Automatic Detection of Low-SNR Seismic Events by Pattern Matching with Automatically Generated Prototypes in an OSI Scenario Benjamin Sick (
[email protected]), Manfred Joswig
Science & Technology 2013
Institute for Geophysics, University of Stuttgart, Germany
Introduction
Detection and Classification with only 1 Mini-Array
The reliable, automatic detection of low-SNR seismic events is not yet feasible without a large amount of false-positive detections, i.e., false alarms. This applies especially to temporal local networks with exposed seismic stations and a-priori unknown noise conditions and event signatures, as in an OSI seismic aftershock monitoring. To overcome this problem, we use a multi-path approach. As a first step, high-SNR events from noise bursts and seismic signals are detected by conventional STA/LTA triggering and coincidence analysis. These events are grouped for similarity to define a set of master events. Thus in step two the events are transformed into noise-adapted sonograms, and further reduced in dimension by principal component analysis (PCA). A self-organizing map (SOM) is used then to create event prototypes by event alignment on a two-dimensional grid based on similarity. Prototypes which are based on noise signals will positively identify repetitive noise sources, while the remaining signal prototypes are used to detect any low-SNR events in the full data set through adapted pattern matching. This method allows to lower the automatic detection threshold significantly while identifying and discarding a large amount of false-positives. The data of only one mini-array is used because the weak signals of interest might not be detected at more stations.
Super-sonogram Compilation Single stations of one SNS are within 200 m of distance which makes it possible to combine the four vertical traces of one SNS into a super-sonogram. Each Pixel of the super-sonogram consists of four sub-pixels, each from one vertical trace of the SNS. Array-wide signal coherency can be checked fast and the data of all stations can be displayed on one screen, Sick et al. (2012). WEST
SUPER-PIXEL
W
NORTH
20 sec
C
E
CENTER
EAST
20 sec SUPERSONOGRAM
Figure 3 : Combination of four single sonograms to one super-sonogram.
North 20 - 100m
West
East
Figure 1 : Tripartite mini-array with 1 central 3-component seismometer and 3 satellite vertical seismometers.
slidequake
noise
150.0 Hz
15.0 Hz
1.6 Hz
8 sec Figure 4 : Super-sonogram pattern examples of the three event classes and typical noise.
Results
At first high-SNR events are detected by STA/LTA on the seismic data with a coincidence of at least 3 of the 4 vertical traces of the mini-array. With these parameters, 102 total triggers are made with 43 of these corresponding to events from the manual bulletin (3 slidequakes, 27 frost-heave events and 13 local earthquakes). At each trigger position a pattern is created in the way described previously. These patterns are then used to construct the self-organizing map (SOM, Kohonen 2001). The SOM groups event classes without prior knowledge, i.e., unsupervised and creates a map of representatives for each event type arranged by proximity of features, giving us a synoptic and topological overview of the triggered events. An analyst can classify regions in the SOM and the labeled events of these regions are used in step two for a pattern recognition to classify the unlabeled events from triggers of a STA/LTA with a much lower threshold.
In total we get 1576 false positive noise detections of which 1007 are classified as such by the pattern comparison. All frost-heave events, almost all local earthquakes and the majority of the slidequakes are classified correctly. 13 slidequakes are classified as local earthquakes which can be accounted to the high similarity of these events.
slidequake
Feature Extraction Actual class
Pattern recognition on raw waveforms with cross-correlation can only be feasible for high SNR events and a compact source region. In case of the search for aftershocks however the expected SNR of the events is so low that the changing noise conditions would influence the cross-correlation significantly. A more noise robust representation for pattern recognition which is called sonograms is used here. Sonograms are based on spectrograms but apply several noise cancellation steps per frequency band which makes patterns robust even in changing noise conditions, Joswig (1993). Power spectral density through short-termfourier-transformation (STFT) I Half-octave frequencyand logarithmic amplitude-scaling I Frequency dependant noise-adaptation, muting and prewhitening I Amplitude normalization to provide an amplitude-invariant clustering I Transformation with Principal Component Analysis (PCA) of the first 5 principal components
frost-heave local
In this study we use data from 14 days of the Heumoes slope which contains 37 local earthquakes, 60 slidequakes and 67 frost-heave events. The frost-heave events are only visible at one mini-array which will limit our analysis to the information of this array.
