Overview

Rasterix provides dask-aware tools for converting vector geometries to raster data.

Functions

rasterize

rasterize() burns geometry indices into a raster. Each pixel is assigned an integer corresponding to the geometry that covers it.

from rasterix.rasterize import rasterize

result = rasterize(ds, geometries, xdim="x", ydim="y")

Use cases:

• Creating categorical rasters from polygons

• Generating zone maps for zonal statistics

• Labeling pixels by region

geometry_mask

geometry_mask() creates a boolean mask indicating which pixels fall within any of the provided geometries.

from rasterix.rasterize import geometry_mask

mask = geometry_mask(ds, geometries, xdim="x", ydim="y")

Use cases:

• Creating masks for spatial selection

• Identifying pixels inside/outside regions

geometry_clip

geometry_clip() clips raster data to geometries, setting pixels outside the geometries to NaN.

from rasterix.rasterize import geometry_clip

clipped = geometry_clip(ds, geometries, xdim="x", ydim="y")

Use cases:

• Extracting data for a specific region

• Removing data outside an area of interest

Engines

Three rasterization engines are available for the functions above:

Engine	GDAL Required	all_touched	Notes
rasterio	Yes	Yes	Most compatible, requires GDAL
rusterize	No	No	Fast, easy to install, no GDAL
exactextract	No	No*	Sub-pixel precision, no GDAL

* exactextract naturally includes any pixel with non-zero coverage, similar to all_touched=True

Choosing an Engine

• rasterio: Best when you need all_touched=True or already have GDAL installed

• rusterize: Best for simple rasterization without GDAL dependency (default when available)

• exactextract: Best when you need sub-pixel precision without GDAL

# Auto-select engine (prefers rusterize, falls back to rasterio)
result = rasterize(ds, geometries, xdim="x", ydim="y")

# Explicitly choose an engine
result = rasterize(ds, geometries, xdim="x", ydim="y", engine="rasterio")
result = rasterize(ds, geometries, xdim="x", ydim="y", engine="rusterize")
result = rasterize(ds, geometries, xdim="x", ydim="y", engine="exactextract")

Out-of-Core Support

All operations support dask arrays and dask-geopandas for out-of-core computation:

import dask_geopandas as dgpd

# Chunked raster + dask geometries
result = rasterize(
    ds.chunk({"y": 100, "x": 100}),
    dgpd.from_geopandas(geometries, npartitions=4),
    xdim="x",
    ydim="y",
)

Coverage Calculation

In addition to the rasterization functions above, coverage() computes the precise fractional area of each pixel covered by each geometry. This is only available with the exactextract engine.

from rasterix.rasterize.exact import coverage

# Fractional coverage (0-1)
cov = coverage(ds, geometries, xdim="x", ydim="y", coverage_weight="fraction")

# Area in square meters
area = coverage(ds, geometries, xdim="x", ydim="y", coverage_weight="area_spherical_m2")

# Binary coverage (0 or 1)
binary = coverage(ds, geometries, xdim="x", ydim="y", coverage_weight="none")

Use cases:

• Area-weighted aggregations

• Accurate zonal statistics

• Sub-pixel analysis

Note

The coverage() function returns a sparse.COO array, which efficiently stores the sparse coverage matrix. This is important because most pixels have zero coverage for any given geometry.