raster2dggs 0.13.0

No project description provided

raster2dggs

Python-based CLI tool to index raster files to DGGS in parallel, writing out to Parquet.

This is the raster equivalent of vector2dggs.

Currently this supports the following DGGSs:

• H3
• rHEALPix
• S2
• A5
• Via DGGAL: ISEA4R, ISEA9R, ISEA3H, ISEA7H, IVEA4R, IVEA9R, IVEA3H, IVEA7H, RTEA4R, RTEA9R, RTEA7H, HEALPix, rHEALPix

And these geocode systems:

• Geohash
• Maidenhead Locator System

Contributions (particularly for additional DGGSs), suggestions, bug reports and strongly worded letters are all welcome.

Contents

• Installation
• Usage
• Sampling strategies
  • Point sampling (default)
  • Overlay (area-based)
  • Windowed resampling
• Visualising output
  • DuckDB
  • GDAL
  • QGIS
• Installation (detailed)
  • For development
  • Tests
  • Generating synthetic sample rasters
  • Experimenting
• Example commands
• Citation

Installation

This tool makes use of optional extras to allow you to install a limited subset of DGGSs.

If you want all possible:

pip install raster2dggs[all]

If you want only a subset, use the pattern pip install raster2dggs[a5] (for one) or pip install raster2dggs[h3,s2,isea4r] (for multiple).

A bare pip install raster2dggs will not install any DGGS backends.

Usage

raster2dggs --help

Usage: raster2dggs [OPTIONS] COMMAND [ARGS]...

Options:
  --version  Show the version and exit.
  --help     Show this message and exit.

Commands:
  a5          Index raster data into the A5 DGGS
  geohash     Index raster data into the Geohash DGGS
  h3          Index raster data into the H3 DGGS
  healpix     Index raster data into the HEALPix DGGS
  isea4r      Index raster data into the ISEA4R DGGS
  isea7h      Index raster data into the ISEA7H DGGS
  isea9r      Index raster data into the ISEA9R DGGS
  ivea4r      Index raster data into the IVEA4R DGGS
  ivea7h      Index raster data into the IVEA7H DGGS
  ivea9r      Index raster data into the IVEA9R DGGS
  maidenhead  Index raster data into the Maidenhead DGGS
  rhp         Index raster data into the rHEALPix DGGS
  rtea4r      Index raster data into the RTEA4R DGGS
  rtea7h      Index raster data into the RTEA7H DGGS
  rtea9r      Index raster data into the RTEA9R DGGS
  s2          Index raster data into the S2 DGGS

raster2dggs h3 --help

Usage: raster2dggs h3 [OPTIONS] RASTER_INPUT OUTPUT_DIRECTORY

  Ingest a raster image and index it to the H3 DGGS.

  RASTER_INPUT is the path to input raster data; prepend with protocol like
  s3:// or hdfs:// for remote data. OUTPUT_DIRECTORY should be a directory,
  not a file, as it will be the write location for an Apache Parquet data
  store, with partitions equivalent to parent cells of target cells at a fixed
  offset. However, this can also be remote (use the appropriate prefix, e.g.
  s3://).

