Swiss Agricultural Landscape Intelligence (SALI) platform

The EOA data infrastructure.

This repository documents how the multi-source Earth observation dataset was produced and where it lives. All data is stored on the shared NAS at \\eo-nas1\data.

📒 Index

Environment setup
Grid creation
Sentinel-2
MeteoSwiss
SwissImage
swissalti3D (DEM)
Landsat
DLR Soilsuite
PlanetLabs
Soil properties — ccsols
Snow depth
Data status
Storage structure

1. Environment setup

Install with:

pip install -r requirements.txt

The Sentinel-2 and Landsat pipelines depend on the minicuber package (commit pinned in the Data status table).

2. Grid creation

The data is saved on the EPSG:32632 grid. The pixels align to those of the Sentinel-2 satellite data in the UTM zone 32. The extent of the grid is defined by the bounding box of the MeteoSuisse data, which extends slightly beyond the administrative borders of Switzerland.

The grid has a resolution of 1280m x 1280m, meaning 128x128 Sentinel-2 pixels (10m resolution) are included per grid cell. To create the grid, run the QGIS model in grid_creator.model3 in QGIS. The inputs are:

a weather file from MeteoSuisse (e.g.O:/Data-Raw/27_Natural_Resources-RE/99_Meteo_Public/MeteoSwiss_netCDF/__griddedData/lv95updated/TminY_ch01r.swiss.lv95_202301010000_202301010000.nc)
an image from the Sentinel-2 tile T32TPS (e.g.~/mnt/eo-nas1/data/satellite/sentinel2/CH_old/2020/S2A_MSIL2A_20200228T101021_N0214_R022_T32TPS_20200228T114852.SAFE/GRANULE/L2A_T32TPS_A024472_20200228T101400/IMG_DATA/R10m/T32TPS_20200228T101021_B03_10m.jp2)

The resulting grid will start at the first pixel of T32TPS covering the weather file. This corresponds to the eastern most point of the data. The grid is then extended to cover the entire weather file. It is provided at the following path:

~/mnt/eo-nas1/eoa-share/projects/012_EO_dataInfrastructure/Project layers/gridface_s2tiles.shp

Sentinel-2 aligned grid covering the bounding box of MeteoSuisse file (shown in background)

Since this grid is a rectangle and contains multiple tiles outside of Switzerland, it is cropped such as to keep only the grid cells that fall over the MeteoSuisse file:

python crop_grid.py

The final grid is saved at

~/mnt/eo-nas1/eoa-share/projects/012_EO_dataInfrastructure/Project layers/gridface_s2tiles_CH.shp

Sentinel-2 aligned grid cropped to MeteoSuisse file

3. Sentinel-2

Temporal coverage: 2016-2025
Spatial resolution: 10 m
CRS: EPSG:32632
Tile size: 128 × 128 pixels (1280 m × 1280 m)

The Sentinel-2 data is downloaded using the grid created above. Each grid tile is 1280m x 1280m, containing 128 x 128 pixels with a resolution of 10m.
For each grid tile, the data is queried using the minicuber code which takes care of reprojecting all data to a common, 10m resolution pixel size and aligned to the coordinates of the grid (EPSG:32632). Details on other processing steps are included in the minicuber documentation.

How to run

python S2/download_pipeline_parallel.py

Key parameters (hard-coded at the top of the script):

output_folder — where zarr files are written
num_cells — number of adjacent tiles merged into one query (reduces Microsoft Planetary Computer round-trips). A value of 2 means 3×3 tiles are fetched at once and then split. Automatically reduced at grid borders.
Downloads run year by year; parallelisation is across years.

Bands and variables

s2_AOT, s2_B01, s2_B02, s2_B03, s2_B04, s2_B05, s2_B06, s2_B07,
s2_B08, s2_B8A, s2_B09, s2_B11, s2_B12, s2_WVP, s2_SCL,
s2_mask (DL cloud mask, see minicuber docs),
product_uri, mean_sensor_zenith, mean_sensor_azimuth,
mean_solar_zenith, mean_solar_azimuth

Data location and format

~/mnt/eo-nas1/data/satellite/sentinel2/raw/CH/

Files: S2_<minx>_<maxy>_<startYYYYMMDD>_<endYYYYMMDD>.zarr
(minx, maxy) = upper-left corner in EPSG:32632. Each zarr is split into two chunks along the longitude axis.

