georaffer

Download German and Czech orthophotos and DSM tiles as GeoTIFF. Supports NRW, RLP, BB, BW, BY, TH, and CZ.

Installation

pip (macOS, Linux x86, Windows)

pip install georaffer

conda-forge (all platforms incl. Linux ARM64)

git clone https://github.com/michaelschleiss/georaffer
cd georaffer
conda create -n georaffer libgdal-jp2openjpeg lazrs-python laspy rasterio -c conda-forge
conda activate georaffer
pip install -e .

Or with pixi:

git clone https://github.com/michaelschleiss/georaffer
cd georaffer
pixi install

Usage

# From 4Seasons dataset
georaffer pygeon /data/4seasons/campaign --output ./tiles

# From CSV with coordinates
georaffer csv coords.csv --cols lon,lat --output ./tiles

# From bounding box
georaffer bbox 6.9,50.9,7.1,51.1 --output ./tiles

# From existing GeoTIFF footprint
georaffer tif ./area.tif --output ./tiles

# From specific tile indices
georaffer tiles 362,5604 --output ./tiles

See georaffer --help for all options. By default, downloads target NRW, RLP, BW, and BY; use --region to select others (e.g., --region bb th cz).

When using the tif command, aligned outputs are written to ./tiles/aligned/ and match the reference GeoTIFF grid (CRS, pixel size, width/height, and bounds). Alignment currently requires all processed tiles to share the same CRS; mixed-zone runs (for example NRW/RLP zone 32 plus BB zone 33) will fail and should be split by zone.

Python API

from georaffer import TileStore

store = TileStore(path="./tiles", regions=["NRW", "RLP"], imagery_from=(2015, None))

# Query tiles at 1km grid coordinates
tiles = store.query(coords=(350, 5600), tile_type="image")

# Retrieve (downloads + converts if needed)
path = store.get(tiles[0], resolution=2000)

# Batch with pipelined download/convert
paths = store.get_many(tiles, resolution=2000)

For single-file conversion without TileStore:

from georaffer.conversion import convert_file

convert_file("raw/image/tile.jp2", "processed", resolution=2000)

File name scheme

Raw downloads keep the provider filenames:

• NRW orthophotos (JP2): dop10rgbi_32_<grid_x>_<grid_y>_<n>_nw_<year>.jp2
• NRW DSM (LAZ): bdom50_32_<grid_x>_<grid_y>_<n>_nw_<year>.laz
• RLP orthophotos (JP2): dop20rgb_32_<grid_x>_<grid_y>_2_rp_<year>.jp2
• RLP DSM (LAZ): bdom20rgbi_32_<grid_x>_<grid_y>_2_rp.laz
• BB DOP/DSM (ZIP): dop_<zone><grid_x>-<grid_y>.zip, bdom_<zone><grid_x>-<grid_y>.zip
• BW DOP/DSM (ZIP): dop20rgb_32_<grid_x>_<grid_y>_2_bw.zip, dom1_32_<grid_x>_<grid_y>_2_bw.zip
• BY DOP (TIF): 32<grid_x>_<grid_y>.tif or 32<grid_x>_<grid_y>_<year>.tif
• BY DSM (TIF): 32<grid_x>_<grid_y>_20_DOM.tif
• TH DOP (ZIP): dop20rgb_32_<grid_x>_<grid_y>_<n>_th_<year>.zip
• TH DSM (ZIP): las_32_<grid_x>_<grid_y>_1_th_<year_range>.zip
• CZ orthophotos (JP2): oi_<grid_x>_<grid_y>_<year>.jp2
• CZ DSM (TIF/LAZ): dmpok_<grid_x>_<grid_y>_<year>.tif, or dmp1g_<grid_x>_<grid_y>_<year>.laz as fallback

Processed GeoTIFFs use a unified UTM-based pattern:

• <region>_<zone>_<easting>_<northing>_<year>.tif

where:

• <region> is nrw, rlp, bb, bw, by, th, or cz
• <zone> is the UTM zone (32 for NRW/RLP/BW/BY/TH, 33 for BB/CZ)
• <easting> and <northing> are UTM coordinates of the tile’s south‑west corner in meters (for example 350000,5600000)
• <year> is the acquisition year inferred from the source filename or LAZ header (falls back to latest when no year is available)

When large source tiles are split into smaller output tiles (for example RLP 2km → 1km grid), filenames still follow this pattern and the easting/northing encode the sub‑tile coordinates, e.g. rlp_32_362500_5604500_2023.tif.

Output directory structure

With --output ./tiles the pipeline creates:

./tiles
├── raw
│   ├── image        # Provider orthophotos (JP2, TIF, or ZIP depending on region)
│   └── dsm          # Provider DSM tiles (LAZ, TIF, or ZIP depending on region)
└── processed
    ├── image
    │   └── <pixels>/      # Orthophoto GeoTIFFs (tile size in pixels, e.g. 2000 for 0.5 m/px on 1 km tiles)
    ├── dsm
    │   └── <pixels>/      # DSM GeoTIFFs with the same convention

<pixels> matches the internal resolution derived from --pixel-size (meters per pixel). For example, with the default --pixel-size 0.5 on a 1 km grid, georaffer produces processed/image/2000/ and processed/dsm/2000/.

Raw tiles remain cached under raw/, so you can re-run conversions with different resolutions without re-downloading.

Adding New Regions

Subclass BaseDownloader and implement:

• parse_catalog() - Parse the region's tile feed/catalog
• grid_to_filename() - Convert grid coordinates to download URL
• utm_to_grid_coords() - Convert UTM to the region's grid system

See georaffer/downloaders/nrw.py as reference.

Data Sources and Licensing

German states publish orthophotos and elevation models as open data—free for commercial and non-commercial use—under EU High-Value Datasets Regulation 2023/138 and the INSPIRE Directive, coordinated nationally by the AdV (Arbeitsgemeinschaft der Vermessungsverwaltungen). Czech data comes from ČÚZK, which has published it as free open data since July 2023.

¹ Attribution: © GeoBasis-DE / <provider>, <year>, dl-de/by-2-0 ² Attribution: Datenquelle: LGL, www.lgl-bw.de ³ Attribution: Bayerische Vermessungsverwaltung – www.geodaten.bayern.de ⁴ Attribution: © GDI-Th ⁵ Attribution: © ČÚZK, CC BY 4.0

License

MIT