search

FIDUCEO / FCDR_HIRS

Other
1 stars 2 forks source link

Outliers mess up self-emission #144

Open gerritholl opened 7 years ago

gerritholl commented 7 years ago

Self-emission training may not appropriate detect outliers. Here, self-emission errors totally messes up radiances and brightness temperatures, such as detected by uncertainty in radiance space but NOT in BT space (see #145 for the latter problem).

Not sure how often it happens, this example is for NOAA-6 1983-03-01.

image

gerritholl commented 7 years ago

Mostly fixed by commit 23bed3d; it doesn't mess up the training anymore, but individual lines with bad temperatures are still going bad

image

gerritholl commented 6 years ago

See also #136.

jmittaz commented 5 years ago

Extra filtering in typhon code may also help with this (see pull request #386)