3-5 seconds of P-wave onset, providing up to 10-15 seconds of warning before destructive S waves; it became operational along the Shinkansen in 1992.[23][24] This prototype demonstrated feasibility for infrastructure protection but was limited to rail corridors, using simple amplitude thresholds for magnitude estimation.[23] Simultaneously, Mexico initiated the conceptual design of its first operational prototype following the devastating 1985 Michoacán earthquake, which highlighted the need for warnings in Mexico City, 300-400 km from coastal subduction zones. In 1988, the Center for Research on Seismic Engineering and Structural Dynamics (CIRES) began developing the Seismic Alert System (SAS), later expanded as SASMEX, using a network of low-cost accelerometers to detect P waves and broadcast sirens, aiming for 30-60 seconds of lead time.[25] Initial tests in the late 1980s focused on threshold-based detection for expected magnitudes above 6.0.[26] Early prototypes like UrEDAS and SAS revealed key challenges, particularly false alarms triggered by non-earthquake signals such as trains, industrial vibrations, or small local quakes misidentified as larger distant events. These issues, noted in 1980s field tests, underscored the need for improved signal discrimination to maintain public trust and system reliability, with false positive rates sometimes exceeding 10% in noisy environments.[27]Major Implementations
The development of major earthquake early warning (EEW) systems began in the 1990s with pioneering implementations focused on critical infrastructure. In Japan, the Urgent Earthquake Detection and Alarm System (UrEDAS) became operational in 1992 specifically for the Shinkansen high-speed rail network, enabling automatic train braking upon detection of initial seismic waves to prevent derailments.[23] This system represented one of the earliest operational EEW deployments, providing seconds of lead time for rail safety along Japan's seismically active corridors.[28] The 2000s saw expansions and new national systems, building on these foundations. Mexico's Seismic Alert System (SASMEX) initiated experimental operations in 1991, with public alerts beginning in Mexico City in 1993 and significant expansions in the early 2000s to cover additional regions along the Pacific coast, incorporating more sensors to detect earthquakes from the Guerrero subduction zone and beyond. These enhancements allowed for broader dissemination of warnings via radio, sirens, and later mobile networks, providing up to 60 seconds of advance notice in distant urban areas.[29] In Taiwan, the Central Weather Bureau's EEW system became operational in 2007, integrating real-time seismic data from over 100 stations to issue alerts nationwide, particularly targeting the densely populated Taipei Basin.[30] During the 2010s, EEW adoption accelerated with pilots and public rollouts in North America and Asia. The United States' ShakeAlert system began its pilot phase in 2015, delivering test alerts to select users in California, Oregon, and Washington, and achieved full public rollout in 2019 through integration with wireless emergency alerts, enabling millions to receive warnings via smartphones.[31] South Korea's Earthquake Early Warning System (EEWS), managed by the Korea Meteorological Administration, launched public operations in 2017 following initial testing, providing alerts for magnitudes above 5.0 with lead times of several seconds across the peninsula.[32] The 2020s have marked further global integration and extensions of existing networks. Israel's TRUAA system went operational in January 2022, leveraging a dense seismic array along the Dead Sea Transform to deliver nationwide alerts via the Home Front Command's infrastructure, offering up to 30 seconds of warning for events above magnitude 4.5.[33] In Canada, British Columbia integrated with the U.S. ShakeAlert framework in 2024, launching a provincial EEW component that uses shared seismic data for real-time alerts to over 5 million residents, enhancing cross-border resilience in the Cascadia region.[34] Alaska completed its expansion plans for ShakeAlert integration in 2025, with a Phase 1 technical implementation plan outlining approximately 450 new stations to provide 10-120 seconds of warning for major events, addressing the state's remote and high-risk seismic zones.[35] As of 2025, according to the United Nations Office for Disaster Risk Reduction (UNDRR) report, 119 countries had established some form of multi-hazard early warning system, more than double the number from 2015, reflecting a global push toward integrated EEW capabilities that combine seismic detection with broader disaster alerts.[36]Technical Components
