Mexico Data Curation¶

This document explains how to process and standardize the raw datasets for the Mexico simulation.

Raw Data Sources¶

• data/mexico/opencellid/334.csv: Cell locations for MCC 334.
  • Source: OpenCellID. 200,641 records, of which Telcel holds 163,116 (334-20), Movistar 14,146 (334-03) and AT&T Mexico 12,461 (334-50).
• data/mexico/agent-unprocessed-raw-datasets/muni_2018gw.*: Municipal boundaries, already in WGS84 geographic coordinates.
  • Source: CONABIO, which republishes the INEGI Marco Geoestadistico. This 40 MB file is used in place of the raw INEGI bundle, which ships at 3.4 GB because it also carries block-level geometry the simulator does not need.
• data/mexico/agent-unprocessed-raw-datasets/conjunto_de_datos_iter_00CSV20.csv: Population by locality, Censo 2020 ITER.
  • Source: INEGI, Censo de Poblacion y Vivienda 2020.

Processing Script¶

data/processing_scripts/standardize_mexico.py¶

1. Reads the ITER file and keeps municipal totals, which are the rows with LOC == 0000 and a non-zero municipality key. The file is UTF-8 with a byte order mark, so it is opened as utf-8-sig.
2. Reads the boundary shapefile with latin-1 encoding for accented names.
3. Joins on CVEGEO, the concatenation of the state and municipality keys, with leading zeros stripped so the key survives the loader's id normalization.
4. Simplifies every polygon to a 200 m tolerance and takes a representative interior point as the centroid, which is safer than a centroid for the concave coastal municipios.
5. Outputs municipalities.csv (2,463 municipios, 125,822,502 inhabitants), regions.geojson and cities.csv.

The ITER municipal rows total 126,014,024 nationally, matching the published Censo 2020 figure. The 191,522 difference against the table written here is the population of municipios that have no polygon in the CONABIO boundary file.

Official Site Data: What Exists, and Why None of It Is Usable¶

Mexico is the only country in the set with no official per-site field, and it is not for lack of looking. Every avenue was checked:

• SNII, Sistema Nacional de Informacion de Infraestructura. Legally this is exactly the right registry: the Ley Federal de Telecomunicaciones requires it to hold georeferenced active and passive infrastructure for every operator. It was launched in June 2024 and initial data upload began in January 2025, but it is operated through IFT's Ventanilla Electronica for obligated parties, with no public bulk download.
• Registro Publico de Concesiones. Its infrastructure section publishes AM, FM and television station data only. Mobile base stations are absent, because Mexican mobile spectrum is concessioned by area with no site-level publication duty, unlike French declarations above 5 W or Canadian licence site uploads.
• Mapa Interactivo de Cobertura 4G. Built from crowdsourced measurements, so it is the same class of evidence as OpenCellID and would not serve as an independent check.
• Mapas de Cobertura Garantizada Movil. Operator-declared polygons, and the right kind of source, but served only through an interactive application that blocks automated access.

The Mexican scenario therefore runs on OpenCellID alone. Its records also rest on a median of 6 measurement samples, the thinnest evidence of any country here.

How Far Off Might It Be?¶

The two countries with both a crowdsourced and an official field give a correction factor. Population more than 10 km from any recorded site:

Both calibrations land near 3, so Mexico's real uncovered population is plausibly around 7 % rather than the 20 % its OpenCellID field implies. That is an inference, not a measurement, and the Mexican results should be read accordingly: trustworthy for the ratio between architectures, which is insensitive to field density, and weak for absolute handover rates.

Scenario Parameters¶

julia --project main.jl national mexico