In order to predict 2024's escapement of Sockeye salmon on 14 rivers in the Bristol Bay area of Alaska, Fraser River Watershed in Canada, and Columbia River Basin in the Western USA, we used models that incorporated the current year's PDO index, and current year's ENSO index, the sum of the previous year's escapement and recruitment (recruitment totals often unavailable after 2017), and trend of previous two years' escapement [escapement(t-1) - escapement(t-2)]. Model 4 selection was based on AIC. Each river was modeled separately. "Model 4", which incorporated all mentioned factors, was most often selected for Bristol Bay stocks and the Columbia River Basin. "Model 3", which did not incorporate the sum of recruitment and escapement, was the next most chosen model and used for most stocks in the Fraser River. In the two cases where the model would give a negative number, the model with a positive output and lowest AIC was chosen, or if all models were negative, the mean of all escapement data was used.

Each river/stock modeled separately [t=current year]. GLM models run in R with glm() function. - "Model 4" most often used, X of 14 stocks: GLM model of escapement[t] ~ sum(escapement[t-1], recruitment[t-1]) + delta(escapement[t-1]-escape[t-2]) + PDO + ENSO - "Model 3" used for X of 14 stocks: GLM model of escapement[t] ~ delta(escapement[t-1]-escape[t-2]) + PDO + ENSO - "Model 1" used for the one case when other, higher ranked models were negative: GLM model of escapement[t] ~ escapement[t-1] + PDO + ENSO - Mean used in the one case where all model outputs were negative.