Fix colrow2lonlat working only for square areadefs

[ameraner]

This PR fixes a bug observed in the AreaDefinition method colrow2lonlat. The function was working properly only for square areadefs, while returning wrong values for other areas.

To reproduce:

from satpy.resample import get_area_def
from satpy.scene import Scene

def lonlat2colrow2lonlat(lon_in, lat_in, area_def, desc):
    """Checks the reciprocity of lonlat2colrow and colrow2lonlat."""

    col, row = area_def.lonlat2colrow(lon_in, lat_in)
    lon_out, lat_out = area_def.colrow2lonlat(col, row)

    print("\nLon out "+desc+": ", lon_out)
    print("Lat out "+desc+": ", lat_out)

    return

lon_in = 30
lat_in = 40
print("Lon in: ", lon_in)
print("Lat in: ", lat_in)

# let's check upright areadefs
area_def = get_area_def("seviri_0deg")
lonlat2colrow2lonlat(lon_in, lat_in, area_def, 'seviri_0deg')

area_def = get_area_def("eurol")
lonlat2colrow2lonlat(lon_in, lat_in, area_def, 'eurol')

# let's check flipped areadefs
test_image = '/you/path/to/a/SEVIRI/native/image.nat'
scn = Scene(reader="seviri_l1b_native", filenames=[test_image])

scn.load(['HRV'])
area_def = scn['HRV'].attrs['area'].defs[1]
lonlat2colrow2lonlat(lon_in, lat_in, area_def, 'nat HRV')

scn.load(['VIS008'])
area_def = scn['VIS008'].attrs['area']
lonlat2colrow2lonlat(lon_in, lat_in, area_def, 'nat vis08')

Output with the current master:

Lon in:  30
Lat in:  40

Lon out seviri_0deg:  29.99159844726278 
Lat out seviri_0deg:  40.006400108344515

Lon out eurol:  -94.68203082968928
Lat out eurol:  27.231622640431453

Lon out nat HRV:  73.08085247495852
Lat out nat HRV:  -17.06078516437714

Lon out nat vis08:  30.037311995180985
Lat out nat vis08:  39.92747931402665

note how the non-square eurol and the (upper) HRV areas have wrong values.

Output after this PR:

Lon in:  30
Lat in:  40

Lon out seviri_0deg:  29.99159844726278 
Lat out seviri_0deg:  40.006400108344515

Lon out eurol:  29.992464774442446
Lat out eurol:  40.01002332383795

Lon out nat HRV:  29.999208764610334
Lat out nat HRV:  39.993233105064675

Lon out nat vis08:  29.991595970000212
Lat out nat vis08:  40.006397985178815

All values are correct.

• Tests added
• Tests passed
• Passes git diff origin/master **/*py | flake8 --diff

[djhoese]

Any idea why the inverse was used in the first place (the y.size - part I mean)? Shouldn't that still be used?

[ameraner]

I think it's because of the test area definition it was using before, plus the fact that it was mixing x and y with cols and rows. In that specific case, the x and y are the inverse of each other, so actually y[y.size - cols]==x[cols] holds, and vice versa.

[djhoese]

I think that makes sense to me. If the old tests still pass then I'm ok with it.

[ameraner]

Yep they do, I actually just reintroduced them! (See Martin's comment below)

[coveralls]

Coverage decreased (-0.006%) to 92.586% when pulling 3a4fb5a on ameraner:fix_areadef_colrow2lonlat into 0e0c485 on pytroll:master.

[codecov]

Codecov Report

Merging #294 into master will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #294      +/-   ##
==========================================
+ Coverage   92.59%   92.60%   +0.01%     
==========================================
  Files          43       43              
  Lines        9057     9077      +20     
==========================================
+ Hits         8386     8406      +20     
  Misses        671      671              

Impacted Files	Coverage Δ
pyresample/geometry.py	83.05% <100.00%> (ø)
pyresample/test/test_geometry.py	97.94% <100.00%> (+0.03%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0e0c485...3a4fb5a. Read the comment docs.

[mraspaud]

Great catch! I just would like to keep the existing test if possible.

[mraspaud]

LGTM