Sensor Networks
Sensor networks form the foundational hardware infrastructure for earthquake early warning (EEW) systems, comprising arrays of specialized instruments deployed to detect and record seismic activity in real time. These networks primarily utilize strong-motion accelerometers, which measure ground accelerations during intense shaking, often reaching up to 3.5 g, making them essential for capturing the damaging phases of earthquakes.[37] Modern implementations increasingly incorporate micro-electro-mechanical systems (MEMS) accelerometers due to their low cost, compact size, and suitability for dense deployment, as seen in systems like Taiwan's P-alert sensors that enable on-site preliminary processing.[38] In contrast, broadband seismometers provide sensitivity to weaker ground motions across a wide frequency range, from hundreds of seconds to hundreds of hertz, with sensitivity to ground motions with a minimum of 10^{-10} m/s², allowing for the identification of initial P-waves that precede stronger shaking.[39] This combination ensures comprehensive coverage from faint precursors to severe jolts, with strong-motion sensors focusing on high-amplitude events and broadband ones on low-amplitude signals for early detection.[40] Network architectures emphasize dense spatial coverage to minimize detection delays, typically featuring stations spaced 20-40 km apart in high-risk zones to achieve lead times of seconds to tens of seconds.[41] Japan's EEW system, operated by the Japan Meteorological Agency, exemplifies this with over 1,000 seismic stations contributing to real-time hypocenter and magnitude calculations, supplemented by additional networks for a total of around 4,000 monitoring points nationwide.[42] Data telemetry is critical for rapid transmission, relying on high-speed fiber optic cables for low-latency delivery in urban areas and satellite links for remote or offshore sites, ensuring sub-second updates to central processing hubs.[43] Since the 2010s, these networks have integrated Global Navigation Satellite Systems (GNSS) stations alongside traditional seismometers, leveraging real-time displacement measurements to refine magnitude estimates for large events (M>7), where seismic saturation can limit accuracy.[44] GNSS data, sampled at high rates (e.g., 1 Hz or more), complements accelerometers by providing absolute ground deformation, enhancing overall system reliability without relying solely on integrated velocity from seismic sensors.[45] Prominent examples include the ShakeAlert system in the western United States, which by the end of 2025 will encompass over 2,000 stations, including seismic and GNSS stations, across California, Oregon, and Washington, enabling public alerts for events as small as M4.0.[46] In California alone, as of May 2025, state-funded expansions reached 99% completion (699 out of 702 stations), with the overall network at 94% completion (1,017 out of 1,115 stations) integrating both strong-motion and broadband sensors to cover urban fault zones like the San Andreas.[47] These deployments prioritize scalability and redundancy, with ongoing additions of GNSS for improved performance in megathrust scenarios, underscoring the evolution toward hybrid sensor ecosystems for robust EEW.[48]Processing Algorithms
Earthquake early warning systems rely on real-time processing algorithms to analyze incoming seismic data, detect events, estimate parameters, and forecast impacts within seconds. The core processing pipeline typically begins with data ingestion from sensor networks, followed by event association to identify coherent seismic activity, magnitude and location estimation, and intensity forecasting to determine potential shaking levels at user locations. This pipeline operates in a distributed computing environment, often using cloud-based infrastructure for low-latency handling of high-volume data streams.[48] Data ingestion involves continuous acquisition and preprocessing of seismic waveforms, including P-wave onsets and ground-motion metrics like peak ground acceleration (PGA). Raw data from multiple stations are timestamped, filtered for noise, and routed to processing centers, where initial triggers flag potential earthquakes based on amplitude thresholds. Event association then integrates these triggers by correlating arrival times and spatial patterns across stations to distinguish true events from noise or unrelated signals; for instance, in the ShakeAlert system, association requires at least four stations within a defined region to contribute data, ensuring robustness against false triggers. This step employs Bayesian or template-matching techniques to build a unified event hypothesis, updating iteratively as more data arrives.