Report of the Working Group on Recruitment Forecasting in a Variable Environment (WGRFE)
reportposted on 2017-10-26, 00:00 authored by ICESICES
The Working Group on Recruitment Forecasting in a Variable Environment (WGRFE) held its initial meeting in Copenhagen, Denmark, 16–20 June 2014. At this meeting, four projects were defined: i) a study examining the impact of recruitment autocorrelation on forecast performance; ii) developing a statistical framework for modelling multi-stage recruitment functions (Paulik diagrams); iii) a review of literature incorporating envi-ronmental drivers in forecasts, focusing on whether forecasts were improved or not; and iv) developing a framework for performing ensemble forecasts, identifying case studies and illustrating the approach.A second meeting of the WGRFE was held in Seattle, Washington, USA, 22–26 June 2015. At this meeting, results from the first two projects (recruitment autocorrelation, and mul-ti-stage recruitment functions) were presented, and the group planned follow-up anal-yses to be conducted before the meeting in 2016. Work was also assigned to WG members towards the third project (review manuscript), and discussions on forecast en-sembles were initiated, identifying a case study for haddock in the NW Atlantic Ocean.WGRFE held its third meeting at the Joint Research Center in Ispra, Italy, 13–17 June 2016. With active participation from regular WGRFE members and JRC scientists, the WG explored the issues of assessment model averaging and multiple model forecasts, focusing specifically on statistical and software frameworks, updating of model weights based on performance, and approaches to defining the set of candidate models. A com-prehensive case study on assessment modelling was presented. A simple ensemble fore-cast case study was presented, and a more complex case study was defined for further exploration at the next WG meeting. In addition, the WG focused on producing text for sections of a review manuscript of published articles that included environmental drivers in forecasts. Final results were presented on incorporating recruitment autocorrelation in forecasts, which suggest that autocorrelation can be estimated fairly well from assess-ment estimates of recruitment deviations and that including autocorrelation typically improved forecast performance, particularly for cases with informative data. This project resulted in a published manuscript (Johnson et al., 2016). Final simulations were also defined for the multi-stage recruitment/Paulik diagram project. Manuscripts for the re-view paper on forecasting with environmental drivers and the Paulik diagram project are planned for submission in the fall of 2016.