Source-Scanning Algorithm Research Articles

SSA2py: A High-Performance Python Implementation of the Source-Scanning Algorithm for Spatiotemporal Seismic Source Imaging

Abstract This article introduces the first version of SSA2py (v.1.0)—an open-source package designed to implement the source-scanning algorithm (SSA). SSA2py is a Python-based, high-performance-oriented package that incorporates the SSA method, which has been effectively applied to numerous earthquakes for imaging the spatiotemporal behavior of the seismic source. The software supports a wide range of data and metadata resources. These include the International Federation of Digital Seismograph Networks Web Services, the SeedLink protocol, and others, ensuring optimal access to waveforms and station metadata. Furthermore, the code may evaluate the quality of accessible waveforms using signal analysis methods, allowing for the most appropriate data selection. The SSA method has been computationally optimized using multiprocessing techniques for efficient central processing unit and graphic processing units executions, enabling considerably accelerated computational processes even for large-scale grid searches. The program is also designed to provide statistical and methodological uncertainties for the executed cases through jackknife, bootstrap, and backprojection array response function tests. After appropriate tuning by the user, SSA2py can be used for detailed earthquake source studies that backprojection technique typically serves as a complementary output to the source inversion result or as a near-real-time tool for successful and quick identification of the style and complexity of the earthquake rupture. With a wide and flexible configuration, the user has complete control over all calculating aspects of SSA2py. This article provides a detailed description of the structure and capabilities of this new package, and its reliability is demonstrated through targeted applications to the 2004 Mw 6.0 Parkfield and 2019 Mw 7.1 Ridgecrest earthquakes. Furthermore, the computational efficiency of SSA2py is validated through rigorous performance tests.

Microseismic Event Location with Dual Vertical DAS Arrays: Insights from the FORGE 2022 Stimulation

Abstract We investigate the resolvability of a microseismic event location given a recording array composed of vertical distributed acoustic sensing (DAS) boreholes. We use a modified source-scanning algorithm that takes into account both P and S waves. We transform the brightness maps it produces into probability density functions (PDFs), over which we carry out a resolution and uncertainty analysis. We apply this approach to microseismic events recorded by two vertical DAS boreholes as part of the Frontier Observatory for Research in Geothermal Energy (FORGE) project. We show that for the specific acquisition geometry in FORGE, the horizontal location of the events cannot be determined, but their depth can, similar to results obtained with a single borehole. Using synthetic examples, we show that the recording array’s geometry is the limiting factor in the determination of the horizontal location. We investigate various possible recording geometries composed of idealized DAS-like vertical boreholes with varying locations and depths. We find that, besides the number of recordingd boreholes, their depth is the main factor influencing the location estimation uncertainty. The number and position of the boreholes mainly influence the spatial distribution of the PDF, whereas the boreholes’ depth mainly influences its size. Despite the simplicity of our analysis, it highlights the influence of the monitoring array design for microseismic events’ locating using vertical DAS arrays.

Imaging the Sources of the March 2021 Seismic Sequence in Thessaly Basin (Central Greece) from Kinematic Slip Inversion and Backprojection of Waveform Envelopes

ABSTRACT In March 2021, a series of three moderate events with moment magnitudes Mw 6.3, 6.0, and 5.5 occurred within a span of 10 days in northern Thessaly, Greece, resulting in a stop–start pattern. The moment tensors (MTs) obtained from the events suggested normal faulting along distinct yet adjacent southeast–northwest-trending faults, indicating a sequential triggering process. We applied two methods to investigate the spatial and temporal characteristics of the source process of the triplet. This approach includes linear slip inversion of regional seismic waveforms and static Global Positioning System offsets, along with backprojection of regional waveforms using the Source-Scanning Algorithm technique. The detailed modeling suggests that three adjacent, subparallel (∼300°), and low-angle, detachment-type structures were responsible for the earthquakes. The stress axes orientation was determined by inverting a dataset of recomputed MTs combined with published solutions of the sequence. The results indicated a clockwise rotation of the local extensional axis by ∼25° compared to the regional ∼north–south extension. This rotation is consistent with the orientation of the modeled seismogenic structures. Inversions for both nodal planes and grid-searching fault geometry and orientation showed that for the first-two strongest events, the activated fault planes dipped to the northeast, whereas an antithetic (southwest dipping) fault was activated during the last weaker event. For the Mw 6.3 event, both the independent analyses indicated an overall rupture duration of ∼10 s, energy bursts above and near the hypocenter, up-dip (toward the southwest) rupture propagation in the early stages of the source process, and a bilateral rupture at later stages.

