Automatic classification of the seismic monitoring data at the Aaknes rock slope, Norway

1
Automatic classification of the seismic
monitoring dataat the Aaknes rock slope, Norway

• T. Fischer1,2, M. Roth3, D. Kuehn3
• (1) Institute of Geophysics, Czech Academy of
  Sciences
• (2) Faculty of Science, Charles University in
  Prague
• (3) NORSAR

2
Outline

• Site characterization
• Seismic network
• Data examples
• Ongoing work
• Spike removal
• Event classification

3
Åknes rock slide location
Rock-avalanche hazard zones in Møre Romsdal
4
Åknes fjord landscape
Åknes
Large rockslide deposits
5
Åknes rock slope
6
Seismic monitoring
Seismic network 8 3-C geophones
Broadband station AKN
7
Back-fracture and seismic network
8
Seismic network
8
9
Network

• 8 three-component geophones
• Network size 250 x 150 m
• Central data acquisition in upper bunker
• Radio link to Hellesylt (13 km distance)
• Internet connection to NORSAR
• Operational since October 2005
• Realtime data transfer and processing

10
Data examples
Rockfalls
Blasts
Lightning/Generator
Earthquakes
Microseismic
Distant
Spikes, noise
Distant
Local short
Local long
ftp//ftp.norsar.no/pub/outgoing/aaknesplot/latest
waveforms/
11
Data examples
Gudvangen rock avalanche 15 June 2010, 0422
GMT 2000 5000 m3 Equivalent to M 2 earthquake
150 km South of Åknes
12
Real-time processing (existing)
http//www.norsar.no/ pc-47-48-Latest-Data.aspx
4 geophones 5 geophones 6 geophones 7 geophones 8
geophones
Last 30 days
12
13
Automatic classification

• Goals
• Remove EM noise (lightnings etc.) - spikes
• Distinguish among 4 event types
• Local short (slip events, locatable) single
  pulse, steep onset, lt1s
• Local long (rockfalls) multiple pulses,
  emergent onset, gt2s
• Distant very long signals with onset
• Noise no clear onset

14
1. Remove spikes

• Spikes detection
• EM spikes synchronous on traces
• Simple stacking of normalized absolute amplitudes
  OK
• Measure maximum amplitude and half-width of the
  pulse

spike
event
15
1. Remove spikes

• Criteria for spikes
• Amax gt 60 NumTr
• Duration lt 10 smpl

16
2. Distinguish event types

• Calculate the signal envelope, sum the envelope
  over traces
• Measured parameters of the stacked signal
  envelope
• Maximum amplitude
• Left and right width at two levels, 33 and 67
• Number of local maxima
• Interval btw. outer maxima
• Training on manually classified data set incl.
  780 events during 2011
• 119 local short
• 526 local long
• 109 distant
• 28 noise

17
Event types - examples
18
Event types - examples
19
Distinguish event types - statistical analysis

• Choice of the statistics
• Aim get the P that the analyzed event belongs to
  one of the event types
• How calculate the P of each studied parameter
• Which statistics
• pdf too discontinuous
• cdf not suitable, P(Xltx)
• folded cdf peaked function
• Plt0.5 -gt P(Xltx)
• Pgt0.5 -gt P(Xltx)

20
Distinguish event types - statistical analysis

• Folded cumulative distribution of the parameters
  
• Shows maximum at median
• Each type characterized by a typical distribution

Maximum amplitude
Training evaluate the mean distribution of the
six parameters of all five event types for the
manually classified training data set
21
Results of distinguishing events

• Automatic characterization applied to 4300
  triggers in 2011
• Success rate
• 81 local short events (slip events)
• 73 local long events (rockfalls, avalanches)
• 80 distant noise events

22
Summary

• Aaknes rock slide is continuously monitored
• Automatic classification of events in progress
• EM spikes successfully removed by stacking traces
• Event classification based on characterization of
  the shape of signal envelope
• Success rate about 80 achived in classifying
  slip events, rockfalls and distant events/noise
• Outlook approximate event localization

23
1. Remove spikes

• Statistical analysis of amplitudes quantile
  ratios
• Stationary signal normal distribution,
  Q99/Q95  1.5
• Spiky signal dominance of high amplitudes high
  Q99/Q95 ratio

24
1. Remove spikes

• Q99/Q95 measured on 917 manually classified
  events gt