Data sampling in an A5cell

[jsorel]

Hello,

I am working on a OGC-API DGGS server implementation.

So far we ported A5 to JAVA in Geotoolkit (it's on a branch for now, so you won't see it in the project yet).

I was wondering about data sampling inside a cell.
In the spec for most defined formats we can read :

An Implementation may support either a DGGS-optimized X representation where a dimension is associated with sub-zones, with coordinates being global zone identifiers or local indices into the deterministic zone order, or a traditional representation interoperable with non-DGGS clients where spatiotemporal axes are individual dimensions.

To obtain a deterministic zone order, would it be possible do have somekind of cell hash iterator for a cell and a given relative depth ?

Thanks

[felixpalmer]

Great to hear you've ported it to Java, looking forward to you releasing it. Just to double-check, have you included all the recent changes in this repo? There have been some significant improvements, although the result has been that the underlying grid has changed, so it is totally non-backcompatible.

To your question, I'm sorry, but I'm not sure I completely understand. Could you give an example of how another system does this, for example h3?

The cellToChildren will give you all the descendants of a given cell, and the order is determined uniquely by the space-filling curve. If you look at how it is implemented, it essentially comes down to iterating from 0 to childrenCount

[Eclesia]

(It's me again, answering with my personnal account)

I ported the code at commit sha1 dc1c21d including the test cases. Yes i noticed you made a lot of changes since then and I plan to update the code in the coming weeks.

For the iteration order I don't know about H3 but in Healpix HIPS service each cell data contains the children cell values which are 8 levels under it. Since healpix use square cells with a refinement of 4 at each depth level, a cell datas contains a 512x512 grid with a predictable ordering. The first cell will have a hash identifier of X and next one X+1, and X+2 and so on..

I guess we could just say , as example for a relative depth of 2 :

• loop on each depth 1 cells and loop on each of it's children cells again.
• resulting in : 1-1, 1-2, 1-3, 1-4, 2-1, 2-2,......4-4
• is there a constant value between each of those cell hash identifier in A5 ? Somekind of regular increment ?

[felixpalmer]

is there a constant value between each of those cell hash identifier in A5 ? Somekind of regular increment ?

Yes, the constant increment depends on the resolution level. At the highest resolution, the increment is 1, then 4, 16 and so on. It is easiest to visualize this in the binary representation https://a5geo.org/examples/hierarchy

Notice how every time you zoom in, the red resolution marker but loves to the right and the child cell takes on one of 4 values

Thus when calculating all the descendants of a cell is is possible to just iterate over a set of values, as I linked to above.

I plan on adding some docs on this, does this answer your question?

[jsorel]

it helps yes, but still have a question, let's take some examples :

I want to make a dggrs image (a DiscreteGlobalGridCoverage to be exact), to make it simple it's an object with two lists :

• a list of zones
• a list of sample values (same size as the zones list)

for efficiency reasons, the sample values is a double[] and the list of zones is dynamic (values computed based on the above parameters, start index, increment, size).

If I wanted to find the start index, increment and size of a coverage :

• of the full globe for depth 6
• a specific zone parent for depth 6

How do you compute the parameters ?

[felixpalmer]

OK, I see. You essentially want to deconstruct the cellToChildren function to return

{startIndex, stride, count}

to allow a compact representation?

I can add a utility function for that, although I’m traveling this week so it will take a little time.

[jsorel]

I updated the java code up to version 0.2.0.

Things coming up as you can see below, with on the fly coverage resampling to A5 :