We used multiview embedding, a form of empirical dynamic modelling. Rather than define mechanistic models, the approach translates time series of data into a path through a multi-dimensional space, whose axes are lagged values of the variables. We modelled each stock separately, and included various relevant environmental covariates. As for last year, we did not have time to refine the details once we had things running.

All code is available at https://github.com/andrew-edwards/sockeyePrize, though is not easily usable by anyone else at the moment and there are various improvements we would make. Most of the covariates were taken from our pacea R package. Attached is the pdf from our Rmarkdown code for the Bonneville stock. Edwards A.M, Rogers L.A., and Holt C.A. (2024). Explaining empirical dynamic modelling using verbal, graphical and mathematical approaches. Ecology and Evolution, 14:e10903, 1-12. https://doi.org/10.1002/ece3.10903 Edwards A.M., Tai T.C., Watson J., Peña M.A., Hilborn A., Hannah C.G., Rooper C.N., Flynn K.L., and Oldford, G.L. (2024). pacea: An R package of Pacific ecosystem information to help facilitate an ecosystem approach to fisheries management. https://github.com/pbs-assess/pacea https://zenodo.org/doi/10.5281/zenodo.13840804 Ye H., and Sugihara, G. (2016). Information leverage in interconnected ecosystems: overcoming the curse of dimensionality. Science, 353:922-925. https://doi.org/10.1126/science.aag0863.