Options:
  -v, --verbosity LVL             Either CRITICAL, ERROR, WARNING, INFO or
                                  DEBUG  [default: INFO]
  -r, --resolution [0-15|smaller-than-pixel|larger-than-pixel|min-diff]
                                  H3 resolution to index. Accepts an integer
                                  in [0, 15] or an auto-detection mode:
                                  'smaller-than-pixel' (first resolution finer
                                  than a pixel), 'larger-than-pixel' (last
                                  resolution coarser than a pixel), or 'min-
                                  diff' (resolution closest to pixel size).
                                  [required]
  -pr, --parent_res INTEGER RANGE
                                  H3 parent resolution to index and aggregate
                                  to. Defaults to max(0, resolution - 6)
                                  [0<=x<=15]
  -b, --band TEXT                 Band(s) to include in the output. Can
                                  specify multiple, e.g. `-b 1 -b 2 -b 4` for
                                  bands 1, 2, and 4 (all unspecified bands are
                                  ignored). If unused, all bands are included
                                  in the output (this is the default
                                  behaviour). Bands can be specified as
                                  numeric indices (1-based indexing) or string
                                  band labels (if present in the input), e.g.
                                  -b B02 -b B07 -b B12.
  -n, --nodata [omit|emit]        'omit' excludes nodata cells from output
                                  (default). 'emit' includes them, writing the
                                  source raster nodata value (or
                                  --nodata-fill if set). Note: non-NaN
                                  emitted values participate in cell
                                  aggregation (see -a/--agg); if this is
                                  undesired, ensure your source nodata is NaN
                                  or override with --nodata-fill.
                                  [default: omit]
  --nodata-fill NUMBER            Override the value written for nodata cells
                                  when --nodata=emit. If omitted, the source
                                  raster nodata value is used (NaN if none is
                                  defined). Coerced to the output dtype. Note:
                                  non-NaN values participate in cell
                                  aggregation (see -a/--agg).
  -c, --compression TEXT          Compression method to use for the output
                                  Parquet files. Options include 'snappy',
                                  'gzip', 'brotli', 'lz4', 'zstd', etc. Use
                                  'none' for no compression.  [default:
                                  snappy]
  -t, --threads INTEGER           Number of threads to use when running in
                                  parallel. The default is determined
                                  dynamically as the total number of available
                                  cores, minus one.
  --point OUTPUT                  [Mutually exclusive with --overlay and
                                  --sample] Assign each pixel to the DGGS
                                  cell containing its centre (default).
                                  OUTPUT: 'value' (scalar per cell, default),
                                  'list' (sorted list of all contributing
                                  pixel values), 'histogram' (value-count
                                  struct).
  --overlay METHOD                [Mutually exclusive with --point and
                                  --sample] Area-based polygon intersection.
                                  METHOD: 'weighted' (area-weighted mean),
                                  'mode' (majority class by overlap area),
                                  'mass-preserve' (area-weighted sum; conserves
                                  total — use when pixel value is a total
                                  count/mass), 'density-preserve' (integrates
                                  density × pixel area; use when pixel value
                                  is a per-area rate), 'fractions' (per-class
                                  area fractions → struct), 'list' (all
                                  overlapping pixel values as a sorted list),
                                  'histogram' (value-count histogram of
                                  overlapping pixels).
  --sample INTERP                 [Mutually exclusive with --point and
                                  --overlay] Sample the raster at each DGGS
                                  cell centre. INTERP: 'nn' (nearest-
                                  neighbour, default), 'bilinear', 'bicubic',
                                  'lanczos'.
  -a, --agg AGGFUNC[,AGGFUNC...]  Aggregation function(s) applied when
                                  multiple raster pixels map to the same DGGS
                                  cell (only relevant for --point). Options:
                                  count, mean, sum, prod, std, var, min, max,
                                  median, mode, majority, nunique, range.
                                  Comma-separate multiple names (e.g. min,max)
                                  to produce a struct column per band.
                                  [default: mean]
  -vct, --valid-coverage-threshold FLOAT RANGE
                                  Minimum fraction of each DGGS cell's
                                  overlapping raster area that must contain
                                  valid (non-nodata) pixels for the cell to
                                  receive a value. Applied per band. 0.0
                                  (default) keeps all cells with any valid
                                  data. Only meaningful for --overlay; ignored
                                  for --overlay mass-preserve (partial sums
                                  are correct values — filtering them would
                                  break mass conservation).  [default: 0.0;
                                  0.0<=x<=1.0]
  -d, --decimals INTEGER|none     Decimal places to round output values. 0 =
                                  integer; negative values round to tens (-1),
                                  hundreds (-2), etc. Use 'none' to disable
                                  rounding.  [default: 1]
  -o, --overwrite
  -co, --compact                  Compact the cells up to the parent
                                  resolution. Compaction is only applied where
                                  all sibling cells share identical values in
                                  every output column.
  -g, --geo [point|polygon|none]  Write output as a GeoParquet (v1.1.0) with
                                  either point or polygon geometry.  [default:
                                  none]
  --tempdir PATH                  Temporary data is created during the
                                  execution of this program. This parameter
                                  allows you to control where this data will
                                  be written.
  --version                       Show the version and exit.
  --help                          Show this message and exit.

