GeoArrow Specification

Version 0.2.

This repository contains a specification for storing geospatial data in Apache Arrow and Arrow-compatible data structures and formats.

The Apache Arrow project specifies a standardized language-independent columnar memory format. It enables shared computational libraries, zero-copy shared memory and streaming messaging, interprocess communication, and is supported by many programming languages and data libraries.

Spatial information can be represented as a collection of discrete objects using points, lines and polygons (i.e., vector data). The Simple Feature Access standard provides a widely used abstraction, defining a set of geometries: Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon, and GeometryCollection. Next to a geometry, simple features can also have non-spatial attributes that describe the feature.

Geospatial data often comes in tabular format, with one or more columns with feature geometries and additional columns with feature attributes. The Arrow columnar memory model is well-suited to store both vector features and their attribute data. The GeoArrow specification defines how the vector features (geometries) can be stored in Arrow (and Arrow-compatible) data structures.

This repository contains the specifications for:

• The memory layout for storing geometries in an Arrow array (format.md)
• The Arrow extension type definitions that ensure type-level metadata (e.g., CRS) is propagated when used in Arrow implementations (extension-types.md)

Defining a standard and efficient way to store geospatial data in the Arrow memory layout enables interoperability between different tools and ensures geospatial tools can leverage the growing Apache Arrow ecosystem:

• Efficient, columnar file formats. Leveraging the performant and compact storage of Apache Parquet as a vector data format in geospatial tools using GeoParquet
• Accelerated between-process geospatial data exchange using Apache Arrow IPC message format and Apache Arrow Flight
• Zero-copy in-process geospatial data transport using the Apache Arrow C Data Interface (e.g., GDAL)
• Shared libraries for geospatial data type representation and computation for query engines that support columnar data formats (e.g., Velox, DuckDB, and Acero)

Relationship with GeoParquet

The GeoParquet specification originally started in this repo, but was moved out into its own repo, leaving this repo to focus on the Arrow-specific specifications (Arrow layout and extension type metadata). Whereas GeoParquet is a file-level metadata specification, GeoArrow is a field-level metadata and memory layout specification that applies in-memory (e.g., an Arrow array), on disk (e.g., using Parquet readers/writers provided by an Arrow implementation), and over the wire (e.g., using the Arrow IPC format).

Implementations

• geoarrow-c: geospatial type system and generic coordinate-shuffling library written in C with bindings in C++, R, and Python
• geoarrow-rs: Rust implementation of the GeoArrow specification and bindings to GeoRust algorithms for efficient spatial operations on GeoArrow memory. See also:
  • Python bindings to geoarrow-rs
  • geoarrow-wasm, JavaScript (WebAssembly) bindings to geoarrow-rs.
• geoarrow-python: Python bindings to geoarrow-c that provide integrations with libraries like pyarrow, pandas, and geopandas.
• geoarrow-r: R bindings to geoarrow-c that provide integrations with libraries like sf and Arrow for geospatial data handling.
• geoarrow-js: Pure TypeScript implementation of GeoArrow, on top of the Arrow JavaScript implementation.

Downstream libraries

• Lonboard: fast, interactive geospatial vector data visualization in Jupyter, building on top of GeoArrow.