[49][48] Magnitude estimation often incorporates finite-fault models to overcome limitations of point-source approximations, particularly for large ruptures. The Earthquake Point-Source Integrated Code (EPIC) algorithm, a point-source method derived from earlier systems like ElarmS, uses the first 0.5 to 4 seconds of P-wave data from nearby stations to estimate epicenter, magnitude, and depth by integrating peak displacement (Pd) and period parameters; it saturates around magnitude 6.5 but provides rapid initial alerts. For larger events, the Finite-Fault Rupture Detector (FinDer) complements EPIC by modeling extended ruptures as line sources, matching observed PGA patterns against precomputed templates to infer fault length and magnitude up to 8.0 or higher, enabling more accurate updates within 10-20 seconds. These models reduce underestimation errors in finite-fault scenarios, with FinDer activating when PGA exceeds 2 cm/s² at multiple stations.[48] The τ_c method provides an alternative for rapid magnitude estimation and lead time prediction, leveraging characteristics of P-wave arrivals. Defined as the ratio of the P-wave displacement spectrum integral to its maximum value over an initial 3-4 second window, τ_c correlates strongly with earthquake size due to its sensitivity to high-frequency content decay; magnitudes are estimated via empirical scaling relations like log(M_w) ≈ 2.0 log(τ_c) + constant, calibrated regionally. By quantifying P-wave arrival delays relative to expected S-wave propagation, τ_c facilitates lead time calculations—typically 5-60 seconds—for sites beyond the epicenter, as lead time equals the difference in travel times scaled by wave speeds (Vp ≈ 6 km/s, Vs ≈ 3.5 km/s). This approach, originally developed for southern California, has been adapted globally, offering stable predictions even with sparse early data. Intensity forecasting translates source parameters into site-specific shaking predictions, primarily using PGA as a proxy for modified Mercalli intensity (MMI). Ground-motion prediction equations (GMPEs), such as those from Abrahamson et al., input epicentral distance, magnitude, and site conditions to estimate PGA grids at 0.2° resolution, categorizing intensities from light (MMI III) to extreme (MMI X+). In systems like ShakeAlert, the eqInfo2GM module generates contour maps of expected PGA and MMI, alerting users if thresholds (e.g., PGA > 0.1g) are exceeded, with updates every few seconds to refine forecasts as rupture evolves.[48] Post-2020 advancements have integrated machine learning to enhance accuracy, particularly for false positive reduction in crowdsourced systems. In the Android Earthquake Alerts framework, convolutional neural networks process accelerometer data from millions of devices, classifying signals as seismic events with >95% precision by learning from labeled datasets; a 2025 study reported detection of over 1,000 global earthquakes (M 4.5+), delivering 18 million alerts monthly while minimizing false alarms through user feedback loops that refine models in real time. These ML techniques, building on earlier works like Kong et al., improve event association in dense networks by 20-30% over traditional thresholds, enabling scalable warnings in data-sparse regions.[50]Global Deployment
Japan
Japan's Earthquake Early Warning (EEW) system, managed by the Japan Meteorological Agency (JMA), forms a core component of the nationwide J-Alert emergency alert infrastructure, delivering rapid notifications to mitigate seismic risks. Launched on October 1, 2007, the system initially provided alerts via television and radio broadcasts, expanding to achieve full nationwide coverage by 2011, encompassing 100% of the population through diverse channels including mobile phones and public address systems. It operates using a dense network of over 1,000 seismic stations to detect initial P-waves, enabling the estimation of earthquake parameters and issuance of warnings before destructive S-waves arrive.[51][52][42] A hallmark of Japan's approach is the seamless integration of EEW with essential infrastructure to automate protective measures. The Urgent Earthquake Detection and Alarm System (UrEDAS), deployed along railway lines since 1992, employs dedicated sensors to trigger immediate halts of Shinkansen bullet trains upon detecting vibrations equivalent to a JMA seismic intensity of 5 lower or higher, corresponding roughly to events with magnitudes exceeding 5 in populated areas. Elevators in high-rise buildings are similarly equipped to respond to EEW signals by stopping at the nearest floor and opening doors for intensities of 5 or greater, averting potential entrapment during shaking. Public broadcasts via J-Alert further amplify reach, activating sirens and interrupting regular programming to urge immediate safety actions.