Estimating uncertainty from multiple sources in acoustic-trawl surveys

Acoustic-trawl (AT) surveys are widely used to map and quantify pelagic animals and help manage several of the planet’s largest fisheries. However, quantifying their uncertainty is challenging. AT surveys sample patchy, non-Gaussian animal densities using systematic, random, and targeted samples at different spatial resolutions, so classical geostatistics cannot provide confidence intervals. Additionally, converting backscatter to biomass requires multiple scaling conversions, each contributing uncertainty. We present a semi-parametric resampling approach to overcome these challenges. This novel method simulates non-Gaussian spatial fields conditioned on observations to account for spatial sampling error, and applies a spatially weighted bootstrap to account for the uncertainty in allocation of backscatter among species and sizes based on trawl samples. We applied it to surveys of walleye pollock ( Gadus chalcogrammus) in the eastern Bering Sea, estimating errors of 4%–13% for biomass and 4%–34% for numerical abundance. These were higher than errors from traditional geostatistical methods, but lower than currently assumed in the stock assessment. Spatial sampling was the largest individual source of uncertainty in this survey, followed by acoustic target strength and echosounder calibration. We discuss the advantages and disadvantages of this approach, and possible future directions for uncertainty quantification in acoustic surveys.