GEOID-Flood

A large-scale multi-modal benchmark for flood segmentation, built from 219 Copernicus EMS Rapid Mapping flood activations across 65 countries (2016-2026).

134,596 rasters / 329 GB across 190 shards. 12,236 tiles: 7,496 train / 1,022 val / 2,417 test, plus 1,301 held-out tiles from activations EMSR857-871 for out-of-distribution evaluation.

Each 1024x1024 tile at 10 m resolution provides co-registered Sentinel-1 GRD and RTC (pre- and post-event, VV/VH, the same acquisitions in both products), a pre-event Sentinel-2 L2A image, a Copernicus GLO-30 DEM, and manually checked three-class labels (0 = background, 1 = permanent water, 2 = flooded water).

v1.1 (current) removes unusable, duplicated and wrongly labelled tiles, fixes Sentinel-1 and Sentinel-2 issues and adds provenance metadata to every file. See CHANGES_v1.1.md. The paper describes v1.0, which stays available at the v1.0 tag (get_data.py --revision v1.0).

Three GEOID-Flood events, all layers of a tile side by side. Flooded water is cyan, permanent water blue, invalid pixels gray.

The 219 activations span 65 countries. Splits are assigned per area of interest and touching AoIs share a split, so no area of interest straddles train, validation and test.

Layout

Rasters ship as uncompressed tar shards of Cloud-Optimized GeoTIFFs, grouped so you can download only the modalities and splits you need:

{tree}/shards/{split}/{layer}/{split}-{layer}-NNNNN-of-NNNNN.tar

Tar members carry the canonical path (EMSR151-1/s1grd/EMSR151-1-0_s1grd_post_*.tif), so extracting any shard into a tree root rebuilds the layout the training configs expect:

tar -xf train-s1grd-00000-of-NNNNN.tar -C data/geoid-flood/

Download

get_data.py fetches shards and unpacks them into the expected structure, deleting each shard as soon as it is unpacked. It checks free space before starting and resumes if interrupted.

pip install huggingface_hub tqdm
python get_data.py --dest data                                     # everything
python get_data.py --dest data --layer s1grd label                 # S1-GRD benchmark
python get_data.py --dest data --tree geoid-flood-heldout \
                   --layer s1rtc label                             # held-out eval only
python get_data.py --list --layer s1grd s2l2a dem label            # preview, no download
python get_data.py --dest data --layer s1grd --workers 8           # more shards in flight

Download v1.1 into a fresh --dest: do not mix v1.0 and v1.1 files.

Shards are fetched over the Xet protocol, several at a time. Peak disk usage is the size of your selection plus roughly --workers x 2 GB for the shards in flight.

Typical selections:

selection	flags	size
everything	(no flags)	~329 GB
S1-GRD single-image benchmark	--layer s1grd label	~91 GB
...train+val only	--layer s1grd label --split train val	~63 GB
early/mid fusion (S1+S2+DEM)	--layer s1grd s2l2a dem label	~241 GB
held-out S1-RTC evaluation	--tree geoid-flood-heldout --layer s1rtc label	~10 GB

Without get_data.py

The shards can also be fetched with the Hugging Face CLI and unpacked with tar. Select layers and splits with --include patterns; every shard is on disk before unpacking, so peak usage is about twice your selection:

pip install -U huggingface_hub
# e.g. the S1-GRD benchmark: metadata + s1grd + label shards of the main tree
hf download links-ads/geoid-flood --repo-type dataset --revision v1.1 --local-dir data \
  --include "geoid-flood/*.csv" --include "geoid-flood/*.parquet" \
  --include "geoid-flood/shards/*/s1grd/*" --include "geoid-flood/shards/*/label/*"
# unpack each shard into its tree, then delete it
cd data && for t in */shards/*/*/*.tar; do tar -xf "$t" -C "${t%%/*}" && rm "$t"; done && rm -r */shards

Modalities

Every tile carries the same nine layers, all co-registered on one 1024x1024 event-UTM grid at 10 m. Four are imagery, five are derived masks. Every GeoTIFF names its bands; Sentinel-1, Sentinel-2 and cloud-mask files list the Copernicus products they were made from in the SOURCE_PRODUCTS tag (a JSON list).

