Chapter 5. Spatial Queries

Table of Contents

The raison d'etre of spatial databases is to perform queries inside the database which would ordinarily require desktop GIS functionality. Using PostGIS effectively requires knowing what spatial functions are available, how to use them in queries, and ensuring that appropriate indexes are in place to provide good performance.

5.1. Determining Spatial Relationships

Spatial relationships indicate how two geometries interact with one another. They are a fundamental capability for querying geometry.

5.1.1. Dimensionally Extended 9-Intersection Model

According to the OpenGIS Simple Features Implementation Specification for SQL, "the basic approach to comparing two geometries is to make pair-wise tests of the intersections between the Interiors, Boundaries and Exteriors of the two geometries and to classify the relationship between the two geometries based on the entries in the resulting 'intersection' matrix."

In the theory of point-set topology, the points in a geometry embedded in 2-dimensional space are categorized into three sets:

Boundary

The boundary of a geometry is the set of geometries of the next lower dimension. For POINTs, which have a dimension of 0, the boundary is the empty set. The boundary of a LINESTRING is the two endpoints. For POLYGONs, the boundary is the linework of the exterior and interior rings.

Interior

The interior of a geometry are those points of a geometry that are not in the boundary. For POINTs, the interior is the point itself. The interior of a LINESTRING is the set of points between the endpoints. For POLYGONs, the interior is the areal surface inside the polygon.

Exterior

The exterior of a geometry is the rest of the space in which the geometry is embedded; in other words, all points not in the interior or on the boundary of the geometry. It is a 2-dimensional non-closed surface.

The Dimensionally Extended 9-Intersection Model (DE-9IM) describes the spatial relationship between two geometries by specifying the dimensions of the 9 intersections between the above sets for each geometry. The intersection dimensions can be formally represented in a 3x3 intersection matrix.

For a geometry g the Interior, Boundary, and Exterior are denoted using the notation I(g), B(g), and E(g). Also, dim(s) denotes the dimension of a set s with the domain of {0,1,2,F}:

  • 0 => point

  • 1 => line

  • 2 => area

  • F => empty set

Using this notation, the intersection matrix for two geometries a and b is:

  Interior Boundary Exterior
Interior dim( I(a) ∩ I(b) ) dim( I(a) ∩ B(b) ) dim( I(a) ∩ E(b) )
Boundary dim( B(a) ∩ I(b) ) dim( B(a) ∩ B(b) ) dim( B(a) ∩ E(b) )
Exterior dim( E(a) ∩ I(b) ) dim( E(a) ∩ B(b) ) dim( E(a) ∩ E(b) )

The following overlapping polygons provide a concrete example. ST_Relate computes the matrix 212101212; it matches the overlap pattern T*T***T** used by ST_Overlaps.

Code
WITH example AS (
  SELECT
    'POLYGON ((140 140,140 122,135 105,126 100,118 99,110 94,100 86,97 73,102 59,98 49,87 38,70 30,55 29,40 30,28 38,20 50,14 66,10 84,6 100,4 119,4 143,6 166,10 180,18 191,28 195,40 190,55 189,67 186,86 179,112 165,124 158,133 148,140 140),(48 177,40 177,30 174,24 166,22 159,25 155,30 153,40 154,45 157,52 162,53 169,51 174,48 177))'::geometry AS a,
    'POLYGON ((75 63,79 50,87 38,95 31,108 25,124 22,140 18,154 11,166 6,176 10,184 21,188 35,190 58,190 82,193 104,190 121,185 139,178 154,166 163,154 171,139 172,124 171,112 165,96 152,92 142,92 126,86 116,79 110,75 104,72 94,73 86,75 76,75 63))'::geometry AS b
), relation AS (
  SELECT a, b, ST_Relate(a, b) AS matrix
  FROM example
)
SELECT matrix,
       ST_RelateMatch(matrix, 'T*T***T**') AS matches_overlap_pattern,
       ST_Overlaps(a, b) AS overlaps
FROM relation;
Output
  matrix   | matches_overlap_pattern | overlaps
-----------+-------------------------+----------
 212101212 | t                       | t
(1 row)
Figure
Geometry figure for visual-de9im-overlapping-polygons

The executable query and generated overlay above are the visual source for the relation. The query below renders the nine cells of the same kind of matrix as a 3-by-3 set of examples. The final exterior/exterior cell is clipped to a finite viewport, since the true exterior/exterior intersection is unbounded.

