Estimating Blainville’s beaked whale density at AUTEC · Presentation Layout Introduction Methods Application Preliminary look at ambient noise in uence on detection function Conclusions

Presentation LayoutIntroduction

MethodsApplication

Preliminary look at ambient noise influence on detection functionConclusions and future work

Estimating Blainville’s beaked whale density atAUTEC

using passive acoustic data

T.A. Marques, J. Ward, L. Thomas, N. DiMarzio, P.L. Tyack,D. Moretti and S. Martin

16-07-2009

T.A. Marques, J. Ward, L. Thomas, N. DiMarzio, P.L. Tyack, D. Moretti and S. MartinEstimating Blainville’s beaked whale density at AUTEC


MethodsApplication


IntroductionBackgroundThe beaked whale case study

Methods

ApplicationNumber of detected clicks - nc

Click production rate - rProportion of true positives - cDetection function - PDensity estimate - D

Preliminary look at ambient noise influence on detection function

Conclusions and future work



MethodsApplication


BackgroundThe beaked whale case study

The problem

I Estimating the abundance of natural populations isfundamental for adequate management

I ”How many are there?” - Despite the simplest question onecan ask about a population, the answer is not simple

I Most common approach for cetaceans is distance sampling(using boats or planes)

I But there are several complicated issues associated with visualsighting based methods

The use of hydrophones has been suggested as an alternative tocollect useful data.



MethodsApplication



AUTEC

The Atlantic Undersea Test and Evaluation Center (AUTEC, in theBahamas) has

I 93 bottom mounted hydrophones(only 82 active in this data set)

I depths of ∼2 km and separated byabout ∼4 km

I covering an area of ∼1536 km2

I recording sound continuously

I automatic detection andclassification to get cetaceandetections



MethodsApplication



AUTEC

The Atlantic Undersea Test and Evaluation Center (AUTEC, in theBahamas) has

I 93 bottom mounted hydrophones(only 82 active in this data set)

I depths of ∼2 km and separated byabout ∼4 km

I covering an area of ∼1536 km2

I recording sound continuously

I automatic detection andclassification to get cetaceandetections 0 5000 10000 15000 20000 25000 30000

−2

00

00

−1

00

00

01

00

00

20

00

03

00

00

easting

no

rth

ing

4748

54

55

62

63

70

71

79

86

87

1 2

3 4 5

6 7

8 9

10 11 12

13 14

1516

17

1819

2021

2223

2425

2627

2829

3031

3233

34

3536

3738

3940

41

4243

4445

464950

5152

5356

5758

5960

6164

6566

6768

6972

7374

7576

7778

8081

8283

8485

8889

909192

93



MethodsApplication



Beaked whales

Here we consider Blainville’s beaked whales (BBW):

I deep long dives, with short periods at the surface

I great number of echolocation clicks during the deeper part ofdives

I very hard to detect visually

0 50000 100000 150000 200000 250000

−1500

−1000

−500

0

Time (seconds since start)

Depth

Data kindly provided by Robin Baird (many thanks!)



MethodsApplication


Formulation I

Population Density: number ofanimals per unit area

D =N

A

I D - density

I N - number of animals

I A - area

If we can not count all theanimals in the covered area:

D =n

aP

I n - number of detected animals

I a - covered area

I P - detection probability



MethodsApplication


Formulation II

Animal density (D) can be estimated1 based on the nc clicksdetected over time interval T over K hydrophones by (cuecounting approach)

D =nc c

KπPw2T r

ondeI r - click production rate

I c - proportion of sounds classified as BBW that were really BBW

I w - maximum distance at which it is possible to detect a click

I P - detection probability of a click in a circle with radius w around thehydrophone

1θ represents an estimator of θT.A. Marques, J. Ward, L. Thomas, N. DiMarzio, P.L. Tyack, D. Moretti and S. MartinEstimating Blainville’s beaked whale density at AUTEC


MethodsApplication


Intuitive approach

D =ncc

kπw2PTr

sounds that were really BBW

(effective detection area) × (clicks produced by an animal duringtime period T )

Note that:

I nc × c represents the number of true BBW soundsI K × π × w2 × P = K × π × ρ2 it is the effective area of

detectionI T × r it is the number of clicks produced by an animal in the

time period T



MethodsApplication


Number of detected clicks - ncClick production rate - rProportion of true positives - cDetection function - PDensity estimate - D

Number of detected clicks

Data: ≈ 6 days of data (from the 26th April till 2nd May 2005) -Previously analyzed by Moretti et al. 2006)

I counts over 4961 minutes

I nc = 2940521 soundsconsidered to be BBW clicks(pooled over the 82hydrophones)

I hydrophone as the samplingunit (CV 5.5%)



MethodsApplication



Available data for estimating r

5 whales, 21 dives, many thousand clicks

0 10000 20000 30000 40000 50000 60000

−140

0−8

00−2

00

Second

Depth x x x x x

x x x x xx

0 10000 20000 30000 40000 50000 60000

−140

0−8

00−2

00

Second

Depth

xx x xx x xx

0 20000 40000 60000

−100

0−6

00−2

00

Second

Depth

xxx

x

x xx

x x xx

0 10000 20000 30000 40000 50000 60000

−100

0−6

00−2

00

Second

Depth x

x x

xx

x xx



MethodsApplication



Results

Weighted average of the number of clicks per second per full deepdive cycle (weighed by time).

