Net-based verification of acoustic techniques used to identify Antarctic krill

Acoustic target classification protocols recently adopted by utilising the Stochastic Distorted Wave Approximation (SDWBA) krill target strength model, are assessed. Three frequency acoustic data and concurrent net information, from 16 net hauls through krill aggregations, were collected during two cruises to the South Georgia region in 1996. For each net-sampled aggregation the differences between acoustic backscatter at and 38 kHz (SV120-3B) and at 200 and 120 kHz (Sv200-120) were calculated. Using the SDWBA model, dB difference identification ranges based on the length-frequencies of krill in each aggregation were calculated Four acoustic target classification algorithms were assessed: (i) '3 freq model' -using SDWBA-derived ranges for Sv120-38 and Sv200-120, (ii) '2 freq model' using an SDWBA-derived range for Sv120-3 , (iii) '2 freq 2-16' - with an Sv120-38 range of 2-16 dB and (iv) '2 freq 2-12' -with an Sv120-38 range of 2-12 dB The '3 freq model' algorithm parameterised using the standard CCAMLR values for orientation distribution correctly identified only 6 of the 16 krill aggregations sampled the nets. contrast the '2 freq model' and freq 2-16' algorithms correctly identified all the aggregations as krill. However, different krill orientation distributions for the two cruises were estimated with the SDWBA model, with cruise specific orientation distributions applied the performance of the '3 freq model' was comparable with the two frequency techniques. Such results indicate that use of the present SDWBA-derived ranges in the CCAMLR 3 frequency target classification protocol without taking account of krill orientation distribution is likely to substantially underestimate krill biomass