Open imagery (Sentinel)¶

Labels need images. Rapid Mapping products are often made from commercial satellites, which are not open:

products = await ems.products("EMSR664")
[(i.sensor_name, i.resolution_class) for i in products[0].images]
# [('RADARSAT-2', 'HR+')]   others: COSMO-SkyMed, PAZ, Pléiades…

So copernicus-ems finds the nearest open acquisitions over each product's AOI, Sentinel-1 (radar: sees through clouds, the usual choice for floods) and Sentinel-2 (optical), in the public STAC catalogue of the Copernicus Data Space Ecosystem. Searching needs no account and no extra dependency; downloading a scene needs a free Data Space account.

From the command line¶

copernicus-ems scenes EMSR664 --aoi 1 --optical --max-cloud 30 --days 2

AOI01_DEL_MONIT01  2023-05-18 16:55  made from RADARSAT-2 (HR+)
     +0.1 h  sentinel-1-grd  1 scenes  AOI 14%   S1A_IW_GRDH_1SDV_20230518T165837_…
    -35.7 h  sentinel-1-grd  1 scenes  AOI 1%    S1A_IW_GRDH_1SDV_20230517T051134_…
AOI01_GRA_PRODUCT  2023-05-22 10:27  made from Pleiades (VHR1)
     -5.1 h  sentinel-1-grd  1 scenes  AOI 100%  S1A_IW_GRDH_1SDV_20230522T051946_…
    +23.7 h  sentinel-2-l2a  3 scenes  AOI 100%  cloud 8%  S2A_MSIL2A_20230523T100601_…

--json prints one object per pass, with the download URLs. The command needs the geo extra for the AOI coverage.

From Python¶

from datetime import timedelta
from copernicus_ems import SENTINEL_1_GRD, SENTINEL_2_L2A, passes
from copernicus_ems.geo import coverage  # [geo] extra

events = await ems.observed_events("EMSR664")
for event in events:
    matches = await ems.scenes(
        event.product,
        collections=[SENTINEL_1_GRD, SENTINEL_2_L2A],
        within=timedelta(days=2),
        max_cloud_cover=30,
    )
    for p in passes(matches):  # slices of one satellite pass, grouped
        print(
            event.product.label,
            f"{p.gap_hours:+.1f} h",
            p.collection,
            f"{coverage(p.geometry, event):.0%} of the AOI",
            [s.download_url for s in p.scenes],
        )

find_scenes / RapidMapping.scenes return scenes, nearest in time first; each SceneMatch.gap is scene time minus the product's observed_time.
passes() groups them by pass (eopf:datatake_id): a catalogue scene is a slice (Sentinel-1) or a tile (Sentinel-2), and one pass often needs several to cover an AOI.
geo.coverage(footprint, event) is the share of the AOI a scene or a pass covers.

Choose by time and coverage¶

The same AOI on EMSR664 shows why both matter:

Product	Nearest pass	AOI covered	Best full pass
AOI01_DEL_MONIT01 (18 May 16:55)	Sentinel-1, +6 min	14%	none within ±2 days
AOI01_GRA_PRODUCT (22 May 10:27)	Sentinel-1, −5.1 h	100%	the same
AOI01_DEL_MONIT03 (21 May 16:50)	Sentinel-1, +12.5 h	100%	the same

The gap is part of the label

Water moves. A scene 12 hours after the observation can show a flood that has grown or receded: keep gap_hours with each image–label pair, and pick a threshold for your use. Sentinel-2 cloud_cover is for the whole tile, not the AOI.

Pairing with the label rasters¶

With a CDSE account, one command

copernicus-ems pairs EMSR664 asks the Sentinel Hub Process API for the backscatter on the labels' grid and writes both: see Sentinel-1 backscatter. The recipe below is for scenes you download yourself.

Read the scene (e.g. its vv / vh Cloud-Optimised GeoTIFF with rasterio, after authenticating to the Data Space), then rasterize the labels on the scene's grid, so pixels line up:

from copernicus_ems.ml import Grid, rasterize

with rasterio.open(path) as src:  # your Sentinel-1 scene, reprojected to UTM
    grid = Grid.from_transform(src.crs, src.transform, src.width, src.height)
labels = rasterize(event, grid=grid)  # 1 flooded, 0 dry, 255 not observed
labels.to_geotiff(Path("pairs/labels.tif"))