Forecasting Pink Salmon Harvest in Southeast Alaska from Juvenile Salmon Abundance and Associated Biophysical Parameters: 2012 Returns and 2013 Forecast
The Southeast Alaska Coastal Monitoring (SECM) project has been sampling juvenile salmon (Oncorhynchusspp.) and associated biophysical parameters in northern Southeast Alaska (SEAK) annually since 1997 to better understand effects of environmental change on salmon production. A pragmatic application of the annual sampling effort is to forecast the abundance of adult salmon returns in subsequent years. Since 2004, peak juvenile pink salmon catch-per-unit-effort (CPUE), adjusted for highly-correlated biophysical parameters, has been used to forecast adult pink salmon harvest (O. gorbuscha) in SEAK. The 2012forecast of 18.8Mfish was 12%lowerthan the actual harvest of 21.3Mfish. Eight of nineforecasts produced over the period 2004-2012have been within 17% of the actual harvest, with an average forecast deviation of 7%. The forecast for 2006 was the exception; while the simple CPUE model indicated a downturn in harvest, the prediction substantiallyoverestimatedthe harvest. These results show that the CPUE information has great utility for forecasting year class strength of SEAK pink salmon, but additional information may be needed to avoid forecast “misses.” For the 2013forecast, model selection included a review of ecosystemindicator variables and considered additional biophysicalparameters to improve the simple single-parameter juvenile CPUE forecast model. The “best” forecast model for 2013 included two parameters, the Icy Strait Temperature Index (ISTI)andjuvenile CPUE. The 2013forecast of 53.8Mfish from this model, using juvenile salmon data collected in 2012,hadan 80% bootstrap confidence interval of 48-60Mfish.
The Southeast Alaska Coastal Monitoring (SECM) project has been sampling juvenile salmon (Oncorhynchusspp.) and associated biophysical parameters in northern Southeast Alaska (SEAK) annually since 1997 to better understand effects of environmental change on salmon production (e.g., Orsi et al. 2011, 2012a, 2013a). A pragmatic application of the information provided by this effort is to forecast the abundance of adult salmon returns in subsequent years. Mortality of juvenile pink (O. gorbuscha) and chum(O. keta) salmon is high and variable during their initial marine residency, and is thought to be a major determinant of year-class strength (Parker 1968; Mortensen et al. 2000; Willette et al. 2001; Wertheimer and Thrower 2007). Sampling juveniles after this period of high initial mortality may therefore provide information that can be used with associated environmental data to more accurately forecast subsequent adult year-class strength.
Pink salmon are a good species to test the utility of indexes of juvenile salmon abundance in marine habitats for forecasting because of their short, two-year life cycle. Sibling recruit models are not available for this species because no leading indicator information exists(i.e., only one age class occursin the fishery). Spawner/recruit models have also performed poorly for predicting pink salmon returns, due to high uncertainty in estimating spawner abundance and high variability in marine survival (Heard 1991; Haeseker et al. 2005). The exponential smoothing model that the Alaska Department of Fish and Game (ADFG) employs using the time series of annual harvests has provided more accurate forecasts of SEAK pink salmon than spawner/recruit analyses(Plotnick and Eggers 2004; Eggers 2006). Wertheimer et al. (2006) documented a highly significant relationship between annual peak juvenile pink salmon catch-per-unit-effort(CPUE)from the SECM research in June or July and the SEAK harvest. TheseCPUE data used asa direct indicator of run strength have been supplemented with associated biophysical data in some years (Wertheimer et al. 2011, 2012), or used as auxiliary data to improve the ADFG exponential smoothing model (Heinl 2012; Piston and Heinl 2013). Recently, efforts have been made to incorporate climate change scenarios into stock assessment models(Hollowed et al. 2011) and to examine relationships of ecosystem metrics to salmon production (Miller et al. 2013; Orsi et al. 2012b, 2013b). The SECM project has developed a 16-yr time series of ecosystem metrics for such applications (Fergusson et al. 2013; Orsi et al. 2012b, 2013b; Sturdevant et al. 2013 a, b). This paper reports on the efficacy of using the SECM time series data for forecasting the 2012SEAK pink salmon harvest and on the development of a prediction model for the 2013forecast.