Inseason Forecasting of Southeastern Alaska Pink Salmon Abundance Based on Sex Ratios and Commercial Catch and Effort Data
Abstract.—We developed a sex ratio index that, together with cumulative catch of all gears or cumulative catch per unit effort of the seine fishery, estimated abundance and catch of pink salmon Oncorhynchus gorbuscha in southern Southeast Alaska during a fishing season. We evaluated three inseason f...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Text |
| Language: | English |
| Subjects: | |
| Online Access: | http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.613.5573 http://www.webpages.uidaho.edu/wlf448/2008/Documents/Inseason Forecasting of Pink Salmon Abundance Based on Sex Ratios.pdf |
| Summary: | Abstract.—We developed a sex ratio index that, together with cumulative catch of all gears or cumulative catch per unit effort of the seine fishery, estimated abundance and catch of pink salmon Oncorhynchus gorbuscha in southern Southeast Alaska during a fishing season. We evaluated three inseason forecast models—linear, nonlinear, and combined—using data from 1983 to 1997. Based on a cross-validation evaluation of forecast accuracy, the nonlinear model generally outperformed the linear and combined models. Cumulative catch per unit effort was a better predictor than cumulative catch in the first 3 weeks (statistical weeks 28–30) of a fishing season, but the relation was reversed for the remaining 5 weeks. Inseason abundance estimations greatly outperformed the preseason forecasts. Incorporating sex ratios into inseason forecast models correctly adjusted the run timings during a large majority of years and thus improved overall forecasts starting in the second week. In the second through fifth weeks (weeks 29–32), the best performing model with sex ratios improved forecasts more than 30 % over the best model without sex ratios; im-provements included averages of absolute percentages of relative forecast errors, absolute devi-ations, and squared residuals. Average absolute percentages of relative forecast errors from the best model were less than 20 % before the midpoint of the run and less than 14 % at and after the |
|---|