FY4A REGC data resampling return all nan

[LiangHongli1]

Describe the bug
When scn.resample FY4A REGC file, it returns all nan with version 3.8.0 and above. It works fine with version lower than 3.7.0. It might have something to do with FY4B compatibility. While version 3.7.0 supports FY4B and the reader name is still agri_l1, it doesn't work with FY4B.
------------ Here is the code. Files can be downloaded from the NSMC web. ----------

from satpy import Scene
from pyresample import load_area
from pyresample.geometry import GridDefinition
from satpy.resample import get_area_def
import numpy as np
f = "FY4A-_AGRI--_N_REGC_1047E_L1-_FDI-_MULT_NOM_20231002233836_20231002234253_4000M_V0001.HDF"
scn = Scene(filenames=[f], reader="agri_fy4a_l1"). # agri_fy4a_l1
scn.load(["C11"])
scn.save_dataset("C11", "fy4b_c11_2023100812.png")
area_projection = load_area("areas.yaml", "china_4000M_webmerc"
                            )
tmp_scn = scn.resample(area_projection, cache_dir="cache")
newdata = tmp_scn["C11"].values
data = scn["C11"].values
print(np.isnan(newdata).all(), np.isnan(data).all())
tmp_scn.save_dataset("C11", "fy4b_c11_china_2023100812_4km.png")

Expected behavior
After resampling, np.isnan(newdata).all() shoud be "False", yet it is "True", with REGC file, satpy version 3.8.0 and above. np.isnan(newdata).all() is False with DISK file.

[mraspaud]

@LiangHongli1 thanks a lot for bringing this issue to our attention! I don't have a REGC file myself, I will ask around, but if you can share the file, that would be great!

[mraspaud]

Also could you share the definition of the area you are resampling to?

[zxdawn]

I have tested with Python 3.9 and it works well with native resampler. Could you (@LiangHongli1 ) share your area definition? That would make it easier to reproduce your problem.

[LiangHongli1]

Thanks for reply! But sorry, I'm unable to upload the files due to its large size, for now.
Here is the area definition:
china_4000M_webmerc:
description: china_4000M_webmerc
projection:
proj: webmerc
ellps: WGS84
lat_0: 30.0
lon_0: 101.0
shape:
height: 1625
width: 1948
area_extent:
lower_left_xy: [-3896182.177765, 669141.057044]
upper_right_xy: [3896182.177765, 7170156.294000]

[zxdawn]

Thanks, I have tested using your area and it works well:

from pyresample.geometry import AreaDefinition

area_id = 'china_4000M_webmerc'
proj_string = '+proj=webmerc +lat_0=30 +lon_0=101 +a=6371228.0 +units=m'
area_extent = (-3896182.177765, 669141.057044, 3896182.177765, 7170156.294000)
shape = (1948, 1625)
area_def = AreaDefinition.from_extent(area_id, proj_string, shape, area_extent)

new_scn = scn.resample(area_def, cache_dir="cache")
new_scn['C11'].plot()

print(new_scn['C11'].isnull().all().values)
# False

[LiangHongli1]

@zxdawn what's the satpy version you are using?

[LiangHongli1]

Well, I've tested the way you load the area, it doesn't work for me either. I'm using python 3.9.6, satpy 0.41.0 on macos apple m2. And the figure doesn't look right. It shoud look like this.

[zxdawn]

Ha, I can reproduce your NaN problem if I use 0.43.0.post0. I tested the newest branch: pip install git+https://github.com/pytroll/satpy.git and it works well.

[LiangHongli1]

Wow, it works for me too. Looks like the newest branch solved this.
Thanks a lot!

[zxdawn]

You're welcome ;) Actually, I don't know which commit fixes this problem ... Maybe @mraspaud knows.

[LiangHongli1]

Probably someone fixes it without knowing it? Still, thanks and I'd be happy to know what causes this problem.

[mraspaud]

Ok, good that it's fixed. Maybe it's time for a new release then. Closing this now as it seems to be solved.