You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Add workaround so that non-finite structured uncertainties do not cause
bad data to enter the harmonisation files or the writing of harmonisation
files to fail, but rather skip those pixels completely for now. This
filter must be removed when in the future a fix for #281 shall have been
introduced.
I also find uncertainties in the Earth counts, for example in NOAA-16 channel 7:
In [30]: ds = xarray.open_dataset("/gws/nopw/j04/fiduceo/Data/FCDR/HIRS/v0.8pre2_no_harm/debug/noaa16/2004/08/03/FIDUCEO_FCDR_L1C_HIRS3_NOAA16_20040803003812_20040803022004_DEBUG_v0.8pre2_no_harm_fv2.0.0.nc")
In [31]: print(ds["u_C_Earth"].sel(calibration_cycle="2004-08-03T01:54", calibrated_channel=7).item())
nan
While debugging the harmonisation input, I found that for shortwave channels (17, 18, 19), there are cases where
[k for k in harm.data_vars.keys() if not harm[k].dtype.kind.startswith("M") and not numpy.isfinite(harm[k]).all().item()]
results in
['nominal_measurand_uncertainty_structured1']
or
['nominal_measurand_uncertainty_independent1']
I need to investigate when and why this happens.
The text was updated successfully, but these errors were encountered: