Does Spatial Autocorrelation Need to be Incorporated When Describing the Diets of Fish?

The contribution of prey to predator diets is often quantified using frequency of occurrence and percent weight.  The original forms of these estimators only adjust for non-independence of prey items within a single stomach.  Fisheries data are inherently spatially dependent with autocorrelation being found among nearby sites and diminishing as sites become further apart.  Treating stomachs collected from a single site as a cluster helps adjust for intra- sample correlation (i.e., fish caught in a net together are feeding on similar prey), but not inter-sample spatial autocorrelation (i.e., fish caught in nearby stations are feeding on more similar prey than distant stations).  The latter may be more important in estuarine surveys where sampling sites are in close proximity.  We introduce a spatially-adjusted estimator that redefines a cluster as all fish collected from spatially-dependent nets.  A Moran’s I is used to determine which nets are spatially-dependent.  Simulated datasets reflecting the observed level of spatial correlation from multiple predators collected in Pamlico Sound, NC are compared using traditional cluster-based estimators and spatially-adjusted estimators.  The estimator with the smallest values for bias, mean absolute deviation, and mean squared error will be considered the most robust and recommended for surveys where spatial autocorrelation exists.