Code
WITH example AS (
  SELECT
    ST_MakeEnvelope(0, 0, 6, 4) AS a,
    ST_MakeEnvelope(3, -1, 8, 3) AS b,
    ST_MakeEnvelope(-1, -2, 9, 5) AS viewport
), parts AS (
  SELECT
    a,
    b,
    ST_Boundary(a) AS ba,
    ST_Boundary(b) AS bb,
    viewport
  FROM example
)
SELECT
  ST_Relate(a, b) AS matrix,
  ST_AsText(ST_Normalize(a)) AS input_a,
  ST_AsText(ST_Normalize(b)) AS input_b,
  ST_AsText(ST_Normalize(ST_Intersection(a, b))) AS "2 I(a) ∩ I(b)",
  ST_AsText(ST_Normalize(ST_Intersection(a, bb))) AS "1 I(a) ∩ B(b)",
  ST_AsText(ST_Normalize(ST_Difference(a, b))) AS "2 I(a) ∩ E(b)",
  ST_AsText(ST_Normalize(ST_Intersection(ba, b))) AS "1 B(a) ∩ I(b)",
  ST_AsText(ST_Normalize(ST_Intersection(ba, bb))) AS "0 B(a) ∩ B(b)",
  ST_AsText(ST_Normalize(ST_Collect(
      'LINESTRING(0 0,0 4,6 4,6 3)'::geometry,
      'LINESTRING(0 0,3 0)'::geometry
  ))) AS "1 B(a) ∩ E(b)",
  ST_AsText(ST_Normalize(ST_Difference(b, a))) AS "2 E(a) ∩ I(b)",
  ST_AsText(ST_Normalize(ST_Collect(
      'LINESTRING(3 -1,8 -1,8 3,6 3)'::geometry,
      'LINESTRING(3 -1,3 0)'::geometry
  ))) AS "1 E(a) ∩ B(b)",
  ST_AsText(ST_Normalize(ST_Difference(viewport, ST_Union(a, b))))
    AS "2 E(a) ∩ E(b) clipped"
FROM parts;
Output
-[ RECORD 1 ]----------+-----------------------------------------------------------------------------
matrix                 | 212101212
input_a                | POLYGON((0 0,0 4,6 4,6 0,0 0))
input_b                | POLYGON((3 -1,3 3,8 3,8 -1,3 -1))
2 I(a) ∩ I(b)          | POLYGON((3 0,3 3,6 3,6 0,3 0))
1 I(a) ∩ B(b)          | LINESTRING(3 0,3 3,6 3)
2 I(a) ∩ E(b)          | POLYGON((0 0,0 4,6 4,6 3,3 3,3 0,0 0))
1 B(a) ∩ I(b)          | LINESTRING(3 0,6 0,6 3)
0 B(a) ∩ B(b)          | MULTIPOINT((6 3),(3 0))
1 B(a) ∩ E(b)          | MULTILINESTRING((0 0,0 4,6 4,6 3),(0 0,3 0))
2 E(a) ∩ I(b)          | POLYGON((3 -1,3 0,6 0,6 3,8 3,8 -1,3 -1))
1 E(a) ∩ B(b)          | MULTILINESTRING((3 -1,8 -1,8 3,6 3),(3 -1,3 0))
2 E(a) ∩ E(b) clipped  | POLYGON((-1 -2,-1 5,9 5,9 -2,-1 -2),(0 0,3 0,3 -1,8 -1,8 3,6 3,6 4,0 4,0 0))
Figure
Geometry figure for visual-de9im-matrix-cells

Reading from left to right and top to bottom, the intersection matrix is represented as the text string '212101212'.

For more information, refer to:

5.1.2. Named Spatial Relationships

To make it easy to determine common spatial relationships, the OGC SFS defines a set of named spatial relationship predicates. PostGIS provides these as the functions ST_Contains, ST_Crosses, ST_Disjoint, ST_Equals, ST_Intersects, ST_Overlaps, ST_Touches, ST_Within. It also defines the non-standard relationship predicates ST_Covers, ST_CoveredBy, and ST_ContainsProperly.

Spatial predicates are usually used as conditions in SQL WHERE or JOIN clauses. The named spatial predicates automatically use a spatial index if one is available, so there is no need to use the bounding box operator && as well. For example:

Code
SELECT city.name, state.name, city.geom
FROM city JOIN state ON ST_Intersects(city.geom, state.geom);

For more details and illustrations, see the PostGIS Workshop.

5.1.3. General Spatial Relationships

In some cases the named spatial relationships are insufficient to provide a desired spatial filter condition. These requirements can be expressed by computing the full DE-9IM intersection matrix with ST_Relate.

