Some comments on the procedure for testing estimators of krill abundance which utilise survey data
Since krill distributional data do show evidence of spatial correlation, estimators of abundance which attempt to model such effects (such as those based on Kriging techniques) may provide improved abundance estimates from survey data. However, computer simulation studies are first required to test whether such estimators, and alternative possible survey designs, are indeed likely to provide improved performance in practice. Such studies require a simple method for computer generation of krill distribution patterns, which are compatible with existing information on the distributions from surveys. ‘Two-level’ models of krill distribution are considered. These achieve overall spatial correlation by placing krill swarms at random within larger aggregation features termed concentrations; these concentrations are then located at random within the survey area. These ‘two-level’ models provide an encouraging improvement in fits to the distribution of inter-swarm distances observed on the 1981 FIBEX survey by MV SA Agulhas. However, evidence of model misspecification remains. Further work is needed before such models can be used as the basis for the simulation studies required - some suggestions are made in this regard.