Evaluation of methods for predicting mean weight-at-age: an application in forecasting yield of four haddock (Melanogrammus aeglefinus) stocks in the Northeast Atlantic

Methods for predicting mean weight-at-age for four haddock (Melanogrammus aeglefinus) stocks in the Northeast Atlantic were evaluated. Their performance in short-term forecasts of yield (catch in weight) was tested in a retrospective analysis conducted for a 12-year period (from 1997 to 2008). For each year, the estimated yield was compared with the observed yield, both being conditional on the recorded catch numbers-at-age. Overall, regardless of the prediction method used, the forecasts generated for Rockall and Icelandic haddock were most accurate, and those for North Sea haddock least accurate. Among the examined methods, there was an overwhelming tendency to overestimate yield. No single method appeared to be superior for all the four stocks. Simple linear models for year classes regressing mean weight on age appeared an effective method for the North Sea and West of Scotland stocks. Some methods based on linear models for expected growth, with weight, age, year or year-class effects as predictors, performed well, which reflected the growth patterns specific to each stock. Application of some of the proposed methods considerably reduced the bias and increased the precision of yield forecasts as compared with the methods currently being used, particularly for North Sea and Icelandic haddock.

► Methods for predicting mean weight-at-age for four haddock stocks are evaluated.
► There is an overall tendency to overestimate mean weight-at-age.
► Linear models regressing weight on age are effective methods for two stocks.
► Some models with weight, age, year or year-class effects as predictors perform well.
► More accurate methods of weight-at-age prediction may considerably improve yield forecasts.