r = 0.41 clicks per second (CV 9.8%)



MethodsApplication



Sampling

Not possible to get c directly, sampling approach was used. Giventhe 6 day data set:

I sample of 30 small time periods (10 minutes each)

I systematically spaced over the 6 day data set (with a randomstart)

I for each period, and for each hydrophone and minute, manualevaluation of which detected sounds were really BBW or not



MethodsApplication



Results

c estimated by the weighted mean of the proportion of truepositives (weights: total number of clicks per period).

I (during sampled periods) 160302 sounds detected andoriginally considered to be BBW

I 78450 (∼50%) identified as being for sure from BBW

I ∼6% of all clicks in a ”mixed” group (BBW + other)

c : 0.549 (CV=1.99%) or 0.489 (CV=2.29%)

(depending on considering ”mixed” clicks to be all or none from BBW)



MethodsApplication



Proposed approach to estimate P

Estimate a detection function: detection probability as a functionof relevant covariates.

I DTag’s over 4 whales (for a totalof 13 deep dives)

I For each click produced by theanimal, record of its detection (orfailure to do so) in surroundinghydrophones

I “Logistic regression” to estimatethe detection function



MethodsApplication



Distances from clicks to hidrophones

Example: distances to produced clicks and detected by hydrophoneAll hydrophones

Fre

quen

cy

2000 3000 4000 5000

050

015

0025

00

Hydrophone 37

Fre

quen

cy

4000 4500 5000 5500

010

020

030

040

050

060

0 Hydrophone 43

Fre

quen

cy

1400 1500 1600 1700 1800 1900 2000

020

060

010

0014

00

Hydrophone 44

Fre

quen

cy

2400 2600 2800 3000 3200

020

040

060

080

0

Hydrophone 49

Fre

quen

cy

2800 3000 3200 3400 3600 3800 4000

020

040

060

080

0

Hydrophone 50

Fre

quen

cy

2800 3000 3200 3400 36000

100

200

300

400

500



MethodsApplication



Fitted model

P(detecting a click is afunction of):

I Distance

I Orientation (relative)

I Pitch (relative)

Integrate out variables bysimulations

P = mean click detectionprobability, in a 8 km radius, is0.032 (CV 15.9%).



MethodsApplication



Density estimate

Combining all this information, the estimated BBW density is:

25.3 (17.3-36.9) or 22.5 (15.4-32.9) BBW per 1000 km2,depending on the proportion of true positives

Moretti et al. 2006: 34.7 or 25.4 BW per 1000 km2



MethodsApplication


Ambient noise

I Characterizing the ambient noise at AUTEC (“sprinkles”)

I Adding ambient noise to hydrophone data from DTag events

I small data subset: 1 Dtag, 4 hydrophones (all AHRPuni-directional)

I 4 levels of noise added, resulting in an ambient noise criteria(ANC)

I Re-run detector and classifier to get detections associatedwith DTag data

I New model for the detection function, now including ANC



MethodsApplication


Ambient noise II

I Unexpected increase in detections with moderate levels ofnoise

I Strange pattern for hydrophone 73



MethodsApplication


Ambient noise III



MethodsApplication


Ambient noise IV

I Get a new density estimate, accounting for ambient noise

I Get ANC within 5 minute periods over the 6 day data set

I Each click in each period gets the detection probability ofdetection conditional on the observed ANC and with othercovariates integrated out

Proof-of-concept → AHRP uni-directional hydrophones only



MethodsApplication


Conclusions

I results consistent with other estimates in the area

I easily extendable to other species (and to terrestrialenvironments)

I but estimating detection function required DTag’s, which arenot available in the supermarket

I need the right (i.e. for the survey period considered) detectionfunction, click rate and true positive proportion

I process can be optimized at several levels



MethodsApplication


Future work

I more independent dive data for modeling the detectionfunction

I incorporate ambient noise in the detection function andcontrast results (work in progress)

I better sound classification (ambiguity / mixed groups)I testing under different scenariosI DECAF: other approaches to estimate the detection function

from passive acoustic data (e.g. SECR, sonar equation, soundpropagation models, etc)

More about this: Marques, T. A., Thomas, L., Ward, J., DiMarzio, N. & P. L. Tyack

(2009). Estimating cetacean population density using fixed passive acoustic sensors:

an example with Blainville’s beaked whales. The Journal of the Acoustical Society of

America. 125: 1982-1994.T.A. Marques, J. Ward, L. Thomas, N. DiMarzio, P.L. Tyack, D. Moretti and S. MartinEstimating Blainville’s beaked whale density at AUTEC

Documents

Estimating Blainville’s beaked whale density at AUTEC · Presentation Layout Introduction Methods Application Preliminary look at ambient noise in uence on detection function Conclusions