[53][54][55] The system's effectiveness was demonstrated during the 2011 Tohoku earthquake (Mw 9.0), where it issued initial alerts approximately 15 to 30 seconds before strong shaking reached coastal areas near the epicenter, providing critical time for individuals to seek cover and for automated systems to activate. This lead time facilitated protective behaviors among millions, such as ducking under furniture, which reduced casualties from falling debris and contributed to averting thousands of potential injuries and deaths. Trains across the network stopped safely, preventing derailments, while factory and pipeline shutdowns minimized secondary hazards like fires.[56][57][58] By 2025, JMA incorporated artificial intelligence enhancements into the EEW framework to accelerate data processing and refine magnitude predictions, reducing alert issuance times by analyzing seismic waveforms with machine learning models trained on historical events. These upgrades improve accuracy for complex ruptures, such as those in subduction zones, ensuring more reliable warnings amid Japan's high seismicity.[59]United States
The ShakeAlert earthquake early warning system is a collaborative effort led by the United States Geological Survey (USGS) in partnership with the states of California, Oregon, and Washington, aimed at detecting significant earthquakes and issuing alerts to enable protective actions before strong shaking arrives.[1] The system became operational for public alerting in California in 2019, with expansions to Oregon and Washington following in subsequent years.[60] ShakeAlert processes data from seismic and geodetic sensors to estimate earthquake magnitude, location, and expected shaking intensity, delivering alerts through multiple channels including wireless emergency alerts, mobile apps, and infrastructure integrations.[61] Coverage extends across the West Coast, encompassing over 50 million residents, with the network approximately 90% built out as of early 2024 and full completion of approximately 1,700 seismic stations achieved by mid-2025.[62][63] In California, the system provides alerts to nearly the entire population via partnerships with public safety agencies and private sector entities.[1] Public delivery includes the MyShake app, developed by UC Berkeley and partners, which uses ShakeAlert data to notify users based on their location and expected shaking intensity.[64] Since August 2020, ShakeAlert has been integrated into the Android operating system, enabling automatic alerts on compatible devices across the covered region without requiring a dedicated app.[50] Funding for ShakeAlert is provided through the National Earthquake Hazards Reduction Program (NEHRP), with Congress appropriating $163.5 million for NEHRP activities in fiscal year 2024, including support for system expansion and operations. This federal-state partnership facilitates shared responsibilities, with USGS handling detection and state agencies managing alert dissemination and public education.[65] Performance evaluations, including analyses of the 2019 Ridgecrest earthquake sequence and its continuing aftershocks through 2024, have shown ShakeAlert providing average lead times of 10 to 50 seconds before strong shaking in tested scenarios, allowing time for actions like dropping, covering, and holding on.[66] These tests underscore the system's reliability for moderate to large events within its operational footprint.[67]Mexico
The Mexican Seismic Alert System (SASMEX) originated with the Seismic Alert System for Mexico City (SAS), which began experimental operations in 1991 using 12 stations along the Guerrero coast to detect subduction zone earthquakes. Public alerts were first issued in 1993, marking it as one of the earliest systems to broadcast warnings to the general population. The network expanded significantly with the addition of the Seismic Alert System for Oaxaca (SASO) in 2003, and by 2018, it included 97 monitoring stations spanning the Pacific subduction zone from Jalisco to Oaxaca.[26][25] SASMEX primarily targets earthquakes along Mexico's Pacific subduction zone, where tectonic plates converge and generate frequent large events. Warnings are disseminated via approximately 12,000 sirens and public loudspeakers in urban areas, as well as radio and television broadcasts, providing lead times of 30 to 90 seconds—or up to 120 seconds for distant coastal ruptures—to cities like Mexico City, Oaxaca, Puebla, and Acapulco. This advance notice allows individuals to take protective measures, such as dropping to the ground or evacuating buildings.