The use of adaptive cluster sampling for hydroacoustic surveys

Resource managers are often required to estimate the size of a wildlife population based on sampling surveys. This problem is especially critical in fisheries, where stock-size estimation forms the basis for key policy decisions. This study looks at design-based methods for a hydroacoustic fisheries survey, with the goal of improving estimation when the target stock has a patchy spatial distribution. In particular, we examine the efficiency and feasibility of a relatively new design-based method known as adaptive cluster sampling (ACS). A simulation experiment looks at the relative efficiency of ACS and traditional sampling designs in a hydroacoustic survey setting. Fish densities with known spatial covariance are generated and subjected to repeated sampling. The distributions of the different estimators are compared.Hydroacoustic data frequently display strong serial correlation along transects and so traditional designs based on one-stage cluster sampling are appropriate. Estimates of total stock size for these designs had a markedly skewed distribution. ACS designs performed better than traditional designs for all stocks with small-scale spatial correlation in fish density, yielding estimates with lower variance. ACS estimators were not skewed and had a lower frequency of large errors. For the most variable stock the use of ACS reduced the coefficient of variation (CV) of the stock size estimate from over 0.9 to around 0.5. Differences between traditional and ACS designs were consistent over multiple realizations of each spatial covariance model.A survey of rainbow smelt (Osmerus mordax) in the eastern basin of Lake Erie was used as a case study for development of a survey design. A field trial showed that use of ACS for the survey is feasible but pointed out some areas for further research. The biggest drawback to use of ACS is uncertainty in the final sample size. This can be partially controlled by applying ACS within a stratified design. ACS retains the unbiased and non-parametric properties of design-based estimation but allows increased sampling in high-density areas that are of greater biological interest. For stocks with an aggregated or patchy spatial distribution ACS can provide a more precise estimate of stock size than traditional survey methods.