herring and mackerel using broadband acoustics A. Pobitzer 1 , E. - - PowerPoint PPT Presentation

Pre-catch sizing of herring and mackerel using broadband acoustics

• A. Pobitzer1, E. Ona2, G. Macaulay2, R. Korneliussen2,1,
• A. Totland2, Y. Heggelund1, I. Eliassen1

1Christian Michelsen Research, 2Institute of Marine Research

2

Motivation for pre-catch sizing

Purse seining

• Traditional catch method

for schooling fish

• Workflow

1. Circumnavigate school setting the net 2. Purse the seine 3. Pull up net

3

Motivation for pre-catch sizing

• Sizing:

price ~ length / weight of fish price ~ size distribution

• Pre-catch:

prevent «slipping» prevent time-consuming sampling

• reduce fish mortality
• comply with regulations
• Challenge: resolve and measure individual fish
• sideways
• at 50 – 100 m range

4

Outline of proposed method

• 1. Resolve single targets in outskirts of schools with split

beam wide band principle and a very narrow beam

• 2. Find size related parameters in TS, pulse form and

backscattered spectrum

• 3. Compute statistical parameters
• 4. Visualize estimated size in echogram

5

Example: Resolve single targets in

• utskirts of schools (sideways)

(LSSS, forthcoming release) Narrow-beam (3o) broadband SB-transducer (160 – 260 kHz), mounted on keel of “G.O. Sars” Mackerel school at 20 – 120 m (sideways) data acquired during cruise in autumn 2014, “G.O. Sars”

6

Size estimation based on TS spread

• Observations
• spread in the TS spectrum ~

constant under rotation

• larger fish ~ more spread
• 3rd quartile – median

stable quantifier of spread

• Empirical model for

𝑀(𝑡𝑞𝑠𝑓𝑏𝑒)

7

Size estimation based on echo duration

• (Stanton et al., 2003):

Δ𝜐 =

2⋅𝑀𝑓𝑔𝑔 𝑑

|cos(𝜚)| (𝜚 unknown)

• Here: 𝑀𝑓𝑔𝑔 unknown
• Use least-square curve fit

for estimation of 𝑀𝑓𝑔𝑔

• Problematic areas
• 𝜚 ≈ 0∘ (acquisition)
• 𝜚 ≈ 90∘ + 𝑙 ⋅ 180∘, 𝑙 ∈ ℤ

(model)

𝑀𝑛𝑓𝑏𝑡𝑣𝑠𝑓𝑒 = 0.34 𝑛 (𝑀𝑓𝑔𝑔 ≈ 𝑀 ⋅ 0.9 = 0.306 𝑛)

8

Size estimation based on dips in frequency spectrum

• Observation:

number of dips in spectra follows 𝑜𝑒𝑗𝑞𝑡(𝜚) = 𝑏|cos(𝜚)|

• Less unstable for 𝜚 ≈ 0∘
• Least square fit for estimation
• f 𝑏 = 𝑜𝑒𝑗𝑞𝑡(0∘)
• Empirical model for

𝑀(𝑏)

• TS spectra for

𝜚 ≈ 90∘ + 𝑙 ⋅ 180∘, 𝑙 ∈ ℤ «flat»  model applicable

9

Robustification of size estimation

• Minimise influence of noise by combining all three

measurements taking median

• 𝜚 ≈ 0∘ problematic for size estimation based on signal

length (and to some extent number of dips)  use the TS spread only

10

Software implementation

• Methodology implemented in prototype
• EK80 data  Track single fish  extract features needed for

three methods  present size probability to user

• Features
• Real-time
• For now herring

and mackerel

11

Detail: Computation of probability

• Probability estimation through Kernel Density Estimation
• KDE ≈ continuous histogram
• Advantage: takes uncertainty in account!
• Measured size μ +

uncertainty σ  𝑂(μ, σ)

• Superposition gives

estimation of probability density function of population

12

Video from software (mackerel)

13

Summary & future work

• DONE
• Proposed pre-catch sizing algorithm based on
• TS
• pulse form
• spectral information
• Currently limited to herring and mackerel
• Prototype implemented and tested in field
• TO DO
• Extension to other species?
• Influence of dorsal incidence angle, depth (swimbladder

compression)?

• …

14

Acknowledgments

• The here presented work is a result of the

DABGRAF-project, founded by The Norwegian Seafood Research Fund – FHF

• Simrad for the use of EK80 prototype and for making a

new split beam transducer

• Institute of Marine Research for the use of “G.O. Sars”

keel on three surveys

15

16