# Methodology

## Dataset identification

**Dataset title:** Mobile Phone GPS Mobility Data for COVID-19 Research in Hermosillo, Sonora, Mexico, 2020  
**Version:** 1.0.0  
**Publisher:** Universidad de Sonora  
**Publishing organization:** Departamento de Matemáticas, Universidad de Sonora  
**Data provider:** LUMEX CONSULTORES, S.C.  
**Maintainer:** Jesús Francisco Espinoza Fierro, jesusfrancisco.espinoza@unison.mx  
**Dataset URL:** https://datos.fi-cen.unison.mx/dataset/covid19-mobility-hmo-2020

## Purpose and scope

This dataset publishes mobile phone GPS mobility records for Hermosillo, Sonora, Mexico, during the 2020 COVID-19 pandemic. It supports reproducible research in mathematical epidemiology, human mobility modeling, urban science, geospatial analysis, and data-driven methodological development.

The dataset preserves the original daily partitioning of the mobility records and also provides a consolidated long-format table. The distribution is designed to support both direct computational analysis and traceable reconstruction of the daily source structure.

## Provenance

The mobility records were made available to Universidad de Sonora through LUMEX CONSULTORES, S.C. under the service contract signed on September 14, 2020. LUMEX CONSULTORES, S.C. subsequently authorized the use of the mobility data for academic research articles and the publication of the data as Open Data in an open access repository.

The underlying mobility records are associated with GPS pings generated by mobile devices when users accessed specific mobile applications with location permissions enabled. The original source documentation describes the raw mobility table as containing ten substantive columns: `id_adv`, `timestamp`, `lat`, `lon`, `gender`, `country`, `age`, `income_level`, `tier1`, and `tier2`. The consolidated file also preserves additional technical fields found in the source exports and adds two curation fields for traceability.

## Source data organization

The source files are daily CSV files named with the pattern `YYYY-MM-DD.csv`. The observed daily files cover 2020-09-18 through 2020-12-13, inclusive, yielding 87 daily files. The consolidation process detected no missing daily files within this observed interval.

The source files were encoded as UTF-8 with a comma delimiter. One schema variant was detected across the 87 files. The consolidated dataset contains 80,582,452 records and 15 columns: two curation fields followed by 13 source/export fields.

## Consolidation procedure

The consolidation procedure performed the following operations:

1. Scanned the source directory for files matching the pattern `YYYY-MM-DD.csv`.
2. Parsed the date from each filename.
3. Detected encoding, delimiter, column names, file size, and SHA-256 checksum for each source file.
4. Verified schema consistency across daily files.
5. Read each source CSV in chunks to avoid excessive memory use.
6. Added `date`, derived from the source filename.
7. Added `source_file`, preserving the original daily filename for every row.
8. Concatenated the records into a single long-format CSV table.
9. Generated an equivalent Parquet file for efficient analytical use.
10. Created a ZIP archive preserving the original daily CSV files.
11. Generated file manifests and SHA-256 checksums for integrity verification.

The consolidation does not aggregate, round, filter, or otherwise modify the source mobility observations. The public analytical table preserves the original device-level mobility records as registered in the source files, with the exception of the two added traceability fields.

## Temporal reference

The `timestamp` field records ping date and time in UTC using the format `YYYY-MM-DD HH:MM:SS`. For local analyses in Hermosillo, Sonora, timestamps should be converted to UTC−07:00. Sonora did not observe daylight saving time during the study period, so this conversion is constant throughout the dataset.

Users should distinguish between:

- `date`: the date derived from the daily source filename, used for source-file traceability;
- `timestamp`: the actual ping timestamp in UTC.

In the related research article, the UTC timestamps span from 2020-09-18 00:00:00 to 2020-12-13 23:59:59. After conversion to Hermosillo local time, the corresponding local observation interval begins on 2020-09-17.

## Spatial reference

The `lat` and `lon` fields are GPS latitude and longitude in decimal degrees, using WGS84 geographic coordinates (EPSG:4326). The related research article reports latitude values approximately ranging from 28.9753 to 29.1960 and longitude values from −111.1002 to −110.8457, corresponding to the Hermosillo urban area considered in the study.

For distance calculations and geospatial modeling in prior analyses, coordinates were transformed to projected coordinate systems such as UTM. For publication, the dataset keeps the source latitude and longitude columns in decimal degrees using WGS84 / EPSG:4326.

