Skip to content

Overlap Group -- Centre

rasterGeoTIFFEPSG:32633tinybundled

One of three small rasters sharing a CRS and a pixel grid, with deliberate partial overlap and a distinct constant value each. This is the middle member, and the second red band -- the alias ambiguity. The three exist only as a group: odc.stac.load and stackstac.stack take a sequence of Items and their whole reason for existing is what happens across it, which no standalone raster case in this corpus could express.

Pixels of overlap_group_centre, a 12x12 raster, with NoData in magenta
1 band, 12x12 px, float32, sentinel NoData. Rendered from the case's actual pixels, contrast-stretched for display; NoData is shown in magenta.
Property Value
Case ID overlap_group_centre
Category raster
Format GeoTIFF
CRS EPSG:32633
Location Southern Italy / Sicily (synthetic, UTM 33N) — 15.00°E, 40.65°N → 15.01°E, 40.65°N
Test tier unit
Size class tiny
Storage class bundled
Redistributable yes
Loader rasterio
Status validated

Use this case

import pytest


@pytest.mark.geocase_case("overlap_group_centre")
def test_overlap_group_centre(geocase_case) -> None:
    data = geocase_case.load()
    assert data is not None

Use GeoCase in your tests

Install the complete set of vector, raster, and NetCDF dependencies:

pip install "geocase[all]"

View GeoCase on PyPI.

What this case checks

Confirm that a consumer compositing this group across its overlaps produces a result whose provenance is readable: each member carries one constant value, so which member won a given pixel is visible by inspection rather than by arithmetic.

Risk types covered

Expected behavior

Assertion Expected
expect_loadable yes
expect_crs yes
expected_epsg 32633
expect_nodata yes
expected_band_count 1
expected_dtype float32
expected_shape [12, 12]
expected_nodata_value -9999.0
nodata_convention sentinel
expected_band_names red

Known answer

Computed from the actual bytes and gated against them. Grade your own output against these.

Quantity Value
Mean over valid pixels 20.0
Mean including NoData -49.5763888889
NoData pixels 1
Bounds (case CRS) [500180.0, 4499460.0, 500540.0, 4499820.0]

Notes

Three small rasters that exist only as a group. They share a directory for the same reason the footprint edge cases and the transform conventions do: their value is entirely in the relationship between them, and no one of them means anything alone.

Why these exist

Every other raster case in this catalog is one standalone file. That made four things unexpressible:

  • stacking order,
  • mosaic compositing,
  • temporal grouping,
  • and two assets in one STAC Item.

odc.stac.load and stackstac.stack both take a sequence of Items, and what happens across that sequence is their whole reason for existing. The 2026-08-31 round-2 validation run could only ever hand them a list of one, so the entire surface went untested — and it is the surface those two libraries are.

The geometry

Case Stack order Constant value Band alias Nodata corner
overlap_group_north 1 10.0 red top-left
overlap_group_centre 2 20.0 red top-right
overlap_group_south 3 30.0 nir bottom-left

Each is 12×12 at 30 m in EPSG:32633, offset from its predecessor by 6 cells diagonally — half its width. Four properties are load-bearing and each is gated in tests/unit/test_raster_groups.py:

  • Partial overlap. Disjoint members never composite; identical members make compositing unobservable. Consecutive members overlap in a quarter of their area.
  • A shared pixel grid. Off-grid members would turn every compositing difference into a resampling difference, which is a different finding on a different axis.
  • Distinct constant values. "Which pixel won" is then readable by inspection rather than by arithmetic.
  • One nodata corner each, at a different corner. A composite's fill behaviour is visible as well as its ordering.

The shared band alias

Two of the three declare common_name: red. That is the ordinary Sentinel-2 shape — several assets legitimately carrying the same alias — and it is what makes a consumer resolving the alias silently to the first candidate visible rather than invisible. odc-stac does exactly that today, and its source carries its own # maybe warn about ambiguity? note.

The two red members carry different constant values (10.0 and 20.0), so resolving to the wrong one is a wrong value, not merely a wrong provenance.

Reading them as STAC Items

from geocase.stac import items_for_cases

items = items_for_cases(
    include_ids=[
        "overlap_group_north",
        "overlap_group_centre",
        "overlap_group_south",
    ],
    assets="per_band",
)

geocase.stac normalises proj:bbox and emits both proj:epsg and proj:code, so the same list is byte-identical input for stackstac and odc-stac. See docs/plans/38-six-consumer-round-2-and-the-stac-adapter.md.

Required capabilities

  • load
  • multi-item-stack
  • nodata-check

Files

Browse this case on GitHub

Source and license

  • Source: geocase-curated
  • License: MIT

Tags

delivery:multi-file geotiff group mosaic overlap raster stac