Rasterizer

rasterizer is a lightweight Python package that speeds up rasterization of geopandas GeoDataFrames by specializing in regular, axis-aligned rectangular grids.

Features

• Rasterize lines into a binary (presence/absence) or length-based grid.
• Rasterize polygons into a binary (presence/absence) or area-based grid.
• Fast because it targets regular rectilinear grids described by 1D x and y cell-center coordinates with constant spacing.
• Fast large-polygon rasterization: exact per-cell clipping for tiny bounding boxes, signed per-edge area accumulation for everything larger, so cost scales with edge-cell crossings instead of vertices times boundary cells.
• Weighted rasterization: Rasterize geometries while weighting the output by a numerical column in the GeoDataFrame.
• Works with geopandas GeoDataFrames.
• Outputs an xarray.DataArray for easy integration with other scientific Python libraries.
• No GDAL dependency for the rasterization algorithm itself.

For detailed usage and API documentation, please see the full documentation.

Installation

You can install the package directly from PyPI:

pip install rasterizer

Usage

Here are some examples of what you can do with rasterizer.

import geopandas as gpd
from rasterizer import rasterize_polygons

polys = gpd.read_file("polygons.gpkg")
area_raster = rasterize_polygons(polys, your_x_grid, your_y_grid, polys.crs, mode="area")

# Enable a tqdm progress bar when processing large geometry collections.
area_raster = rasterize_polygons(
    polys,
    your_x_grid,
    your_y_grid,
    polys.crs,
    mode="area",
    progress_bar=True,
)

Rasterizing Lines

You can rasterize lines in either binary or length mode.

Binary Mode	Length Mode

Rasterizing Polygons

You can rasterize polygons in either binary or area mode.

For polygon workloads, rasterizer picks an internal strategy per polygon. Small polygon bounding boxes are handled with exact per-cell clipping. Larger ones use a signed per-edge area accumulation: each edge distributes exact area and cover contributions over the cells it crosses, and a row sweep turns the running cover into interior fill. Boundary cells keep exact fractional areas while the total cost scales with edge-cell crossings plus bbox size instead of re-clipping the polygon for every boundary cell.

Binary Mode	Area Mode

Large Dataset Showcase

This real-world example uses 606,667 building polygons on a 10 m Lambert-93 grid covering Paris. The area rasterization step completes in 3.9 s on a regular laptop used as the local documentation machine for a 2804 x 1978 grid.

import geopandas as gpd
import numpy as np
from rasterizer import rasterize_polygons

buildings = gpd.read_file(
    "BDT_3-5_GPKG_LAMB93_D075-ED2026-03-15.gpkg",
    layer="batiment",
    columns=[],
)

xmin, ymin, xmax, ymax = buildings.total_bounds
x = np.arange(xmin, xmax, 10.0)
y = np.arange(ymin, ymax, 10.0)

coverage = rasterize_polygons(buildings, x=x, y=y, crs=buildings.crs, mode="area")

The full walkthrough, including the benchmark context and reproduction script, is available in the large dataset showcase documentation.

Why rasterizer

This package provides functionalities that are not present in rasterio.features, such as area and length-based rasterization. It is also lighter and faster than using more general GDAL-based solutions because it is specialized for regular rectilinear grids instead of arbitrary raster layouts. GDAL's rasterization only burns values per pixel; it cannot return exact fractional area or length contributions without an expensive workaround. The common workaround is to rasterize at a much finer resolution and then downsample with averaging, which approximates the true area/length but is not exact and can be slow, e.g.:

gdal_rasterize -burn 1 -tr 1 1 -ot Float32 -of GTiff input.gpkg tmp_fine.tif
gdalwarp -tr 10 10 -r average tmp_fine.tif out_area_approx.tif

Doing this purely in geopandas by generating one polygon per grid cell and overlaying it with the input geometry is also slow because it creates a huge number of tiny geometries, triggers expensive overlay operations, and scales poorly with grid size.

That speed-up comes with a deliberate constraint: rasterizer is built for regular, axis-aligned rectangular grids, not for arbitrary affine transforms or irregular meshes.