SOLV-Code
commented
9 months ago The Chasco et al (2021) paper looks at this explicitly, using multiple correlated environmental covariates to identify cumulative effects and interactions. They found that all four seasonal PDO covariates had an effect on wild smolt-to-adult survival of Snake River Chinook, but the summer PDO covariate was the strongest. None of the PDO versions showed up as a key driver for the hatchery-origin fish.
What might explain the observation that summer PDO affected wild Chinook but not hatchery Chinook? The discussion section of the paper covers potential explanations.
PDO covariates describe ocean conditions, but are mapped to salmon data for a time period where the fish are not in the ocean
For the DFO Fraser Sockeye forecast, the winter PDO covariate is matched to the winter preceding ocean entry. I think this means it is using a large-scale ocean condition indicator as an indirect covariate linked to freshwater conditions in winter and spring (PDO linked to snowpack?, timing of freshet?, air temperature?). If that is the case, why not use these freshwater indicators directly as covariates? Is it because the large-scale PDO covariate encompasses a whole suite of interconnected mechanisms and a cumulative effect on salmon survival, whereas the relationship between survival and any individual localized variable is very noisy and can't be used for modelling?
For the NOAA Ocean Conditions Index, the winter PDO covariate and summer PDO covariate and summer SST are assigned to the same calendar year in the overall aggregate index. If the overall index is then matched to ocean entry year for a specific stock, the interpretation of the winter PDO matches up with how it is used in the Fraser Sockeye Forecast work.
This also means that the Summer PDO covariate is assumed to interact with salmon survival through a very different mechanism, because it reflects ocean conditions during their initial marine migration heading north.
If the assumed mechanism for Winter and Summer PDO is different, is the assumed direction of the effect still the same?
If the direction of the effect is the same, why include both in the aggregate index, given that the signal is the same for most years (see comparison plot)