Sampling strategies

Three mutually exclusive modes control how pixel values are mapped to DGGS cells:

• --point (default) — index each pixel centre to a DGGS cell
• --overlay METHOD — compute area-weighted intersections between pixels and DGGS cells
• --sample — sample the raster at each DGGS cell centre

Point sampling (default) — --point

Each raster pixel centre is indexed to its containing DGGS cell. When multiple pixels fall in the same cell, -a/--agg determines how they are combined (default: mean). Produces sparse output (gaps) when the DGGS resolution is finer than the raster.

Output modes (pass as --point OUTPUT):

• (no arg / --point value) — scalar per cell per band; use --agg min,max etc. for multi-agg struct output
• --point list — sorted list of all contributing pixel values: list<T> per band. --agg is ignored.
• --point histogram — value-count histogram: struct<values: list<T>, counts: list<int64>> per band. --agg is ignored.

# Default: mean of all contributing pixels
raster2dggs h3 input.tif output/ -r 9

# Min and max in a single pass
raster2dggs h3 input.tif output/ -r 9 --agg min,max

# Sorted list of all pixel values per cell
raster2dggs h3 input.tif output/ -r 7 --point list -d 2

# Histogram of pixel values per cell
raster2dggs h3 input.tif output/ -r 7 --point histogram -d 0

Overlay (area-based) — --overlay METHOD

Uses exactextract to compute exact pixel–cell intersection areas. METHOD is required:

--overlay METHOD	Output schema	Use for
weighted	Scalar T per band	Intensive quantities: temperature, elevation, concentration, fraction cover — value per unit area, averaged across the cell
mode	Scalar T per band	Categorical rasters: land cover, soil type, zone IDs, masks
mass-preserve	Scalar T per band	Extensive totals: population count, emissions — pixel value is already a total; sum is conserved
density-preserve	Scalar T per band	Density rasters (W/m², kg/km²) — integrates density × pixel area to give the cell total; geographic CRS uses geodesic pixel areas
fractions	struct<classes: list<int64>, fractions: list<float64>> per band	Class area fractions within each DGGS cell
list	list<T> per band	All overlapping pixel values as a sorted list (collect mode)
histogram	struct<values: list<T>, counts: list<int64>> per band	Histogram of overlapping pixel values

# Area-weighted mean (intensive quantities)
raster2dggs h3 input.tif output/ -r 8 --overlay weighted

# Majority-class (categorical rasters)
raster2dggs h3 landcover.tif output/ -r 8 --overlay mode

# Mass-conserving sum (population counts)
raster2dggs h3 popcount.tif output/ -r 8 --overlay mass-preserve

# Density integration (W/m² → total W per DGGS cell)
raster2dggs h3 power_density.tif output/ -r 8 --overlay density-preserve

# Per-class area fractions
raster2dggs h3 landcover.tif output/ -r 8 --overlay fractions

# Collect all overlapping pixel values as a list
raster2dggs h3 input.tif output/ -r 8 --overlay list

# Histogram of all overlapping pixel values
raster2dggs h3 input.tif output/ -r 8 --overlay histogram

Performance note

--overlay uses exactextract to compute pixel–cell area intersections. For each raster window, exactextract reads only the raster blocks needed to cover that window's DGGS cells. If the raster fits in memory, increasing GDAL's block cache allows blocks read for early windows to remain cached for later ones, reducing redundant I/O:

GDAL_CACHEMAX=512 raster2dggs h3 input.tif output/ -r 8 --overlay weighted

The value is in megabytes. The default is 64 MB. For large rasters or high DGGS resolutions where each window covers many cells, a larger cache can significantly reduce processing time.

Valid-data coverage threshold (-vct / --valid-coverage-threshold)

By default (-vct 0.0) any cell with at least one valid pixel in its overlap area receives a value. Use --valid-coverage-threshold to require a minimum fraction of the cell's raster-overlapping area to have valid (non-nodata) data:

# Discard cells where fewer than 50% of overlapping pixels are valid
raster2dggs h3 input.tif output/ -r 8 --overlay mode -vct 0.5

The threshold is applied per band: a cell may receive a valid value for one band and be nulled for another if that band has sparser nodata. Nulled values are then handled by --nodata — the default omit drops rows where any band is null; emit keeps them as NaN (or --nodata-fill if set).

This option has no effect for --overlay mass-preserve. Partial sums produced by mass-preserve are correct values representing the fraction of mass within the cell–raster intersection; filtering them out would break the mass-conservation guarantee. --overlay density-preserve respects the threshold normally — a cell with insufficient valid coverage will be nulled.

