skip to main content

Title: A Distributed Multi-Sensor Machine Learning Approach to Earthquake Early Warning
Our research aims to improve the accuracy of Earthquake Early Warning (EEW) systems by means of machine learning. EEW systems are designed to detect and characterize medium and large earthquakes before their damaging effects reach a certain location. Traditional EEW methods based on seismometers fail to accurately identify large earthquakes due to their sensitivity to the ground motion velocity. The recently introduced high-precision GPS stations, on the other hand, are ineffective to identify medium earthquakes due to its propensity to produce noisy data. In addition, GPS stations and seismometers may be deployed in large numbers across different locations and may produce a significant volume of data consequently, affecting the response time and the robustness of EEW systems.In practice, EEW can be seen as a typical classification problem in the machine learning field: multi-sensor data are given in input, and earthquake severity is the classification result. In this paper, we introduce the Distributed Multi-Sensor Earthquake Early Warning (DMSEEW) system, a novel machine learning-based approach that combines data from both types of sensors (GPS stations and seismometers) to detect medium and large earthquakes. DMSEEW is based on a new stacking ensemble method which has been evaluated on a real-world dataset validated with more » geoscientists. The system builds on a geographically distributed infrastructure, ensuring an efficient computation in terms of response time and robustness to partial infrastructure failures. Our experiments show that DMSEEW is more accurate than the traditional seismometer-only approach and the combined-sensors (GPS and seismometers) approach that adopts the rule of relative strength. « less
; ; ; ; ; ; ; ; ; ;
Award ID(s):
1640834 1835692 1745246
Publication Date:
Journal Name:
Proceedings of the AAAI Conference on Artificial Intelligence
Page Range or eLocation-ID:
403 to 411
Sponsoring Org:
National Science Foundation
More Like this
  1. SUMMARY We present the first 16 months of data returned from a mobile array of 16 freely floating diving instruments, named mermaid for Mobile Earthquake Recording in Marine Areas by Independent Divers, launched in French Polynesia in late 2018. Our 16 are a subset of the 50 mermaid deployed over a number of cruises in this vast and understudied oceanic province as part of the collaborative South Pacific Plume Imaging and Modeling (SPPIM) project, under the aegis of the international EarthScope-Oceans consortium. Our objective is the hydroacoustic recording, from within the oceanic water column, of the seismic wavefield generated by earthquakes worldwide, and the nearly real-time transmission by satellite of these data, collected above and in the periphery of the South Pacific Superswell. This region, characterized by anomalously elevated oceanic crust and myriad seamounts, is believed to be the surface expression of deeply rooted mantle upwellings. Tomographically imaging Earth’s mantle under the South Pacific with data from these novel instruments requires a careful examination of the earthquake-to-mermaid traveltimes of the high-frequency P-wave detections within the windows selected for reporting by the discrimination algorithms on board. We discuss a workflow suitable for a fast-growing mobile sensor database to pick the relevant arrivals,more »match them to known earthquakes in global earthquake catalogues, calculate their traveltime residuals with respect to global seismic reference models, characterize their quality and estimate their uncertainty. We detail seismicity rates as recorded by mermaid over 16 months, quantify the completeness of our catalogue and discuss magnitude–distance relations of detectability for our network. The projected lifespan of an individual mermaid is 5 yr, allowing us to estimate the final size of the data set that will be available for future study. To prove their utility for seismic tomography we compare mermaid data quality against ‘traditional’ land seismometers and their low-cost Raspberry Shake counterparts, using waveforms recovered from instrumented island stations in the geographic neighbourhood of our floats. Finally, we provide the first analyses of traveltime anomalies for the new ray paths sampling the mantle under the South Pacific.« less
  2. Long‐period ground motions from large (Mw≥7.0) subduction‐zone earthquakes are a real threat for large‐scale human‐made structures. The Nankai subduction zone, Japan, is expected to host a major megathrust earthquake in the near future and has therefore been instrumented with offshore and onshore permanent seismic networks. We use the ambient seismic field continuously recorded at these stations to simulate the long‐period (4–10 s) ground motions from past and future potential offshore earthquakes. First, we compute impulse response functions (IRFs) between an ocean‐bottom seismometer of the Dense Oceanfloor Network System for Earthquakes and Tsunamis (DONET) network, which is located offshore on the accretionary wedge, and 60 onshore Hi‐net stations using seismic interferometry by deconvolution. As this technique only preserves the relative amplitude information of the IRFs, we use a moderate Mw 5.5 event to calibrate the amplitudes to absolute levels. After calibration, the IRFs are used together with a uniform stress‐drop source model to simulate the long‐period ground motions of the 2004 