layer	bands	passes	size	what it is
s1grd	2 x float32	pre + post	90.5 GB	Sentinel-1 GRD backscatter as linear sigma0, bands VV, VH. One acquisition before the event and one after. The loader converts to dB (10*log10) on read. Stored rounded to 10 mantissa bits (ZSTD), at most 0.0021 dB from the original after conversion to dB.
s1rtc	2 x float32	pre + post	88.4 GB	The same two acquisitions, radiometrically terrain-corrected, also linear sigma0, same storage. NaN marks pixels outside the valid swath.
s2l2a	12 x uint16	pre	142.5 GB	Pre-event Sentinel-2 L2A surface reflectance, bands B01…B12 (incl. B8A). A per-pixel mosaic: each pixel takes the scene closest to the CEMS pre-event mapping date with valid data. Harmonized to processing baseline 04.00 (no +1000 offset); reflectance = DN x 0.0001, recorded as the band scale. 0 = no data.
dem	1 x float32	static	7.6 GB	Copernicus GLO-30 elevation in metres, resampled from 30 m onto the 10 m tile grid.
label	1 x uint8	static	0.1 GB	The training target. Manually checked three classes: 0 background, 1 permanent water, 2 flooded water. 255 marks pixels outside the mapped area and is the ignore index.
cloudmask	1 x uint8	pre	0.1 GB	Cloud and shadow over the S2 image, from OmniCloudMask: 0 clear, 1 thick cloud, 2 thin cloud, 3 shadow. Not folded into label, so this is the only per-pixel record of cloud.
floodmask	1 x uint8	static	0.1 GB	Binary CEMS Rapid Mapping flood delineation used to build label. Retains extent the three-class label dropped.
permwater	1 x uint8	static	0.1 GB	Binary permanent-water extent used to build label. No released config reads it.
validity	1 x uint8	static	0.1 GB	Binary per-pixel validity: 1 where the tile was imaged and mapped. Largely redundant with label == 255.
			329 GB	134,596 rasters

sample/ holds two complete event-AoIs from activation EMSR712 -- EMSR712-10 (train split) and EMSR712-3 (test split) -- with all nine layers for all 47 tiles, 8,076 chip rows, so the loader can be exercised end to end without a full download. It extracts to the same canonical paths, so the only override needed is --data.init_args.metadata_filename data_tiles_s256_st128_sample.csv.

Normalization statistics

Recomputed on the v1.1 train split, over valid pixels. S1 in dB as the loader reads it (10*log10 of linear sigma0), pre and post pooled; S2 in DN (multiply by 0.0001 for reflectance).

product	band	mean	std
s1grd	VV	-10.358	4.919
s1grd	VH	-17.455	6.184
s1rtc	VV	-10.586	4.651
s1rtc	VH	-17.680	5.936

S2 band	B01	B02	B03	B04	B05	B06	B07	B08	B8A	B09	B11	B12
mean	555.5	668.2	954.0	1036.1	1465.4	2315.6	2633.1	2759.9	2857.4	2892.9	2349.3	1640.7
std	800.9	843.6	853.7	1019.8	991.2	1049.8	1163.4	1220.7	1205.8	1261.6	1177.8	1101.8

Metadata

file	role
data_tiles_s256_st128.csv	the only metadata the dataloader reads. Enumerates 256x256 chips at stride 128 (train) / 256 (val, test), with valid_proportion, positive_proportion, cloud_cover and split per chip
tile_catalog.parquet	the 1024x1024 tile inventory, one row per shipped tile: geometry, UTM CRS, event and delineation times, invalid_pixel_frac (share of invalid pixels in validity), split, countries, countries_set, macro_areas_set, and the Copernicus products of each tile (s1_pre_products, s1_post_products, s2_pre_products, the same names as the GeoTIFF tags). Nothing reads it at training time; query it to select events by geography or date

The tile catalog, the chip CSV and the shipped rasters list the same 12,236 tiles.

Known limitations

• label is not masked under clouds; use cloudmask if you need to.
• Some RTC files are smoother than 10 m (lower-resolution processing of the right acquisition).

Citation

See CITATION.cff in the code repository.

License

The GEOID-Flood compilation, splits and labels: CC BY 4.0. Code: MIT, in the code repository. The dem layer is licensed separately (below).

Contains modified Copernicus data. Carry these notices forward when redistributing:

• Modified Copernicus Sentinel-1 and Sentinel-2 data (2016-2026).
• Copernicus Emergency Management Service Rapid Mapping products, © European Union.
• The dem layer is a resampled Copernicus WorldDEM-30 (instance COP-DEM-GLO-30-F), used under the Copernicus WorldDEM-30 licence.