# hhg9 / Hex9
This is `Hex9` is an **ongoing project** exploring (and developing) a novel
*hierarchical hexagonal grid* (`HHG`) system for global projections.
It is well-suited to population mapping, environmental modeling, heat-mapping,
hex-binning, and other geospatial analyses.

Examples and further documentation may be found on the git repository.

#### Changelog
```
0.1.2a0 Better PostGIS SQL functions - offering support in PostGIS and QGIS.
        Current focus is now on PROJ9.8 - CRS and projections.
        A few deep bug-fixes. All examples should be working.
        However, many examples are not very interesting!
        A Composition class has been added for overlaying layers.
0.1.1a2 Fixed missing warp file in distro.
0.1.0a6 Warp in place, offering good authalicity.
0.1.0a5 More bug-fixes - especially regarding grids.
        A new example, showing multiple hexagons over Chiginagak volcano, AK.
        A little more documentation.
0.1.0a4 Couple of small bug-fixes,
        A couple of new features
0.1.0a3 Rewrite of the hex address and hex polygon code.
        Hex addresses are not backward compatible, so several files
        (documentation) will need to be revised.
        Examples have all been refactored, and tested accordingly.
0.1.0a2 Improved some documentation; added more licence declr.
0.1.0a1 Modifying Numpy minimum - bitwise_count dependency.
        Removed redundant methods in neighbours.
        hex_layer is still unreliable
0.1.0a0 Initial Package
```
#### Why hexagonal tiling
* Hexagonal tiling is unique among regular tilings: all neighbors share an edge.
* Hexagonal tiling corresponds to optimal circle packing.
* However...
  * Hexagons cannot be tiled with hexagons.
  * The sphere cannot be covered by hexagonal tilings
* So, although ideal HHG have a long history in geospatial modeling...
  * Most global HHG are either flat or approximate, or not entirely hexagonal.
  * Existing approaches involve trade-offs, such as
    * Approximating the sphere with slight distortions.
    * Limiting the number of hierarchical layers.
    * Supporting only partial support for transitions between layers.
    * Deriving the hexagonal grid from spherical geometry
    * Needing precomputed databases of fixed points.
    * Requiring additional polygon types (commonly pentagons) for closure
* **Hex9** presents a *new* approach to the “holy grail” of HHG; It aims to
  reduce some of these constraints while remaining early-stage and not yet
  production-ready (Autumn 2025).
  Where does it fail?
  The main concern is that the current Hex9 octahedral projection is not amazingly
  equal-area, especially across a global level.  However, it's hexagons are (for the main part)
  pretty round, and the area is pretty stable (±7% or so, globally), however local variation is
  far more stable.  For example, when examining nations,
  the variation is not so strong  (ok maybe not so much Russia, or Canada).
  The authalicity example reveals global variation very clearly.

### Why Hex9
#### Grid-Projection Decoupling
The Hex9 grid is *fully decoupled from the underlying global projection*.
The logical hex grid exists independently of any coordinate reference system and
can be applied within any octahedral global projection.
While Hex9 comes with a derived projection for this project, the grid itself is
projection-agnostic, while the derived projection relies on the Hex9 grid
 for root-finding operations.

This separation ensures:
 * The grid can be reused across projections without loss of structure.
 * Analyses and visualizations remain consistent, regardless of map
distortions in the underlying projection.

#### Accuracy
Hex9 supports near-lossless forward and inverse mappings between
grid addresses and geodetic coordinates.
Accuracy is maintained even at extreme resolution: At layer 30, hexagons are accurate to
about 1µm. Example round-trip accuracy for several landmarks:

*
The grid addresses below might be non-canonical (still prone to change).
Current Grid address semantics are as follows.
The Great Pyramid at Giza is identified at 29°58'45.817792004858"N, 31°8'3.457294813097"E
This can be projected to a (reversible) hex-grid address at any given layer.
Here it is at layer 36.
`0070143470686461861005464283175018506`
Broken down..
`0-0701434706864618610054642831750185-06`
The first digit is the root hexagon at layer 0.  This is one of 12 hexagons that cover the
earth's surface, so the range of values are 0...B (where A,B = hexagons 10,11 respectively).
07...185  These are the sub-hexagons at each respective layer; so as there are 35 of those, along with
the root hexagon, we know this address is at layer 36.
`06` - these act as 'metadata tail', and provide the minimum amount of metadata to convert the address back to
it's longitude/latitude. (The calculations for this are found in h9/addressing.py).

