Plane from trace

[claudio525]

Build on top of #14, so that needs to be merged first.

• Adds function to create a Plane object from a fault trace definition. Useful for the old NSHM fault definitions.
• Increase tolerance in orientation check to prevent flipping when dip is very close to 90 (

  source_modelling/source_modelling/sources.py

  Line 286 in d22d5f0

  if not np.isclose(orientation, 0, atol=1e-3) and orientation < 0:

  )

[claudio525]

The latest commit has two changes:

• The order of the trace points is no longer important. The dip direction is used to determine the strike, which is all handled by the post_init. Made this change as my initial interpretation of the NHM fault definitions was incorrect, i.e. I had assumed that the order of the trace points has meaning, however turns out it does not.
• It now operates on nztm data instead of lat,lon to ensure a consistent dip direction, which is required for creating a Fault object.

[lispandfound]

@claudio525 I have made a few changes:

1. Vectorised the mathematics,
2. Updated the docstrings,
3. Improved the tests to use pytest.approx (and also modify the tests for the failure cases on fault construction),
4. Changed the dip direction consistency check to work directly on the NZTM corner representation (actually checking the dip direction vectors are parallel). This should remove the numerical issues related to the calculation of dip direction in this method.

[claudio525]

@lispandfound I was working on this last night and this morning without realising that you had made changes, so have just combined our changes.

[lispandfound]

@claudio525 I made a couple of changes:

1. I move the dip direction check to after the plane rearrangement for strike. For really long faults (such as the alpine fault), dip direction is not quite consistent because it's a measured bearing from north and the relative location of north changes ever so slightly over really long faults -- like the Alpine fault for example. Now dip direction is checked after rearranging the planes into a line, and then only adjacent faults are checked. This is the only way I've been able to get the alpine fault nshm examples to load properly.
2. Improve the fault construction failure tests by making them use from_centroid_strike_dip instead of corners so we can understand their behaviour better.

[lispandfound]

Only one small change left!

[sungeunbae]

The value of abs varies from 1e-6 (strike_nztm, dip) to 10 (dip_dir). Was it determined empirically?

[lispandfound]

The value of abs varies from 1e-6 (strike_nztm, dip) to 10 (dip_dir). Was it determined empirically?

The tolerance on dip direction is 0.1. Yes some of these values have higher tolerances because they are subject to greater variation and numerical error. The great circle bearings in particular are very finicky to get right.

[lispandfound]

The value of abs varies from 1e-6 (strike_nztm, dip) to 10 (dip_dir). Was it determined empirically?

The tolerance on dip direction is 0.1. Yes some of these values have higher tolerances because they are subject to greater variation and numerical error. The great circle bearings in particular are very finicky to get right.