Mexico full data process finds general poor detection. 43% of urban blocks has no buildingFootprints.

Mexico data process

Mexico full data process reports general poor detection and null buildings structures detection on 43% of urban blocks.

• Download data https://github.com/microsoft/GlobalMLBuildingFootprints https://minedbuildings.blob.core.windows.net/global-buildings/2022-05-02/Mexico.zip

• unzip file Mexico.zip

• Load info into postgis DB

  • /Applications/Postgres.app/Contents/Versions/13/bin/ogr2ogr -f "PostgreSQL" PG:"dbname=dbname user=user" /Volumes/PROJECT_DATA/microsoft_GlobalMLBuildingFootprints/Mexico.geojsonl -nln mexicoblds
  • 40 minutes
  • At this point data is useless. There no way to identify information related to a known area over the country

• Load info prom INEGI's state & municipal polygons
  • /Applications/Postgres.app/Contents/Versions/13/bin/ogr2ogr -f "PostgreSQL" PG:"dbname=dbname user=user" /Volumes/PROJECT_DATA/INEGI/CARTOGRAFIA/2020/MARCO_GEOESTADISTICO_CENSO2020/889463807469_s/MG_2020_Integrado/munici/mexmun.geojsonl -nln mexshpmun
  • less than a minute
• INEGI's state & municipal geo information is the basic area division

• Create centroids for each polygon buildings
  • create table mexicobldscen as ( SELECT ogc_fid, ST_centroid(wkb_geometry) as centroide FROM mexicoblds );
  • 2 minutes

• Assign municipal and state code to each polygon building via centroids
  • create table mexicoblds_mun as (
    select a.ogc_fid, a.centroide,
    b.cvegeo,
    b.cve_ent,
    b.cve_mun,
    b.nomgeo,
    c.wkb_geometry
    from mexicobldscen a
    inner join mexshpmun b
    on ST_Intersects(a.centroide, b.wkb_geometry)
    inner join mexicoblds c
    on a.ogc_fid = c.ogc_fid);
  • 35 minutes
  • Municipal and State code produces a rich Geo_object for easy public's data query.

  ---

• Let's try something complex: Assign state, municipal and zip codes via centroid
  • create table mex_bld_mun_cp as (
    select a.ogc_fid, a.centroide,
    b.cvegeo,
    b.cve_ent,
    b.cve_mun,
    b.nomgeo,
    d.cp,
    c.wkb_geometry
    from mexicobldscen a
    inner join mexshpmun b
    on ST_Intersects(a.centroide, b.wkb_geometry)
    inner join mexicoblds c
    on a.ogc_fid = c.ogc_fid
    left outer join mexcodpos d
    on ST_Intersects(a.centroide, d.wkb_geometry));
  • 1 hour and 8 minutes
  • The new data can respond queries by State, municipal or ZIP codes.

• Let's try assign the smaller INEGI's geo polygon representations:
  - Localidades - urban areas -
  - AGEBS - Geo Statistical Basic Areas
  - Manzana Urbana - Blocks
  • create table mex_bld_mun_cp_mza as (
    select
    a.ogc_fid,
    a.centroide,
    m.cvegeo,
    m.cve_ent,
    m.cve_mun,
    m.cve_loc,
    m.cve_ageb,
    m.cve_mza,
    m.ambito,
    m.tipomza,
    a.cp,
    a.wkb_geometry
    from mex_bld_mun_cp a
    inner join mexmza m
    on ST_Intersects(a.centroide, m.wkb_geometry) );
• 47 minutes (recycle previous process)

Results

Poor buildings footprints detection in most urban areas in Mexico.

Data reports 23,769,529 buildings footprints.

There are 19,356,595 buildings footprints detected in 1,367,908 INEGI´s urban blocks.

4,412,934 buildings footprints can not be assigned to geo urban blocks.

1,062,208 INEGI´s urban blocks has no buildings footprints.

The buildings footprints process detection do not report urban structures on 43.7% of geo urban blocks.

No accuracy evaluation was performed in this excercise.

Start by reproducing the Mexico.zip import and the documented PostGIS queries against the INEGI polygon tables. Compare the 43.7% of urban blocks without footprints with the source data and query results. Before implementation, clarify whether the expected outcome is improved detection, corrected geographic assignment, or better reporting of missing buildings.