The Workshop on evaluating survey information on Celtic Sea gadoids (WKESIG) provides ex-pertise for the survey indices that contribute to the Benchmark Workshop on Celtic Sea Stocks (WKCELTIC). Three of the largest stocks assessed by the Working Group for the Celtic Seas Ecoregion (WGCSE), namely Cod 7e_k, Haddock 7b_k and Whiting 7b_k, form part of a signifi-cant mixed demersal fishery in the Celtic Sea. The assessments rely heavily on survey indices, which in all cases use combined survey indices between the Irish Ground Fish Surveys (IE-IGFS) and the French Surveys (EVHOE).
A number of survey data issues were highlighted for potential review by the Working Group on Improving Use of Survey Data for Assessment and Advice (WGISDAA). Key issues include: (1) how survey data that vary across indices are standardized and combined, (2) unavoidable gaps in survey coverage that have occurred in recent years, and (3) estimates of uncertainty that are not routinely calculated for these index calculations.
Having explored variability of survey effort, it was agreed data gaps would be addressed to allow production of indices by swept area for the benchmark. Additional data such as speed, trawl settling time, and sea conditions were discussed as useful measures of observation error. A few modern modelling approaches were also discussed and should be explored further.
Migration patterns should be included as supplementary information in the benchmark, but it was not clear how this information could be integrated into assessments yet in a quantitative way.
Modelling of average length keys (ALK) data highlighted the precision gains achievable when spatial variability can be accounted for. This ongoing work is a progression of the methods cur-rently used for ALK ‘fill-ins’ for current indices, so the working group will continue to follow its development.
Dealing with overall data gaps was addressed by presentation of a case study on Celtic Sea whit-ing index calculation using a vector autoregressive spatio-temporal model (VAST). It was shown that the model had good stability and predictive accuracy even when large amounts of input data were removed from a given year, to simulate vessel breakdown for example. However, the ability to model the missing values was heavily influenced by how ‘average’ a survey year was. In unusually high or low abundance years the capacity to ‘model your way out’ of data gaps is reduced significantly. Overall the modelling approach offered potential to address spatial vari-ability of the data, estimate uncertainty about observation and process error separately and also reduce the impact of unavoidable data gaps.