A novel deep-learning image condition for locating earthquake

SUMMARY Migration-based earthquake location methods may encounter the polarity reversal issue due to the non-explosive components of seismic sources, leading to an unfocused migration image. Such a problem usually makes it difficult to accurately retrieve the optimal location from the migrated source image. In this study, by taking advantage of the general pattern recognition ability of the convolutional neural network, we propose a novel deep-learning image condition (DLIC) to address this issue. The proposed DLIC measures the goodness of waveform alignments for both P and S waves, and it follows the geophysical principle of seismic imaging that the best-aligned waveforms represent fully a best-imaged source location. A synthetic test shows that the DLIC can effectively overcome the polarity reversal issues. Real data applications to southern California show that the DLIC can enhance the focusing of the migrated source image over the classic source scanning algorithm. Further tests show that the DLIC applies to continuous seismic data, to regions with few previously recorded earthquakes, and has the potential to locate small earthquakes. The proposed DLIC shall benefit the migration-based source location methods.

Coseismic and Postseismic Imaging of a Composite Fault System: The Samos 2020 Mw 7.0 Sequence

ABSTRACT The 2020 Mw 7.0 Samos earthquake (eastern Aegean Sea) provides a rare chance to study the ongoing interaction of normal and strike-slip faulting during an aftershock sequence in a suspected transfer zone with differential deformation style. After assessing them for possible site-effect influence, we first backproject local strong-motion data to elicit the high-frequency (HF) radiation for the main seismic event using the Source Scanning Algorithm. Our results indicate the existence of multiple HF radiators during the main rupture. Second, we construct a new, high-precision catalog for the aftershock sequence utilizing a deep-learning-based detector and picker. This new catalog contains thousands of additional seismic events compared with the routine catalogs built by local agencies and is accompanied by an enhanced dataset of fault-plane solutions determined with automated techniques. Complex fault networks activated during the aftershock period are mapped in detail, indicating the existence of an active transtensional, possibly asymmetric, basin offshore Samos Island. This detailed seismological characterization of Samos sequence leads to an improved understanding of the controlling mechanisms that operate concurrently during a sequence in a transtensional type of system.

Revisiting the Ridgecrest Aftershock Catalog Using a Modified Source-Scanning Algorithm Applied to Multiple Dense Local Arrays

Abstract We develop and implement a modified source-scanning algorithm (SSA) method to estimate the absolute event locations in the presence of velocity model errors, commonly found around fault zones. We split the array into subarrays of clustered receivers, for which relative travel-time errors are smaller. We apply a conventional SSA using both P and S waves to each subarray and combine the estimations using a probabilistic scheme to yield locations that are robust to velocity model errors. We also compute uncertainty estimations for the locations. We apply the method to 688 aftershocks recorded by 197 short-period geophones deployed as part of the Ridgecrest dense array. The dense receiver deployment allows for the recording of spatially coherent seismic arrivals. We compare 339 locations to a relocated catalog built using a sparser regional array and the same 1D velocity model. In general, locations are consistent despite the different methodology and recorded data. We qualitatively compare location estimations using the alignment of time-shifted seismograms, utilizing the spatial coherency of the dense subarrays. Our locations yield, in most cases, better alignment and are 2 km deeper on average. For events in the northern part of the study area, our locations are shifted to the north-east. We discuss various potential causes for the differences between estimations and investigate the possibility of velocity-driven biases in our locations. We also attempt to approximate the scale of lateral velocity heterogeneity near the fault in the northern part of the region. Although our location method is tailored to the Ridgecrest dense array, it demonstrates that using dense arrays may help mitigate the effect of velocity model errors on the absolute locations.

Combining Deep Learning and the Source-Scanning Algorithm for Improved Seismic Monitoring

ABSTRACTIn this study, we develop an integrated framework for simultaneous detection of seismic events and picking phase arrival times, phase association, and locating earthquakes. The proposed model combines the accuracy of convolutional neural networks for classification tasks and the efficiency of waveform-based algorithms for identifying coherent seismic arrivals. We find that our model strongly dominates the classic techniques, especially in identifying small magnitude earthquakes. We apply our model to one month of continuous seismic data recorded in western Canada for monitoring seismic activity associated with fluid injection operations. In comparison with previously developed deep-learning models, our technique reveals a nearly identical performance without human interaction during the entire process of picking the phase arrival times and locating the associated events.