[68][29] In the 2017 Puebla earthquake (Mw 7.1 on September 19), SASMEX issued warnings to Mexico City about 20 seconds before the strongest shaking arrived, facilitating evacuations in some areas despite the event's proximity to the capital. The system's performance during this intraslab earthquake highlighted its value in enabling rapid responses, though the short lead time limited broader impacts.[69][70] SASMEX faces limitations in southern coverage, particularly for earthquakes originating beyond the Oaxaca sensors or those occurring intraslab rather than at the interface, which can result in minimal or no warning for nearby populations. Expansions in the 2010s increased station density along the coast, and 2020s initiatives include further enhancements to integrate with regional networks in Central America, aiming to improve cross-border alert capabilities amid shared seismic risks.[26][71]Taiwan
Taiwan's earthquake early warning system, operated by the Central Weather Administration (CWA), was launched in 2007 as the first nationwide operational system, utilizing over 100 telemetered strong-motion stations to achieve island-wide coverage and deliver warnings of 10 to 30 seconds in advance for distant areas.[72][73] This centralized setup is particularly tailored to Taiwan's geographic vulnerabilities, where frequent earthquakes often overlap with typhoon seasons, enabling coordinated disaster response within the CWA's broader framework for seismic and meteorological hazards.[74] The system's design emphasizes rapid detection across the compact island, contrasting with larger-scale regional networks elsewhere. The system integrates earthquake alerts with tsunami warnings, issuing unified multi-hazard notifications through diverse channels including mobile phone applications, television broadcasts, and public address systems to maximize public reach and response efficacy.[75] This delivery mechanism ensures that alerts are disseminated promptly to populated areas, supporting actions like halting trains and elevators, as demonstrated in routine operations since inception.[76] In the April 3, 2024, Hualien earthquake (Mw 7.4), the system provided up to 20 seconds of advance warning via mobile apps and broadcasts, allowing individuals to seek cover and secure surroundings, which contributed to limiting fatalities to 19 and injuries to over 1,100 despite intense shaking near the epicenter.[77][78] Examples include interruptions of live TV broadcasts and hospital staff protecting patients, underscoring the alerts' role in mitigating harm.[78] Post-2020 upgrades, part of Phase V enhancements (2016-2021), have improved the system's ability to refine warnings dynamically and support post-event assessments.[72]Other National Systems
South Korea's Korean Earthquake Early Warning System (KEEWS), operational since 2017 under the Korea Meteorological Administration, detects earthquakes of magnitude 5.0 or greater and issues public alerts within 50 seconds of origin, prioritizing urban areas prone to high population density and infrastructure vulnerability.[79][80] The system integrates data from a nationwide seismic network to provide rapid notifications via mobile apps, television, and sirens, enabling actions like halting trains and elevators in densely populated cities such as Seoul.[81] By 2025, KEEWS has demonstrated reliability through automated warnings for multiple events, including offshore earthquakes, contributing to minimized casualties in urban settings.[82] China's earthquake early warning efforts began with a regional pilot in Sichuan Province following the 2008 Wenchuan earthquake, evolving into evaluations during seismic sequences like the 2019–2020 Changning events, where the system provided alerts seconds before strong shaking.[83] By 2025, the nationwide system has expanded to cover over 90% of high-risk areas, serving 148 million users through 70,000 terminals and delivering alerts in 6–7 seconds in priority zones via integrated ground sensors and emerging satellite technologies for enhanced real-time data transmission.[84][85] The system's performance was validated in the 2022 M6.1 Lushan earthquake, where it issued timely warnings, underscoring its role in mitigating impacts across seismically active regions like the Tibetan Plateau.[86] Israel's TRUAA (Tazrika Rishonit Umitzuyanut Al Sefat Ha'Aretz, or Early and Rapid Warning System) became operational in 2022, leveraging an upgraded seismic network of 120 stations deployed primarily along the Dead Sea Transform fault to provide public alerts integrated with national emergency infrastructure.