Geostatistical assessment of sampling designs for Portuguese bottom trawl surveys

New sampling designs for the Autumn Portuguese bottom trawl survey (ptBTS) were investigated to explore alternative spatial configurations and possible increments on sample size. The currently used stratified random design and five proposals of systematic based designs were assessed by a simulation study, adopting a geostatistical approach based on likelihood methods of inference. The construction of the designs was based on “informal” method to reflect the practical constraints of bottom trawl surveys. The proposed designs were a regular design with 28 locations (S28), two regular designs with extra regular added locations with 44 (S44) and 47 (S47) locations, a design which overlaps the regular and stratified random design currently used with 45 locations (S45) and an high density regular design with 108 locations (S108), used just as a benchmark. The designs were assessed by computing bias, relative bias, mean square error and coverages of confidence intervals. Additionally a variance ratio statistic between each study designs and a corresponding random design with the same sample size was computed to separate out the effects of different sample sizes and spatial configurations. The best performance design was S45 with lower variance, higher coverage for confidence intervals and lower variance ratio. This result can be explained by the fact that this design combines good parameter estimation properties of the random designs with good prediction properties of regular designs. In general coverages of confidence intervals were lower than the nominal 95% level reflecting an underestimation of variance. Another interesting fact was the lower coverages of confidence intervals computed by sampling statistics for the random designs, for increasing spatial correlation and sample size. This result illustrates that in the presence of spatial correlation, sampling statistics will underestimate variances according to the combined effect of spatial correlation and sampling density.