Some properties of diagnostics of GLMM model tuning for standardising CPUE indices in the Area 48 using the CCAMLR fishery statistics database

The data of krill fishery in Subarea 48 for 1986 – 2008 presented to CAMLR database by haul-by-haul data were used. Standardizing CPUE was performed by applying the generalized linear model with random effects (GLMM). The analysis of diagnostics of GLMM tuning indicated that the model with Tweedie’s distribution satisfactorily describes the set of data collected in Subareas 48.1, 48.2 and 48.3. However, there were many hauls to be interpreted as “outliers”, i.e. the points from the sample where residuals significantly deviated from the normal distribution. The number of krill hauls, where Pearson’s residuals corresponded to quintiles of normal distribution more then 2, constituted 2.3% of the total number of krill hauls in the CAMLR database for 1986 – 2008. It was shown that extremely high CPUE values are results from converting high catch values obtained for a short time haul (5-10-15 min) into catches per hour. Such high CPUE values constituted 100 – 600t/hour and more (up to 1000 t/hour). The» outliers by different fleets (countries) in the Subareas 48.1-48.3 were analyzed. The procedures of CPUE indices estimation for krill fishery description claim special attention. The authors discussed several approaches to improve diagnostics of GLMM tuning for krill fishery analysis.