Elsevier, Fisheries Research, 2(76), p. 252-265
DOI: 10.1016/j.fishres.2005.06.014
Full text: Download
Geostatistical methods have been applied to the problem of accurately mapping animal densities derived from trawl surveys. Sample data are often sparse, highly skewed in distribution and quite unlike the examples used to investigate the adequacy of the methodological options available. We analysed the data from a trawl survey of the portunid crab Liocarcinus depurator using two approaches: (a) removal of outliers and (b) logarithmic transformation of the densities. Within each approach we compared a range of options for both the estimation of the underlying spatial structure (variogram) and modelling of crab density through kriging.The results indicated that log-transformation produced the least robust and most unrealistic assessment of L. depurator spatial distribution. Removing outliers gave consistent estimates, regardless of small changes in methodology except when inappropriate spatial models were applied (exponential and Gaussian models did not fit the variogram well). Differences in the number of lags used to build the variogram or the number of outliers removed from the data had more effect on the spatial model parameters than did most of the procedural alterations.Density estimates from kriging highlighted the difference between the two approaches. For example, estimates of the coefficient of variation were most unrealistic from the log-transformation approach but were roughly half that of the original sample when the spatial model was fitted with data without outliers.In general, the failure of the methods to reflect the original sample were related to the assumptions underlying the methods. Thus, the log-transformation approach produces a peculiar distribution of densities, with zero densities creating considerable departure from log-normality. The resulting parameter and density estimates were thus erratic and unrealistic. Removal of outliers helped uncover the spatial structure in the crab population and led to very realistic parameter and density estimates. However, the lack of symmetry in the distribution led to unrealistic (negative) minimum density estimates when kriging forced a symmetrical distribution on the data.L. depurator populations along the Spanish coast showed high spatial dependency with densities aggregated in patches. Patch sizes were estimated to have diameter of around 20 km. Density decreases with depth and this was adequately represented by the ‘removal of outlier approach’, using depth as a covariate.