Empirical comparisons of survey designs in acoustic surveys of Gulf of Maine-Georges Bank Atlantic herring

Three different hydroacoustic survey designs (parallel, zigzag, and stratified random) were used to survey Atlantic herring (Clupea harengus) on Georges Bank during 2000 and 2001. The surveys covered the pre-spawning distribution of herring from the northern edge of Georges Bank to Nantucket Shoals, spanning a linear distance of roughly 300 km. The goal was to establish a practical survey design and methods for analysis of population estimates of herring in the northwest Atlantic Ocean. Analyses of acoustic transect data using a classical (design based) approach and a model based (geostatistical) approach were completed, so that the properties of the three survey designs could be compared. Herring were distributed across the entirety of the western side of Georges Bank between the 50–150 m isobaths in all surveys during both years. Mean herring backscatter (sA) was similar for the three surveys during 2000 and 2001; the mean from the parallel design was the highest in both years, due to a few very large observations and also the timing of the surveys. Coefficients of variation for the three surveys ranged from 23% for the zigzag design in 2001 to 44% for the stratified random design in 2000 using design based estimators and between 10% for the parallel design in 2001 to 17% for the stratified random design in 2000 using model-based methods. Herring biomass (geostatistical estimates) analyzed using meta-analysis, ranged between 1.27 and 1.58 million mt during 2000 and between 1.60 and 2.14 million metric tons during 2001. Bootstrap results suggest that herring biomass is generally well estimated by all three designs. The zigzag design provided the lowest average CV over the two years, while the parallel design in 2001 produced the lowest CV for the six surveys. Overall, the systematic designs were robust in capturing the relatively strong spatial signal in the herring distribution data, while the stratified random design performed relatively poorly in this regard in both years.