An estimate of error for the CCAMLR 2000 survey estimate of krill biomass

Combined sampling and measurement error is estimated for the CCAMLR 2000 acoustic estimate of krill abundance in the Scotia Sea. First, some potential sources of uncertainty in generic echo-integration surveys are reviewed. Then, specific to the CCAMLR 2000 survey, some of the primary sources of measurement error is explored. The error in system calibration is evaluated in relation to the effects of variations in water temperature and salinity on sound speed, sound absorption, and acoustic-beam characteristics. Variation in krill target strength is estimated using a distorted-wave Born approximation model fitted with measured distributions of animal lengths and orientations. The variable effectiveness of two-frequency species classification methods is also investigated using the same scattering model. Most of these components of measurement uncertainty are frequency-dependent and covariant. Ultimately, the total random error in the CCAMLR 2000 acoustic estimate of krill abundance is estimated from a Monte Carlo simulation which assumes independent estimates of krill biomass are derived from acoustic backscatter measurements at three frequencies (38, 120, and 200 kHz). The overall coefficient of variation (10.2less than or equal toCVless than or equal to11.6%; 95% CI) is not significantly different from the sampling variance alone (CV = 11.4%). That is, the measurement variance is negligible relative to the sampling variance due to the large number of measurements averaged to derive the ultimate biomass estimate. Some potential sources of bias (e.g., stemming from uncertainties in the target strength model, the krill length-to-weight model, the species classification method, bubble attenuation, signal thresholding, and survey area definition) may be more appreciable components of measurement uncertainty. (C) 2004 Published by Elsevier Ltd.