Weighting and smoothing of stomach content data as input for MSVPA with particular reference to the Barents Sea

Abstract

Multispecies Virtual Population Analysis (MSVPA) is based on parameterization of the average relative food compositions for all possible prey-age predator-age combinations in the model by year and quarter. This sets high demands on stomach sampling programmes in terms of spatial coverage of the predator population and of sampling intensity for individual cohorts. In practice, there are many sources of error in the input data and large variances, which call for a smoothing procedure to avoid outliers in the MSVPA output. We investigate the potential of geostatistics (specifically kriging) in improving (1) estimates of total and partial stomach content weights from spatially non-uniformly distributed samples and (2) smoothing of the average stomach content weights over the two-dimensional input array of predator age and years. The examples shown indicate that kriging provides an efficient method to deal with geographical variability in food composition and predator abundance as well as to fill gaps and make extrapolations within a two-dimensional array in temporal space characterized by many empty cells or cells not sufficiently well sampled.