Feedback on verification_RPSS notebook
Is this what you would expect when assessing RPSS? @awr @fpv07
Is the way I compute the terciles correct? I dont do any anomalies. Raw values. category_edges
for RPSS depend on longitude, latitude and weekofyear
.
- Taking last 20 years to calculate weekly
category_edges
for RPSS based on the observational distribution (as currently done in code block 15)? - Or should it be different RPSS
category_edges
for observations and forecast separated, e.g. terciles of forecasts to bin the forecasts and terciles of observations to bin the observations (as commented out in the code block 16)?
Note: I dont understand yet why RPS for 2t
is identical for benchmark
and ML_forecast
. Also 2t
should have some land data for Antarctica.