Windowed resampling — --sample

For each DGGS cell, samples the raster at the cell centre using windowed I/O. Pass the resampling kernel as --sample INTERP (default: nn):

INTERP	Description
nn (default)	Nearest-neighbour — suitable for categorical rasters
bilinear	Bilinear (2×2 stencil) — smooth continuous fields
bicubic	Bicubic/Keys (4×4 stencil) — higher-quality smooth fields
lanczos	Lanczos-3 (6×6 stencil) — highest quality, slowest

# NN sample (default — works for both continuous and categorical)
raster2dggs h3 input.tif output/ -r 9 --sample

# Bilinear sample for a smooth continuous field (DEM, temperature)
raster2dggs h3 dem.tif output/ -r 9 --sample bilinear

# NN sample for a categorical raster (landcover)
raster2dggs h3 landcover.tif output/ -r 9 --sample -d 0

--agg is ignored for --sample. Supports --compact.

Visualising output

Output is in the Apache Parquet format, hive partitioned with the parent resolution as partition key. The example below is with -pr 3 with the H3 DGGS.

tree /home/user/example.pq

/home/user/example.pq
├── h3_03=83bb09fffffffff
│   └── part.0.parquet
└── h3_03=83bb0dfffffffff
    └── part.0.parquet

Output can also be written to GeoParquet (v1.1.0) by including the -g/--geo parameter, which accepts:

• polygon for cells represented as boundary polygons
• point for cells represented as centre points
• none for standard Parquet output (not GeoParquet) ← this is the default if -g/--geo is not used

GeoParquet output is useful if you want to use the spatial representations of the DGGS cells in traditional spatial analysis, or if you merely want to visualise the output.

Below are some ways to read and visualise it.

DuckDB

$ duckdb
DuckDB v1.4.1 (Andium) b390a7c376
Enter ".help" for usage hints.
Connected to a transient in-memory database.
Use ".open FILENAME" to reopen on a persistent database.
D INSTALL spatial;
D LOAD spatial;
D SELECT * FROM read_parquet('se_island.pq') LIMIT 7;
┌────────┬────────┬────────┬────────────────────────────────────────────────────────────────────────────────┬─────────────┬─────────┐
│ band_1 │ band_2 │ band_3 │                                    geometry                                    │    s2_19    │  s2_08  │
│ float  │ float  │ float  │                                    geometry                                    │   varchar   │ varchar │
├────────┼────────┼────────┼────────────────────────────────────────────────────────────────────────────────┼─────────────┼─────────┤
│    0.0 │    0.0 │    0.0 │ POLYGON ((-176.17946725380486 -44.33542073938414, -176.17946725380486 -44.33…  │ 72b47e01e24 │ 72b47   │
│    0.0 │    0.0 │    0.0 │ POLYGON ((-176.18439390505398 -44.33543749229784, -176.18439390505398 -44.33…  │ 72b47e02a14 │ 72b47   │
│    0.0 │    0.1 │    0.1 │ POLYGON ((-176.18550630891403 -44.33547457195554, -176.18550630891403 -44.33…  │ 72b47e1d54c │ 72b47   │
│    0.0 │    0.0 │    0.0 │ POLYGON ((-176.17819578278952 -44.33537828938332, -176.17819578278952 -44.33…  │ 72b47e01d64 │ 72b47   │
│    0.1 │    0.1 │    0.3 │ POLYGON ((-176.18344039674218 -44.335553297533835, -176.18344039674218 -44.3…  │ 72b47e0282c │ 72b47   │
│    0.0 │    0.0 │    0.0 │ POLYGON ((-176.17899045588274 -44.335404822417665, -176.17899045588274 -44.3…  │ 72b47e01dfc │ 72b47   │
│    0.1 │    0.1 │    0.3 │ POLYGON ((-176.1832814769592 -44.33554799806149, -176.1832814769592 -44.3356…  │ 72b47e02824 │ 72b47   │
└────────┴────────┴────────┴────────────────────────────────────────────────────────────────────────────────┴─────────────┴─────────┘

Output value columns may also be arrays (double[] or int64[]) or structs, not just scalar values, depending on the options you pass to the tool (e.g. --agg min,max (multiple aggregations) or --point list/--point histogram) and the relative size of the DGGS cells and raster cells.