Source properties of the 2019ML6.3 Hualien, Taiwan, earthquake, determined by the local strong motion networks

SUMMARYThe 2019 ML 6.3 Hualien earthquake struck the northern Longitudinal Valley (LV) and generated not only large strong motions (intensity of 7, as defined by the Central Weather Bureau, Taiwan) locally but also widespread strong shaking in metropolises in northern Taiwan. In this study, we analyse strong motion records from local seismic networks to understand the source properties of the 2019 event. We first obtain the centroid location of the 2019 event using the source-scanning algorithm (SSA) technique by applying the unfiltered records. The determined centroid location is 121.55°E, 24.10°N, with a depth of 22.5 km. This location is 5.5 km north–northwest of and 3.8 km deeper than the Central Weather Bureau hypocentre, suggesting that the 2019 event occurred on the high-angle west-dipping plane of the focal solution. The centroid time delay is 3.35 s. Then, we obtain strong motion generation areas (SMGAs) of the 2019 event using the empirical Green's function method by considering the broad-band waveforms (0.4–10 Hz). Unlike other moderate-sized earthquakes in Taiwan, which have one SMGA, we determine that there were two SMGAs in the 2019 event. SMGA1 initiated at the CWB hypocentre with a size of 4.00 km2, and SMGA2 initiated at the centroid location determined by the SSA approach with a size of 3.63 km2. Such small areas cause high stress drops of 13.7 and 27.4 MPa for SMGA1 and SMGA2, respectively. We infer that the localized high stress drop of SMGAs is one of the important factors responsible for high peak-ground accelerations (PGAs) in Taiwan in addition to a strong directivity effect coupled with the radiation pattern reported by the previous study. Furthermore, previous moderate-sized earthquakes on an active structure called the Xiulin segment revealed similar source properties with a high stress drop and generated large PGA locally as well as in the metropolises of northern Taiwan. Considering the stored moment deficit, the probability of a future large earthquake in the northern LV region remains high. It is essential to consider seismic hazard assessment and mitigation for this not-well-known but high-seismic-potential region.

Local stress field inverted for a shale gas play based on focal mechanisms determined from the joint source scanning algorithm

The joint source scanning algorithm (SSA) scans locations and focal mechanisms of microseismic events simultaneously. Compared to the traditional source scanning algorithm, it yields much more events with extra information of focal mechanisms. The availability of more events and focal mechanisms make it possible to invert for a 2D gridded stress field. As a byproduct of hydrofracturing monitoring, the method offers a new way to extract stress field as a substitute to other more expensive technologies. This method is applied to a hydraulic fracturing dataset collected from one shale gas production field in the southeast of the Sichuan basin. A damped stress inversion is conducted to obtain a 2D stress field. five hydraulic-fracturing induced fractures can be determined from the result. The events associated with these fractures generally have relatively low stacked energy and are limited to the depth of horizontal well. One existing fault (possibly associated with the axis of the central Sichuan uplift) is also determined and the events associated with the existing fault generally have higher stacked energy and are more densely populated. The existing fault may also serve as a structural boundary where the rocks to the NW side are easier to be fractured while events on the other side are sparse with low stacked energy. The existing fault also divides the stress field into two regimes: the maximum compressional stress field to the NW and SE of the fault line are dominantly in NW-SE and N-S directions, respectively.

Attenuation contrast in mantle wedge across the volcanic front of northeastern Japan that controls propagations of high-frequency S-wave later phases

Distinct later phases of waves with rich high-frequency (> 8 Hz) components were observed for intraslab earthquakes that occurred at intermediate depths, particularly at depths exceeding 100 km, in the northeastern (NE) Japan subduction zone. These high-frequency later phases (HFLPs) showed anomalously large peak-amplitude delays, up to ~ 50 s after direct S-wave arrivals at stations in the backarc region. Using a source-scanning algorithm, we investigated the locations of passing points affecting the propagation of HFLPs. The passing points were estimated to be in the forearc region in the entire NE Japan, indicating that HFLPs are scattered waves that pass through the forearc region. The propagating HFLPs seem to bypass the backarc mantle wedge, as a consequence of the distinct attenuation contrast in the mantle wedge across the volcanic front in NE Japan. These HFLP observations suggest that the high-attenuation zone in the backarc mantle wedge controls propagations of the high-frequency waves of intraslab earthquakes, in addition to the scatterers possibly locate in the forearc region.