To test a particular spatial relationship, an intersection matrix pattern is used. This is the matrix representation augmented with the additional symbols {T,*}:

  • T => intersection dimension is non-empty; i.e. is in {0,1,2}

  • * => don't care

Using intersection matrix patterns, specific spatial relationships can be evaluated succinctly. The ST_Relate and ST_RelateMatch functions can both test these patterns.

These road segments share a line segment. ST_Crosses does not find this relationship for linear features, because it only returns true when their interiors intersect at a point. The pattern 1*1***1** tests for a one-dimensional intersection instead.

Code
WITH roads AS (
  SELECT
    'LINESTRING(10 10,40 90,70 110,140 110,170 130,190 190)'::geometry AS "road A",
    'LINESTRING(10 190,50 130,90 110,130 110,160 70,180 10)'::geometry AS "road B"
), relation AS (
  SELECT "road A", "road B", ST_Relate("road A", "road B") AS matrix
  FROM roads
)
SELECT
  ST_Intersection("road A", "road B") AS overlap,
  matrix,
  ST_Crosses("road A", "road B") AS "ST_Crosses",
  ST_RelateMatch(matrix, '1*1***1**') AS "matches 1*1***1**"
FROM relation;
Output
          overlap           |  matrix   | ST_Crosses | matches 1*1***1**
----------------------------+-----------+------------+-------------------
 LINESTRING(90 110,130 110) | 1F1FF0102 | f          | t
(1 row)
Figure
Geometry figure for visual-relate-overlapping-roads

The next example finds a wharf which runs from the lake interior onto its shoreline, with one endpoint on the shoreline. The other wharves provide nearby counterexamples in the figure. The pattern 102101FF2 expresses the complete required relationship in one test.

Code
WITH data AS (
  SELECT
    'POLYGON((-20 10,30 30,80 70,110 80,145 85,190 110,300 220,230 220,-20 220,-20 10))'::geometry AS lake,
    'MULTILINESTRING((120 180,145 85,110 80),(10 130,30 70),(90 140,110 80,94.81118881118876 75.83496503496498),(150 160,180 80))'::geometry AS wharves
), wharf AS (
  SELECT lake, (item).path[1] AS id, (item).geom
  FROM data CROSS JOIN LATERAL ST_Dump(wharves) AS item
), relation AS (
  SELECT id, geom, ST_Relate(lake, geom) AS matrix
  FROM wharf
)
SELECT
  id AS wharf,
  geom AS "matching wharf",
  matrix
FROM relation
WHERE ST_RelateMatch(matrix, '102101FF2')
ORDER BY id;
Output
 wharf |           matching wharf           |  matrix
-------+-----------------------------------+-----------
     1 | LINESTRING(120 180,145 85,110 80) | 102101FF2
(1 row)
Figure
Geometry figure for visual-relate-lake-wharves

5.2. Using Spatial Indexes

When constructing queries using spatial conditions, for best performance it is important to ensure that a spatial index is used, if one exists (see Section 4.9, “Spatial Indexes”). To do this, a spatial operator or index-aware function must be used in a WHERE or ON clause of the query.

Spatial operators include the bounding box operators (of which the most commonly used is &&; see Section 7.10.1, “Bounding Box Operators” for the full list) and the distance operators used in nearest-neighbor queries (the most common being <->; see Section 7.10.2, “Distance Operators” for the full list.)

Index-aware functions automatically add a bounding box operator to the spatial condition. Index-aware functions include the named spatial relationship predicates ST_Contains, ST_ContainsProperly, ST_CoveredBy, ST_Covers, ST_Crosses, ST_Intersects, ST_Overlaps, ST_Touches, ST_Within, ST_Within, and ST_3DIntersects, and the distance predicates ST_DWithin, ST_DFullyWithin, ST_3DDFullyWithin, and ST_3DDWithin .)

Functions such as ST_Distance do not use indexes to optimize their operation. For example, the following query would be quite slow on a large table:

Code
SELECT geom
FROM geom_table
WHERE ST_Distance(geom, 'SRID=312;POINT(100000 200000)') < 100

This query selects all the geometries in geom_table which are within 100 units of the point (100000, 200000). It will be slow because it is calculating the distance between each point in the table and the specified point, ie. one ST_Distance() calculation is computed for every row in the table.

The number of rows processed can be reduced substantially by using the index-aware function ST_DWithin:

Code
SELECT geom
FROM geom_table
WHERE ST_DWithin(geom, 'SRID=312;POINT(100000 200000)', 100)

This query selects the same geometries, but it does it in a more efficient way. This is enabled by ST_DWithin() using the && operator internally on an expanded bounding box of the query geometry. If there is a spatial index on geom, the query planner will recognize that it can use the index to reduce the number of rows scanned before calculating the distance. The spatial index allows retrieving only records with geometries whose bounding boxes overlap the expanded extent and hence which might be within the required distance. The actual distance is then computed to confirm whether to include the record in the result set.