In the case of struct outputs, it should be noted that there's no real consequence of using a struct (i.e. band_1.min, band_1.max) over a series of flat columns (i.e. band_1_min, band_1_max), since Parquet uses Dremel shredding for nested types: a struct(min double, max double) column is physically stored as two separate column chunks (band_1.min, band_1.max) with definition/repetition level metadata. So the on-disk layout is identical to flat columns. Consequences:

• Compression: identical. The min values are stored contiguously together, max values together, same as flat columns. Encoding schemes (dictionary, RLE, delta) apply the same way.
• Column pruning / projection pushdown: also identical for modern readers. DuckDB's SELECT band_1.min FROM ... reads only the min sub-column chunk, same as SELECT band_1_min FROM ... would with flat columns.
• Definition level overhead: structs add a small amount of metadata to encode nullability at each nesting level. For non-nullable structs with non-nullable fields this is negligible: a few bytes per row group.

Examples:

--point list -d 1:

D SELECT band_1 FROM read_parquet('./tests/data/output/larger-than-pixel/temp_mean_wgs84-poly.geoparquet') LIMIT 7;
┌────────────────────────────────────────────┐
│                   band_1                   │
│                  double[]                  │
├────────────────────────────────────────────┤
│ [15.9, 15.9, 16.1, 16.1, 16.3, 16.3]       │
│ [16.0, 16.0, 16.0, 16.1, 16.1, 16.2, 16.2] │
│ [16.4, 16.6, 16.7, 16.7, 17.0, 17.1]       │
│ [18.0, 18.2, 18.3, 18.4, 18.5]             │
│ [16.4, 16.5, 16.5, 16.6, 16.8, 16.8]       │
│ [17.4, 17.6, 17.7, 17.9, 18.0, 18.2]       │
│ [16.1, 16.2, 16.2, 16.3, 16.4, 16.5]       │
└────────────────────────────────────────────┘

--point histogram -d 0:

D SELECT band_1 FROM read_parquet('./tests/data/output/larger-than-pixel/temp_mean_wgs84-poly.geoparquet') LIMIT 7;
┌────────────────────────────────────────────┐
│                   band_1                   │
│ struct("values" bigint[], counts bigint[]) │
├────────────────────────────────────────────┤
│ {'values': [16], 'counts': [6]}            │
│ {'values': [16], 'counts': [7]}            │
│ {'values': [16, 17], 'counts': [1, 5]}     │
│ {'values': [18], 'counts': [5]}            │
│ {'values': [16, 17], 'counts': [2, 4]}     │
│ {'values': [17, 18], 'counts': [1, 5]}     │
│ {'values': [16, 17], 'counts': [5, 1]}     │
└────────────────────────────────────────────┘

--agg min,max,majority,mode -d 0:

D SELECT band_1 FROM read_parquet('./tests/data/output/larger-than-pixel/temp_mean_wgs84-poly.geoparquet') LIMIT 7;
┌────────────────────────────────────────────────────────────────┐
│                             band_1                             │
│ struct(min bigint, max bigint, majority bigint, "mode" bigint) │
├────────────────────────────────────────────────────────────────┤
│ {'min': 16, 'max': 16, 'majority': 16, 'mode': 16}             │
│ {'min': 16, 'max': 16, 'majority': 16, 'mode': 16}             │
│ {'min': 16, 'max': 17, 'majority': 17, 'mode': 17}             │
│ {'min': 18, 'max': 18, 'majority': 18, 'mode': 18}             │
│ {'min': 16, 'max': 17, 'majority': 17, 'mode': 17}             │
│ {'min': -9999, 'max': 18, 'majority': 18, 'mode': 18}          │
│ {'min': 16, 'max': 17, 'majority': 16, 'mode': 16}             │
└────────────────────────────────────────────────────────────────┘

GDAL

ogrinfo -so -al ./se_island.pq
INFO: Open of `se_island.pq'
      using driver `Parquet' successful.