Parallel Source Scanning Algorithm using GPUs

The use of methods using waveform stacking are nowadays more common in microseismic monitoring applications because they avoid manual or automatic phase picking. The Source Scanning Algorithm (SSA) is a widely known technique in which the source location is estimated using a brightness function obtained from stacking the normalized absolute amplitude seismograms recorded at several stations. The SSA has the advantage of the straightforwardness of its implementation but has the inconvenience of being computationally costly even for small-scale experiments. Our approach is then to parallelize the sequential SSA using graphics processing units (GPUs), and we named this parallel version pSSA. We have parallelized the Stacking step of the SSA Algorithm because this is by far the most computationally demanding Step. This can be done efficiently because of the spatial independence of the data. In our test cases we performed sequential and parallel computations of the SSA and pSSA in two different platforms. Additionally, we compared the performance of pSSA with a parallel implementation using OpenMP. We demonstrate that pSSA has produced speedups up to 125× as compared to the sequential version. We implemented a client–server architecture to receive and process the data. This architecture can treat with various simultaneous clients and also with out-of-order data packets. This allows for re-sending lost or corrupted data. We anticipate that pSSA has the impact of allowing SSA like algorithm to be used in microseismic experiment design and the use of on-site real-time denoising techniques, as well as the potential of being used in traffic light warning systems for fluid injection operations.

Reliability evaluation of volcanic tremor source location determination using cross-correlation functions

Summary Source location determination of volcanic tremor has been a challenge in seismology due to the waveform complexity and difficulties in reading P- and S-wave arrival times. We present a method for locating volcanic tremor recorded at a seismic network distributed around a volcano. The method combines the source-scanning algorithm and cross-correlation analysis. Tremor records are processed using a technique adopted from ambient seismic interferometry to obtain stacked cross-correlation functions (CCFs) for all station pairs, which are expected to show high amplitudes at the lag time that corresponds to the travel time difference between the stations. The best seismic source location is determined from the maximum of the sum of envelope amplitudes of CCFs at predicted travel time differences between all pairs of stations. This method does not compute theoretical amplitudes, assume an initial hypocenter location, or measure the arrival times. To quantitatively evaluate the accuracy of the location determination, we examine the method by using the vertical component seismic data of volcano-tectonic (VT) earthquakes recorded at six seismic stations at Izu-Oshima volcano. The VTs have been previously located by using arrival times of P- and S-waves, and the hypocenters are used as the reference for evaluation and error estimation of the method. The results show that the misfit, which is the distance between our estimated sources and the references, is about 2 km or less when using CCFs at the frequency band of 4–16 Hz which contains the dominant frequencies of direct S-waves. To test whether the method can be used for volcanic tremor, we simulate the tremors by combining the observed VTs that occurred randomly in time in a localized region. The simulated tremors are determined with location errors of approximately 1 km or less, when the sources of VTs are located within a distance of 1 km and CCFs are calculated for a minimum data length of about 2 minutes. The volcanic tremor location method we present here can be used as an alternative tool for volcano monitoring, especially to locate tremors and seismic events with no clear phase arrival.

On the accuracy and efficiency of the joint source scanning algorithm for hydraulic fracturing monitoring

We have developed the joint source scanning algorithm (JSSA) to determine the locations and focal mechanisms (FMs) of microseismic events simultaneously. However, the computational expense of using JSSA is too high to meet the requirements of real-time monitoring in industrial production. We have developed several scanning schemas to reduce computation time. A multistage scanning schema can significantly improve efficiency while retaining accuracy. For the optimized joint inversion method, a series of tests has been carried out using actual field data and synthetic data to evaluate the accuracy of the method, as well as its dependence on the noise level, source depths, FMs, and other factors. The surface-based arrays better constrain horizontal location errors ([Formula: see text]) and angular errors of P-axes (within 10° for [Formula: see text]). For sources with varying rakes, dips, strikes, and depths, the errors are mostly controlled by the partition of positive and negative polarities in different quadrants. More evenly partitioned polarities in different quadrants yield better results for locations and FMs. Nevertheless, even when some FMs have bad resolutions, the optimized JSSA method can still significantly improve location accuracies.