For more information and examples see the PostGIS Workshop.

5.3. Examples of Spatial SQL

The examples in this section make use of a table of linear roads, and a table of polygonal municipality boundaries. The definition of the bc_roads table is:

Output
Column    | Type              | Description
----------+-------------------+-------------------
gid       | integer           | Unique ID
name      | character varying | Road Name
geom      | geometry          | Location Geometry (Linestring)

The definition of the bc_municipality table is:

Output
Column   | Type              | Description
---------+-------------------+-------------------
gid      | integer           | Unique ID
code     | integer           | Unique ID
name     | character varying | City / Town Name
geom     | geometry          | Location Geometry (Polygon)

5.3.1.

What is the total length of all roads, expressed in kilometers?

You can answer this question with a very simple piece of SQL:

Code
SELECT sum(ST_Length(geom))/1000 AS km_roads FROM bc_roads;
Output
km_roads
------------------
70842.1243039643

5.3.2.

How large is the city of Prince George, in hectares?

This query combines an attribute condition (on the municipality name) with a spatial calculation (of the polygon area):

Code
SELECT
  ST_Area(geom)/10000 AS hectares
FROM bc_municipality
WHERE name = 'PRINCE GEORGE';
Output
hectares
------------------
32657.9103824927

5.3.3.

What is the largest municipality in the province, by area?

This query uses a spatial measurement as an ordering value. There are several ways of approaching this problem, but the most efficient is below:

Code
SELECT
  name,
  ST_Area(geom)/10000 AS hectares
FROM bc_municipality
ORDER BY hectares DESC
LIMIT 1;
Output
name           | hectares
---------------+-----------------
TUMBLER RIDGE  | 155020.02556131

Note that in order to answer this query we have to calculate the area of every polygon. If we were doing this a lot it would make sense to add an area column to the table that could be indexed for performance. By ordering the results in a descending direction, and them using the PostgreSQL "LIMIT" command we can easily select just the largest value without using an aggregate function like MAX().

5.3.4.

What is the length of roads fully contained within each municipality?

This is an example of a "spatial join", which brings together data from two tables (with a join) using a spatial interaction ("contained") as the join condition (rather than the usual relational approach of joining on a common key):

Code
SELECT
  m.name,
  sum(ST_Length(r.geom))/1000 as roads_km
FROM bc_roads AS r
JOIN bc_municipality AS m
  ON ST_Contains(m.geom, r.geom)
GROUP BY m.name
ORDER BY roads_km;
Output
name                        | roads_km
----------------------------+----------
SURREY                      | 1539.476
VANCOUVER                   | 1450.331
LANGLEY DISTRICT            |  833.793
BURNABY                     |  773.769
PRINCE GEORGE               |  694.376
...

This query takes a while, because every road in the table is summarized into the final result (about 250K roads for the example table). For smaller datasets (several thousand records on several hundred) the response can be very fast.

5.3.5.

Create a new table with all the roads within the city of Prince George.

This is an example of an "overlay", which takes in two tables and outputs a new table that consists of spatially clipped or cut resultants. Unlike the "spatial join" demonstrated above, this query creates new geometries. An overlay is like a turbo-charged spatial join, and is useful for more exact analysis work:

Code
CREATE TABLE pg_roads as
SELECT
  ST_Intersection(r.geom, m.geom) AS intersection_geom,
  ST_Length(r.geom) AS rd_orig_length,
  r.*
FROM bc_roads AS r
JOIN bc_municipality AS m
  ON ST_Intersects(r.geom, m.geom)
WHERE
  m.name = 'PRINCE GEORGE';

5.3.6.

What is the length in kilometers of "Douglas St" in Victoria?

Code
SELECT
  sum(ST_Length(r.geom))/1000 AS kilometers
FROM bc_roads r
JOIN bc_municipality m
  ON ST_Intersects(m.geom, r.geom
WHERE
  r.name = 'Douglas St'
  AND m.name = 'VICTORIA';
Output
kilometers
------------------
4.89151904172838

5.3.7.

What is the largest municipality polygon that has a hole?

Code
SELECT gid, name, ST_Area(geom) AS area
FROM bc_municipality
WHERE ST_NRings(geom) > 1
ORDER BY area DESC LIMIT 1;
Output
gid  | name         | area
-----+--------------+------------------
12   | SPALLUMCHEEN | 257374619.430216