[brentwilder]

This looks to address #16 by ensuring interpolation only occurs on the striping artifacts. In this PR the skew() method is returned to the same transformation as carried out in the ARS repo.

This update leverages scipy.ndimage's median_filter method . To determine if it's a line I use a somewhat arbitrary value of 3pix, passed twice (vertically, horizontally) through median filter.

horizon_cos = horizon_cos - median_filter(horizon_cos - median_filter(
                             horizon_cos, size=(3, 1)), size=(1, 3))

dx=50m

---

Changing the size parameter above dramatically impacts the amount filtered by the method. Eyeballing it between the Lakes test and also Kings Canyon Topo file I have locally, this appeared to give the most stable results (size of 3x1 then 1x3).

This is a nice middle ground, as yes maybe there are still some small striping present, but it's also not interpolating heavily.

dx=100m

[brentwilder]

Note to self: if this works we will need to update the test data.

[jomey]

I say go ahead with the test updates. Then we will tackle the ones in SMRF and AWSM

[brentwilder]

Discussing this with Joe, there is a small slowdown due to the image filtering technique here. However, in the context of iSnobal domain for a given day this only add about ~0.3 seconds to total run time per day.

Raw numbers for Kings Canyon domain (dx=100m)

• with filter : 0.0345 sec/azimuth 
• no filter:  0.0215 sec/azimuth

[jomey]

We still need to update the pyproject.toml. See the "unresolved comments"

[brentwilder]

Huh, I already pinned scipy (and removed from build)... I must have did something strange when trying to resolve your comments @jomey ?