There is also a 'key' version of the address.  This is used for identifying which points are within a specific hexagon
when hex-binning (a main purpose for this sort of grid). In this case, the 'metadata tail' is reduced.
In this case, the great pyramid is:
`0070143470686461861005464283175018500`
Importantly, and despite best efforts,
the hexagon key of a higher layer cannot be solely derived by chopping the string.
The c2 identity is an important aspect for decoding ids 6,7,8
Likewise the octant face mode is very useful.
For the first 11 layers, the pyramid address key is as follows.
0:00
1:002
2:0072
3:00704
4:007012
5:0070140
6:00701430
7:007014344
8:0070143474
9:00701434700
10:007014347064
*

**Great Pyramid**
```
29°58'45.817792004858"N, 31°8'3.457294813097"E (Reference Coordinates)
29°58'45.817792004871"N, 31°8'3.457294813071"E (Roundtrip Coordinates)
0070143470686461861005464283175018506 (L35 Grid Address)
∂0.984436nm (roundtrip via GCD<->Hex9 Label) in Geodesic distance (nanometres)
```

**Stonehenge**
```
51°10'43.672800075871"N, 1°49'34.283450385600"W (Reference Coordinates)
51°10'43.672800075845"N, 1°49'34.283450385640"W (Roundtrip via Grid Address)
4352164061084274326815104253457062812 (L35 Grid Address)
∂1.314516nm delta (Geodesic.DISTANCE)
```

**Moai on Rapa Nui**
```
27°7'32.827199567155"S, 109°16'36.740870832014"W (Reference Coordinates)
27°7'32.827199567142"S, 109°16'36.740870832014"W (Roundtrip via Grid Address)
b501888670665528318830382731266380195 (L35 Grid Address)  
0.352774nm  delta (Geodesic.DISTANCE)
```

**North Pole** (edge case)
```
90°0'0.000000000000"N, 0°0'0.000000000000"E (Reference Coordinates)
89°59'59.999999999898"N, 12°16'35.957736079345"E (Roundtrip via Grid Address)
422222222222222222222222222222224832a (Grid Address)
∂3.174534nm  delta (Geodesic.DISTANCE)
```
#### Intuitive uint64 Addresses
Hex9 supports various `uint64` addresses in a directly intuitive manner.

For example, one of the Nazca Spirals - at `14.679806S, 75.101925W` has the
uint64 address `0xb404124850835306` (in hexadecimal).
'b' here represents hexagon 11
This is (by default) a 'Layer 13' Address; The hexagons are in bytes 0..13
The final 2 bytes are the meta, used to convert the address back to another location.
To convert the address to an identifying hex key, this can be done by masking the final
byte with 0x70 (or using a far more reliable internal method).

```
0x40412...
  ↑↑↑↑↑
  ||||└─── Layer 4, hexagon 1
  |||└──── Layer 3, hexagon 4
  ||└───── Layer 2, hexagon 0
  |└────── Layer 1, hexagon 4
  └─────── Hexagon 'B'
```
When the uint64 address is depicted in hexadecimal, the global address is
revealed and may be readily eye-balled with a crib - see the following image
that traces the first 5 regions 8,2,5,4,A - each one covering 1/9th the area of
the preceding layer.

#### Performance
Hex9 efficiently handles large datasets. For example, *25 million sparse
GCD points can be mapped in far less than 20 minutes* on standard desktop hardware.

#### Summary
Thanks to its decoupled, fractal-based structure,
Hex9 allows direct projection of spatial data onto hexagonal grids.
It has strong roundtrip integrity.
It is *reasonably* authalic, with a global RSME of about 7%, but smooth authalic change - meaning that at local levels,
data sampling is representative.