Add Coordinate Resolution

[CFoye-Creare]

Description

New feature branch to address issue #502. We are looking to implement a helper function to return the resolution of coordinate systems at different levels of accuracy.

Proposed Changes

For a Coordinate object, a method horizontal_resolution to return the resolution of horizontal coordinate system. This should support both stacked and unstacked coordinates. We are offering three kinds of resolutions:

• "nominal" <-- this is wrong but cheap to calculate. Give a 'nominal' resolution over the entire domain
• "summary" <--- Returns average and std.dev
• "full" <--- returns the exact distance between every point, or every dx, dy if coordinates are unstacked We also let the user specify their desired return units. We support any unit that podpac.units supports.

  Stacked Coordinates

  If the horizontal coordinates are stacked, then we should return information based upon the set of points.

• "nominal" resolution: returns the average min distance from each point to the next
• "summary" resolution: returns the average min distance from each point to the next, and the standard deviation of the min distances
• "full" resolution: returns every distance between every point in a distance matrix

  Unstacked Coordiantes

  If the horizontal coordinates are unstacked, then we should return information based upon the grid space

• "nominal" resolution: returns the average dx,dy.
• "summary" resolution: returns average dx,dy and their standard deviations
• "full" resolution: returns the exact dx,dy of each grid square.

[CFoye-Creare]

Currently assumes the inputted data will be in valid lat/lon degrees

[CFoye-Creare]

• [x] Add Support for input units other than degrees
• [x] Add Support for all ellipsoids

[mpu-creare]

Could you also edit the MR and add a short description of the changes so we can remember what they were next time we look at this. Also add a link to the relevant issue.

[CFoye-Creare]

See https://github.com/creare-com/podpac/issues/502#issuecomment-1425955172 for some info on how I implemented the unstacked coords this time.

[jmilloy]

I haven't been following this but I get all the notificatons and took a brief look today. I'm not going to make a thorough review, but I notice that this new method has logical cases for different types of coordinates which is something I tried pretty hard to avoid when developing the coordinates module. So something like this would be more in line with the style of the rest of the module:

def horizontal_resolution(self, units, type):
    resolution_dict = {}
    for dim, c in self.items():
        if 'lat' in c.dims or 'lon' in c.dims:
            resolution_dict[dim] = c.horizontal_resolution(units, type)

or maybe

def horizonal_resolution(self, units, type):
    return {dim: c.horizontal_resolution(units, type) for dim, c in self.items() if c.is_horizontal}

Then the four methods are defined each in Coordinates1d or StackedCoordinates instead. Similarly, I think there are some benefits to then testing these methods individually in test_coordinates1d and test_stacked_coordinates and then lightly testing the Coordinates horizontal_resolution class that pulls them together. Anyways, just a note.

[mpu-creare]

@jmilloy this would typically work but to get the resolution you need both the latitude and longitude, even in the unstacked case -- so it can't be handled in Coordinates1d. Unless you have a better suggestion?

I really do appreciate the note on the design.

[jmilloy]

@mpu-creare Yeah, I wondered about that and wasn't thorough enough. I'd consider passing the lat in, which would ignored in some cases, but make it very clear what's needed.

for dim, c in self.items():
    resolution[dim] = c.horizontal_resolution(units=units, type=type, lat=self['lat'])

In fact, you can technically make Coordinates with a lon dimension but not a lat dimension, right? And then calling horizontal_resolution on coordinates without a 'lat' should be caught and raised as an error, right? I didn't look super closely but I don't think it is. I think making it it explicit helps point that out and reminds you to test for it.

[jmilloy]

(Also, don't use type as an argument/variable)

[CFoye-Creare]

@jmilloy restype work?

[CFoye-Creare]

Wrote a check for unstacked "lat" or unstacked "lon" without a counterpart. Now throwing a ValueError. Also deleted some lingering commented code, and changed the "type" parameter to "restype".

[CFoye-Creare]

I realized that the lat value actually matters for longitude as well, because the distances are greater near the equator due to the bump at the equator.

[mpu-creare]

@CFoye-Creare Nice first PR. Thanks for navigating all the trickiness behind this implementation. Merging.

Closes #502 .