gdal_rasterize

What is gdal_rasterize?

gdal_rasterize converts vector features — points, lines, or polygons — into pixel values in a raster. It can create a new raster with an explicit extent and resolution, or burn features into an existing raster, updating pixels either with a constant value or with values taken from an attribute. This is the canonical tool for building masks, categorical rasters from polygon layers, and sparse point-density rasters.

gdal_rasterize [options] <src_datasource> <dst_filename>

Commonly used options:

• -b <band> — target band(s) in an existing raster
• -burn <value> — constant burn value (repeat for multiple bands)
• -a <attribute> — burn attribute from the vector source
• -3d — use Z coordinate as burn value
• -add — accumulate into existing values rather than overwrite
• -l <layer> / -sql <statement> / -where <expr> — source selection
• -te <xmin> <ymin> <xmax> <ymax> / -tr <xres> <yres> / -ts <xsize> <ysize> — extent, resolution, or pixel size for new outputs
• -tap — snap grid to target resolution
• -a_nodata <value> / -init <value> — NoData or initialisation value
• -at — enable "all-touched" mode so every pixel touched by geometry is burned
• -of <format> / -ot <type> / -co <NAME>=<VALUE> — output driver, type, creation options

When would you use gdal_rasterize?

Use gdal_rasterize whenever vector data needs to be represented as pixels. Typical jobs: creating a binary mask raster from a watershed boundary (gdal_rasterize -burn 1 -tr 10 10 -te ... boundary.gpkg mask.tif), building a categorical land-use raster from a polygon layer by burning a class-code attribute (-a class_code), or producing a sparse presence raster from point observations. Pair with gdal_proximity to convert that sparse raster into a distance-to-features surface.

For analysis that requires alignment with an existing raster, create the target with gdal_create or copy the reference grid, then burn into it: gdal_rasterize -a class_code -b 1 polygons.gpkg reference.tif. The -at (all-touched) mode is essential for thin features that would otherwise miss many intersecting pixels under the default centre-point rule.

FAQs

When should I use -at (all-touched) mode?

Enable -at whenever the source features are narrow relative to the pixel size — linear features like roads or streams, small polygons, or point data. Without -at, GDAL burns only pixels whose centres fall inside a geometry, which will miss most pixels crossed by a road or a one-metre-wide polygon on a 10-metre grid. The trade-off is that -at over-represents area for polygons.

How do I create a new raster sized to the vector extent?

Pass -te <xmin> <ymin> <xmax> <ymax> and either -tr <xres> <yres> or -ts <xsize> <ysize>. Snap the grid to a meaningful origin with -tap so outputs align with other rasters at the same resolution. Set -a_nodata and -init to control the initial fill of non-burned pixels.

Can I burn different values per feature?

Yes — use -a <attribute> and the source field's numeric value is burned for each feature. Combine with -sql or -where to subset first. For multi-band outputs, repeat -burn per band or use -a on several invocations with different -b targets.

Why does my rasterised polygon layer have stair-step artefacts?

Rasterisation is inherently discrete; boundaries are approximated to the nearest pixel. Use a finer target resolution if the artefact matters, or consider anti-aliased rendering in a cartographic stack. -at reduces omission but does not produce smooth edges — it just switches the rule for which pixels count as "inside".