amoeba / 2019-yukon-forecasting Goto Github PK

Right now, refitting is subjective. Ideally, we'd have something pretty objective. This would define when we refit (sum of squares > some value; ratio of positive to negative residuals) and provide an objective algorithm for picking a new mu and s.

Per Mundy:

then we need to develop forecasting approach that starts with preseason "prior" based on linear model relations, then proceeds to incorporate wind driven destabilization of temp-salinity front of river plume using wind velocity or proxies of drivers of mixing of river plume with sea water.

We have talked about how forecast uncertainty goes up in warm years, but this isn't quantified in a useful way. Look at how errors correlate with AMATC, MSSTC, and PICE to see if there's really a relationship and possibly if there's something useful there.