Closed gerritholl closed 5 years ago
Looking at the underlying radiances in the debug FCDR, it appears the uncertainty is 4 orders of magnitude smaller than for independent or structured, so brightness temperature uncertainties get 'rounded' to zero due to storage digitisation. The above was based on a test run based on the old harmonisation run from April 2018, which was wrong. It may be that this problem gets resolved when using the output of the improved harmonisation run from October 2018.
Out[133]:
<xarray.DataArray 'u_R_Earth_random' ()>
array(1.319945e-15)
Coordinates:
calibrated_channel float32 8.0
In [134]: dsd["u_R_Earth_nonrandom"].sel(calibrated_channel=8).median()
Out[134]:
<xarray.DataArray 'u_R_Earth_nonrandom' ()>
array(5.89319e-15)
Coordinates:
calibrated_channel float32 8.0
In [135]: dsd["u_R_Earth_harm"].sel(calibrated_channel=8).median()
Out[135]:
<xarray.DataArray 'u_R_Earth_harm' ()>
array(1.55e-19)
Coordinates:
calibrated_channel float32 8.0
I seem to be misinterpreting the harmonisation parameters. See commit 24d27d18246faaf1e638c92d32b753309a5ee090 where they were all set to very small values, effectively zero. May need to check with @octoflar on whether something has changed or whether I'm doing something wrong.
All
u_common
common uncertainty are coming out as zero in the easy FCDR.