Himawari AHI ARCO Index

This Dataset repo is the lightweight index and documentation surface for the Himawari AHI ARCO collection. The large Zarr payloads live in the public Hugging Face Storage Bucket:

hf://buckets/Gary0404/himawari-arco/arco_v1/

The repo keeps small metadata files such as *.manifest.json and *.coverage.json. New formal ARCO payloads should be published to the bucket, not to this Dataset repo.

Available Bucket Payloads

B03:
  bucket prefix: hf://buckets/Gary0404/himawari-arco/arco_v1/B03/
  months: 2023-01 through 2025-11
  chunks: time/y/x = 6 x 1024 x 1024

B08:
  bucket prefix: hf://buckets/Gary0404/himawari-arco/arco_v1/B08/
  months: 2023-01 through 2025-12
  chunks: time/y/x = 6 x 1024 x 1024

B13/B09/B10 formal products are planned as bucket-first products after canary validation. Pilot payloads may still appear in this repo until they are either migrated or retired.

Direct Bucket Access

import xarray as xr
from huggingface_hub import HfFileSystem

band = "B08"
month = "2024-07"
store = f"himawari_ahi_{band}_{month}_arco_v1.zarr"

fs = HfFileSystem()
mapper = fs.get_mapper(
    f"buckets/Gary0404/himawari-arco/arco_v1/{band}/{store}"
)
ds = xr.open_zarr(mapper, consolidated=True, chunks={})

patch = ds[band].isel(
    time=slice(0, 24),
    y=slice(1024, 1152),
    x=slice(1536, 1664),
).load()
print(ds.sizes, patch.shape)

Store Contents

Each monthly ARCO Zarr store contains:

{band}(time, y, x): Himawari AHI band values
source_present(time): 1 if a source observation exists for the target time
source_time_index(time): compact source time index, or -1 for missing source time
time/y/x: coordinates
spatial_ref: geostationary projection metadata

Missing source times are kept on the regular 10-minute UTC time grid. For those times, the band array reads as NaN, source_present is 0, and source_time_index is -1.

Repository Role

This repo is intentionally small:

keep here:
  README.md
  .gitattributes
  *.manifest.json
  *.coverage.json
  small validation or example files

do not keep here as the steady state:
  full monthly *.zarr payload directories

The large B03/B08 Zarr payloads have been copied and validated in the bucket. They are removed from this repo so the Dataset repo remains a lightweight catalog instead of a second large storage copy.