Pad grids automatically before applying transformations by leouieda · Pull Request #608 · fatiando/harmonica

leouieda · 2025-11-19T10:17:11Z

Rework the apply_filter function to make it apply and then remove padding when applying a filter automatically. The padding can be turned off or controlled by passing keyword arguments. The default padding mode is "edge" which works well with our tests and examples. Dropping non-dimensional coordinates is now handled by apply_filter as well instead of the FFT functions. Only drop non-dimensional coordinates on upward continuation since that's the only transformation that changes coordinates. Update the tests for transformations, in particular the tilt, Gaussian filters, and reduction to the pole which now test against a synthetic instead of the Montaj results from previously. Remove the unused Montaj testing grid.

Relevant issues/PRs: Fixes #390 and fixes #630

Changes how keyword arguments are handled by this function since we now need one set for the padding and one for the filter.

They are failing but I don't know why yet

We have a better test now against a synthetic and this test didn't work very well before, since it required removing a mean from the Oasis result for some unspecified reason.

Was using the potential but it's too smooth and causes problems at the edges. Switch to gz instead and it works when looking at a smaller region inside the area.

leouieda · 2026-03-30T18:22:19Z

To do:

Update documentation examples everywhere
Update docstrings
Allow transformation functions to receive the padding arguments

Introduce an argument to apply_filter for dropping coordinates. We only really need to do this for upward continuation. The other filters don't alter the location of points. Update the filter tests to use synthetics instead of Montaj results.

leouieda · 2026-04-02T13:12:19Z

@santisoler @mdtanker I'm done here and could use a quick review. The transformations are much better now and so are the tests. @mdtanker I ended up sticking with the "edge" mode since it seemed to work better in the tests and docs than the "reflect" mode. Not sure why.

santisoler

Thanks for pushing this forward, @leouieda!

I think it's looking great. I left a few suggestions to follow numpydoc style for optional arguments. I just went with , optional, but we can also use default: ... if you like it better.

I also added a few suggestions to force to pass optional arguments as keyword arguments. I know that would break backward compatibility, but I don't think a lot of users are actually using them, and the change wouldn't be too hard to update on their side. Let me know what do you think.

Feel free to merge afterwards!

santisoler · 2026-04-02T16:31:41Z

harmonica/filters/_utils.py

    fft_filter : func
        Callable that builds the filter in the frequency domain.
-    kwargs :
+    filter_kwargs : dict


Suggested change

filter_kwargs : dict

filter_kwargs : dict or None, optional

santisoler · 2026-04-02T16:31:48Z

harmonica/filters/_utils.py

        Any additional keyword argument that should be passed to the
-        ``fft_filter``.
+        ``fft_filter`` in the form of a dictionary.
+    pad : bool


Suggested change

pad : bool

pad : bool, optional

santisoler · 2026-04-02T16:32:06Z

harmonica/filters/_utils.py

+        and applying the filter and remove it after the inverse Fourier Transform.
+        Adding padding usually helps reduce edge effects from signal truncation.
+        Default is True.
+    pad_kwargs : dict


Suggested change

pad_kwargs : dict

pad_kwargs : dict or None, optional

santisoler · 2026-04-02T16:32:28Z

harmonica/filters/_utils.py

+        :meth:`xarray.DataArray.pad` function. If none are given, the default
+        padding of 25% the dimensions of the grid will be added using the
+        "edge" method.
+    drop_coords : bool


Suggested change

drop_coords : bool

drop_coords : bool, optional

santisoler · 2026-04-02T16:34:14Z

harmonica/filters/_utils.py

+def apply_filter(
+    grid, fft_filter, filter_kwargs=None, pad=True, pad_kwargs=None, drop_coords=False
+):


I'd be inclined to force passing optional arguments as keyword arguments:

Suggested change

def apply_filter(

grid, fft_filter, filter_kwargs=None, pad=True, pad_kwargs=None, drop_coords=False

):

def apply_filter(

grid, fft_filter, *, filter_kwargs=None, pad=True, pad_kwargs=None, drop_coords=False

