MDS in geo: An example with Mapillary

Two articles by Michaël Galien on #Geotribu showcase how to leverage a Modern Data Stack (#MDS) for processing geospatial data from #Mapillary. The workflow features #dbt, Apache #Airflow, #PostGIS, #SQL, and #Python, highlighting a robust open-source approach to geospatial data engineering.
Author
Published

June 9, 2025

Geotribu, a community dedicated to “free and open geomatics”, has an interesting two-part article series (in French) by Michaël Galien giving an example of using the MDS1 in the geospatial domain. The example centers around Mapillary2 data:

Mapillary features extracted and processed across 8 different tasks (source: CC-BY Michaël Galien)

The stack shown involves dbt, Airflow, PostgreSQL/PostGIS, SQL, and Python (but the articles also mention alternative technologies for building an MDS). And relevant data engineering tools and paradigms are touched upon, such as DAGs4, orchestration, lineage, data models, and model documentation.

Footnotes

  1. “MDS” stands for “Modern Data Stack”, a term used to describe a set of tools and technologies for data integration, processing, and analysis. It typically includes components like ELT processes (extract, load, transform), data warehouses, data lakes, and analytics platforms.↩︎

  2. Mapillary is a platform for street-level imagery and data collection. It was launched by Mapillary AB in Sweden in 2013. In 2020, it was acquired by Facebook, now Meta.↩︎

  3. The medallion architecture is a data processing framework that posits three layers for data: bronze (raw data), silver (cleaned and transformed data), and gold (aggregated and ready for analysis).↩︎

  4. Directed Acyclic Graphs↩︎