looking at the code for basic regression

the variable looks like:

while is should be like mean_wdir:

There is an error with predicted wind direction angle using "pdf" (i.e. distribution classification) method:

If we look at

We want:

• all the subswath files generated in VV
• all the L1B tiles present in the L2A file (could be filled with NaN if on land)
• all the variables in the files even if on land.

The processor should not output something else than vv polarization subswath files.

example for a similar processor:
l2wave/2023/307/S1A_IW_WAV__2SDV_20231103T063230_20231103T063257_051049_0627C3_C45D_E00.SAFE/l2a-s1a-iw1-wav-vv-20231103t063230-20231103t063255-051049-0627c3-004-e00.nc

add .pbs and prun script from project_rmarquart

currently:

1. SLC ->XSARSLC xtiling-> L1B -> tiles corners
2. SLC ->sigma0 calibrated+tiles corners (L1B) -> XSARSLC interface-> patches
3. patches->L2A predictions->wdir

it could be more direct (all in L2A processor):

1. SLC->xsarslc dependance->sigma0+tiles-> in memory patches->L2A predictions->wdir

what would be needed?

• integrate interface.py in L2A wind dir processor
• add xsarslc as optional dependency
• skip production of patches
• change the code to use L1B datatree in memory instead of the files

• notebook with predictions "pdf" and "regression" for product S1A_IW_XSP__1SDV_20210418T212223_20210418T212253_037510_046C42_6560.SAFE:IW1

• for pdf
• for regression

default values for sigma0 patches in original network developed by @rmarquarlops are: 44x44 pixels.
These 44x44 pixels matrices are achieved using interface.py on a 17.6km² tiles at 400 m resolution in azimuth and range:

def get_low_res_tiles_from_L1BSLC(file_path, xspectra = 'intra', posting = {'sample':400,'line':400},
                                  tile_width = {'sample':17600.,'line':17600.}, window='GAUSSIAN', **kwargs):

but it seems that using Level-1B generated with a 17600m² size is leading to have sigma0 patches with NaN on the edges:

While we can get 44x44 sigma0 matrix starting from Level-1B produced on 17.7km² tiles without any NaN:

Question for @rmarquarlops : Should we start from Level-1B product with larger tiles (i.e. 17.7km² instead of 17.6km²)?
Having NaN in the sigma0 matrix does not prevent to do the prediction but the results are completely different compare to matrix without NaN.