[87][88] This high-density setup, with sensors spaced approximately 10 km apart, focuses on the Dead Sea and Carmel-Zefira fault systems capable of producing magnitude 7.5 events, enabling warnings within seconds for densely populated areas like Jerusalem and Tel Aviv.[89] TRUAA's real-time processing has supported seismicity monitoring and public notifications during swarms, such as the 2022 Jordan Valley events, enhancing resilience in this tectonically active zone.[90][33] In Canada, the Earthquake Early Warning system launched in British Columbia in spring 2024 as an extension of the U.S. ShakeAlert framework, utilizing existing seismic networks to deliver alerts via cell phones, radio, and television to over 10 million residents in western regions before strong shaking arrives.[6][91] Expansion to Quebec and Ontario is scheduled for late 2025, targeting eastern seismic hazards with integrated processing algorithms for rapid magnitude estimation and ground-shaking predictions.[92] This initiative, funded by federal investments exceeding $36 million, emphasizes cross-border collaboration to protect infrastructure like pipelines and urban centers in Vancouver and Toronto.[93] India's earthquake early warning pilot in the seismically vulnerable Northeast region, initiated in the early 2020s through the National Centre for Seismology, focuses on Himalayan zones using observed and simulated data from magnitude 6.1–8.3 events to develop scaling relations for timely alerts.[94][95] The prototype system, still in early development stages as of 2025, collaborates with international partners like the U.S. Geological Survey to test real-time warnings in areas like Uttarakhand and the Northeast, aiming to reduce risks from frequent moderate earthquakes.[96][97] Romania operates a government-supported operational earthquake early warning system, including the Rapid Earthquake Warning System (REWS), which has been expanded through EU-funded projects to monitor crustal movements via a network of 19 GPS stations and provide alerts in high-risk areas like Bucharest.[43][98] Recent enhancements incorporate operational earthquake forecasting to predict aftershocks, supported by European investments in seismic risk management for the Vrancea seismic zone.[99] In Italy, regional opt-in systems like the PRESTo-based EEWS in southern regions deliver personalized alerts through mobile apps to voluntary users, funded partly by EU Horizon programs for multi-hazard integration along the Apennines.[100][101] These initiatives emphasize user participation and rapid ground-shaking predictions, with EU support enabling upgrades to existing networks for events like those in central Italy.[102]Crowdsourced and Mobile Systems
MyShake
MyShake is a smartphone-based earthquake early warning application developed by the University of California, Berkeley's Seismology Lab, leveraging the built-in accelerometers of mobile devices to create a crowdsourced seismic network.[103] Launched on February 12, 2016, the app employs an artificial neural network to distinguish earthquake signals from everyday human motions, such as walking or driving, by analyzing features like interquartile range, zero-crossing rate, and cumulative absolute velocity from the accelerometer data.[104] Upon detecting potential seismic activity, the phone uploads the waveform to a central processing center, where a network detection algorithm confirms the event by requiring triggers from multiple devices within a localized area, enabling rapid estimation of magnitude, location, and expected shaking intensity.[103] This methodology allows MyShake to contribute to earthquake detection globally, particularly in regions lacking dense traditional seismic infrastructure.[105] In operation, MyShake integrates with the ShakeAlert system to deliver timely alerts to users in California, Oregon, and Washington, providing seconds to tens of seconds of warning before strong shaking arrives for earthquakes of magnitude 4.5 or greater.[106] By November 2025, the app had surpassed 4 million downloads, primarily among users in California, where it focuses on enhancing public safety through real-time notifications and post-event information, though its design supports expansion to over 80 countries with active users.[107] The app's global potential lies in its ability to form ad hoc seismic networks wherever smartphones are prevalent, supplementing official systems in underserved areas.[108] A 2024 study validated MyShake's effectiveness by compiling a database of over 1,600 ground-shaking waveforms recorded between 2019 and 2023, demonstrating that crowdsourced smartphone data closely matches traditional free-field ground-motion models and improves spatial resolution for peak ground acceleration predictions.