Mapping, Estimating Biomass, and Optimizing Sampling Programs for Spatially Autocorrelated Data: Case Study of the Northern Shrimp (Pandalus borealis)

Abstract

The methodology for mapping and for global and cutoff estimation of autocorrelated exploitable resources is presented, based on stationary geostatistical methods. Use and performance of these methods in marine ecology are illustrated with an application to northern shrimp (Pandalus borealis) abundance data, collected in 1989 at 137 stations in the western Gulf of St. Lawrence. Nonstationarity of the biomass data, a proportional increase of the local variance with the local mean, and the presence of outliers all violated the stationarity assumption and strongly hindered the modeling of the spatial structure. Cross-validation tests showed that kriging estimates were better when interpolating within very local neighborhoods using a small number of points. Kriging always performed better than polygonal tessellation. A stratification scheme produced better estimations than the whole-region approach using traditional or relative variograms. The spatial organization of the shrimp biomass was composed of a trend superimposed onto mesoscale patches of 30–50 km in diameter. The area under study contained about 22 000 tonnes of northern shrimp; 70% of this biomass was concentrated in less than 30% of its surface. The spatial information is used to derive guidelines for optimizing future sampling programs.