Using large databases of salmon time series (NWFSC) and marine fish productivity (RAM Legacy), Eric Ward, Eli holmes and Jim Thorson (FRAM, NWFSC) are evaluating the forecast performance of traditional ARIMA time-series models versus non-parametric regression (such as generalized additive models) and non-linear projection models (such as SMAP models) for short-term forecasts using only time-series data. Short-term forecasts are used in many management decisions, for example in harvest decisions, and often data are unable to develop mechanistic models. Time-series models may provide an alternative in such situations. Comparisons of time-series model forecasts to more traditional stock assessments forecasts are in progress.
Ward, E.J., Holmes, E.E., Thorson, J.T., and B. Collen. 2013. Comparison of parametric and non-parametric methods for short-term population forecasting. Submitted.