meteoswiss / biascorrection Goto Github PK

Mean debiasing and spr_err doesn't work properly with single values, maybe 1/n sqrt(1/n-1) issue

Clarify missing value treatment.
Use the same mask on fcst and obs to mask all values as missing before passing on to bias correction function.
return missing values if method works only for non-missing data
make bias correction methods missing value proof (e.g. compute mean(..., na.rm=T))
how to deal with different missing values in input and output forecasts?

MASS
parallel
check auf R3.3.2

• Empirical statistical QQmap (as in Themessl 2011, 2012 and Wilcke 2014).
• qq from 1st to 99th percentile and constant correction beyond
• Raety (2014) and Rajczaks (2015) suggest 91 over 31-day moving windows
• use rollRcpp to get faster implementation of moving quantiles

Check with toy model