Layer name: se_island
Geometry: Polygon
Feature Count: 18390
Extent: (-176.185824, -44.356933) - (-176.159915, -44.335364)
Layer SRS WKT:
GEOGCRS["WGS 84",
    ENSEMBLE["World Geodetic System 1984 ensemble",
        MEMBER["World Geodetic System 1984 (Transit)"],
        MEMBER["World Geodetic System 1984 (G730)"],
        MEMBER["World Geodetic System 1984 (G873)"],
        MEMBER["World Geodetic System 1984 (G1150)"],
        MEMBER["World Geodetic System 1984 (G1674)"],
        MEMBER["World Geodetic System 1984 (G1762)"],
        MEMBER["World Geodetic System 1984 (G2139)"],
        MEMBER["World Geodetic System 1984 (G2296)"],
        ELLIPSOID["WGS 84",6378137,298.257223563,
            LENGTHUNIT["metre",1]],
        ENSEMBLEACCURACY[2.0]],
    PRIMEM["Greenwich",0,
        ANGLEUNIT["degree",0.0174532925199433]],
    CS[ellipsoidal,2],
        AXIS["geodetic latitude (Lat)",north,
            ORDER[1],
            ANGLEUNIT["degree",0.0174532925199433]],
        AXIS["geodetic longitude (Lon)",east,
            ORDER[2],
            ANGLEUNIT["degree",0.0174532925199433]],
    USAGE[
        SCOPE["Horizontal component of 3D system."],
        AREA["World."],
        BBOX[-90,-180,90,180]],
    ID["EPSG",4326]]
Data axis to CRS axis mapping: 2,1
Geometry Column = geometry
band_1: Real(Float32) (0.0)
band_2: Real(Float32) (0.0)
band_3: Real(Float32) (0.0)
s2_19: String (0.0)
s2_08: String (0.0)

QGIS

qgis sample.pq

With some styling applied:

Installation (detailed)

PyPi:

pip install raster2dggs[all]

Conda environment:

name: raster2dggs
channels:
  - conda-forge
channel_priority: strict
dependencies:
  - python>=3.11,<3.12
  - pip=23.1.*
  - gdal>=3.8.5
  - pyproj=3.6.*
  - pip:
    - raster2dggs[all]>=0.9.0

For development

In brief, to get started:

• Install Poetry
• Install GDAL
  • If you're on Windows, pip install gdal may be necessary before running the subsequent commands.
  • On Linux, install GDAL 3.6+ according to your platform-specific instructions, including development headers, i.e. libgdal-dev.
• Create the virtual environment with poetry install. This will install necessary dependencies.
• Subsequently, the virtual environment can be re-activated with poetry env activate.

If you run poetry install -E all --with dev, the CLI tool will be aliased so you can simply use raster2dggs rather than poetry run raster2dggs, which is the alternative if you do not poetry install -E all --with dev.

For partial backend support you can consider poetry install --with dev -E h3 -E a5 etc. To check what is installed: poetry show --tree.

Code formatting

Please run black . before committing.

Tests

Tests are included. To run them, set up a poetry environment, then run from the project root:

pytest -v --durations=10 --tb=short

• -v — one line per test with pass/fail status
• --durations=10 — reports the 10 slowest tests at the end
• --tb=short — compact tracebacks on failure
• Add -x to stop on the first failure when debugging

To run a subset of tests:

pytest -k h3                        # all tests whose ID contains "h3"
pytest -k "h3 or rhp"
pytest -k "sample and rhp"
pytest tests/classes/test_sample_nn.py          # sample transfer smoke tests
pytest "tests/classes/test_cli_integration.py::TestAllDGGS::test_command[h3-polygon-co]"  # exact parametrised case

Test data are included at tests/data/.

Generating synthetic sample rasters

make_samples.py generates a small suite of synthetic GeoTIFF rasters for experimentation. It only requires numpy and rasterio (plus optional scipy for better smoothing):

python make_samples.py --outdir sample_rasters --seed 42

This writes six rasters to sample_rasters/:

