Report of the Working Group on methods of fish stock assessment (WGMG)

The purpose of the Working Group on Methods of Fish Stock Assessments (WGMG) is to develop and critically evaluate the models and software code used in assessments, forecasts and management simulations, and to suggest ways in which these might be improved. WGMG meets to address particular concerns raised by ACFM and the Resource Management Committee of ICES. The issues covered by each meeting are a function both of the Terms of Reference, and of the interests and expertise of the participants.
The 2007 meeting of WGMG was held at the Northeast Fisheries Science Center (NEFSC), NOAA, Woods Hole, USA. The principal reason for this was to draw on existing expertise at NEFSC on detecting and accounting for retrospective bias in fish stock assessments. The ToRs for the meeting were extremely wide, and covered many problems currently encountered in fisheries assessment and management science. With the time available WGMG could not address all the ToRs, so following an opening series of presentations of previous and current work, the group was divided into three subgroups to work on more focussed issues.
Subgroup A looked at methods for running management strategy evaluations (MSEs), and started designing simulations to assess how management advice might be affected by errors in assessments (in particular, retrospective bias). Subgroup B investigated ways in which the uncertainty in outputs from assessment models could be estimated. As a starting point, this was done by comparing Bayesian and bootstrap estimates of uncertainty arising from a comparatively simple surplus production model. Subgroup C looked further into the problem of retrospective assessment bias; that is, where each successive annual assessment substantially alters the perception of historical stock in a systematic way (either consistently increasing or decreasing it).  
Subgroup A reached three main conclusions. Firstly, WGMG is not yet in a position to answer the questions of whether and how management should proceed in the presence of retrospective bias. The presence of such bias should lead to more cautious management, but how to implement this and how cautious such management should be is less clear. This is due principally to the complexity of programming management-strategy evaluations, but answers to these questions are certainly feasible using current approaches. Secondly, any management-strategy evaluation toolbox must allow for assessments to be run “live” as part of the evaluation loop. And thirdly, managers will get management decisions wrong if these are based on biased advice. This last point may seem obvious, but the analyses presented by Subgroup A highlights the issue with great clarity.
Assessments will always have problems of one sort or another, and it is important that MSEs are able to accommodate this fact. The function of WGMG in this regard is then to provide methods to do this. This endeavour therefore links the work of all three Subgroups.
Subgroup B provided important advances in the implementation of MCMC algorithms for model fitting, and went a considerable distance in generating comparisons of uncertainty estimates from bootstrap and Bayesian methods, with observation and/or process error, using a number of different datasets with different problems (one-way trips, under-reporting and changes in survey catchability). They were able to explore the varying reactions of models to these situations, but firm conclusions remained elusive due to considerable problems in software coding. The Section should be viewed as a strong advance in a work-in-progress. Nonetheless, it seems that not accounting for process errors can lead to a biased view of the true uncertainty in stock estimates based on approximate populations’ models. Reliable methods that account for process and measurement errors simultaneously in stock assessment models are not yet available.