Improvement of Greater Argentine Stock Assessment

An ongoing commercial market sampling for the Dutch greater argentine fishery is carried out by WMR in collaboration with the pelagic fishing fleet of the Netherlands. Start of this sampling programme goes back to year 1990. The aim of this programme is to achieve a haul-level sampling each year, stratified in area and time within the targeted ICES division. Later, market sampling data is pooled with respect to the quarter and the ICES divisions and the inferred age structure to is allocated to the catch, a classical ALK is generated from the collected age-length data which describes the probability of age given length (Fridriksson 1934). The sample sizes in this long time series can be relatively low in some years and the coverage can be relatively limited in space and time. The additional data sets, collected outside of this program, such as length composition data collected by self-sampling of PFA, maybe useful for filling some of these gaps and improve the level of accuracy of the estimations. However, this approach requires - A detailed understanding of the characteristics of the data set and the patterns in the market sampling time series. - A careful comparison between the datasets while assessing the quality and consistency of the measurements - Adoption of an unbiased method to incorporate age structure information from one dataset to the other. In the first section of the report, the characteristics of the two datasets are explained such as the sampling strategy, sample processing, sample size as well as the demographic changes over the years. Secondly the results of the comparison between the two datasets are presented. The initial aim was to focus on the samples from the two different sampling programs overlapping in space and time. However the level of overlap was not as close as expected. Nevertheless the closest samples in space and time selected and compared. It was found that the resemblance between the datasets did not show a pattern that can be explained with the distance or time between the samples. This is most likely due to stochasticity of the length distributions. A secondary problem that was tackled with was the lack of age information in the PFA data sets. Although length distributions can be an indirect indicator of the age structure a cohort estimation can not be acquired from the length distribution due to slow growth and long life span of the species. One solution would be inferring such missing ages using the available age/length information. Currently, classical age/length keys are commonly used to extrapolate otolith readings to length composition. Unfortunately the classical ALK obtained with data from one year/area cannot be applied to another year/area, because the underlying structure of the population may change due to variation in recruitment, survival or growth. An alternative is the use of an inverse key, which describes the probability of length given age, and has the benefit that this probability is not affected by variation in recruitment and survival. An inverse key can be applied to other years or areas, but under the assumption that size at age does not vary between years or areas. Another alternative is the combined forward-inverse key (hereafter called ‘combined key’) which combines the forward and inverse key into one approach (Hoenig et al. 2002). It uses Bayes’ rule to implement the forward and inverse key into one maximum likelihood framework through the probability of length at age. In this work, we applied the combined key using an R package implementation ( ‘ALKr’) by Murta et al. (2016). For this exercise, we used market age/length samples to build a combined key for greater greater argentine and compared it with the results obtained by the classical ALK. We further used the constructed combined key to estimate the age composition for the catch of samples by the PFA based on their length measurements. The resulting combined key is a promising solution for estimating the age composition in the PFA self-sampling data set, from which the length composition could be used as an input. However before that it is necessary to ensure that the this length composition time series is free from systematic biases with reasonable accuracy. Therefore a comprehensive check was carried out for the PFA length frequency data while using the market sampling dataset as a reference. The results are presented in this report. Lastly stock assessment tests were performed for the greater argentine by with and without incorporation of the PFA data, and the results are discussed.