Using Meteorological and Oceanographic Data to Forecast Fishery Performance: Research into Methods of Operational Forecasting of Salmon Migration Timing

Alaska's salmon fishery comprises about 80% of the total wild-caught North American salmon harvest. An important component of this harvest is the Yukon River Chinook (king) salmon fishery. Salmon are anadromous— they are born in fresh water, migrate downstream to the ocean where they mature, and three to eight years later return to their freshwater location of birth to reproduce.

The 3200 km-long Yukon River drainage of interior Alaska and Canada is roughly 25% larger than the state of Texas, and Chinook salmon spawn along nearly the entire length. This fish stock is exploited extensively by commercial and sport fishermen. Furthermore, it is a vital subsistence food source for many indigenous people in the Yukon River drainage. Treaty obligations also exist between the US and Canada to ensure adequate Chinook salmon returns to Canadian spawning grounds. Fishery managers must balance the requirement for spawners against the need to permit the harvests that provide for sustainable Yukon River Chinook salmon fisheries. Thus, there exists a compelling need to provide fishery managers both with current and historical fish escapement observations and with physically-based objective models to forecast escapement timing.

Salmon fishery managers base regulations for lower Yukon River Chinook salmon on the history of the timing of adult Chinook salmon migrating into the river. Salmon transit fishing areas in the lower Yukon River shortly after leaving the northern Bering Sea in route to freshwater spawning grounds. Catch records for the fishing areas have been used to develop quantitative measures of migratory timing, or phenology, over a fifty-two year time period, 1961—2012. Although the migration enters the river during a predictable window of time between the end of May and the end of July, in any given year, migration percentiles can occur on any date within a window of twenty days. Given the natural variability in timing, late migrations may be mistaken for very small migrations requiring stringent control of fishing effort.  Conversely, early migrations of small size may be mistaken for large migrations requiring little control of fishing effort. Both kinds of mistakes can result in damage either to fish populations or to participants in the fishery. This is the primary source of uncertainty in the fishery regulatory process and narrowing this uncertainty is the goal of our physically-based modeling efforts.

Before returning salmon can leave the Bering Sea and enter the Yukon River delta, they must become "osmotically competent", that is, they must undergo a physiological adaptation that allows them to survive in freshwater. The exact circumstances of the marine environment in which pre-spawning salmon gain this competency have been measured only infrequently, however the physiological mechanism of the process is well enough understood at the molecular level to permit hypothesis development. Our working hypothesis is that both the ability to enter freshwater and the speed of the average competent individual on entering freshwater are dependent on the nature of the horizontal salinity gradient in the sea near the Yukon River delta.  The seaward geographic extent of the gradient, and the rate of change in salinity within the gradient depend on mixing of the freshwater discharge into the surrounding seawater. It is in this zone of brackish water that the physiological process leading to competency must be initiated, and it is where the salmon are hypothesized to encounter gradients in temperature and the chemical signatures of the river that guide the fish into the river. Several geophysical factors– both atmospheric and oceanic– interact to drive the water mixing process and the varying physical nature of the mixing zone each year, making them likely influences in determining the yearly timing of returns. For example, it is well established that annual timing of Chinook salmon in the lower Yukon River is significantly correlated with spring meteorological conditions such as April surface atmospheric temperatures in Nome, AK (PAOM).

Oceanographic measurements are difficult to acquire in this challenging environment, but a robust, if sparse, climate record exists for the Bering Sea region. Using climate data for the Northern Bering Sea– observed and simulated surface temperatures, sea surface temperatures (SSTs), wind speed and direction, and fractional sea-ice coverage – we have created linear and nonlinear statistical models to predict yearly timing of key percentiles of the migration.

Currently an operational return forecast is made at the end of May, implementing a linear statistical model using monthly mean surface temperatures at PAOM and NCEP reanalysis for a region near the Yukon River Delta. In application, the 2010–2012 operational forecast has narrowed the possible percentiles of the migration to dates within a window of about three days, about 17 days narrower than the long-term average. Verifying the model in hindcasting mode for the period 1978—2011, the predicted window is cut to about seven days.

A newer, more comprehensive, nonlinear statistical approach has recently been developed, using observations from the village of Emmonak (PAEM) along with the inputs of the linear model mentioned above to estimate the return timing.  This method proceeds in two steps.  The first uses the "Eureqa" genetic programming system developed by Michael Schmidt at Cornell University's  Computational Biology Program.  PAEM values of  March and April vector-averaged (resultant) winds, April mean air temperature; SST, and yearly mean fractional sea ice cover are used to estimate key migration percentiles (measured as days since June 1st) and the maximum catch per unit effort (CPUE) obtained from test fishing. Data from 1978 through 2011 for instance were used to predict the 2012 percentiles and maximum CPUE.  In the second step, the percentiles and maximum CPUE are used to fit a four parameter general logistics model to predict the timing and strength of the migration. This alternative nonlinear modeling approach, coupled with the use of the more proximal PAEM observations, significantly improved the precision of timing forecasts over linear model timing forecasts for the 2012 season. We demonstrate in hindcasts that the nonlinear modeling approach can significantly improve the skill of the forecast. Improved forecasts will both improve the ability of fishery managers to conduct sustainable fisheries.