Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

PACE OCI L2 AOP V3_1 pre-training corpus — part 2

This is the second half of one corpus. It was split across two repositories only because a single Hugging Face account ran out of storage. The shards never overlap, and the file layout and naming are identical, so you can merge the two repos by copying train/ and val/ from each into the same directory.

repository holds
TalhaMah56/PACE_v3_1 train/shard_14503.tartrain/shard_24202.tar (9,700 shards). Full dataset card, measured limitations and provenance are there.
TalhaMah56/PACE_v3_1_part2 (this repo) train/shard_00000–00359 (360 shards) and train/shard_24203–24497 (295 shards). No val/.

Complete corpus

split shards range size
train/ 24,498 shard_00000shard_24497 ~20.5 TB
val/ 1,332 shard_00000shard_01331 ~1.1 TB

The two splits are separated by granule hash (5 %), so no acquisition appears in both. That also means you cannot rebuild val/ from train shards.

⚠️ What is here, and what is not

The upload stopped on 2026-09-14 when the account hit its storage limit. The two repos together hold 10,355 of the 24,498 train shards and none of the 1,332 val shards:

split full corpus PACE_v3_1 this repo on neither
train/ shard_00000–24497 shard_14503–24202 shard_00000–00359, shard_24203–24497 shard_00360–14502 (14,143 shards)
val/ shard_00000–01331 all 1,332

Every shard in either repo was checked against the local copy before that copy was deleted (sizes match, and a random sample matched byte for byte). The shards on neither repo still exist offline; open a discussion if you need them.

Read this before using the data

  • Product version V3_1 has been withdrawn by NASA. OB.DAAC now serves only version 3.2, and V3_1 cannot be downloaded again. Going from V3_1 to V3_2 changes Rrs by a median of 3.3 % per band and 1.03° median SAM, so a model trained here and tested on current PACE files has a hidden train/test radiometry mismatch.
  • The corpus has measured problems: the quality score mostly measures missing pixels, turbid water is 1.6 % of tiles, invalid pixels are stored as zero, and l2_flags and Rrs_unc were not kept. All of these are described with numbers on the PACE_v3_1 card.

Format

Each .tar shard holds patches: Rrs as int16 with a scale factor (172 bands), a one-bit per-pixel validity mask, and per-patch metadata (lat/lon, year/day-of-year, solar zenith, flag fractions, per-granule quality terms).

Other datasets from the same project

repository what it is
TalhaMah56/PACE_v3_1 part 1 of this corpus
TalhaMah56/pace-s2-emit-triplets PACE × Sentinel-2 × EMIT scenes on a shared grid
TalhaMah56/pace-s2-emit-sr-tiles super-resolution training tiles cut from those scenes

Attribution

NASA PACE OCI Level-2 Apparent Optical Properties (V3_1), courtesy of the NASA Ocean Biology Distributed Active Archive Center (OB.DAAC). PACE data are NASA public-domain products.

Downloads last month
92