05.08.2013 Views

Blue whale calling in the Rottnest trench-2000, Western ... - ANP

Blue whale calling in the Rottnest trench-2000, Western ... - ANP

Blue whale calling in the Rottnest trench-2000, Western ... - ANP

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

dB for first spectral peak and 3 dB for second spectral peak) <strong>the</strong> correlation po<strong>in</strong>t was deemed<br />

correct.<br />

This gave <strong>the</strong> presence of a component. To give <strong>the</strong> number of dist<strong>in</strong>ct <strong>call<strong>in</strong>g</strong> animals with<strong>in</strong><br />

each 90 s block, <strong>the</strong> number of type II components was counted (s<strong>in</strong>ce it was not possible for a<br />

s<strong>in</strong>gle animal to produce more than one type II component <strong>in</strong> a 90 s block) and <strong>the</strong> number of<br />

type II components which fell with<strong>in</strong> 32.5 s of <strong>the</strong> block start added to this (s<strong>in</strong>ce a type III<br />

component followed around 22 s after a type II component and ran for approximately 23 s, thus a<br />

lead<strong>in</strong>g type III component <strong>in</strong>dicated a different caller). Type I components which fell <strong>in</strong> <strong>the</strong> last<br />

45 s of <strong>the</strong> block also would <strong>in</strong>dicate a different caller, but were ignored <strong>in</strong> this version of <strong>the</strong><br />

search technique as <strong>the</strong> correlation technique was not tuned for correct type I component<br />

identification.<br />

An example plot for a poor correlation, show<strong>in</strong>g <strong>the</strong> block spectrogram, result<strong>in</strong>g correlation<br />

vectors, and <strong>the</strong> component presence is shown on Figure 38. In this example <strong>the</strong> lead<strong>in</strong>g portion<br />

of <strong>the</strong> type I component has been falsely recognised as a type III component. This was only a<br />

problem for high SNR type I components. The type I recognition was not optimised and has<br />

falsely recognised a second type I component.<br />

The technique was run across 4827 records of <strong>the</strong> bluey logger and cross checked aga<strong>in</strong>st 500<br />

records which were manually counted for <strong>the</strong> number of callers. The result<strong>in</strong>g errors of <strong>the</strong><br />

number of callers manually counted m<strong>in</strong>us <strong>the</strong> number of callers automatically recognised is<br />

shown by <strong>the</strong> distribution plot of Figure 39. The correct number of callers accounted for 35% of<br />

<strong>the</strong> sample, with <strong>the</strong> majority of <strong>the</strong> rema<strong>in</strong><strong>in</strong>g blocks under-sampled. Reasons for <strong>the</strong><br />

discrepancies were:<br />

• <strong>the</strong> lead<strong>in</strong>g portion of <strong>the</strong> type I component be<strong>in</strong>g mistaken for a type II component - only an<br />

issue for <strong>the</strong> few high SNR type I components;<br />

• very low component level or SNR - <strong>the</strong> most common error but not considered important <strong>in</strong><br />

<strong>the</strong> acoustic census<strong>in</strong>g technique described below, s<strong>in</strong>ce detection was limited on received<br />

level to those calls likely to be with<strong>in</strong> <strong>the</strong> <strong>Rottnest</strong> <strong>trench</strong> region only, thus low SNR<br />

components were rejected from count<strong>in</strong>g anyway;<br />

• occasional mistake of type II for a particular sequence of close 20 Hz 'click<strong>in</strong>g' signal<br />

(section 3.5) - although this was rare and <strong>the</strong> algorithm was generally robust <strong>in</strong> reject<strong>in</strong>g <strong>the</strong><br />

click<strong>in</strong>g signals;<br />

• propagation effects caus<strong>in</strong>g pronounced variability <strong>in</strong> <strong>the</strong> received level throughout a<br />

component (ie. Figure 13) - mistakes here were worst for components with high lead<strong>in</strong>g and<br />

end sections and low middle sections;<br />

• a particular comb<strong>in</strong>ation of vessel tones be<strong>in</strong>g mistaken for a component - this was rare as it<br />

required <strong>the</strong> right mix of tones;<br />

• propagation effects caus<strong>in</strong>g variability <strong>in</strong> <strong>the</strong> received level of <strong>the</strong> different spectral<br />

components - <strong>the</strong>re were some blocks where components were rejected as <strong>the</strong> harmonics<br />

were lost due to different propagation of <strong>the</strong> different frequencies.<br />

It is believed <strong>the</strong> correlation technique can be improved considerably by several adjustments,<br />

particularly by differences <strong>in</strong> <strong>the</strong> normalisation methods and vary<strong>in</strong>g <strong>the</strong> relative <strong>in</strong>tensity of<br />

component harmonics, so that <strong>the</strong>y play a greater role <strong>in</strong> <strong>the</strong> <strong>in</strong>itial correlation.<br />

36

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!