Joint inversion of source location and focal mechanism of microseismicity

The seismic focal mechanism (FM) is an effective property to indicate source physics, as well as stress and strain distribution in regional, local, and microscales. We have developed an algorithm to jointly invert for the FM and source locations. For a given velocity structure, all possible combinations of source locations ([Formula: see text], [Formula: see text], and [Formula: see text]) and FM (strike, dip, and rake) were used to compute traveltimes and polarities of waveforms. Correcting normal moveout times and polarities and stacking all waveforms, the ([Formula: see text], [Formula: see text], [Formula: see text], strike, dip, and rake) combination that gave the strongest stacking power was identified as the optimal solution. Compared with the traditional source scanning algorithm (SSA) that only scanned source locations, this algorithm was thereby called the joint source scanning algorithm (jSSA). The jSSA method was tested rigorously, and it was applied to a hydraulic fracturing data set. Our work determined several advantages against the SSA method: (1) The jSSA method could identify many shear sources that could not be detected by the SSA method due to polarity variation; (2) the jSSA almost always yielded more events than the SSA method, and the added events could often provide much better characterization of the hydraulic fracturing; (3) the statistics of source mechanisms could provide additional knowledge on the orientation of fractures, as well as the local and regional stress and strain field; and (4) for those events that were detected by both methods, the stacking power of jSSA was always higher than that obtained in SSA.

Detecting and locating volcanic tremors on the Klyuchevskoy group of volcanoes (Kamchatka) based on correlations of continuous seismic records

We analyse daily cross-correlation computed from continuous records by permanent stations operating in vicinity of the Klyuchevskoy group of volcanoes (Kamchatka). Seismic waves generated by volcanic tremors are clearly seen on the cross-correlations between some pairs of stations as strong signals at frequencies between 0.2 and 2 Hz and with traveltimes typically shorter than those corresponding to interstation propagation. First, we develop a 2-D source-scanning algorithm based on summation of the envelops of cross-correlations to detect seismic tremors and to determine locations from which the strong seismic energy is continuously emitted. In an alternative approach, we explore the distinctive character of the cross-correlation waveforms corresponding to tremors emitted by different volcanoes and develop a phase-matching method for detecting volcanic tremors. Application of these methods allows us to detect and to distinguish tremors generated by the Klyuchevskoy and the Tolbachik, volcanoes and to monitor evolution of their intensity in time.

Analysis of the source scanning algorithm with a new P-wave picker

We analyze the location of earthquakes in near regional networks using complete seismic records. The method is based on the source scanning algorithm (SSA) of Kao and Shan (2004), but similarly to Grigoli et al. (2013), seismograms are substituted by a P-wave picker trace. The picker traces in a network are repeatedly stacked using grid of trial source positions, and hypocenter is identified with the point providing the best stack (the largest brightness). The first innovation of this paper is a new picker, measuring the ratio of the summed absolute values of seismogram in the right and left part of a moving time window, the RPA/LPA picker. The brightness maps based on this picker are clearer than those based on the STA/LTA picker. The second innovation is a simple theoretical model of the brightness maps. It makes it easy to identify how individual stations contribute to form the brightness spot. It is shown on synthetic tests that the performance of the method depends on focal mechanism, progressively improving from normal to reverse and strike-slip events. The method is successfully applied to four events of different mechanisms and depths, recorded at different ranges of epicentral distance by either broad-band sensors or accelerographs. The events have been located close to previously published epicenters. The brightness maps provide an estimate of the relative uncertainty of the (non-linear) location problem. The uncertainty estimate is also applicable without measured arrival times, “without earthquakes”, thus useful when designing or upgrading seismic networks for better location performance.

