Data model

GeoDataFrames works with ordinary DataFrames and with Tables.jl-compatible tables. A spatial table has one or more columns that contain GeoInterface.jl-compatible geometries. The geometry-column names and coordinate reference system are recorded as table metadata; see Metadata and CRS.

GeometryVector is an implementation detail used for geometry columns in some read results. It is a mutable vector wrapper that can retain a spatial index. Tables supplied to write do not need to use GeometryVector.

Geometry columns and table metadata

The table records which columns contain geometries and its coordinate reference system as metadata. This keeps spatial context with the table without requiring each geometry value to carry the same information. See metadata and CRS for its rationale and the metadata reference for the keys and accessors.

Read-time geometry storage

Reading can place a geometry column in a GeometryVector, a mutable wrapper around an underlying vector, A. Its optional index field holds a cached spatial tree. Supported mutations clear that cache, so it cannot describe stale geometries. GeometryVector is an implementation detail of read results: a table supplied to write need only provide GeoInterface-compatible geometry columns. See spatial indexes for the supported index workflow.

Interoperability

Because a spatial table is a DataFrame, code that accepts a DataFrame can use it directly. I/O also accepts Tables.jl- compatible input, so a table need not use a GeoDataFrames-specific container before it is written. For reading and writing procedures, see read and write vector data.