OPTIMAL SAMPLING AND TRADITIONAL VERSUS MODEL-BASED DATA-ANALYSIS IN ACOUSTIC FISH STOCK ASSESSMENT IN LAKE-VESIJARVI

Optimal sampling strategy and data analysis methods were studied for p elagic fish stock assessment in eutrophic Lake Vesijarvi in southern F inland based on 10 surveys conducted in 1993. The systematic sampling scheme was applied with parallel transects. The precision of the fish density estimates in unstratified and post-stratified sampling were co mpared with each other and with the precision that could be obtained b y an optimum effort allocation. Because the frequency distribution of fish density in the primary sampling units is typically skewed, a mode l-based estimation was also applied in which the distribution is appro ximated with a known distribution. The best normalizing transformation was found with Box's and Cox's test. These model-based estimates were compared with the estimates derived from non-transformed data. Post-s tratified sampling by depth areas produced more precise estimates than unstratified sampling (reduction in the variances about 40%), reduced the bias caused by spatial autocorrelation and eliminated the bias ca used by unproportional sampling of depth areas. The most precise estim ates (reduction in the variances compared with post-stratification abo ut 50%) would be given by optimum allocation, which suggests that more effort should be allocated to the deep areas. The optimal fraction of sample size to deep areas increased in the course of the summer (abou t 20% in June and over 50% in August). The model-based estimation prod uced a slight gain in precision with the post-stratified data. But, su rprisingly, with unstratified data and with optimum allocation more pr ecise estimates were given by initial untransformed data. In conclusio n, we do not recommend the model-based estimation (including logarithm ic transformation) in lakes if a suitable survey time is restricted by fish behaviour (diel migrations, schooling behaviour), because of dif ficulties in obtaining adequate sample size to determine appropriate t ransformation.