Why the two main forecasts have a 5 million-fish difference

Jun 28, 2019

Forecasting the salmon run is an evolving science. This year, there is a 5 million-fish gap between the ADF&G and UW sockeye salmon forecasts this year. Here’s why. 

Salmon in the Wood River
Credit Alex Hager/KDLG

This year, 44.6 million sockeye salmon will return to Bristol Bay – if you trust the University of Washington forecast. 

But the University of Washington forecast isn’t the only one out there. 

“This year’s forecast is 40.18 million,” said Gregory Buck, Fish and Game’s fishery biologist and regional research coordinator for Bristol Bay. 

Buck says the almost 5 million sockeye difference between the ADF&G and UW forecasts has a pretty simple explanation: an accumulation of small data interpretation differences. 

Curry Cunningham, a quantitative biologist who works with the UW Alaska salmon forecast, generally agrees. 

“We’re all looking at the same data,” Cunningham said. “We just use slightly different statistical methods to do so.” 

Salmon forecasting basically boils down to a lot of data analysis and number crunching. 

Salmon from the same cohort, also called a brood year, spend different ratios of time in fresh versus ocean water before coming back to spawn. By analyzing returns of younger fish that were born the same year, biologists and ecologists can predict how many fish from older age classes will return in the coming years. 

“The best proxy for freshwater and early marine survival is how well members of the same cohort survived when they went out to sea,” Cunningham elaborated. “We primarily are just forecasting one year ahead, looking at the returns of a younger age class in a younger year or the prior two to three years at most.”

Cunningham said the 5 million fish difference between the ADF&G forecast and his own at the University of Washington probably has something to do with the fact that UW has worked new analysis models into their salmon forecasting. 

In the past, salmon forecasts looked at returns of age classes and environmental conditions as relatively static. But the environment’s effect on fish movement is actually much more variable. 

“In recent years we’ve increasingly relied on what are known as dynamic linear models,” Cunningham explained. “[These models] account for the fact that there are changes in average productivity of these systems over time as well as changes in the relationships between these sibling age classes.”

In addition to changing age class relationships and overall system productivity, environmental changes and ocean temperatures can also make a salmon run devite from its forecast. UW’s dynamic linear models aim to lessen that gap. 

Despite sometimes large yearly differences, both the UW and ADF&G forecasts have similar percent errors. The UW forecasts have an absolute median percent error of 14.6% since about 2000, while ADF&G’s have been between 10 and 15% for the past few years. 

As for whether fishermen trust the forecast, it seems like the prevailing philosophy is that they never know how a season is going to go until it’s going. Every fisherman around the Naknek boatyards last week had their own way of trying to predict how the run will go, from abundance of tundra cotton, to Port Moller numbers, to water temperature, to last year’s run, to absolute refusal to speculate until fish are being delivered to a tender. 

So far, both Buck and Cunningham say the run is shaping up to be right on track with the previous few years. Not early or late, not larger or smaller than forecast, but just on time and just the right size. 

As for whether the 5 million-fish difference will actually show, biologists and fishermen alike will have to wait until the end of this year’s commercial fishing season. 

You can contact the author at sage@kdlg.org or 907-842-2200.