skip to main content

Title: Long Short-Term Memory Networks for Pattern Recognition of Synthetical Complete Earthquake Catalog
Exploring the spatiotemporal distribution of earthquake activity, especially earthquake migration of fault systems, can greatly to understand the basic mechanics of earthquakes and the assessment of earthquake risk. By establishing a three-dimensional strike-slip fault model, to derive the stress response and fault slip along the fault under regional stress conditions. Our study helps to create a long-term, complete earthquake catalog. We modelled Long-Short Term Memory (LSTM) networks for pattern recognition of the synthetical earthquake catalog. The performance of the models was compared using the mean-square error (MSE). Our results showed clearly the application of LSTM showed a meaningful result of 0.08% in the MSE values. Our best model can predict the time and magnitude of the earthquakes with a magnitude greater than Mw = 6.5 with a similar clustering period. These results showed conclusively that applying LSTM in a spatiotemporal series prediction provides a potential application in the study of earthquake mechanics and forecasting of major earthquake events.  more » « less
Award ID(s):
Author(s) / Creator(s):
; ; ; ; ; ;
Date Published:
Journal Name:
Page Range / eLocation ID:
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract

    The Raton Basin has been an area of injection induced seismicity for the past two decades. Previously, the reactivated fault zone structures and spatiotemporal response of seismicity to evolving injection have been poorly constrained due to sparse publicly available seismic monitoring. The application of a machine‐learning phase picker to 4 years of continuous seismic data from a local array enables the detection and location of ∼38,000 earthquakes. The events from 2016 to 2020 are ∼2.5–6 km below sea level and range from ML < −1 to 4.2. Most earthquakes occur within previously identified ∼N‐S zones of seismicity, however our new catalog illuminates that these zones are composed of many short faults with variable orientations. The two most active zones, the Vermejo Park and Tercio zones, are potentially linked by small intermediate faults. In total, we find ∼60 short (<3 km long) basement faults with strikes from WNW to NNE. Faulting mechanisms are predominantly normal but some variability, including reverse dip‐slip and oblique‐slip, is observed. The Trinidad fault zone, which previously hosted a Mw5.3 earthquake in 2011, is quiescent during 2016–2020, likely in response to both slow accumulation of tectonic strain after the 2011 sequence, and the significant decrease (80% reduction) in nearby wastewater injection from 2012 to 2016. Unlike some other regions, where induced seismicity was triggered in response to higher injection rates, the Raton Basin's frequency‐magnitude and spatiotemporal statistics are not distinguishable from tectonic seismicity. The similarity suggests that seismicity in the Raton Basin is predominantly releasing tectonic stress.

    more » « less
  2. Abstract

    Earthquake stress drop is an important source parameter that directly links to strong ground motion and fundamental questions in earthquake physics. Stress drop estimations may contain significant uncertainties due to such factors as variations in material properties and data limitations, which limit the applications of stress drop interpretations. Using a high‐resolution borehole network, we estimate stress drop for 4551 (M0‐4) earthquakes on the San Andreas Fault at Parkfield, California, between 2001 and 2016 using spectral decomposition and an improved stacking method. To evaluate the influence of spatiotemporal variations of material properties on stress drop estimations, we apply different strategies to account for spatial variations of velocity and attenuation changes, and divide earthquakes into three separate time periods to correct temporal variations of attenuation. These results show that appropriate corrections can significantly reduce the scatter in stress drop estimates, and decrease apparent depth and magnitude dependence. We find that insufficient bandwidth can cause systematic underestimation of stress drop estimates and increased scatter. The stress drop measurements from the high‐frequency borehole recordings exhibit complex stable spatial patterns with no clear correlation with the nature of fault slip, or the slip distribution of the 2004 M6 earthquake. Temporal variations are significantly smaller, less well resolved and varying spatially. They do not affect the long‐term stress drop spatial variations, suggesting local material properties may control the spatial heterogeneity of stress drop.