):

santisoler · 2026-04-02T16:48:23Z

harmonica/_transformations.py

        evenly spaced (regular grid). Its dimensions should be in the following
        order: *northing*, *easting*. Its coordinates should be defined in the
        same units.
+    pad : bool


Suggested change

pad : bool

pad : bool, optional

santisoler · 2026-04-02T16:48:28Z

harmonica/_transformations.py

+        and applying the filter and remove it after the inverse Fourier Transform.
+        Adding padding usually helps reduce edge effects from signal truncation.
+        Default is True.
+    pad_kwargs : dict or None


Suggested change

pad_kwargs : dict or None

pad_kwargs : dict or None, optional

santisoler · 2026-04-02T16:48:36Z

harmonica/_transformations.py



-def tilt_angle(grid):
+def tilt_angle(grid, pad=True, pad_kwargs=None):


Suggested change

def tilt_angle(grid, pad=True, pad_kwargs=None):

def tilt_angle(grid, *, pad=True, pad_kwargs=None):

santisoler · 2026-04-02T16:48:43Z

harmonica/_transformations.py

        evenly spaced (regular grid). Its dimensions should be in the following
        order: *northing*, *easting*. Its coordinates should be defined in the
        same units.
+    pad : bool


Suggested change

pad : bool

pad : bool, optional

santisoler · 2026-04-02T16:48:49Z

harmonica/_transformations.py

+        and applying the filter and remove it after the inverse Fourier Transform.
+        Adding padding usually helps reduce edge effects from signal truncation.
+        Default is True.
+    pad_kwargs : dict or None


Suggested change

pad_kwargs : dict or None

pad_kwargs : dict or None, optional

leouieda mentioned this pull request Nov 19, 2025

Put padding inside the transformation function #390

Open

leouieda added 9 commits March 30, 2026 14:19

Better comments in apply filter and a test notebook

2476017

Try to mute ifft lag warnings

25c1a97

Add and remove padding automatically in apply_filter

e443ece

Changes how keyword arguments are handled by this function since we now need one set for the padding and one for the filter.

Ignore a Ruff warning about replacing dict comp with fromkeys

238951c

Add more tests and update older ones

bbf70f1

They are failing but I don't know why yet

Fix the reduction to the pole test

27b4858

Remove Oasis Montaj test for reduction to the pole

38923f6

We have a better test now against a synthetic and this test didn't work very well before, since it required removing a mean from the Oasis result for some unspecified reason.

Come up with a better test for gaussian filters

4d2dd30

Fix tilt angle test

2f886f6

Was using the potential but it's too smooth and causes problems at the edges. Switch to gz instead and it works when looking at a smaller region inside the area.

leouieda force-pushed the padding-inside branch from 5fd8ce8 to 2f886f6 Compare March 30, 2026 18:21

leouieda added 6 commits March 30, 2026 19:44

Merge branch 'main' into padding-inside

175d35a

Adjust Euler tests after the new filters

7169f6e

Fix broken tests and add missing tests

d841ce3

Fix broken formatting

6a3863c

Update docstrings and add option to control padding in transformations

5f5b797

Update gallery examples to use built-in padding

8551905

leouieda added this to the v0.8.0 milestone Apr 1, 2026

leouieda added 3 commits April 2, 2026 09:49

Remove test notebook

a99ea83

Remove Montaj testing grid

98ffd81

leouieda marked this pull request as ready for review April 2, 2026 12:53

leouieda requested review from mdtanker and santisoler April 2, 2026 13:11

santisoler requested changes Apr 2, 2026

View reviewed changes

	pad_kwargs : dict or None
	pad_kwargs : dict or None, optional



		def tilt_angle(grid):
		def tilt_angle(grid, pad=True, pad_kwargs=None):

	def tilt_angle(grid, pad=True, pad_kwargs=None):
	def tilt_angle(grid, *, pad=True, pad_kwargs=None):

Conversation

leouieda commented Nov 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

leouieda commented Mar 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

leouieda commented Apr 2, 2026

Uh oh!

santisoler left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

leouieda commented Nov 19, 2025 •

edited

Loading

leouieda commented Mar 30, 2026 •

edited

Loading