[109] This validation highlighted enhanced detection capabilities in low-seismic-station-density regions, such as urban or remote areas, where the dense sampling from numerous devices allows for better interpolation of shaking patterns and identification of site-specific effects that sparse station networks might miss.[109] MyShake facilitates data sharing by collecting and anonymizing accelerometer recordings from participating devices, using unique device identifiers without personal information to ensure privacy while enabling contributions to broader seismic research.[110] These anonymized crowdsourced datasets are shared with official networks, augmenting traditional monitoring by providing high-density observations in populated areas and improving overall earthquake early warning accuracy and ShakeMap generation.[111]Android Earthquake Alerts
Android Earthquake Alerts is a system developed by Google that leverages the accelerometers in Android smartphones and smartwatches to detect seismic activity and deliver early warnings to users before strong shaking arrives. Launched initially in the United States in 2020 as part of Android's safety features, the system began expanding internationally in April 2021, starting with countries like New Zealand and Greece, and reached 98 countries by the end of 2023. By 2025, it utilizes a network of over 2 billion devices worldwide, enabling opt-in participation where users can enable earthquake detection and alerts through device settings, provided location services and internet connectivity are active.[112][113][50] The core technology relies on on-device artificial intelligence to analyze accelerometer data in real time, distinguishing primary (P) waves—which travel faster and cause less damage—from secondary (S) waves that produce stronger shaking. This processing occurs locally on the device to ensure low latency, with detected events reported to Google's servers for verification and alert dissemination if confirmed. A 2025 study published in Science evaluated the system's performance as a real-time global network, finding it capable of detecting earthquakes ranging from magnitude 1.9 to 7.8, with an average of 312 detections per month between 2021 and 2024, rivaling traditional seismometer networks in accuracy for magnitudes above 4.5.[50][113][50] Google partners with authoritative seismic agencies to enhance reliability and integrate data, including the United States Geological Survey (USGS) for ShakeAlert-powered alerts in California, Oregon, and Washington, as well as the European-Mediterranean Seismological Centre (EMSC) for European operations. These collaborations allow the system to supplement official warnings in regions with established infrastructure while providing standalone alerts elsewhere; by 2025, opt-in alerts are available in over 50 countries, with broader detection coverage across 98 nations and territories.[114][113][50] The system's impact is evident in its ability to detect and alert for events in under-monitored areas, issuing over 1,279 alerts for Android-detected earthquakes as of March 2024, including numerous 2024 events in regions lacking dense seismic networks, such as parts of Asia and Africa. These detections contribute valuable crowdsourced data back to official systems like USGS and EMSC for post-event analysis and model refinement, expanding global early warning access from 250 million people in 2019 to 2.5 billion by 2025 and providing users with 5 to 60 seconds of warning in 36% of cases where shaking was felt.[50][113][50]Earthquake Network
The Earthquake Network is a citizen science initiative that leverages smartphones to create a crowdsourced earthquake early warning system, launched in 2013 by Italian statistician Francesco Finazzi of the University of Bergamo.[115][116] The project began development in late 2012 and has since grown into a global network, with the app downloaded over 20 million times across Android, iOS, and Huawei platforms as of 2025.[117][118] This volunteer-based model relies on users opting in to contribute accelerometer data from their devices, forming a dense, real-time sensor array without dependence on traditional seismic infrastructure.[119] Functionally, the system detects ground motion through smartphone accelerometers when phones are stationary and charging, triggering immediate analysis on central servers to identify potential earthquakes.[120] Upon verification—using statistical algorithms to filter false positives and estimate magnitude, location, and epicentral intensity—alerts are pushed to nearby users within seconds, often providing warnings before destructive S-waves arrive. This rapid response is achieved through a decentralized detection phase followed by centralized processing, allowing notifications as short as 5-10 seconds after initial P-wave detection in well-covered areas.