Lorna Taylor,

Hafrannsóknastofnun, Reykjavík - Marine Research Institute, Reykjavik

Parameter estimation issues for a statistical fisheries model

In fisheries management there is increasing interest in ecosystem effects of fishing and ecosystem based management. Population dynamics models are therefore required which are able to consider factors such as multispecies effects, technical interactions and closed areas. The complexity of these models is determined not only by the characteristics of the population but also by the availability of data. Gadget is a statistical modelling framework which can be used to create forward simulation multispecies, multiarea, multifleet models. Parameters can be estimated through the combination of a variety of data types in weighted negative log likelihood components. Statistical models allow testing of the importance and existence of interactions. Use of such models, however, introduces problems such as how to weight the components and which parameters can be estimated given the availability of data. As different weighting schemes can pull the solution in different directions, it is essential that the weights are not ad hoc. Using a simple model of a cod population a method to calculate objective component weights is evaluated. The importance of the initial parameters and the types of data included in the objective function are also discussed. While these examples are specific to gadget and fisheries, the statistical issues are more widely applicable.