As OSM has become more consistent in terms of completeness, OSM data has become the foundation for many applications. Among these uses is the creation of base maps. In this article, by “base map” we mean both topographic maps (i.e., those where topographic features are the main focus of the map) and base maps that accompany thematic data, even though this distinction may sometimes seem artificial (are commercial points of interest on Google Maps part of the map’s topographic features, or are they thematic data surrounded by a base map that allows them to be located?). Indeed, we can cite several projects based on OSM data that involve the distribution of a pan-scalar base map through various geographic services (Figure 1):

• (a) Mapbox: a vector tiles base map with the ability to customize its style via the Mapbox Studio application.
• (b) CARTO: known for its Voyager, Positron, and Darkmatter base maps, often used to display thematic data due to their minimalist appearance.
• (c) Stamen: known for its artistic base maps such as Toner (black and white) and Watercolor.

Regardless of the quality of these products, they all share a minimal use of cartographic generalization techniques. Two main reasons can be cited to explain this assertion. First, with the development of multi-scale maps, it was believed that the ability to navigate between scales using the zoom and pan functions offered by these maps would eliminate the primary reason why generalization is necessary [1]. Indeed, when designing a paper map covering a vast area—and thus at a single scale—generalization becomes essential to reduce the graphical complexity of cartographic elements (roads, buildings, thematic data, etc.), thereby making the map readable. A zoomable map therefore allows the scale to be changed if the data representation is too complex. On the other hand, the use of generalization techniques still faces technical limitations. First, certain cartographic generalization techniques can be resource-intensive. For example, a complex displacement algorithm that manages multiple cartographic elements such as rivers, roads, and buildings to avoid collisions between them has a processing time that can be considerable, especially for datasets covering a vast area. Secondly, most current GIS and mapping software offer only a limited number of cartographic generalization algorithms, often restricted to the simplification of lines. Some projects were launched to address this issue, such as the Java CartAGen platform (designed as a plug-in for a GIS software developed and used by the French National Cartographic Agency), but these are no longer maintained and are not user-friendly. For these two reasons, cartographic generalization has been primarily limited to basic selection and simplification operators. Taking the example of road network representation, most of the base maps mentioned above select only major roads based on their attributes at small scales (level in the road network hierarchy), but without modifying the complex elements of their shape (winding roads in mountainous areas, complex structures at major road intersections, etc.).

Regarding the first reason cited to explain the low use of map generalization, the literature shows that generalizing data seems to remain useful in multi-scale interactive maps [1]. As for the second reason, the availability of the CartAGen tool could encourage the adoption of advanced cartographic generalization techniques by base map providers relying on OSM data. CartAGen is an open-source Python library containing over 80 functions or methods for using generalization algorithms [2]. These algorithms are drawn from the scientific literature in the field of cartographic generalization and sometimes involve porting algorithms from other languages to Python. The library was created and is currently maintained by a research team at the IGN (the french national mapping agency), with the aim of supporting research on interactive maps and building an archive of the literature on cartographic generalization algorithms. Beyond its research applications, CartAGen aims to be used by a wider range of cartographers. CartAGen’s innovative features lie in its interoperability and ease of use. The choice of Python, a language widely adopted by the GIS community, allows CartAGen to be used in various environments, such as Python scripts, Jupyter Notebook, and a QGIS plugin currently available. Integrations with other software could also be considered (ArcGIS plugin, R package, etc.). More specifically, CartAGen relies on several Python libraries already commonly used for managing geometry and spatial data formats (Geopandas and Shapely), which enhances its interoperability.

This talk at SOTM 2026 will explore several examples of generalizing OSM data using CartAGen, with the goal of designing a high quality pan-scalar map. We will focus particularly on cases where cartographic generalization reduces conflicts between map features, improves their visual hierarchy, limits clutter effect, and allows for the display of more information on the map. Indeed, these principles are likely to reduce the disorientation one may feel when navigating a pan-scalar map, thereby improving user satisfaction.

By enabling cartographers to produce better (base) maps, CartAGen could have a direct impact on how OSM data is utilized in mapping applications. Improvements in the quality of OSM data have paved the way for the emergence of mainstream mapping applications (COmaps, Organic Maps, Cartes, etc.). These open-source projects provide not only base maps but also services such as advanced route-planning features. Despite the widespread use of closed-source maps (notably the dominant position of Google Maps), CartAGen could contribute to this improvement in OSM data quality, thereby promoting the diversification of mapping projects.