Geoscience and seismology have utilized the most advanced technologies and equipment to monitor seismic events globally from the past few decades. With the enormous amount of data, modern GPU-powered deep learning presents a promising approach to analyze data and discover patterns. In recent years, there are plenty of successful deep learning models for picking seismic waves. However, forecasting extreme earthquakes, which can cause disasters, is still an underdeveloped topic in history. Relevant research in spatiotemporal dynamics mining and forecasting has revealed some successful predictions, a crucial topic in many scientific research fields. Most studies of them have many successful applications of using deep neural networks. In Geology and Earth science studies, earthquake prediction is one of the world's most challenging problems, about which cutting-edge deep learning technologies may help discover some valuable patterns. In this project, we propose a deep learning modeling approach, namely \tseqpre, to mine spatiotemporal patterns from data to nowcast extreme earthquakes by discovering visual dynamics in regional coarse-grained spatial grids over time. In this modeling approach, we use synthetic deep learning neural networks with domain knowledge in geoscience and seismology to exploit earthquake patterns for prediction using convolutional long short-term memory neural networks. Our experiments show a strong correlation between location prediction and magnitude prediction for earthquakes in Southern California. Ablation studies and visualization validate the effectiveness of the proposed modeling method.
Biocomplexity Institute and Initiative, University of Virginia（弗吉尼亚大学生物复杂性研究所与计划）；Computer Science Dept., University of Virginia（弗吉尼亚大学计算机科学系）；Jet Propulsion Laboratory, California Institute of Technology（加州理工学院喷气推进实验室）；Santa Fe Institute（圣菲研究所）；Indiana University（印第安纳大学）
We review previous approaches to nowcasting earthquakes and introduce new approaches based on deep learning using three distinct models based on recurrent neural networks and transformers. We discuss different choices for observables and measures presenting promising initial results for a region of Southern California from 1950-2020. Earthquake activity is predicted as a function of 0.1-degree spatial bins for time periods varying from two weeks to four years. The overall quality is measured by the Nash Sutcliffe Efficiency comparing the deviation of nowcast and observation with the variance o…
This paper describes the use of the idea of natural time to propose a new method for characterizing the seismic risk to the world's major cities at risk of earthquakes. Rather than focus on forecasting, which is the computation of probabilities of future events, we define the term seismic nowcasting, which is the computation of the current state of seismic hazard in a defined geographic region.
Time Series Foundation Models and Deep Learning Architectures for Earthquake Temporal and Spatial Nowcasting
University of Virginia（弗吉尼亚大学）；University of California, Davis（加州大学戴维斯分校）；Santa Fe Institute（圣达菲研究所）；Jet Propulsion Laboratory（喷气推进实验室）；California Institute of Technology（加州理工学院）；University of California, Irvine（加州大学欧文分校）
Yes. Interval statistics have been used to conclude that major earthquakes are random events in time and cannot be anticipated or predicted. Machine learning is a powerful new technique that enhances our ability to understand the information content of earthquake catalogs. We show that catalogs contain significant information on current hazard and future predictability for large earthquakes.