Note — known data quality issues:

When num_cells > 1, timestamps that don't span all tiles in the block are filled with NaN. These all-NaN timestamps should be dropped before use.

product_uri records the originating Sentinel-2 L2A product. Atmospheric correction was applied upstream — see Copernicus S2 products.

Potential improvement: parallelise across grid tiles rather than years for future single-year updates.

4. MeteoSwiss

Temporal coverage: 1961–2025 (varies by variable)
Original spatial resolution: 1 km
Processed resolution: 10 m (nearest-neighbour upsampling — see warning below)
CRS: EPSG:32632

Raw data location

O:/Data-Raw/27_Natural_Resources-RE/99_Meteo_Public/MeteoSwiss_netCDF/__griddedData/lv95updated/

How to run

python Meteo/meteo_to_cube_parallel.py

Reprojects the native LV95 (EPSG:2056) netCDF files to EPSG:32632 and regrid to the S2-aligned 10 m grid.

Variables

Variable	Description	Unit
`Rhires`	Daily precipitation	mm
`Srel`	Daily relative sunshine duration	%
`Tabs`	Daily mean air temperature	°C
`Tmin`	Daily minimum air temperature	°C
`Tmax`	Daily maximum air temperature	°C

For full product documentation see the MeteoSwiss product page.

Data location and format

~/mnt/eo-nas1/data/meteo/<datavar>/MeteoSwiss_<datavar>D_<minx>_<maxy>_<year>0101_<year>1231.zarr

Warning: Nearest-neighbour interpolation was used for upsampling from 1 km to 10 m. The data structurally resembles 1 km resolution. For true 10 m meteorological data, reprocess with bilinear interpolation.

5. SwissImage (Swisstopo)

Spatial resolution: 10 cm or 2 m
CRS (raw): EPSG:2056 → reprojected to EPSG:32632
Product: swissimage-dop10

Download raw TIF files

python SwissImage/si_download.py --urls_path path/to/urls.csv --downloads_path path/to/output/folder

URL lists:

SwissImage/ch.swisstopo.swissimage-dop10-DOp5jXFT.csv — 10 cm resolution
SwissImage/ch.swisstopo.swissimage-dop10-vWuyN4vG.csv — 2 m resolution

Raw TIF filename structure: swissimage-dop10_<YEAR>_<MINX>_<MINY>_<RESOLUTION>_2056.tif
(MINX, MINY in km, EPSG:2056; one file = 1 km × 1 km, RGB.)

Raw files stored at:

~/mnt/eo-nas1/data/swisstopo/SwissImage/raw/10cm
~/mnt/eo-nas1/data/swisstopo/SwissImage/raw/2m

Reproject and align to S2 grid

python SwissImage/si_to_cube_parallel.py

Reprojects to EPSG:32632 and resamples to align to the custom Sentinel-2 grid (preserving 10 cm or 2 m resolution).

Data location and format

~/mnt/eo-nas1/data/swisstopo/SwissImage/cubes/10cm/
~/mnt/eo-nas1/data/swisstopo/SwissImage/cubes/2m/

6. swissalti3D (Swisstopo DEM)

Original resolution: 2 m
CRS (raw): EPSG:2056 → reprojected to EPSG:32632
Product: swissalti3D

How to run

python DEM/dem_to_cube_inverse.py

Reprojects from EPSG:2056 to EPSG:32632 using nearest-neighbour interpolation, aligned to the custom grid.

Data location and format

~/mnt/eo-nas1/data/swisstopo/DEM/sa3D_<minx>_<maxy>.zarr

(minx, maxy) = top-left corner in EPSG:32632.

Warning: The DEM source uses 65535 as a NoData value. This was not always correctly handled during reprojection, so grid tiles at the Swiss border may have artificially elevated values. These tiles most likely all fall outside the Swiss boundary. Fix: pass the correct NoData flag during reprojection.

7. Landsat

Temporal coverage: 1982–2012 (L4-5), 2013–present (L8-9)
Spatial resolution: 30 m
CRS: EPSG:32632
Tile size: 128 × 128 pixels (3840 m × 3840 m)

Landsat uses a separate grid from S2 (different pixel spacing and origin). S2 and Landsat can be reconciled by restructuring S2 data to 30 m.

