DuckDBSpatial

sf-compatible spatial operations on DuckDB-backed data — out-of-core geometries, stored as GeoParquet.

Overview

DuckDBSpatial extends DuckDBDataFrame with sf-compatible spatial methods that run on DuckDB-backed columns and tables, powered by DuckDB's native spatial extension. It exposes much of the sf spatial vocabulary --- predicates (st_intersects, st_within, st_contains), measurements (st_area, st_length), and geometry transforms (st_centroid, st_buffer, st_union) --- but the geometries stay on disk in columnar Parquet and operations are recorded as lazy SQL, so you can filter and transform spatial layers without loading them into memory.

It reads and writes GeoParquet 1.0, so the same files interoperate with GeoPandas, GDAL, QGIS, and DuckDB. Within the BiocDuckDB suite it is the spatial layer: it serves GEOMETRY columns for DuckDBDataFrame and underpins the MultiAssaySpatialExperiment on-disk format in BiocDuckDB.

Installation

# once available from Bioconductor:
if (!require("BiocManager")) install.packages("BiocManager")
BiocManager::install("DuckDBSpatial")

The DuckDB spatial extension is fetched and cached automatically the first time a spatial operation runs.

Quick start

library(DuckDBSpatial)
library(sf)

# open a GeoParquet layer as a lazy DuckDBDataFrame
spatial_path <- system.file("extdata", "spatial", package = "DuckDBSpatial")
df <- DuckDBDataFrame(spatial_path)

geom <- df[["geometry"]]
head(st_area(geom))                       # lazy; ST_Area pushed into DuckDB

query_pt <- st_sfc(st_point(c(30, 10)))
st_filter(df, query_pt)                    # keep rows intersecting the query

For plain x/y coordinate columns (e.g. transcript centroids), the layer* helpers run predicates without a geometry column:

layerSubsetByGeometry(pts, poly, coords = c("x", "y"))

GeoParquet I/O

ddb <- readGeoParquet("layer.parquet")     # lazy DuckDBDataFrame w/ GEOMETRY column
writeGeoParquet(sf_object, "out.parquet")  # GeoParquet 1.0 (WKB + metadata)

writeGeoParquet() conforms to GeoParquet 1.0.0 (WKB encoding, geometry-type and bounding-box metadata) and needs the suggested nanoparquet and jsonlite packages.

When to use DuckDBSpatial

A good fit when the spatial layer is larger than memory, when the workload is filter-heavy (selecting features by region or predicate before the expensive step), or when the data lives on disk as GeoParquet shared with other spatial tooling. An in-memory sf object remains preferable for small layers and interactive geometry editing.

Documentation

• Introduction to DuckDBSpatial — motivation, lazy spatial columns, predicates and filtering, layer engines, and GeoParquet I/O (vignettes/DuckDBSpatial.Rmd).
• Design and extension of DuckDBSpatial — the sf-generic to DuckDB spatial SQL translation, the two dispatch paths, GeoParquet 1.0 I/O, and the BiocDuckDB integration, for developers (vignettes/DuckDBSpatial-design.Rmd).

License

MIT License. Copyright Genentech, Inc., 2026.