Acoustic Classification Methods Research Articles

Comparing acoustic classification methods to estimate krill biomass in the Georges Bank region from 1999 to 2012

AbstractKrill are a critical component of northern‐latitude ecosystems, yet are poorly monitored in the Gulf of Maine and Georges Bank regions of the northeast United States continental shelf. We applied five acoustic classification methods (multifrequency single‐beam index [MFSBI], dB‐differencing, multifrequency index [MFI], and two variations of the Z‐score method) to acoustic data collected during surveys conducted annually in the Georges Bank region in order to map krill spatial distribution from 1999 to 2012. Material properties (sound speed and mass density) of krill and length data from net samples collected in the Gulf of Maine and Georges Bank regions were used in conjunction with distorted wave Born approximation (DWBA) predicted target strengths to scale acoustic data to estimates of biomass for the Georges Bank region. All methods require a priori information, ranging from training sets (i.e., empirically based classification) to validated acoustic models of target strength (i.e., theoretically based classification) for the species and acoustic frequencies of interest. Incorporating predicted backscatter from theoretical models can improve confidence in classification, especially when clean training sets are not available, but does come at the cost of a validated acoustic backscatter model. The biomass estimates from the four acoustic classification methods were comparable among each other, with the MFSBI method consistently estimating lowest biomass and the empirically based Z‐score consistently estimating highest biomass. Comparisons to net‐based estimates were inconsistent, most likely due to net inefficiency (low catchability), targeting of layers, and scaling point estimates to regions.