# SkyClimate: existing-data release and V4 demo

This is the dataset we can publish now, not a claim that the full augmented Atlas is finished.
The parent datasets and their scientific claims are unchanged. No missing month was filled,
no new atmospheric calculation was run, and no calibration was deployed for this publication.

## What is included

Cloud cover (total, low, middle, high) and total column water vapour (TCWV): global ERA5,
2018–2025, hourly, 0.25° source grid. All 96 monthly histories, spatial summaries and the
existing full-catalogue cloud-only and TCWV temporal descriptors are accessible.
TCWV is the integrated atmospheric water-vapour column, not cloud cover and not a
site-altitude-adjusted PWV estimate. Low/middle/high cloud are model layers, not resolved 3D clouds.

Aerosols: global CAMS, 2018–2025, every three hours, 0.75° grid, at 500, 550, 1064,
1550 and 1640 nm. The complete 96 monthly histories, 120 statistical/descriptor tiles
and descriptor normalisation are included. 550 and 1640 nm are native channels;
other wavelengths use the retained spectral-transfer method. AOD is dimensionless
column extinction (scattering plus absorption), not a directional link loss in dB.
The frozen global 1550/1640 tests passed; 500/550/1064 did not, and the mountainous-site
test did not pass at any band. These are not generally validated or locally calibrated products.

V4 turbulence: **January 2018 reference-summary demo only**, 248 three-hourly samples,
global 0.25°. The six medians and their finite sample counts cover integrated Cn²,
r0, theta0, tau0, reference Rytov variance (optical fields at 1550 nm), and PWV above
the model surface. The original authority remains phase-F evaluation-only: this attachment
does not admit a full engineering turbulence release. Surface-layer turbulence is
under-represented; r0 may be optimistic. Reference Rytov is not an arbitrary-terminal
scintillation prediction. This demo is neither eight-year climatology nor a CFLOS product.

## Open and download

Open [the GeoLibre project](./skyclimate.geolibre.json) in GeoSpatial Studio or GeoLibre.
Its 16 initial layers cover all variables; [STAC](./catalog.json) lists all 112
overall statistical views. Each COG preserves the full source-grid values and missingness
with lossless compression; overview pixels are nearest-neighbour display aids.
Cell centres include the poles; the small half-cell edge beyond ±90° is only raster encoding.
No smoothing or resolution improvement is claimed. The Explorer's PNG/u16 map caches are
separate, quantised display files and are not the analytical data.

[native-data.json](./native-data.json) is the index of all hourly and three-hourly packs
and the three separate descriptor families. For each group, read its `index` JSON, then
append the object's relative name to `public_data_prefix` under this release URL:
`https://data.dynamikorbits.com/skyclimate/existing-data-2026-09-06-v2/`. Each object has its original SHA-256 and byte count. Access is anonymous
HTTPS, GET/HEAD, byte ranges and CORS; there is no bucket-listing or credential endpoint.
The public gateway fails closed if an object's stored checksum metadata or size differs.
The indexes reference existing R2 objects; there is no redundant full-corpus download or copy.
These exact keys are retained public-release dependencies, not disposable derived scratch.
Scripted clients should send an explicit application User-Agent (for example
`SkyClimate-client/1.0`); the existing Cloudflare policy rejects Python urllib's
default bot-like identifier. No access token is required.

The current analytical packs are the existing Parquet/binary temporal layout and NetCDF
statistics, **not newly converted Zarr**. Parquet row-group reads make one spatial tile
available without downloading a whole month. See each parent manifest for dtype, UTC
axis, quantisation and missing-value rules before decoding its binary `payload` column.
For hourly packs use `series/<variable>/tiles-XXXX.parquet` and its explicit tile row;
for aerosols use `series-aod1550.parquet` (or the other band) and the required row group.
Always verify the row's payload SHA-256 before reshaping. The reader implementations are
`r2_hourly_temporal_source.read_r2_hourly_series_tile` and
`aerosol_spectral_dossier._read_aerosol_temporal_tile` in the linked source repository.
Do not treat missing values as zeros or replicate coarse cells into finer observations.
The historic `exact-sites.parquet` pilot is deliberately not exposed; all global grid
cells remain accessible and descriptors use the full retained catalogue, not 186 sites.

The cloud and TCWV descriptors each have 178 nonempty spatial shards, with `candidates`,
`cells`, `descriptors` and `bitsets` files plus a global normalisation. Aerosol descriptors
use 120 native-grid tiles and their own normalisation. Read candidate-to-source-cell
joins and support flags; H3 resolution 5 is a catalogue key, not atmospheric resolution.
These deterministic retrieval descriptors are not learned embeddings, site rankings,
or a universal vector joining unrelated physical families.

## Licence and attribution

The combined SkyClimate distribution is [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/).
Commercial use requires a separate licence from Dynamik Orbits; contact the maintainer
through [the project repository](https://github.com/DynamikOrbits/optical-ground-siting).
Original Copernicus source rights remain distinct and unchanged. Contains modified
Copernicus Climate Change Service information [2026] and Copernicus Atmosphere Monitoring
Service information [2026]. Neither the European Commission nor ECMWF is responsible
for any use that may be made of the Copernicus information or data this product contains.
AERONET is validation context, not republished station observations; acknowledge its
principal investigators and follow its data-use policy for separately obtained station data.

## Deferred, without changing this release

Full 2018–2025 V4 turbulence, RF/rain/wind extensions, further calibration and new paid
compute are deferred. Retained restart plans, budget lessons, provenance and original
method identities remain in the repository. Publishability is not scientific validation.
SkyDome owns directional rays and CFLOS; neither is claimed here.
