The number of countries in the world can change (South Sudan created 2011, Germany reunification), so can their boundaries. This document explain how the pre-generated files can be updated.

Nominatim imports two pre-generated files

• data/country_name.sql (country code, name, default language, partition)
• data/country_osm_grid.sql (country code, geometry)

before creating places in the database. This helps with fast lookups and missing data (e.g. if the data the user wants to import doesn't contain any country places).

Thus the country grid is mainly Fallback Country Boundaries.

Each place is assigned a country_code and partition. Partitions derive from country_code.

Each place is assigned a two letter country_code based on its location, e.g. gb for Great Britain. Or NULL if no suitable country is found (usually it's in open water then).

In sql/functions.sql: get_country_code(geometry) the place's center is checked against

1. country places already imported from the user's data file. Places are imported by rank low-to-high. Lowest rank 2 is countries so most places should be matched. Still the data file might be incomplete.
2. if unmatched: OSM grid boundaries
3. if still unmatched: OSM grid boundaries, but allow a small distance

Each place is assigned partition, which is a number 0..250. 0 is fallback/other.

During place indexing (sql/functions.sql: placex_insert()) a place is assigned the partition based on its country code (sql/functions.sql: get_partition(country_code)). It checks in the country_name table.

Most countries have their own partition, some share a partition. Thus partition counts vary greatly.

Several database tables are split by partition to allow queries to run against less indices and improve caching.

• location_area_large_<partition>
• search_name_<partition>
• location_road_<partition>

Export from existing database table plus manual changes. country_default_language_code most taken from https://wiki.openstreetmap.org/wiki/Nominatim/Country_Codes, see utils/country_languages.php.

country_grid.sql creates a base table with simplified country polygons from a global Nominatim database. The polygons are split by rectilinear lines for faster point-in-polygon lookup.

The simplification of the polygons is guided by Nominatim's requirement that it should minimize the error that a OSM object is placed in the wrong country. This means simplification can be greater on water areas where there are fewer OSM object and needs to remain quite close to the original in inhabited areas. For example, the boundaries around Baarle-Nassau need to be kept precise, while the boundary of Northern Canada will only need a few vertex points.

country_fills.sql contains manual simplifications around some areas that are known to cause issues with the subsequent automatic simplification and deduplication.

Where OSM countries overlap because an area is disputed, the same strategy as within Nominatim itself is used and the country with the smaller area wins. Exceptions can be added by editing country_fills.sql and declaring the owner country for the area.

The script can be run in a fully automatic fashion to create a new table country_osm_grid. However, it is advisable to carefully check the result because the countries in the Nominatim database are not necessarily in a clean state. PostGIS version 3.4 or higher is required for the ST_CoverageSimplify function.

To visualize one country as geojson feature collection, e.g. for loading into geojson.io:

-- http://www.postgresonline.com/journal/archives/267-Creating-GeoJSON-Feature-Collections-with-JSON-and-PostGIS-functions.html

SELECT row_to_json(fc)
FROM (
  SELECT 'FeatureCollection' As type, array_to_json(array_agg(f)) As features
  FROM (
    SELECT 'Feature' As type,
    ST_AsGeoJSON(lg.geometry)::json As geometry,
    row_to_json((country_code, area)) As properties
    FROM country_osm_grid As lg where country_code='mx'
  ) As f
) As fc;

cat /tmp/query.sql | psql -At nominatim > /tmp/mexico.quad.geojson

To create a new country_grid.sql.gz, run the following:

pg_dump  -Ox -t country_osm_grid <DATABASE NAME> | grep -v '^SET.*;' | grep -v '^SELECT.*;' | grep -v '^--' > country_grid.sql

Now add the following license header:

-- SPDX-License-Identifier:  ODbL-1.0
--
-- Copyright OpenStreetMap contributors
--
-- Simplified OSM country grid (data as of <DATE OF NOMINATIM DATABASE>)

and pack using gzip -9.

The source code is available under a GPLv2 license.