File	Semantics	Nodata pattern	CRS
landcover_utm33.tif	piecewise_constant — 6 landcover classes	Scattered holes + missing stripe	UTM 33N
frac_treecover_utm33.tif	fraction_cover — tree cover 0–1	Coastline-shaped mask	UTM 33N
popcount_webmerc.tif	count_total — heavy-tailed counts	Rotated rectangle polygon	Web Mercator
temp_mean_wgs84.tif	cell_average — continuous temperature	Edge band + scattered pixels	WGS84
zone_ids_laea.tif	piecewise_constant — Voronoi zone IDs	Islands + sliver patches	Europe LAEA
multiband_per_band_nodata_wgs84.tif	4-band float32	Nodata at different pixels per band	WGS84
swath_wgs84.tif	point_center_strict — 3-band simulated swath	~85% nodata outside diagonal strip; bbox ~6× larger than data footprint	WGS84 (diagonal NE–SW strip within 120–160°E, 20–60°N; swath widens northward via geodesic cross-track mask)
swath_polar_stereo.tif	point_center_strict — 3-band simulated swath	Thin nodata margins at swath edges	Arctic polar stereographic (custom CRS); rectangular grid in CRS appears as a curved arc across ~38–80°N, ~148–172°E in WGS84 — exercises CRS reprojection code path
swath_u_shape.tif	point_center_strict — 1-band continuous	No nodata	EPSG:3031 (Antarctic polar stereo); all four corners at ~21–26°N in WGS84, but the bottom edge centre is at ~11°N — a naive corner-only bbox misses ~10° of southward extent

The multi-band raster is specifically designed to exercise per-band nodata handling: a pixel that is nodata in one band can be valid in another.

Experimenting

Two sample files have been uploaded to an S3 bucket with s3:GetObject public permission.

• s3://raster2dggs-test-data/Sen2_Test.tif (sample Sentinel 2 imagery, 10 bands, rectangular, Int16, LZW compression, ~10x10m pixels, 68.6 MB)
• s3://raster2dggs-test-data/TestDEM.tif (sample LiDAR-derived DEM, 1 band, irregular shape with null data, Float32, uncompressed, 10x10m pixels, 183.5 MB)

You may use these for experimentation. However you can also use local files too, which will be faster. A good, small (5 MB) sample image is available here.

A small test file is also available at tests/data/se-island.tif.

You can also generate a suite of synthetic rasters locally using make_samples.py — see Generating synthetic sample rasters above.

Example commands

Index to H3 at resolution 11, integer output:

raster2dggs h3 --resolution 11 -d 0 s3://raster2dggs-test-data/Sen2_Test.tif ./tests/data/output/11/Sen2_Test

Same raster to rHEALPix:

raster2dggs rhp --resolution 11 -d 0 s3://raster2dggs-test-data/Sen2_Test.tif ./tests/data/output/11/Sen2_Test_rhp

DEM indexed to H3, median aggregation, GeoParquet polygon output:

raster2dggs h3 --resolution 13 --compression zstd --agg median -d 1 --geo polygon s3://raster2dggs-test-data/TestDEM.tif ./tests/data/output/13/TestDEM

Auto-select resolution (first H3 resolution finer than the raster pixel size):

raster2dggs h3 --resolution smaller-than-pixel input.tif ./output

Multi-aggregation struct output — min, max, and mean per band in one pass:

raster2dggs h3 --resolution 9 --agg min,max,mean -d 1 input.tif ./output

Collect all contributing pixel values per cell as a sorted list:

raster2dggs h3 --resolution 7 --point list -d 2 input.tif ./output

Histogram of contributing pixel values per cell:

raster2dggs h3 --resolution 7 --point histogram -d 0 input.tif ./output

Nearest-neighbour sampling for a continuous field (e.g. DEM, temperature grid):

raster2dggs h3 --resolution 9 --sample input.tif ./output

Bilinear sampling for a continuous field:

raster2dggs h3 --resolution 9 --sample bilinear input.tif ./output

Nearest-neighbour sampling for a categorical raster (e.g. landcover):

raster2dggs h3 --resolution 9 --sample -d 0 landcover.tif ./output

Area-weighted mean for a continuous raster:

raster2dggs h3 --resolution 8 --overlay weighted input.tif ./output

Majority-class for a categorical raster:

raster2dggs h3 --resolution 8 --overlay mode landcover.tif ./output

Mass-conserving sum for population counts:

raster2dggs h3 --resolution 8 --overlay mass-preserve popcount.tif ./output

Emit nodata cells rather than omitting them, replacing the nodata value with −1:

raster2dggs h3 --resolution 9 --nodata emit --nodata-fill -1 input.tif ./output

Citation

Citation metadata is maintained in CITATION.cff. GitHub renders this as a "Cite this repository" button on the repository page (top-right of the About panel), which provides ready-to-copy BibTeX and APA formats.