The OpenStreetMap (OSM) database offers a wealth of collaborative geographic data, providing diverse thematic perspectives on a global scale. This study focuses specifically on global mining activity, a domain of critical economic, environmental, and societal importance. By applying an ad hoc GIS methodology, we generated an initial data layer featuring one point per potential mining site.

Our approach began with selecting specific OSM tags related to mining. In addition to community-recommended tags (via the OSM Wiki), we utilized the Taginfo API to identify and include additional relevant tags, particularly those identifying ores extracted through mining activities. Data extraction was performed using the Overpass Turbo API, yielding a total of 29,233 polygon entities, 13,630 lines, and 58,628 points, within 3 different layers. Many sites extracted using the rules described above include tags indicating their heritage status and/or abandonment. It is therefore necessary to exclude them from our database. We then applied a rigorous filtering process to exclude historical sites (e.g., destroyed, abandoned, or in ruins) and tourist attractions using a large SQL query with more than 200 arguments.

To further refine the dataset, we implemented a shape-based filter to detect "atypical" and complex geometries unrelated to mining activity. For example, in Florida, entities tagged as man_made=spoil_heap appear to be linked to canals and may correspond to embankments. We employed Miller’s Circularity Ratio [1] as a compactness index. Finally, the layers were simplified using spatial selections; centroids were calculated for polygons and lines, and the layers were then aggregated to produce a single layer where each potential site is represented by a unique point. Lastly, any duplicates were removed.

This process yielded a dataset of 39,291 points. For validation, we compared our results with two benchmark international studies: Maus et al. (2022) [2], which identified 44,929 sites, and Tang and Werner (2023) [3], with 74,548 sites. These studies rely on satellite image interpretation.

A visual analysis reveals that OSM’s spatial distribution differs significantly from these benchmarks – which themselves exhibit notable discrepancies. OSM data appears relatively consistent across the Americas but is overrepresented in Europe and largely underrepresented throughout the rest of the world. Furthermore, spatial correlation is low: only 7.2% of OSM sites fall within Tang and Werner’s (2023) polygons, and 8.9% within those of Maus et al. (2022). Notably, since the two benchmark studies only share a ~48% overlap, these results must be viewed within the context of general global mapping uncertainties.

While the quantitative results are below expectations, they highlight critical areas for discussion. Our extraction yielded fewer sites than the benchmarks but also identified numerous sites absent from them. We propose four hypotheses for these discrepancies:

• Insufficient completeness of OSM regarding mining themes.
• Inconsistent data entry, where contributors use incorrect geometries or incomplete tagging.
• Filtering inefficiencies inherent to the tags used in our methodology.
• Unique OSM entries representing valid sites missed by other global studies.

These are typical biases. OSM is a quintessential example of Volunteered Geographic Information (VGI) [4] a nature that inherently raises questions regarding the quality of the provided data [5]. Indeed, the specific characteristics of such data can sometimes be difficult to grasp for the “citizen-sensor” [4], who is often a non-specialist. Furthermore, one might surmise that the mining theme is particularly challenging to grasp in this context.

Several factors significantly degrade the output quality:

• The “Quarry” vs. “Mine” Ambiguity: The quarry tag is frequently misused for mining operations. For example, 81% of entities tagged with resource=gold are classified as quarries, despite gold extraction being a mining activity. Furthermore, 68% of quarries (156,014 entities) lack any resource tag, making it difficult to distinguish them from actual mines.
• The Historical Bias: In Europe, the overrepresentation of sites is largely due to abandoned or historical locations that lack the proper historical or disused status tags.
• Semantic Misuse: We identified significant interpretation and translation errors. Examples include: 1. A French site named “Ancienne mine de” (in English: “Former mine of...”) tagged as man_made=adit but lacking historical indicators; 2. A historic wine cellar (cultural heritage) tagged solely as man_made=adit; 3. Caves documented by speleologists using the man_made=adit tag.

Despite these challenges, OSM likely contains valid mining sites not captured by existing scientific inventories. Future improvements could involve using the name key to filter for adjectives indicating historical status (e.g., “former”, “abandoned”, “ancient”). However, this would require a complex multilingual approach to cover the 10 million+ OSM users.

This research underscores the need for better contributor practices and more comprehensive tagging. While these initial tests produced results below initial expectations, they established a functional, replicable GIS methodology. Automating this process will be the next step, allowing for iterative testing and refinement to improve the accuracy and efficiency of global mining mapping.