1st step: Template Creation
120°
Center
Data from OSI training campaigns is restricted to internal use which is why we take data from research projects with similar requirements as the OSI SAMS would face. Here we use a dataset from a permanent seismic network of 2 years for the analysis of slidequakes on a creeping landslide in Vorarlberg, Austria (Walter et al. 2011) with the following properties similar to SAMS challenges: I Detected events on the landslide with magnitudes down to M -3.2. L I Low event rate of approximately 1 slidequake in 5 days in average. I Multiple noise sources on the slope: ski-slope (lift, snowcat), holiday village (people, cars), agriculture (cattle, tractor) etc. I Ground truth manual bulletin for verification of automatic analysis with mainly three types of events: 1. Slidequakes, small fracture processes from the movement of the slope 2. Frost-heave events, near-surface local events from freezing processes which are only detected at one mini-array at a time 3. Local earthquakes
N
20 sec
Seismic aftershock monitoring system (SAMS) The seismic aftershock monitoring system (SAMS) of an OSI comprises up to fifty mini-arrays (Seismic Navigating System, SNS) which can be deployed in the 1000 km2 inspection area. The soughtafter events can have a magnitude as low as ML -2.0, and a duration of just a few seconds which makes it particularly hard to discover them in the large data set.
OSI Scenario and Event Examples
frost-heave
Predicted local class eq
slidequake
frost-heave
42
0
0
60
13 0
noise
1.0
9
0.9
0
2 µm/s
STFT
0.6
local eq
0
0
26
2
noise
246
125
198 1007
0.4 0.3 0.2 0.1 0.0
Figure 7 : Confusion matrix for classification with 1 mini-array.
150.0 Hz
Increasing energy
15.0 Hz
1.6 Hz
PCA
Figure 6 : The SOM with each node visualized as the according pattern prototype which is an average of the closest event patterns. In this view an analyst can recognize event categories and label them appropriately and thereby classify the underlying patterns. These labeled patterns are used in step 2 to classify unknown events.
0.7
0.5
I
Figure 5 : Grid of the SOM, each black dot represents 1 SOM node. The colored rectangles represent patterns which are most similar to the node prototype (blue=slidequake, pink=frost-heave, red=local earthquake). The thickness of the rectangular border around the nodes indicates the similarity to a neighbouring node (thicker lines = less similar).
0.8
150.0 Hz
15.0 Hz
2nd step: Pattern Recognition In the second step the STA/LTA parameters are changed to detect also low-SNR events. This results in 1728 detections of which 152 are actual events from the manual bulletin. Patterns are created again at all detections and compared by cross correlation and with additional varying amplitude shifts based on the area of maximum energy in the pattern, Joswig (1993). This allows to detect also patterns where the trigger time is shifted and also takes into account similar patterns with different amplitudes.
1.6 Hz
8 sec
Figure 2 : Transformation from seismograms to sonograms and feature extraction by PCA.
Institute for Geophysics, University of Stuttgart, Germany
References Joswig, M. (1993). Automated Seismogram Analysis for the Tripartite Bug Array: An Introduction. Computers & Geosciences, 19(2):203–206. Kohonen, T. (2001). Self-organizing maps. Springer Ser. Inf. Sci., 30:501 pp. Sick, B., Walter, M., and Joswig, M. (2012). Visual Event Screening of Continuous Seismic Data by Supersonograms. Pure and Applied Geophysics, pages 1–11. Walter, M., Walser, M., and Joswig, M. (2011). Mapping rainfall-triggered fracture processes, and seismic determination of landslide volume at the creeping Heumoes slope. Vadose Zone Journal, 10(2):487–495.
For most of the wrong classifications, it is also impossible to manually classify the events if restricted to one mini-array. The 9 slidequakes which are discarded as noise events could be recognized by a coincidence analysis over multiple mini-arrays which could run on all “noise” detections. Furthermore only 3 slidequake were triggered in the first step and only these were available as templates. In total we get a recognition rate of 80.6 % (86.2 % without the noise class) in bad noise conditions with very low-SNR events and only using one mini-array with four single stations. Acknowldgements: We would like to thank Marco Walter who constructed the permanent seismic network on the Heumoes slope and created the ground truth bulletin by manual analysis of the data with 3 mini-arrays and the software NanoseismicSuite (Sick et al. 2012). The software is also the official software of SAMS.
http://www.geophys.uni-stuttgart.de
