A problem with salmon forecasts is that they rely on recent years' data. For some reason, the managers love to throw out older data...

Anyway, we know that ocean conditions and subsequent returns are cyclic. Which means you always play catch-up. When the conditions improve, you are using the down years to forecast and vice-versa.

Lastly, and I may have beaten this horse before, I have seen a beautiful model that well represents Oregon coho returns. Uses four ocean variables that are temporally sequential. The problem is that any one can switch the impact. Three bads followed by a good give you a good run; three goods followed by a bad and it sucks. Their conclusion was that management by actual in-season update works best, but that precludes raping and pillaging the ocean.