Earthquake Detection

Notebooks: codes/earthquake_detection.ipynb

How to detect earthquakes?

  • Amplitude threshold
  • STA/LTA
  • Template matching / Matched filter
  • Deep learning

Amplitude threshold

  • PGA (Peak Ground Acceleration)
  • PGV (Peak Ground Velocity)
  • Displacement

Recent M4.5 earthquake

The Modified Mercalli Intensity Scale (MMI)

Instrumental Intensity Acceleration (g) Velocity (cm/s) Perceived shaking Potential damage
I < 0.000464 < 0.0215 Not felt None
II–III 0.000464 – 0.00297 0.135 – 1.41 Weak None
IV 0.00297 – 0.0276 1.41 – 4.65 Light None
V 0.0276 – 0.115 4.65 – 9.64 Moderate Very light
Instrumental Intensity Acceleration (g) Velocity (cm/s) Perceived shaking Potential damage
VI 0.115 – 0.215 9.64 – 20 Strong Light
VII 0.215 – 0.401 20 – 41.4 Very strong Moderate
VIII 0.401 – 0.747 41.4 – 85.8 Severe Moderate to heavy
IX 0.747 – 1.39 85.8 – 178 Violent Heavy
X+ > 1.39 > 178 Extreme Very heavy
PGA Magnitude Depth Fatalities Earthquake
3.23g 7.8 15 km 2 2016 Kaikoura earthquake
2.7g 9.1 30 km 19,759 2011 Tōhoku earthquake
1.92g 7.7 8 km 2,415 1999 Jiji earthquake
1.82g 6.7 18 km 57 1994 Northridge earthquake
1.62g 7.8 10 km 57,658 2023 Turkey–Syria earthquake
0.65 6.9 19 km 63 1989 Loma Prieta earthquake

Amplitude threshold

  • Pros:
    • Simple and fast
    • Physical parameter
    • Directly related to shaking/damage
  • Cons:
    • Limit to large earthquakes
    • Need backgroud noise level for small earthquakes
  • Improvments:
    • How to make the threshold adaptive to the background noise level?

STA/LTA

  • STA/LTA = Short-Term Average / Long-Term Average

STA/LTA

STA/LTA

  • Pros:

    • Simple and fast
    • More sensitive than amplitude threshold
    • More robust for noisy data

  • Cons:

    • More parameters for tuning
    • Prone to false detections

Template matching / Matched filter

Review of convolution and cross-correlation in last lecture: cross-correlation

Notebook: cross-correlation

(QTM) Quake Template Matching

Template matching / Matched filter

  • Pros:
    • Robust to noise
    • More sensitive to small earthquakes
  • Cons:
    • High computational cost
    • Need existing catalog to build templates
    • Limited to waveform similarity with templates

FAST (Fingerprint And Similarity Thresholding)

  • Pros:
    • Sensitive to small earthquakes
    • Computational efficient
  • Cons:
    • Detect all repeating signals
    • Complex to implement

Deep learning

  • Generalized similarity search

Convolutional Neural Network for Earthquake detection and location

Residual Network of Convolutional and Recurrent Units for Earthquake Signal Detection

Deep learning

  • Pros:
    • Robust to noise
    • Sensitive to small earthquakes
    • Fast prediction
  • Cons:
    • Need large amount of labeled data
    • Black box
    • Generalization ability

---

### Download continuous seismic data ![height:500px](https://docs.obspy.org/_images/waveform_plotting_tutorial_4.png)

![](https://www.researchgate.net/profile/Eloi-Batlle/publication/228609631/figure/fig1/AS:669574030708741@1536650297568/Audio-Fingerprint-Framework_W640.jpg)