feat(Gridintersect): new grid intersection options

[dbrakenhoff]

Features:

• Add min_area_fraction kwarg for polygon intersections, keeping only results that cover a certain fraction of a grid cell (as discussed in GridIntersect fails with specific shapely Polygon due to change of nrow, ncol in the discretization object #1216)
• Add contains_centroid kwarg for polygon intersections, keeping only results that contain cell centroid in intersection result (as discussed in GridIntersect fails with specific shapely Polygon due to change of nrow, ncol in the discretization object #1216)
• Add return_all_intersections kwarg, which return all multiple intersection results for points and linestrings on cell boundaries (as discussed but not quite implemented in GridIntersect does not include cells with identical vertices, even with DISU #1198 and feat(GridIntersect): add optional kwarg to .intersect() to keep all c… #1214)

Bug fixes:

• fixes bug in gridintersect when intersecting linestrings starting in vertex with method="structured" #1467
• fixes GridIntersect fails with specific shapely Polygon due to change of nrow, ncol in the discretization object #1216

Other improvements:

• Update example notebook
• Add tests for new features and bug in gridintersect when intersecting linestrings starting in vertex with method="structured" #1467
• Deal with some shapely DeprecationWarnings

[dbrakenhoff]

I also wanted to deal with the shapely DeprecationWarning "ShapelyDeprecationWarning: Setting custom attributes on geometry objects is deprecated, and will raise an AttributeError in Shapely 2.0".

But I was wondering what the policy is on backward compatibility for optional dependencies?

The new methods needed to silence this warning are available from shapely 1.8.0 onwards. Do we continue supporting shapely<1.8 for a while, or can we break the code for older versions of shapely?

Let me know, and i will commit a fix accordingly.

[codecov]

Codecov Report

Merging #1468 (6746000) into develop (1757470) will decrease coverage by 44.926%.
The diff coverage is 3.973%.

@@              Coverage Diff               @@
##           develop     #1468        +/-   ##
==============================================
- Coverage   74.970%   30.044%   -44.927%     
==============================================
  Files          249       249                
  Lines        53645     52276      -1369     
==============================================
- Hits         40218     15706     -24512     
- Misses       13427     36570     +23143     

Impacted Files	Coverage Δ
flopy/utils/gridintersect.py	7.455% <3.973%> (-75.677%)	⬇️
flopy/mf6/utils/testutils.py	0.000% <0.000%> (-93.334%)	⬇️
flopy/mfusg/mfusgsms.py	6.190% <0.000%> (-91.941%)	⬇️
flopy/mf6/utils/postprocessing.py	9.259% <0.000%> (-88.889%)	⬇️
flopy/modflow/mfevt.py	7.537% <0.000%> (-87.463%)	⬇️
flopy/utils/mtlistfile.py	4.620% <0.000%> (-85.348%)	⬇️
flopy/mf6/utils/lakpak_utils.py	2.097% <0.000%> (-85.315%)	⬇️
flopy/modflow/mfhyd.py	13.223% <0.000%> (-85.124%)	⬇️
flopy/utils/swroutputfile.py	10.714% <0.000%> (-85.079%)	⬇️
flopy/modflowlgr/mflgr.py	6.271% <0.000%> (-84.040%)	⬇️
... and 206 more

[jlarsen-usgs]

The PR looks good overall. The only question I have is about the

sys.path.insert(1, "..")

call in t065_test_gridintersect.py. Is there a support script that is located within that path that is not explicitly installed into python?

[jlarsen-usgs]

Approving this PR. @w-bonelli is going to remove the sys.path.insert() call from t065 in his upcoming PR.

[dbrakenhoff]

Sorry, the insert was just to get the code to use my fork instead of the installed flopy package. Thanks for dealing with that!

Any advice on dealing with that shapely deprecation warning?

[jlarsen-usgs]

@dbrakenhoff

It looks like we won't be able to use p.name, which is used to store the cellid, once shapely 2.0.0 is released. Would it break anything or cause performance issues to instead store a dictionary of {name : shapely geometry object} and then update the internals to work with a convention like this instead of storing the cellid directly in the object?

[dbrakenhoff]

That would certainly be an option, but not sure if that would hurt performance.

I'd prefer to go the method Shapely 2.0 suggests:

geoms, items = self._get_gridshapes()  # modify method to return geoms and cellids
self.strtree = strtree.STRtree(geoms, items)
# then get either geoms or cellids internally with: 
self.strtree.query_geoms()
self.strtree.query_items()

Only trouble is then we'd need to add logic for Shapely < 1.8 users if that is something we still want to support. I'd be okay with requiring shapely > 1.8 since it's an optional dependency anyway, but not sure how you guys think about that.

PS. this is probably something for a new issue.

[christianlangevin]

I support requiring shapely > 1.8 unless it wasn't available for one of our supported operating systems. It looks like win, Mac, and linux are all supported.

[jdhughes-dev]

@dbrakenhoff I would say it would be fine to support shapely > 1.8 now in develop and the in the next flopy release