How ADF&G forecasts salmon runs

ADF&G builds its preseason salmon forecasts from the return records of past runs, not from anything happening in the water this spring. The two workhorses are sibling models, which relate this year's likely return to how earlier age classes from the same spawning years came back, and spawner-return relationships, which tie a run to the number of fish left to spawn several years earlier. The forecast is a projection off history, and it is the benchmark the season is then measured against.

Salmon come back by age, and that is the key

A salmon run is a mix of age classes. Sockeye from one spawning year, one brood year, return across several later years, some at age four, more at age five, some at age six. Because siblings from the same brood come home in different years, the fish that returned last year carry information about the fish due this year. If the younger siblings came back strong, the older siblings behind them often follow. That relationship, read across many brood years, is the sibling model.

Sibling models and spawner-return curves

In recent years ADF&G and its partners have leaned on dynamic linear models, a form of sibling model that lets the relationship between younger and older age classes shift over time rather than holding it fixed. The other leg is the spawner-return curve, sometimes called a stock-recruit relationship: the number of spawners a run put in the gravel years ago sets a rough expectation for how many adults it should produce now. Bristol Bay's sockeye forecast, the one the whole industry watches, is produced by ADF&G together with University of Washington researchers using methods in this family.

Forecasts miss, and ADF&G says so

A forecast is an estimate, and salmon are hard to estimate. ADF&G is candid about the track record: over the past two decades its Bristol Bay sockeye forecasts have run about 14 percent below the actual return on average, with individual years off by much more in both directions. That is not a knock on the method, it is the nature of projecting a wild run years ahead. It is also the reason a forecast is a starting line and not a promise.

Forecast versus actual: the season's scoreboard

The table above is ADF&G's April forecast by area and species, the denominator for everything on this site. The scoreboard divides the live Blue Sheet harvest by these numbers to get a percent of forecast, a pace read on how far each area has come against what was projected. Early in the run the percent is low because the fisheries have barely opened. By the end it shows how the actual harvest stacked up against the spring projection. The forecast never moves through the season, which is exactly what makes it a fair yardstick.

These are ADF&G's forecasts, reported here as published. The Run does not adjust them, second-guess them or issue forecasts of its own. For how the harvest side of that comparison is counted, see how to read the Blue Sheet, and for the goals managers protect while the run comes in, see escapement in plain English.