I fit models of run size by river as a function of various lags of run size, escapement, and run size by age class using nonparametric regression, specifically the separable length-scale Gaussian process (GP) model as implemented in `GPEDM` (Munch and Rogers 2022). For the Fraser and Bristol Bay river systems, I used a hierarchical model structure. From a non-exhaustive search, I selected the best model for each river based on leave-future-out mean absolute percent error over the years 2012-2023.

See attached word document for description. Models are attached as R scripts. I've included the Bristol Bay historical forecasts.