Grid creation

The grid originates from the upper-left corner of Landsat path 195 / row 27 (already in EPSG:32632). Tiles are 3840 m × 3840 m.

python landsat/landsat_grid.py   # create rectangle grid
python crop_grid.py              # crop to Switzerland

Final grid:

~/mnt/eo-nas1/eoa-share/projects/012_EO_dataInfrastructure/Project layers/grid_landsat_CH.shp

How to run

Choose the script matching the satellite generation:

Script	Satellites	Years
`landsat/download_pipeline_parallel_45.py`	Landsat 4 & 5 (TM)	1982–2012
`landsat/download_pipeline_parallel_7.py`	Landsat 7 (ETM+)	1999–2022
`landsat/download_pipeline_parallel.py`	Landsat 8 & 9 (OLI/TIRS)	2013–present

Key parameter to update in scripts: the target year range (hard-coded near the top of each file).

Bands and variables

Landsat 4–5 (TM): TM_B1 (blue), TM_B2 (green), TM_B3 (red), TM_B4 (NIR), TM_B5 (SWIR1), TM_B6 (LWIR/surface temp), TM_B7 (SWIR2), SR_ATMOS_OPACITY

Landsat 8–9 (OLI/TIRS): OLI_B1 (coastal aerosol), OLI_B2 (blue), OLI_B3 (green), OLI_B4 (red), OLI_B5 (NIR), OLI_B6 (SWIR1), OLI_B7 (SWIR2), TIRS_B10 (LWIR11/surface temp)

Quality / QA bands (all generations): ST_QA, QA_RADSAT, QA_PIXEL, QA_AEROSOL, SR_CLOUD_QA

Surface temperature ancillary: ST_ATRAN, ST_DRAD, ST_TRAD, ST_URAD, ST_EMIS, ST_EMSD, ST_CDIST

Metadata: such as scene id, original projetion, orbit path and row...

Each variable has metadata with description, scale/offset to convert to reflectance, and fill value. When raw data is in EPSG:32631 it is reprojected to EPSG:32632.

See USGS band designations for details.

Data location and format

~/mnt/eo-nas1/data/satellite/landsat/raw/CH/45/   # Landsat 4-5
~/mnt/eo-nas1/data/satellite/landsat/raw/CH/89/   # Landsat 8-9

Files: LS_<minx>_<maxy>_<startYYYYMMDD>_<endYYYYMMDD>.zarr

Note: Same NaN-timestamp issue as S2 (see section 3) may occur. Drop all-NaN timestamps before use.

8. DLR Soilsuite

Temporal coverage: 2018–2022 composite (5-year)
Source: DLR EO Center soilsuite — bare soil composite from Sentinel-2
CRS: EPSG:32632 (reprojected from source)

How to run

python DLR_soilsuite/download_src.py    # download raw TIFs
python DLR_soilsuite/SRC_to_cube.py     # reproject to EPSG:32632 and save as zarr

Data location and format

~/mnt/eo-nas1/data/satellite/sentinel2/DLR_soilsuite/SRC_<left>_<top>.zarr

(left, top) = top-left corner aligned to the S2 grid.

9. PlanetLabs

Product: PlanetScope 8-band (PSB.SD), TIF clipped to AOI
Cloud cover filter: < 60% scene cloud coverage
CRS: native PlanetScope projection (not reprojected)

AOI geometries (GeoJSON) are stored in PlanetLabs/geoms/. To download a new AOI, edit the AOI and date range in:

python PlanetLabs/PlanetScopeDownload.py

Requires a Planet API key (set in PlanetLabs/.env — do not commit this file).

10. Soil properties — ccsols

Source: Swiss Competence Centre for Soils (KOBO) — ccsols
Original resolution: 30 m, EPSG:2056
Processed resolution: 10 m, EPSG:32632 (nearest-neighbour interpolation)

Raw raster layers: ~/mnt/eo-nas1/data/soil/ccsols/Daten_2024-01

How to run

python Soil/soil_to_cube.py

Variables (each at depths 0–30, 30–60, 60–120 cm)

Variable	Description	Unit
`CECpot_depth_*`	Cation exchange capacity	mmolc/kg
`clay_depth_*`	Clay content	%
`pH_depth_*`	pH	–
`sand_depth_*`	Sand content	%
`silt_depth_*`	Silt content	%
`soc_depth_*`	Soil organic carbon	%

Data location and format

~/mnt/eo-nas1/data/soil/ccsols/datacubes/ccsols_<minx>_<maxy>.zarr

11. Snow depth

Temporal coverage: 1970s–2023 (raw), 2015–2023 (processed cubes)
Original resolution: 1 km, EPSG:2056
Processed resolution: 10 m, EPSG:32632 (nearest-neighbour)

Raw data (daily snow depth):

~/mnt/eo-nas1/eoa-share/projects/012_EO_dataInfrastructure/DataInfrastructure/Meteo/HSCLQMD_ch01h.swiss.lv95_WY_1962_2023.nc

How to run

python Meteo/snow.py

Data location and format

~/mnt/eo-nas1/data/meteo/snowdepth/snowdepth_<minx>_<maxy>.zarr

Note: Nearest-neighbour upsampling used (same caveat as MeteoSwiss — see section 5).