## Interpretation of GPS pings

A ping should be interpreted as a GPS observation associated with a device at the time and location recorded by the mobile application ecosystem. It should not be treated as deterministic evidence that an identifiable person was located at an exact point at an exact instant.

Consecutive pings associated with the same device may show short-range spatial variability over seconds or minutes. This variability is consistent with GPS positioning uncertainty, application-level sampling behavior, signal conditions, and source-data processing effects. The related scientific analyses used these observations at aggregated spatial scales such as AGEBs, larger urban zones, or patch-based regions, where small-scale variability does not materially affect the conclusions.

## Variables

The consolidated dataset includes the following groups of fields:

- **Curation fields:** `date`, `source_file`.
- **Source mobility fields:** `id_adv`, `timestamp`, `lat`, `lon`, `gender`, `country`, `age`, `income_level`, `tier1`, `tier2`.
- **Technical/source-export fields:** `unnamed_column_1`, `file`, `IDL`.

Detailed variable-level descriptions, types, examples, missingness counts, allowed values where available, and responsible-use guidance are provided in `metadata/data_dictionary.csv`.

## Data quality checks

The quality-control process verified the following:

- 87 daily CSV files were detected.
- Observed daily coverage is 2020-09-18 through 2020-12-13.
- No missing file dates were detected within the observed interval.
- 80,582,452 records were consolidated.
- 13 original/export columns were detected.
- 15 total columns are present in the consolidated table after adding `date` and `source_file`.
- One schema variant was detected across all daily files.
- Parquet generation completed successfully.
- SHA-256 checksums were generated for integrity verification.

## Relationship to published studies

The 2025 Computational Urban Science article used mobile phone GPS pings for Hermosillo to estimate residence-occupation matrices and urban patch occupation times using Brownian bridge models. It reports the same total number of records and describes the device-level identifiers, timestamps, and coordinate fields used in the analysis.

The 2023 Entropy article used mobile phone sensing data and confirmed COVID-19 data to estimate mobility and residence parameters in a multi-patch epidemic model, including grouped urban zones of Hermosillo. The article describes the use of approximately 80 million GPS-position and timestamp records associated with nearly 300,000 devices, with alphanumeric IDs used to protect user privacy.

This dataset makes the underlying mobility records available for independent reuse, subject to the license, citation requirements, and responsible-use restrictions stated in the dataset documentation.

## Limitations

This dataset is not a probability sample of all inhabitants of Hermosillo. Pings are generated only when devices interact with specific mobile applications with location permissions enabled. Therefore, sampling intensity varies across devices and time. Some devices may have only a small number of pings, while others may have many records.

Demographic fields such as `gender`, `age`, `country`, and `income_level` are source-provided categorical attributes associated with the device or user profile. They should not be interpreted as independently verified demographic survey responses.

The `tier1` and `tier2` fields are provider-assigned categorical or audience/interest taxonomy fields. Their exact public taxonomy version is not distributed as part of this dataset. Users should treat them as categorical source codes unless they have independent documentation from the original provider.

The dataset alone should not be used to infer causal effects of mobility on epidemiological outcomes without additional modeling assumptions, epidemiological data, spatial aggregation, and uncertainty analysis.

## Responsible-use restrictions

Users must not attempt to identify, re-identify, contact, profile, locate, or infer sensitive personal routines of any individual, household, workplace, or device owner. Users should avoid publishing maps, tables, or examples that isolate very small numbers of devices or expose sensitive locations at fine spatial-temporal resolution.

Publications and derivative works should use spatial and temporal aggregation appropriate to the scientific question and privacy context. When reporting analyses, users should describe filtering criteria, aggregation units, coordinate transformations, and uncertainty assumptions.

## Reproducibility

To reproduce the dataset structure from the original daily CSV files, use `scripts/build_ckan_mobility_package.py`, the consolidation script maintained by the dataset creators. The script scans the daily source directory, adds traceability fields, builds the consolidated CSV and Parquet resources, creates the ZIP archive of original daily files, and writes data-resource integrity files. Final documentation and metadata files are maintained separately as part of the CKAN release package.

After downloading resources, users should verify file integrity with SHA-256 checksums before analysis.

## License and citation

This dataset is released under CC BY 4.0. Users must cite the dataset creators, identify Universidad de Sonora as the publisher, and acknowledge LUMEX CONSULTORES, S.C. as the provider of the original mobility data source.