Use of Statistical Models for the Estimation of Abundance from Groundfish Trawl Survey Data

Abstract

Estimates of fish abundance from stratified random trawl surveys are highly variable and a number of estimators from various statistical models have been suggested to provide more precise estimates. However, model-based estimates of the survey finite population mean which are not based on the sample mean, can be biased and nonrobust to deviations from the model. This is demonstrated in particular for estimates based on the Δ-distribution. A criterion known as asymptotic design consistency (ADC) is presented for selecting those models that can provide estimates of the finite population mean which are asymptotically robust to deviations from the model. The concept of a predictive estimate is presented as a means of incorporating models into an estimate of the finite population mean which can provide more information than the sample mean. Predictive estimates use statistical models to relate the abundance measured in the sample to covariates measured over the whole survey area. This paper demonstrates that consistent relationships exist between the catch of age 4 cod (Gadus morhua) in the survey trawl and concurrently measured hydrographic covariates which can be used to construct model-based ADC predictive estimates of the finite population mean.