Data status

The download history is tracked here:

Date	Jobs	Notes
31.05.2024	Run S2 downloading	Package versions: sen2nbar==2023.8.1 minicuber (commit version)
18.07.2024	Completed S2 download
26.07.2024	Downloaded SwissImage 2m
29.07.2024	Downloaded SwissImage 0.1m
09.09.2024	Processed SwissImage 0.1m and 2m to cubes
09.09.2024	Downloaded S2 2016	Package versions: sen2nbar==2023.8.1 minicuber (commit version)
25.11.2024	Processed MeteoSwiss variables to cubes
02.12.2024	Processed swissalti3D to cubes
28.02.2025	Downloaded Landsat 8-9 and processed to cubes	Package versions: sen2nbar==2023.8.1 minicuber (commit version)
20.03.2025	Downloaded S2 for 2024	Package versions: sen2nbar==2023.8.1 minicuber (commit version)
05.2025	Downloaded and processed DLR soilsuite
25.07.2025	Processed MeteoSwiss 2024 variables to cubes
25.07.2025	Downloaded PlanetLabs data for some AOIs
28.07.2025	Downloaded S2 until 2025-07 for some regions	Package versions: minicuber (commit version) with local sen2nbar provided in minicuber repo
17.10.2025	Processed ccsols data
11.11.2025	Processed snowdepth data
07.04.2026	Downloaded S2 for 2025	Package versions: sen2nbar==2023.8.1 minicuber (commit version)

Storage structure

📁 \\eo-nas1\data
├── satellite
│   ├── sentinel2
│   │   ├── raw
│   │   │   └── CH                    # raw S2 zarr cubes
│   │   ├── coreg
│   │   │   └── CH                    # coregistered S2 zarr cubes
│   │   └── DLR_soilsuite             # DLR bare-soil composite
│   ├── PlanetLabs
│   │   └── raw
│   │       └── <AOI_name>
│   └── landsat
│       └── raw
│           └── CH
│               ├── 45                # Landsat 4-5
│               └── 89                # Landsat 8-9
│
├── swisstopo
│   ├── SwissImage
│   │   ├── raw
│   │   │   ├── 10cm
│   │   │   └── 2m
│   │   └── cubes
│   │       ├── 10cm
│   │       └── 2m
│   └── DEM                           # swissalti3D zarr cubes
│
├── meteo
│   ├── Rhires
│   ├── Srel
│   ├── Tabs
│   ├── Tmax
│   ├── Tmin
│   └── snowdepth
│
└── soil
    └── ccsols
        ├── Daten_2024-01             # raw rasters
        └── datacubes                 # processed zarr cubes

Tools

Code examples and function to extract the data and process the zarr files with the xarray package can be found here

Name		Name	Last commit message	Last commit date
Latest commit History 108 Commits
DEM		DEM
DLR_soilsuite		DLR_soilsuite
Meteo		Meteo
PlanetLabs		PlanetLabs
S2		S2
Soil		Soil
SwissImage		SwissImage
img		img
landsat		landsat
notebooks		notebooks
README.md		README.md
crop_grid.py		crop_grid.py
grid_creator.model3		grid_creator.model3
requirements.txt		requirements.txt

Folders and files

Latest commit

History

Repository files navigation

Swiss Agricultural Landscape Intelligence (SALI) platform

📒 Index

1. Environment setup

2. Grid creation

3. Sentinel-2

How to run

Bands and variables

Data location and format

4. MeteoSwiss

Raw data location

How to run

Variables

Data location and format

5. SwissImage (Swisstopo)

Download raw TIF files

Reproject and align to S2 grid

Data location and format

6. swissalti3D (Swisstopo DEM)

How to run

Data location and format

7. Landsat

Grid creation

How to run

Bands and variables

Data location and format

8. DLR Soilsuite

How to run

Data location and format

9. PlanetLabs

10. Soil properties — ccsols

How to run

Variables (each at depths 0–30, 30–60, 60–120 cm)

Data location and format

11. Snow depth

How to run

Data location and format

Data status

Storage structure

Tools

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages