Skip to content

rioxarray — Geospatial xarray

Goal: Open GeoTIFFs as xarray objects with correct CRS, bounds, and dimensions using rioxarray, then clip, align a DEM, and write outputs.

Lessons 1–2 streamed rasters from STAC. This lesson uses files on disk so clip, reproject_match, and write stay in the foreground.

What is rioxarray?

rioxarray extends xarray so that raster I/O and spatial metadata are first-class:

  • Open rasters (GeoTIFF, COG, etc.) from a path or URL into a DataArray or Dataset.
  • Attach CRS and transform via the .rio accessor.
  • Reproject, clip, align, and write rasters using the same accessor.

Under the hood it uses rasterio for reading/writing and GDAL for formats.

Sample Data

The lesson notebook reads GeoTIFFs from data/:

File Variable
prec_daily_2024_ERA5LAND_native.tif ERA5-Land daily precipitation (m), one band per day in 2024
temp_daily_2024_ERA5LAND_native.tif ERA5-Land daily 2 m temperature (K)
copernicus_dem_30m.tif Copernicus DEM, 30 m
NL_provinces.shp Province polygons (clip to Utrecht) 
import rioxarray as rxr

prec = rxr.open_rasterio("data/prec_daily_2024_ERA5LAND_native.tif", masked=True)
print(prec.rio.crs, prec.rio.bounds(), prec.rio.resolution())

Default behavior:

  • Dimensions: often band, y, x (daily stacks use band for time until you rename it).
  • Coordinates: x and y come from the geotransform.
  • CRS / bounds / transform: data.rio.crs, data.rio.bounds(), data.rio.transform().

masked=True turns NoData into NaN.

The .rio accessor

Method / property Purpose
.rio.crs Coordinate reference system
.rio.bounds() (left, bottom, right, top)
.rio.resolution() Pixel size
.rio.transform() Affine transform (pixel ↔ geo)
.rio.clip_box() Clip to a bounding box
.rio.clip() Clip with a geometry (e.g. GeoDataFrame)
.rio.reproject() Reproject to another CRS
.rio.reproject_match() Warp onto another raster's grid
.rio.to_raster() Write to GeoTIFF/COG

Daily climate bands are mapped to a time coordinate. The 30 m DEM does not share that grid — reproject_match warps it onto one ERA5 slice:

dem = dem.rio.reproject_match(prec.isel(time=0))
ds = xr.Dataset({"prec": prec, "temp": temp, "dem": dem})

Then clip a rectangle or a polygon (Utrecht), convert kelvin → °C, mask wet days (prec > 0.001 m), and write GeoTIFF / NetCDF.

Chunked (lazy) opening

For large rasters and Dask-based workflows, open with chunks so the underlying array is a Dask array:

data = rioxarray.open_rasterio(path, chunks="auto")  # or chunks={"x": 1024, "y": 1024}

Operations then stay lazy until .compute() or plotting is triggered. This is the bridge to Lesson 4 (Dask).

Summary

Concept Detail
rioxarray Geospatial extension for xarray: open rasters, CRS, bounds, clip, reproject
Opening rioxarray.open_rasterio(path) → DataArray with band, y, x
Align reproject_match puts the DEM on the ERA5 grid
Clip clip_box (rectangle) or clip (polygon)
.rio Accessor for CRS, bounds, transform, clip, reproject, to_raster

Next, Dask runs parallel, chunked analytics on these structures.

Download the notebook for this lesson: notebooks/03-rioxarray.ipynb.