xarray-grass

A GRASS backend for Xarray. Explore all your GRASS rasters with Xarray. Import zarr or NetCDF into your GRASS database.

Installation

Install the package using uv or pip:

uv add xarray-grass

You need to install GRASS independently.

Loading GRASS data as an Xarray Dataset

import xarray as xr
import grass_session

import grass.script as gscript


with grass_session.Session(
    gisdb="/home/laurent/data/grassdata/",
    location="nc_spm_08_grass7_test",
    mapset="PERMANENT",
):
    # Make modis_lst mapset accessible
    gscript.run_command("g.mapsets", mapset="modis_lst", operation="add")
    test_ds = xr.open_dataset(
        "",  # No need to pass a path, the information is taken from the active grass session
        engine="xarray_grass",  # If no path is given, then the engine must be specified
        raster=["boundary_county_500m", "elevation"],
        strds="LST_Day_monthly@modis_lst",
    )

    print(test_ds)

Search path not modified
<xarray.Dataset> Size: 6MB
Dimensions:                     (y: 165, x: 179, start_time_LST_Day_monthly: 24)
Coordinates:
  * y                           (y) float32 660B 2.196e+05 ... 2.212e+05
  * x                           (x) float32 716B 6.377e+05 ... 6.395e+05
  * start_time_LST_Day_monthly  (start_time_LST_Day_monthly) datetime64[ns] 192B ...
    end_time_LST_Day_monthly    (start_time_LST_Day_monthly) datetime64[ns] 192B ...
Data variables:
    boundary_county_500m        (y, x) float64 236kB ...
    elevation                   (y, x) float32 118kB ...
    LST_Day_monthly             (start_time_LST_Day_monthly, y, x) int64 6MB ...
Attributes:
    crs_wkt:      PROJCRS["NAD83(HARN) / North Carolina",BASEGEOGCRS["NAD83(H...
    Conventions:  CF-1.13-draft
    history:      2025-11-02 23:51:57.141257+00:00: Created with xarray-grass...
    source:       GRASS database. project: , mapset:

xarray-grass requires an active GRASS session to work. Here we're using the grass-session package to set it.

You can choose which maps you want to load with the raster, raster_3d, strds and str3ds parameters to open_dataset. Those accept either a single string or an iterable. If none of those are specified, the whole mapset will be loaded, ignoring single maps that are already registered in either a strds or str3ds; those maps will be loaded into the Xarray Dataset as part of the GRASS Space Time Dataset. Any time-stamp associated to a single map not registered in a stds is ignored.

The extent and resolution of the resulting Dataset is defined by the region setting of GRASS, set with the g.region GRASS tool. Note that in GRASS the 3D resolution is independent from the 2D resolution. Therefore, 2D and 3D maps loaded in Xarray will not share the same dimensions and coordinates. The coordinates in the Xarray Dataset correspond to the center of the GRASS cell.

In GRASS, the time dimension of various STDSs is not homogeneous, as it is for the spatial coordinates. To account for this, xarray-grass will create one time dimension for each STDS loaded.

CF conventions attributes mapping

DataArray attributes

The attributes of the coordinates are in line with CF Conventions.

Dataset attributes

The attributes set at the dataset level are:

• crs_wkt from the g.proj command
• Conventions, the CF Convention version
• history, the time of creation and version of xarray-grass
• source, the name of the current grass project and mapset

For a GRASS XY (unprojected) location, g.proj does not provide a CRS and returns the display label XY location (unprojected). xarray-grass represents such a location as a WKT2 engineering CRS named XY location (unprojected), with a Cartesian east/north coordinate system measured in metres. This keeps the location explicitly local rather than assigning a geographic or projected CRS, while ensuring that crs_wkt can be parsed by WKT consumers such as pyproj.CRS.from_wkt.

Writing an Xarray Dataset or DataArray to GRASS

Continuing the script fro above, we can now write back the STRDS to GRASS.

There are two requirements of note to write to GRASS using xarray-grass:

• The Dataset or SataArray needs a crs_wkt attribute with the CRS information in the WKT format.
• The dims parameter needs to map every dimensions which is non standard to its standard name. The standard names are [x, y, z and start_time]

    from xarray_grass import to_grass

    # Let's write the modis time series back into the current (PERMANENT) mapset

    da_modis = test_ds["LST_Day_monthly"]
    # xarray-grass needs the CRS information to write to GRASS
    da_modis.attrs["crs_wkt"] = test_ds.attrs["crs_wkt"]

    to_grass(
        dataset=da_modis,
        dims={
            "LST_Day_monthly": {"start_time": "start_time_LST_Day_monthly"},
        },
        overwrite=False,
    )

The above print statement should return this:

<xarray.Dataset> Size: 3MB
Dimensions:                     (y: 165, x: 179, start_time_LST_Day_monthly: 24)
Coordinates:
  * y                           (y) float32 660B 2.196e+05 ... 2.212e+05
  * x                           (x) float32 716B 6.377e+05 ... 6.395e+05
  * start_time_LST_Day_monthly  (start_time_LST_Day_monthly) datetime64[ns] 192B ...
    end_time_LST_Day_monthly    (start_time_LST_Day_monthly) datetime64[ns] 192B ...
Data variables:
    boundary_county_500m        (y, x) float64 236kB ...
    elevation                   (y, x) float32 118kB ...
    LST_Day_monthly             (start_time_LST_Day_monthly, y, x) int32 3MB ...
Attributes:
    crs_wkt:      PROJCRS["NAD83(HARN) / North Carolina",BASEGEOGCRS["NAD83(H...
    Conventions:  CF-1.13-draft
    history:      2025-11-01 02:10:24.652838+00:00: Created with xarray-grass...
    source:       GRASS database. project: <nc_spm_08_grass7_test>, mapset:<P...

Roadmap

Goals for version 1.0

• Load a single raster map
• Load a single Space-time Raster Dataset (strds)
• Load a single raster_3d map
• Load a single str3ds
• Load a combination of all the above
• Load a full mapset
• Support for the drop_variables parameter
• Write from xarray to GRASS
  • Write to a 2D raster
  • Write to STRDS
  • Write to 3D raster
  • Write to STR3DS
  • Transpose if dimensions are not in the expected order
  • Support time units for relative time
  • Support end_time
  • Accept non homogeneous 3D resolution in NS and EW dimensions (GRASS 8.5)
• Lazy loading of STDS on the time dimension
• Properly test with lat-lon location

Stretch goals

• Read CRS definitions from CF compatible fields
• Lazy load on the spatial dimension