Preseason Forecasts of Ocean Escapements of Columbia River Chinook Salmon (Oncorhynchus tshawytscha) Populations
Preseason forecasts for ocean escapements of Columbia River summer and fall Chinook Salmon Oncorhynchus tshawytscha populations are not only scientifically but also politically important. These forecasts affect salmon fishery quotas in the northeast Pacific Ocean. At present, the preseason forecasts are made by the U.S. v. Oregon Technical Advisory Committee (TAC), and are used as an input value for the Pacific Salmon Commission (PSC) and Pacific Fishery Management Council (PFMC) models that calculate fish abundance in the ocean. Annual harvest quotas for the Chinook Salmon ocean fisheries are determined on the basis of these ocean abundance estimates. Although the traditional forecasts by the TAC were not seriously biased, they tended to be under-forecasts, and also uncertainty in the forecasts such as prediction interval is not measured. The main objective of this study was to develop robust forecast models and to express the forecast uncertainty. Fish populations of concern were: Upriver Bright (URB), Bonneville Upriver Bright (BUB), Pool Upriver Bright (PUB), Lower River Wild (LRW), Lower River Hatchery (LRH), Bonneville Pool Hatchery (BPH), and Upper Columbia River summer (UCS). In addition to data on fish sibling runs (age specific runs from the same brood year) that the TAC uses for the traditional forecast, we explored various data including sea surface temperature, upwelling, and coded-wire-tag (CWT) records. The exploration indicated that ocean recovery rate of CWT fish and winter sea surface temperatures (SSTs) in fish’s first ocean year from coast waters of British Columbia, Canada significantly contributed to forecasts. Then, using these data for each population, we built both ordinary regression models and autoregressive error models. To evaluate these models, we made extensive hind-casting forecasts of population runs in 10 years of 1994-2005 with different sample sizes except for PUB and UCS populations. Temporal range of data on PUB and UCS 2 populations was not as long as those on the other populations, and five year (2000-2004) and three year (2002-2004) run forecasts were made respectively. We tested candidate models based on hind-casting forecast performance in terms of relative error, frequency of prediction interval’s coverage, model residuals, and model parsimony. Results indicated that an autoregressive model outperformed the traditional (ordinary regression) model for all populations studied. The ordinary regression model was apparently confounded due to autocorrelation in the model residuals, and routine inclusion of the intercept term in the ordinary regression model did not contribute to forecast in some instances. Inclusion of data on SSTs and CWT ocean recovery rate provided additional improvement in forecasts for some, though not all populations. Future years’ forecasts would improve if instead of the current routine ordinary regression model, population-specific autoregressive models were adopted. The form of each population’s model could also be reexamined on a regular basis as data accumulates over time.
Hyun, S., S. Ellis, and P. Roger. 2006. Preseason forecasts of ocean escapements of Columbia River Chinook Salmon (Oncorhynchus tshawytscha) populations. PSC Southern Fund Project. Columbia River Inter-Tribal Fish Commission Technical Report 6-11. Portland, Oregon. 67p.