We explore in depth the validity of a recently proposed scaling law for earthquake interevent time distributions in the case of the Southern California, using the waveform cross-correlation catalog of Shearer et al. Two statistical tests are used: on the one hand, the standard two-sample Kolmogorov-Smirnov test is in agreement with the scaling of the distributions. On the other hand, the one-sample Kolmogorov-Smirnov statistic complemented with Monte Carlo simulation of the inter-event times, as done by Clauset et al., supports the validity of the gamma distribution as a simple model of the sc…
Modern, powerful techniques for the residual analysis of spatial-temporal point process models are reviewed and compared. These methods are applied to California earthquake forecast models used in the Collaboratory for the Study of Earthquake Predictability (CSEP). Assessments of these earthquake forecasting models have previously been performed using simple, low-power means such as the L-test and N-test. We instead propose residual methods based on rescaling, thinning, superposition, weighted K-functions and deviance residuals. Rescaled residuals can be useful for assessing the overall fit of…
We present two models for estimating the probabilities of future earthquakes in California, to be tested in the Collaboratory for the Study of Earthquake Predictability (CSEP). The first, time-independent model, modified from Helmstetter et al. (2007), provides five-year forecasts for magnitudes m > 4.95. We show that large quakes occur on average near the locations of small m > 2 events, so that a high-resolution estimate of the spatial distribution of future large quakes is obtained from the locations of the numerous small events. We employ an adaptive spatial kernel of optimized bandwidth a…
The ETAS model is widely employed to model the spatio-temporal distribution of earthquakes, generally using spatially invariant parameters. We propose an efficient method for the estimation of spatially varying parameters, using the Expectation-Maximization (EM) algorithm and spatial Voronoi tessellation ensembles. We use the Bayesian Information Criterion (BIC) to rank inverted models given their likelihood and complexity and select the best models to finally compute an ensemble model at any location. Using a synthetic catalog, we also check that the proposed method correctly inverts the know…
One of the main interests in seismology is the formulation of models able to describe the clustering in time occurrence of earthquakes. Analysis of the Southern California Catalog shows magnitude clustering in correspondence to temporal clustering. Here we propose a dynamical scaling hypothesis in which time is rescaled in terms of magnitude. This hypothesis is introduced in the context of a generalized trigger model and gives account for clustering in time and magnitude for earthquake occurrence. The model is able to generate a synthetic catalog reproducing magnitude and inter-even time distr…
Recent studies in the literature have introduced a new approach to earthquake forecasting based on representing the space-time patterns of localized seismicity by a time-dependent system state vector in a real-valued Hilbert space and deducing information about future space-time fluctuations from the phase angle of the state vector. While the success rate of this Pattern Informatics (PI) method has been encouraging, the method is still in its infancy. Procedural analysis, statistical testing, parameter sensitivity investigation and optimization all still need to be performed. In this paper, we…
Based on waveform data from a profile of aftershocks following the north-south trace of the June 28, 1992 Landers rupture across the Mojave desert, we construct a new velocity model for the Mojave region which features a thin, slow crust. Using this model, we obtain source parameters, including depth and duration, for each of the aftershocks in the profile, and in addition, any significant (M>3.7) Joshua Tree--Landers aftershock between April, 1992 and October, 1994 for which coherent TERRAscope data were available. In all, we determine source parameters and stress-drops for 45 significant (M_…
The Multifractal Stress-Activated (MSA) model is a statistical model of triggered seismicity based on mechanical and thermodynamic principles. It predicts that, above a triggering magnitude cut-off $M_0$, the exponent $p$ of the Omori law for the seismic decay of aftershocks is a linear increasing function $p(M) =a M+b$ of the main shock magnitude $M$. We previously reported empirical support for this prediction, using the Southern California SCEC catalog. Here, we confirm this law using an updated, longer version of the same catalog, as well as new methods to estimate $p$. One of this methods…
Upon employing the analysis in a new time domain, termed natural time, it has been recently demonstrated that a remarkable change of seismicity emerges before major mainshocks in California. What constitutes this change is that the fluctuations of the order parameter of seismicity exhibit a clearly detectable minimum. This is identified by using a natural time window sliding event by event through the time series of the earthquakes in a wide area and comprising a number of events that would occur on the average within a few months or so. Here, we suggest a method to estimate the epicentral are…
By analyzing a southern California earthquake catalog, Davidsen and Paczuski [Phys. Rev. Lett. 94, 048501 (2005)] claim to have found evidence contradicting the theory of aftershock zone scaling in favor of scale-free statistics. We present four elements showing that Davidsen and Paczuski's results may be insensitive to the existence of physical length scales associated with aftershock zones or mainshock rupture lengths, so that their claim is unsubstantiated. (i) Their exponent smaller than 1 for a pdf implies that the power law statistics they report is at best an intermediate asymptotic; (i…
We introduce a new approach to short-term earthquake prediction based on the concept of selforganization of seismically active fault networks. That approach is named "Reverse Detection of Precursors" (RDP), since it considers precursors in reverse order of their appearance. This makes it possible to detect precursors undetectable by direct analysis. Possible mechanisms underlying RDP are outlined. RDP is described with a concrete example: we consider as short-term precursors the newly introduced chains of earthquakes reflecting the rise of an earthquake correlation range; and detect (retrospec…
Earthquake network captures the complexity of seismicity in a peculiar manner. Given a seismic data, the procedure of constructing an earthquake network proposed in [S. Abe, N, Suzuki, Europhys. Lett. 65 (2004) 581] contains as a single parameter the size of the cells, into which a geographical region under consideration is divided. Then, the characteristics of the network depend on the cell size, in general. Here, the dependency of the clustering coefficient, C, of network on the cell size is studied. Remarkably, C of the earthquake networks constructed from the seismic data taken from Califo…
We present a RELM forecast of future earthquakes in California that is primarily based on the pattern informatics (PI) method. This method identifies regions that have systematic fluctuations in seismicity, and it has been demonstrated to be successful. A PI forecast map originally published on 19 February 2002 for southern California successfully forecast the locations of sixteen of eighteen M>5 earthquakes during the past three years. The method has also been successfully applied to Japan and on a worldwide basis. An alternative approach to earthquake forecasting is the relative intensity (R…
We propose two methods to calibrate the parameters of the epidemic-type aftershock sequence (ETAS) model based on expectation maximization (EM) while accounting for temporal variation of catalog completeness. The first method allows for model calibration on long-term earthquake catalogs with temporal variation of the completeness magnitude, $m_c$. This calibration technique is beneficial for long-term probabilistic seismic hazard assessment (PSHA), which is often based on a mixture of instrumental and historical catalogs. The second method generalizes the concept of $m_c$, considering rate- an…
Earthquake prediction is one of the most pursued problems in geoscience. Different geological and seismological approaches exist for the prediction of the earthquake and its subsequent land change. However, in many cases, they fail in their mission. In this paper, we address the well-established earthquake prediction problem by a novel approach. We use a four-dimensional location-time machine learning scheme to estimate the time of earthquake and its land change. We present a study for the Ridgecrest, California 2019 earthquake prediction. We show the accuracy of our method is around 14 centim…
Earthquake network is known to be of the small-world type. The values of the network characteristics, however, depend not only on the cell size (i.e., the scale of coarse graining needed for constructing the network) but also on the size of a seismic data set. Here, discovery of a scaling law for the clustering coefficient in terms of the data size, which is refereed to here as finite data-size scaling, is reported. Its universality is shown to be supported by the detailed analysis of the data taken from California, Japan and Iran. Effects of setting threshold of magnitude are also discussed.
The properties of earthquake networks have been studied so far mainly for the seismic data sets taken from California, Japan and Iran, and features common in these regions have been reported in the literature. Here, an earthquake network is constructed and analyzed for the Chilean data to examine if the scale-free and small-world properties of the earthquake networks constructed in the other geographical regions can also be found in seismicity in Chile. It is shown that the result is affirmative: in all the regions both the exponent "gamma" of the power-law connectivity distribution and the cl…
Reliable earthquake detection and seismic phase classification is often challenging especially in the circumstances of low magnitude events or poor signal-to-noise ratio. With improved seismometers and better global coverage, a sharp increase in the volume of recorded seismic data is witnessed. This makes the handling of the seismic data rather daunting based on traditional approaches and therefore fuels the need for a more robust and reliable method. In this study, we investigate two deep learningbased models, termed 1D ResidualNeuralNetwork (ResNet) and multi-branch ResNet, for tackling the …
Seismic phase association connects earthquake arrival time measurements to their causative sources. Effective association must determine the number of discrete events, their location and origin times, and it must differentiate real arrivals from measurement artifacts. The advent of deep learning pickers, which provide high rates of picks from closely overlapping small magnitude earthquakes, motivates revisiting the phase association problem and approaching it using the methods of deep learning. We have developed a Graph Neural Network associator that simultaneously predicts both source space-t…
