zarrita-hcc

zarrita codecs for the two Zarr v3 serializers of healpix-compress that store Sentinel-2 reflectances, decode-only:

• hcc-sentinel2-pyramid (lossy: the parent level as lossless digital numbers, the children predicted from it plus a quantised detail, all JPEG XL coded);
• hcc-sentinel2 (lossless digital numbers in JPEG XL).

Chunks decode to float32 or float64 reflectances, with no-data cells as NaN. JPEG XL is decoded by jxl-oxide-wasm.

Use

npm install github:GRID4EARTH/zarrita-hcc#v0.1.0

import { registry } from "zarrita";
import { registerHcc } from "zarrita-hcc";

registerHcc(registry);

registerHcc registers both codec names lazily: the codec code and the JPEG XL WASM (1.7 MB, 616 kB gzipped) are only loaded the first time an array uses them. In the browser the WASM is fetched from next to the jxl-oxide-wasm module (new URL(..., import.meta.url), which bundlers such as Vite rewrite into an asset URL). Under Node, hand the bytes over first:

import { readFileSync } from "node:fs";
import { createRequire } from "node:module";
import { dirname, join } from "node:path";
import { setJxlWasmSource } from "zarrita-hcc";

const require = createRequire(import.meta.url);
const jxlDir = dirname(require.resolve("jxl-oxide-wasm/package.json"));
setJxlWasmSource(readFileSync(join(jxlDir, "jxl_oxide_wasm_bg.wasm")));

Develop

npm install
npm test
npm run typecheck
npm run build # writes dist/, which is committed

dist/ is committed so that the package can be installed straight from GitHub; rebuild it with every change to src/.

The decoders are tested against vectors produced by the Python reference implementation. scripts/make_hcc_fixtures.py regenerates them into tests/__fixtures__/ from a Python environment where hcc (healpix-compress) and imagecodecs are installed; see its docstring. Without --store it fetches nothing and decodes the real chunks already in the fixtures again.

License

Apache-2.0, like healpix-compress.