Experiences in Implementing Temporally Varying Parameters in Real Assessments

Over the past 20 years I have helped develop a range of age-structured  stock assessments in the Laurentian Great Lakes, including for multiple stocks of Chinook Salmon, Lake Trout, Lake Whitefish, Walleye, and Yellow Perch.  During this time period the lakes have experienced large environmental and food web changes.  These changes have led to alterations in parameters such as selectivity, catchability, and natural mortality.  Given the need to allow for temporally varying parameters I participated in simulation studies that evaluated approaches for allowing for such variation, and the results suggested that random walks could capture such changes, with little cost when such changes were not occurring.  For real stocks, however, multiple parameters are potentially changing simultaneously.  When this is the case modeling multiple parameters by random walks can lead to problematic results, because temporal changes may be captured by the wrong parameter.  This can sometimes be identified via retrospective patterns.  As an alternative we have worked to capture temporal changes via functions of predictors, rather than as stochastic processes.  These issues and thoughts for the next generation of stock assessments will be illustrated with examples from the assessments.