Better gestion of error for recalibration

[vincelhx]

No description provided.

[vincelhx]

changed rasters chunks

[vincelhx]

in the last commit I changed RCM calibration :

we did {varname}_raw = {varname}_raw.where({varname}_raw > 0) .
But it was a problem in the case of denoising was already applied in DN. It was forcing the future "{varname}_raw" (modified in add_denoised) values to be nan ;

Now steps are made in the good order :

-calibration
-denoising + renaming if necessary ({varname}_raw is with noise while {varname} is without noise

• then force raw vars to be > 0

@agrouaze should i do the same logic for S1 & RS2 ?

[agrouaze]

in the last commit I changed RCM calibration :

we did {varname}_raw = {varname}_raw.where({varname}_raw > 0) . But it was a problem in the case of denoising was already applied in DN. It was forcing the future "{varname}_raw" (modified in add_denoised) values to be nan ;

Now steps are made in the good order :

-calibration -denoising + renaming if necessary ({varname}_raw is with noise while {varname} is without noise

* then force raw vars to be > 0

@agrouaze should i do the same logic for S1 & RS2 ?

@alxmouche said "it would be better to keep negative value in the sigma0 (resp. beta0 and gamma0)", also it would be good to keep the 3 variables at the 3 steps in the measurement dataset:

1. digital number
2. sigma0 calibrated
3. sigma0 calibrated+denoised

Then this protocol about the negative values should be described in the sphinx documentation.
I don't see any reason why it would be different from a mission to another.

It also related to #193 and #242 .

[vincelhx]

In the last code I wrote, all the missions perform 'force raw vars to be > 0' at the end of the calibration and denoising process. However, since RCM — with sometimes an already denoised product — is processed differently, the code had to be modified

[agrouaze]

In the last code I wrote, all the missions perform 'force raw vars to be > 0' at the end of the calibration and denoising process. However, since RCM — with sometimes an already denoised product — is processed differently, the code had to be modified

What is the "last code"? I cannot find any mention of this sentence "force raw vars to be > 0".
The sphinx documentation does not explain the calibration nor the denoising nor this last question "do we let negative values", I think this is important, so if you don't add it, I will do it in another PR.
sphinx documentation is aligned with the code except for the special case of "already denoised".
I will make sure with Canadian agency that the 2 quantities:
a ) sigma0 = denoise(calibrate(DN))
b ) sigma0 = calibrate(denoise(DN))
are equivalent.
Then documentation will be updated.

[vincelhx]

#245 will be in marge of making an homogenous code between the three sensors