Spatial databases
A spatial database is a database optimized to store and query data representing objects defined in a geometric space — points, lines, and polygons, along with more complex shapes such as multi-polygons and geometry collections. Where a conventional database treats each column as an atomic value, a spatial database treats a geometry as a first-class value with its own operators, indexes, and constraints.
Spatial databases are usually implemented as extensions to general-purpose database systems rather than as standalone engines. Most commonly, relational databases provide the underlying storage, query planner, and transaction machinery, and a spatial extension layer adds a geometry type, spatial functions, and spatial indexes. The Open Geospatial Consortium (OGC) Simple Features for SQL standard, and the corresponding SQL/MM Part 3: Spatial clause of the SQL standard, codify the geometry types, predicates, and functions that a conforming spatial database should expose.
Spatial queries
The queries a spatial database is built to answer go beyond ordinary key lookups. The common families are the following.
- Containment and within. Does one geometry contain another, or lie inside another? Used to find which administrative boundary a given point falls in.
- Intersection and overlap. Do two geometries share any points? Used to find features crossed by a proposed road or pipeline.
- Distance and proximity. How far apart are two geometries, or which features lie within a given radius of a point? Backs the "find nearby" queries behind mapping and ridesharing applications.
- Nearest-neighbour. Which K features are closest to a given point, ordered by distance?
- Bounding-box and window queries. Which geometries intersect a rectangular region of interest?
Spatial indexing
Conventional database indexes such as B-trees order values along a single dimension. Coordinates are two-dimensional (or three, with elevation; four, with time), so a B-tree on latitude and longitude separately cannot answer a two-dimensional range query without scanning one dimension wholesale. Spatial databases use index structures that partition space itself.
- R-trees group nearby geometries into minimum bounding rectangles (MBRs) and nest those rectangles recursively. A query for geometries intersecting a region descends only the branches whose MBRs overlap the query region, pruning the rest. R-trees are the default in PostGIS, Oracle Spatial, and SQLite SpatiaLite.
- Quadtrees recursively subdivide a two-dimensional region into four quadrants, stopping when a cell holds few enough geometries. They suit uniformly distributed point data.
- Space-filling curves such as the Z-order (Morton) curve linearise two-dimensional coordinates into a single sortable key, letting a standard B-tree answer spatial range queries approximately. Geohashing is one application of this idea.
Coordinate reference systems
A coordinate pair only means something relative to a coordinate reference
system (CRS) — the model that maps coordinates to positions on the Earth.
Spatial databases store an integer identifier, the spatial reference system
identifier (SRID), alongside each geometry to record which CRS it uses. The
EPSG registry is the conventional source of SRID values, eg. EPSG:4326 for
WGS 84 latitude/longitude, or EPSG:3857 for the Web Mercator projection used
by online map tiles.
Operations between geometries require both to be in the same CRS, or one must
be transformed into the other’s. Spatial databases provide functions such as
ST_Transform for this. Getting the CRS wrong is a common and silent source
of error — distance computed between two points stored in projected meters
differs by orders of magnitude from distance between the same points stored
in geographic degrees.
Geographic databases
A geographic database (or geodatabase) is a category of spatial database in which the spatial data represents geographic data (or geodata) – data representing locations on Earth. Geodatabases are the storage layer of geographic information systems (GIS). The distinction is mostly one of domain. A spatial database may hold the floor plan of a building or the layout of a VLSI chip, neither of which is geographic, while a geodatabase is concerned with the Earth.
Notable implementations
- PostGIS is the spatial extension for PostgreSQL, and the de facto open source spatial database. It implements the OGC Simple Features standard and adds raster, topology, and geography (geodetic) types.
- Oracle Spatial (licensed separately from Oracle Database) provides spatial types, indexing, and analysis aimed at enterprise GIS workloads.
- SpatiaLite is the spatial extension for SQLite, providing a lightweight, embeddable spatial database.
- SQL Server Spatial ships built into Microsoft SQL Server as the
geometryandgeographytypes. - Several NoSQL stores, including MongoDB and Elasticsearch, support geo-queries and geo-indexes, though without the full OGC type and predicate model.