Delineating complex spatiotemporal distribution of earthquake aftershocks: an improved Source-Scanning Algorithm

SUMMARY Conventional earthquake location methods depend critically on the correct identification of seismic phases and their arrival times from seismograms. Accurate phase picking is particularly difficult for aftershocks that occur closely in time and space, mostly because of the ambiguity of correlating the same phase at different stations. In this study, we introduce an improved Source-Scanning Algorithm (ISSA) for the purpose of delineating the complex distribution of aftershocks without time-consuming and labour-intensive phase-picking procedures. The improvements include the application of a ground motion analyser to separate P and S waves, the automatic adjustment of time windows for ‘brightness’ calculation based on the scanning resolution and a modified brightness function to combine constraints from multiple phases. Synthetic experiments simulating a challenging scenario are conducted to demonstrate the robustness of the ISSA. The method is applied to a field data set selected from the ocean-bottom-seismograph records of an offshore aftershock sequence southwest of Taiwan. Although visual inspection of the seismograms is ambiguous, our ISSA analysis clearly delineates two events that can best explain the observed waveform pattern.

Localization of T-wave energy on land revealed by a dense seismic network in Japan

SUMMARY We have investigated the wave propagation characteristics of a seismic T wave on land in northeastern Japan using the seismograms recorded by a dense seismic network. The analysed T wave is a unique one that was generated by a large intermediate-depth earthquake beneath the network that backpropagated from the Emperor Seamount Chain in the northern Pacific. The T wave has a predominant frequency of 1.5 Hz, which is lower than the previously reported values. Spatiotemporal variation of an rms envelope bandpass filtered with a passband of 1–2 Hz demonstrates multiple arrivals of the T wave and inhomogeneous distribution of amplitude. Judging from the spatial pattern of first arrival times and the polarization characteristics, such as the relatively large dip angle and high planarity, we regard the seismic T wave as a Rayleigh wave. A possible conversion area, estimated by a method similar to the source-scanning algorithm, is located to the northeast of Honshu Island and may have a dimension of ∼100 km. The most prominent feature of the spatiotemporal variation of the rms amplitude is the localization of T-wave energy in the northeastern part of Honshu Island. This localization can be attributed either to the close proximity of the conversion area to land or to the large extent of the area. However, these factors are only supplementary because they cannot explain a sharp contrast in amplitude between the two areas with comparable distance but different azimuth from the conversion point. An additional and essential factor in explaining the observation is the non-isotropic radiation from the conversion points. Since the conversion occurred on the continental slope, the maximum amplitude of converted energy is expected to appear either in the propagation direction of incoming acoustic waves or in the steepest direction of the local topography. These factors, in particular a convex shape of bathymetry, probably caused the focusing of converted energy into the northeastern part of Honshu Island.

Delineating the rupture planes of an earthquake doublet using Source-Scanning Algorithm: application to the 2005 March 3 Ilan Doublet, northeast Taiwan

SUMMARY Correct identification of the fault plane(s) associated with an earthquake doublet is a very challenging problem because the pair of events often occurs in close space and time with almost the same magnitude. Most long-period waveforms of an earthquake doublet are severely tangled and thus unsuitable for conventional waveform inversion methods. In this study, we try to resolve this issue by utilizing the recently developed Source-Scanning Algorithm (SSA). The SSA systematically searches the model space for seismic sources whose times and locations are most compatible with the observed arrivals of large amplitudes on seismograms. The identification of a seismic source is based on the brightness function, which is defined as the summation of the normalized waveform amplitudes at the predicted arrival times at all stations. By illuminating the spatiotemporal distribution of asperities during an earthquake's source process, we are able to constrain the orientation of the rupture propagation that, in turn, leads to the identification of the fault plane. A series of synthetic experiments are performed to test SSA's resolution under various scenarios including different directions of rupture propagation, imperfect station coverage and short origin time difference between the two events of a doublet. Because only short-period records are needed in the analysis, the proposed method is best suited for an earthquake doublet with a short time gap between the two events. Using the 2005 Ilan doublet (the origin time difference is only 70 s) that occurred in northeast Taiwan as an example, we show that the trace of the brightest spots moves towards the west and infer the E–W-striking plane to be the actual fault plane.

Research on the Movement of Vibration Source of Train by Means of SSA

AbstractIn this paper, we study the train‐induced vibration recorded by seismographs during our experiment in May 2004 in Yanzhou, Shandong Province by means of Source‐Scanning Algorithm (SSA). The result shows that there are 3 distinct nest‐shaped fine structures made by the movement of train during three very short periods. We get three vibration source images by piecewise separation. The images show that the train is heading forward from NW to SE when passing through this district. The example convinces us that the source of train can be dealt with as a lot of separate little vibration sources by means of some suitable pretreatment. It also shows the potentials of SSA in dealing with the data lacking seismic phases like this.