How to convert polygon to H3 cells where the cells completely cover the shape ?

[michaelbrichko]

Hi,

The functionality that library provides is to approximate polygon shape with H3 cells where every cell center resides inside the shape but what I want is to cover the shape completely with H3 cells such that I have a minimal set of H3 cells which cover the shape completely.

The option that I can think of is given H3 cells approximation, add their immidiete neighbors to the result.
I have two issues with that:

1. Some neigbors are completely doesn't intersect with the shape, so they are absolutely redundant.
2. It's much more compute

Is there a better option to solve this?
Thanks!!!

[isaacbrodsky]

This is being actively worked on in #796, as the OVERLAPPING mode. There are workarounds involving buffering the H3 cells from the current algorithm but it sounds like you have already explored most of those options.

[michaelbrichko]

A follow up question please. With already existing tools, how would you suggest to calculate how many more rings to add to polygon in order to be sure that eventual set of cells overlap with polygon+radius in meters. I suspect the main part of solving this is with how many cells the radius in meters should be covered? Given specific resolution, I can calculate hexagon edge length. Deviding buffer length by hexagon edge should be it. But this is a bit combersome, plus there is a chance to stumble on pentagon. What would be better way to solve this ?

Thanks you!

[michaelbrichko]

Or perhaps somehow generate 2 cells buffer meters apart (somehow) and then check cells count produced by "gridPathCells".
I'm looking for assurence for overlapping, not approximation, with minum cells as possible

[michaelbrichko]

Thinking a bit more about it, I think my workarounds aren't good enough, I suspect they will produce too much excess cells.
Let me know please if you think if radius\buffer in meters support in the "horizon" or there is some other good enough workaround

[nrabinowitz]

TBH, the best option here is probably to buffer the polygon beforehand and then fill the buffered area. That's what we'd likely do within the library if we wanted to offer a real-distance buffer, but it's much more likely that if we offer a buffer at all it would be a grid-distance buffer. In general, discrete global grids are going to be tricky to work with when you need this kind of exact measurement, because of variations in shape, area, and orientation of grid cells. Better to do the measured transformations on the vector geometry, then move to the grid, rather than the other way around.

[michaelbrichko]

Makes sense, thank you!

[diehl]

@nrabinowitz If I'm reading the exchange above correctly, we should see the capability to generate a complete hexagonal cover of an arbitrary polygon soon. When do you anticipate this capability will land in h3-py? I'm wondering if it's worth waiting for as opposed to developing a workaround in the interim, as I have a need for this capability now. Grateful to see that this is in flight!

[nrabinowitz]

@ajfriend should have a better perspective on h3-py. The functionality has landed in H3 core but we haven't released it yet.

[diehl]

Roger that @nrabinowitz - much appreciated

[diehl]

@ajfriend if you have a sense of the ETA for this capability in h3-py, I'd love to hear it.