[121] Unlike institutional systems, Earthquake Network emphasizes user privacy, with data anonymized and only seismic events shared publicly, fostering a standalone, community-driven approach that complements but operates independently of platforms like MyShake.[122] The network provides global coverage, though detection reliability is highest in regions with high user density, such as Europe and Asia, where millions of active installations enable robust monitoring.[123] Verified earthquake events detected by the system are integrated into international databases, including contributions to the U.S. Geological Survey (USGS) for enhanced global seismicity reporting.[124] A notable real-world application occurred during the February 6, 2023, M7.8 Pazarcık earthquake in Turkey-Syria, where the network detected the event through user phone data and issued alerts up to 58 seconds before strong shaking reached some areas, with initial user reports arriving before official governmental notifications.[120] This incident underscored the potential of crowdsourced systems to fill gaps in official early warning coverage, particularly in regions with limited traditional sensors.[120]Effectiveness and Impacts
Performance Evaluations
Performance evaluations of earthquake early warning (EEW) systems focus on key metrics such as detection accuracy, false alarm rates, and lead times, which determine their reliability and utility in providing actionable warnings. For instance, the ShakeAlert system in the United States, operational across the West Coast, achieved a detection rate of approximately 87% for earthquakes of magnitude 4.5 or greater between 2019 and 2023, successfully identifying 46 out of 53 such events within its alerting region. False alarm rates for these magnitudes were low at 7.4%, with no false alerts for events of magnitude 5.0 or higher during this period. Median lead times for initial alerts ranged from 4 to 8 seconds in densely instrumented areas, extending to 10–23 seconds of warning time at sites experiencing moderate to strong shaking (Modified Mercalli Intensity VI–VIII).[125] In Japan, the nationwide EEW system, managed by the Japan Meteorological Agency, has undergone significant refinements following the 2011 Tohoku-Oki earthquake, leading to improved performance metrics. Post-2016 updates, including the integration of the Intensity Prediction with FinDer (IPF) and Propagation of Local Undetermined Magnitude (PLUM) algorithms, resulted in prediction success rates of 83.6% to 84.6% for maximum shaking estimates through 2020. These enhancements reduced underprediction for large inland events and improved lead times, with the addition of the S-net seafloor network shortening detection times by a median of 3.9 seconds (and up to 8.8 seconds at the 75th percentile) for offshore earthquakes. False alarm rates have been minimized through hybrid processing, though specific quantitative rates post-2011 remain below 10% in operational reviews.[126] Comparisons between nationwide seismic network-based systems and crowdsourced mobile platforms highlight differences in scale and efficacy. ShakeAlert and Japan's system provide consistent performance in instrumented regions, with detection accuracies exceeding 80% for moderate magnitudes and lead times of 5–20 seconds, but coverage is geographically limited. In contrast, the Android Earthquake Alerts system, leveraging over 2.5 billion smartphones globally, detected an average of 312 earthquakes per month (magnitudes 1.9–7.8) from 2021 to 2024, issuing alerts for magnitude 4.5 or greater events in 98 countries. User feedback indicated that 85% of alerts corresponded to felt shaking, with 36% delivered before shaking onset, though false positives arise from non-seismic noise; overall efficacy in 2025 remains high for global reach but varies by user density.[50] The economic impact of these systems underscores their value, with the global EEW market valued at USD 1.35 billion in 2024 and projected to reach USD 2.10 billion by 2032, driven by adoption in high-risk regions and integration with infrastructure like transportation and utilities.[127]| System | Detection Accuracy (M ≥ 4.5) | False Alarm Rate | Median Lead Time |
|---|---|---|---|
| ShakeAlert (US) | ~87% (2019–2023) | 7.4% | 4–8 s (initial); 10–23 s (warning) |
| Japan EEW | 83.6–84.6% (post-2016) | <10% | 3.9 s improvement (offshore) |
| Android Alerts | 85% alignment with shaking | Variable (noise-related) | 36% before onset |