Automated Model-based Localization of Marine Mammals Near Hawaii

1
Automated Model-Based Localization of Marine
Mammals
Christopher O. Tiemann Michael B. Porter Science
Applications International Corporation John A.
Hildebrand Scripps Institution of Oceanography

2
Traditional Passive Acoustic Localization Methods

• Hyperbolic fixing Assumption of direct acoustic
  path
• and
  constant soundspeed
• Matched-field processing Sensitive to
  environment

• Advantages of Model-Based
• Localization Technique
• Acoustic propagation model provides accuracy
• Robust against environmental and acoustic
  variability
• Graphical display with inherent confidence
  metrics
• Applicable to sparse arrays
• Fast for real-time processing without user
  interaction

3
Algorithm has been tested with real acoustic data
from two locations
Robust against differences in environment and
species
PMRF Deep water Humpback whale calls .2-4
kHz 2 sec duration Sperm whale
clicks Hydrophone array
San Clemente Shallow water Blue whale calls
10-20 Hz 20 sec duration Seismometer array
4
Array Geometries
Pacific Missile Range Facility Hydrophone
Positions
San Clemente Seismometer Positions
5
Spectrograms from PMRF Channels 2 and 4
3/22/01 201630
dB
Time-Lag
dB
6
San Clemente Seismometer Spectrograms
Seismometer 1 08/28/01 1136
Sensors measured 3-axis velocity plus pressure
Blue whale type A and B calls observed
4 receivers 11 days of data 128 Hz sample rate
7
Algorithm Overview
1) Predict direct and reflected acoustic
path travel times and time-lags
2) Pair-wise cross- correlation measures
time-lag
3) Compare predicted vs measured time-lags
for likelihood scores
4) Summed scores form ambiguity surface
indicating mammal position and confidence
8
Ch. 2, 3/22/01 201630
Spectrogram Correlation

• Pixilate spectrograms
• to binary intensity
• (black white)

Ch. 4, 3/22/01 201630
2) Correlate via logical AND and
count of overlapping pixels
9
Spectral correlations provide more consistent
time-lag estimates than do waveform correlations

Time-lag between PMRF Ch. 2 4, 3/22/01 201600
Time-lag between PMRF Ch. 2 4, 3/22/01 201600
10
Phase-Only Correlation

• Measures time-lag between receiver pairs
• Product of two whitened spectra
• Frequency-band specific
• Advantages over waveform or spectrogram
  correlation
• Over time, see change in bearing to persistent
  sources

Pair-wise Time-lag between Seismometers 1 and
4 08/28/01 08/30/01
11
Ambiguity Surface Construction
PMRF 3/22/01 2016
1) Discard low-score time-lags 2) Compare
predicted vs measured time-lags for all
candidate source positions 3) Sum
likelihood contributions from all hydrophone
pairs
12
Whale Tracking
Ambiguity surface peaks from consecutive
localizations follow movement of source
San Clemente
13
Tracking Examples

• Sources can be localized far outside array
• Tracks give clues to animal behavior

08/28/01 0252-0452
08/28/01 0933-1350
08/29/01 0255-0450
14
Tracking Examples
Whale movement can be followed with time-lapse
movies. Click on a figure to play.
San Clemente 08/28/01 0252 0443
San Clemente 08/28/01 0933 1350
15
Depth Estimation
Repeat modeling and surface construction for
several depths Surface peak defocuses at
incorrect depths
Sperm whale localization at PMRF 03/10/02 1153
200 m depth
800 m depth
UTM North (km)
UTM East (km)
UTM East (km)
16
Multiple Sources

• Singing whales
• Time-lag from single correlation peak limits
• one localization per receiver pair
• Different receiver pairs can localize different
  sources
• on same ambiguity surface

• Clicking whales
• Pair-wise click association tool measures
  time-lag
• Can track multiple whales simultaneously

PMRF receiver 501 waveform, 03/10/02 1152, with
clicks identified
Amplitude
Time (sec)
17
Verification

• Goal to verify accuracy of localization
  algorithm
• Low probability of concurrent visual and
  acoustic localization
• of same individual

Sperm Whale Localizations at PMRF 03/10/02

• Matched acoustics to
• visual sighting
• of sperm whale pod
• at PMRF
• Have data from
• controlled-source
• localization
• experiment at AUTEC

1154-1156
1155
1153-1156
1158
18
Conclusions

• Model-based algorithm benefits
• Portable to other distributed array shapes,
• environments, and sources of interest
• Robust against environmental variability
• Suitable for automated real-time processing
• Modular design
• Future work
• Test on other ranges, species and vs. controlled
  source
• Add species identification tool
• Long-term, real-time range monitoring and alert
  generation