Mw 7.2 intraplate earthquake. For both events, the residuals of the 5\\% damped spectral acceleration (SA) computed from the horizontal and vertical components of the observed and simulated waveforms exhibit almost no bias and acceptable uncertainties. We also compare the observed SA valuesmore »of the Mw 7.2 event to those from the subduction‐zone BC Hydro ground‐motion model (GMM) and find that our simulations perform better than the model. Finally, we simulate the long‐period ground motions of a hypothetical Mw 8.0 subduction earthquake that could occur along the Nankai trough. For this event, our simulations generally exhibit stronger long‐period ground motions than those predicted by the BC Hydro GMM. This study suggests that the ambient seismic field recorded by the ever‐increasing number of ocean‐bottom seismometers can be used to simulate the long‐period ground motions from large megathrust earthquakes.« less
  3. SUMMARY In the Gulf of California, Mexico, the relative motion across the North America–Pacific boundary is accommodated by a series of marine transform faults and spreading centres. About 40 M> 6 earthquakes have occurred in the region since 1960. On 2009 August 3, an Mw 6.9 earthquake occurred near Canal de Ballenas in the region. The earthquake was a strike-slip event with a shallow hypocentre that is likely close to the seafloor. In contrast to an adjacent M7 earthquake, this earthquake triggered a ground-motion-based earthquake early warning algorithm being tested in southern California (∼600 km away). This observation suggests that the abnormally large ground motions and dynamic strains observed for this earthquake relate to its rupture properties. To investigate this possibility, we image the rupture process and resolve the slip distribution of the event using a P-wave backprojection approach and a teleseismic, finite-fault inversion method. Results from these two independent analyses indicate a relatively simple, unilateral rupture propagation directed along-strike in the northward direction. However, the average rupture speed is estimated around 4 km s−1, suggesting a possible supershear rupture. The supershear speed is also supported by a Rayleigh wave Mach cone analysis, although uncertainties in local velocity structure preclude a definitive conclusion. Themore »Canal de Ballenas earthquake dynamically triggered seismicity at multiple sites in California, with triggering response characteristics varying from location-to-location. For instance, some of the triggered earthquakes in California occurred up to 24 hr later, suggesting that nonlinear triggering mechanisms likely have modulated their occurrence.« less
  4. Monitoring systems have hundreds or thousands of distributed sensors gathering and transmitting real-time streaming data. The early detection of events in these systems, such as an earthquake in a seismic monitoring system, is the base for essential tasks as warning generations. To detect such events is usual to compute pairwise correlation across the disparate signals generated by the sensors. Since the data sources (e.g., sensors) are spatially separated, it is essential to consider the lagged correlation between the signals. Besides, many applications require to process a specific band of frequencies depending on the event’s type, demanding a pre-processing step of filtering before computing correlations. Due to the high speed of data generation and a large number of sensors in these systems, the operations of filtering and lagged cross-correlation need to be efficient to provide real-time responses without data losses. This article proposes a technique named FilCorr that efficiently computes both operations in one single step. We achieve an order of magnitude speedup by maintaining frequency transforms over sliding windows. Our method is exact, devoid of sensitive parameters, and easily parallelizable. Besides our algorithm, we also provide a publicly available real-time system named Seisviz that employs FilCorr in its core mechanismmore »for monitoring a seismometer network. We demonstrate that our technique is suitable for several monitoring applications as seismic signal monitoring, motion monitoring, and neural activity monitoring.« less
  5. With the prevalence of smartphones, pedestrians and joggers today often walk or run while listening to music. Since they are deprived of their auditory senses that would have provided important cues to dangers, they are at a much greater risk of being hit by cars or other vehicles. In this paper, we build a wearable system that uses multi-channel audio sensors embedded in a headset to help detect and locate cars from their honks, engine and tire noises, and warn pedestrians of imminent dangers of approaching cars. We demonstrate that using a segmented architecture consisting of headset-mounted audio sensors, a front-end hardware platform that performs signal processing and feature extraction, and machine learning based classification on a smartphone, we are able to provide early danger detection in real-time, from up to 60m away, and alert the user with low latency and high accuracy. To further reduce power consumption of the battery-powered wearable headset, we implement a custom-designed integrated circuit that is able to compute delays between multiple channels of audio with nW power consumption. A regression-based method for sound source localization, AvPR, is proposed and used in combination with the IC to improve the granularity and robustness of localization.