University of British Columbia（不列颠哥伦比亚大学）；GD Goenka University（GD戈恩卡大学）
Earthquake prediction has been a challenging research area for many decades, where the future occurrence of this highly uncertain calamity is predicted. In this paper, several parametric and non-parametric features were calculated, where the non-parametric features were calculated using the parametric features. $8$ seismic features were calculated using Gutenberg-Richter law, the total recurrence, and the seismic energy release. Additionally, criterions such as Maximum Relevance and Maximum Redundancy were applied to choose the pertinent features. These features along with others were used as …
We report an empirical determination of the probability density functions $P_{\text{data}}(r)$ of the number $r$ of earthquakes in finite space-time windows for the California catalog. We find a stable power law tail $P_{\text{data}}(r) \sim 1/r^{1+μ}$ with exponent $μ\approx 1.6$ for all space ($5 \times 5$ to $20 \times 20$ km$^2$) and time intervals (0.1 to 1000 days). These observations, as well as the non-universal dependence on space-time windows for all different space-time windows simultaneously, are explained by solving one of the most used reference model in seismology (ETAS), which …
We report an empirical determination of the probability density functions P(r) of the number r of earthquakes in finite space-time windows for the California catalog, over fixed spatial boxes 5 x 5 km^2 and time intervals dt =1, 10, 100 and 1000 days. We find a stable power law tail P(r) ~ 1/r^{1+mu} with exponent mu \approx 1.6 for all time intervals. These observations are explained by a simple stochastic branching process previously studied by many authors, the ETAS (epidemic-type aftershock sequence) model which assumes that each earthquake can trigger other earthquakes (``aftershocks''). …
To optimally monitor earthquake-generating processes, seismologists have sought to lower detection sensitivities ever since instrumental seismic networks were started about a century ago. Recently, it has become possible to search continuous waveform archives for replicas of previously recorded events (template matching), which has led to at least an order of magnitude increase in the number of detected earthquakes and greatly sharpened our view of geological structures. Earthquake catalogs produced in this fashion, however, are heavily biased in that they are completely blind to events for wh…
Seismic phase association is the task of grouping phase arrival picks across a seismic network into subsets with common origins. Building on recent successes in this area with machine learning tools, we introduce a neural mixture model association algorithm (Neuma), which incorporates physics-informed neural networks and mixture models to address this challenging problem. Our formulation assumes explicitly that a dataset contains real phase picks from earthquakes and noise picks resulting from phase picking mistakes and fake picks. The problem statement is then to assign each observation to ei…
We present a new method of data clustering applied to earthquake catalogs, with the goal of reconstructing the seismically active part of fault networks. We first use an original method to separate clustered events from uncorrelated seismicity using the distribution of volumes of tetrahedra defined by closest neighbor events in the original and randomized seismic catalogs. The spatial disorder of the complex geometry of fault networks is then taken into account by defining faults as probabilistic anisotropic kernels, whose structures are motivated by properties of discontinuous tectonic deform…
As the number of seismic sensors grows, it is becoming increasingly difficult for analysts to pick seismic phases manually and comprehensively, yet such efforts are fundamental to earthquake monitoring. Despite years of improvements in automatic phase picking, it is difficult to match the performance of experienced analysts. A more subtle issue is that different seismic analysts may pick phases differently, which can introduce bias into earthquake locations. We present a deep-neural-network-based arrival-time picking method called "PhaseNet" that picks the arrival times of both P and S waves. …
Seismic phase association is a fundamental task in seismology that pertains to linking together phase detections on different sensors that originate from a common earthquake. It is widely employed to detect earthquakes on permanent and temporary seismic networks, and underlies most seismicity catalogs produced around the world. This task can be challenging because the number of sources is unknown, events frequently overlap in time, or can occur simultaneously in different parts of a network. We present PhaseLink, a framework based on recent advances in deep learning for grid-free earthquake ph…
We inquire the statistical nature and dynamics of shallow and deep seismogenesis along major plate margins of the NW Circum-Pacific Belt, by examining whether earthquakes are generated by Poisson processes and are independent (uncorrelated), or by Complex processes and are dependent (correlated). Analysis is based of Non Extensive Statistical Physics and the complete and homogeneous catalogue of the Japan Meteorological Agency for the period 2002-2016.5. Emphasis is given to background seismicity recovered by removing aftershocks with stochastic declustering. Long-term correlation and long-ran…
