Finalise target

[aaronspring]

I think we already agreed on many specifics of the target, but some are still unclear.

• valid_time: OND
• test years: 2005,2007,2018,2004,2006
• area: Addis, Nairobi Square (exact lon lat box required)
• variable: CHIRPS binary drought/no drought with 33% edge based on all years (which exactly) climatology
• reference: climatology, persistence, linear model (should all be beaten)
• metrics(s):
  • AUC-ROC
  • Brier Score
  • only one or how to combine more than one?

Next step:

• upload target netcdf: to zenodo and put link to gist/notebook in metadata

@semvijverberg or should this be discussed somewhere else

[aaronspring]

How many years is the 33% threshold calculated?
Only 20 seems a bit low. We could enhance to earlier years and do a rolling mean to smooth the dayofyear curve.

marpyr/Lorentz_workshop#16 (comment)