    more » « less
  3. ABSTRACT Although the Brune source model describes earthquake moment release as a single pulse, it is widely used in studies of complex earthquakes with multiple episodes of high moment release (i.e., multiple subevents). In this study, we investigate how corner frequency estimates of earthquakes with multiple subevents are biased if they are based on the Brune source model. By assuming complex sources as a sum of multiple Brune sources, we analyze 1640 source time functions of Mw 5.5–8.0 earthquakes in the seismic source characteristic retrieved from deconvolving teleseismic body waves catalog to estimate the corner frequencies, onset times, and seismic moments of subevents. We identify more subevents for strike-slip earthquakes than dip-slip earthquakes, and the number of resolvable subevents increases with magnitude. We find that earthquake corner frequency correlates best with the corner frequency of the subevent with the highest moment release (i.e., the largest subsevent). This suggests that, when the Brune model is used, the estimated corner frequency and, therefore, the stress drop of a complex earthquake is determined primarily by the largest subevent rather than the total rupture area. Our results imply that, in addition to the simplified assumption of a radial rupture area with a constant rupture velocity, the stress variation of asperities, rather than the average stress change of the whole fault, contributes to the large variance of stress-drop estimates. 
    more » « less
  4. The 2019 Ridgecrest, CA earthquake sequence has provided a unique opportunity and a rich dataset to understand earthquake source properties and near-fault structure. Using the high-quality seismic data provided by the SCEC Stress Drop Validation group, we first estimate the corner frequency of M2.0-4.5 earthquakes by applying the spectral ratio method based on empirical Green’s function (Liu et al., 2020). We relate corner frequency estimates to stress drops assuming the Brune source model and circular cracks. Our preliminary results show increasing median stress drops with magnitude for both P and S waves, from 1 MPa for M2.0 events to 10 MPa for M4.0 events, though the limited frequency bandwidth may cause underestimation for small events. The estimated moment magnitude is proportional to the catalog magnitude by a factor of 0.72, which is close to 0.74 estimated by Trugman (2020) for the Ridgecrest earthquake sequence. In the second part of the study, we examine the impact of fault zone structure on the azimuthal variation of the source spectra. Using kinematic simulations and observations of the 2003 Big Bear earthquake sequence, Huang et al. (2016) showed that fault damage zones can act as an effective wave guide and cause high-frequency wave amplification along directions close to fault strike. We use clusters of M1.5-3 earthquakes in the Ridgecrest region to further examine the azimuthal variation of the stacked source spectra and investigate if the near-source structure can affect our corner frequency estimates. We aim to develop robust methods that utilize high-quality seismic data to illuminate earthquake source processes and fault zone properties. 
    more » « less
  5. Abstract

    Physics‐based numerical modeling of earthquake source processes strives to predict quantities of interest for seismic hazard, such as the probability of an earthquake rupture jumping between fault segments. How to assess the predictive power of numerical models remains a topic of ongoing debate. Here, we investigate how sensitive the outcomes of numerical simulations of sequences of earthquakes and aseismic slip are to choices in numerical discretization and treatment of inertial effects, using a simplified 2‐D crustal fault model with two co‐planar segments separated by a creeping barrier. Our simulations demonstrate that simplifying inertial effects and using oversized cells significantly affects the resulting earthquake sequences, including the rate of two‐segment ruptures. We find that fault models with different properties and modeling assumptions can produce similar frequency‐magnitude statistics and static stress drops but have different rates of two‐segment ruptures. For sufficiently long faults, we find that long‐term sequences of events can substantially differ even among simulations that are well resolved by standard considerations. In such simulations, some outcomes, such as static stress drops, are similar among adequately resolved simulations, whereas others, such as the rate of two‐segment ruptures, can be highly sensitive to numerical procedures and modeling assumptions. While it is possible that the response of models with additional ingredients ‐Realistic fault geometry, fluid effects, etc. ‐Would be less sensitive to numerical procedures, our results emphasize the need to examine the potential dependence of simulation outcomes on the modeling procedures and resolution, particularly when assessing their predictive value for seismic hazard assessment.

    more » « less