By Chloë Farr and Emilia Hurd
UVic Libraries holds 68,000 loose maps, as well as multi-sheet atlases and fire insurance plan books. Historic maps, visually and informationally complex, are time consuming to document so that they can be easily found. A year ago, Chloë Farr started to investigate the use of vision-language models for creating metadata for maps as part of the Peter and Ana Lowens Fellowship program.
Chloë was introduced to Emilia in the Eco-Visualization Studio, as she is using historical fire insurance plans to investigate the ways industrial development and urbanization have impacted shoreline environments and relationships. The volume of maps alongside their cartographic detail typically require manual georeferencing, vector feature creation, and digitization of attributes such as construction materials, infrastructure and industrial land uses.
Georeferencing of the page alone may take up to 30 minutes if done with precision, before any meaningful data are pulled out. A (largely) automated method could make it possible to georeference thousands of maps, unlocking time-based analyses at urban scales previously requiring significant manual processing.
Fig. 1. Goad Fire insurance plan map index from the City of Victoria, 1891. Goad, C. E. (1891). Victoria, British Columbia [Fire Insurance Plans] [Map]. University of Victoria Libraries. https://doi.org/10.58066/12DC-5G44
Fig. 2. Example of resultant tiled map from manual georeferencing. City of Victoria, 1891. https://doi.org/10.58066/12DC-5G44
Background of Material
Cities of the 19th and 20th centuries experienced dramatic change as growth of manufacturing and industrial sectors brought with it urban expansion, population growth and densification. Wood construction, increasing density, and widespread burning of oil for heating and industrial processing were among the most significant contributors to growing fire risk, which in many cases, led to catastrophic fires, further accelerating cycles of urban transformation. These conditions contributed to a boom in fire insurance companies. Fire insurance plans – multi-sheet atlases, reissued every few years – were drawn to track and assess risk and losses at the building, block, and urban scales.
Fig. 3. Fire insurance map index from the City of Victoria, 1891. https://doi.org/10.58066/12DC-5G44
Cartographic detail, ubiquity and frequent updates mean that archival fire insurance plans can act as evidence of urban dynamics such as expansion, changing land uses, and architectural techniques and materials, to forms of settler colonialism that include land alienation, ecological damage, and industrial legacies such as contamination. To date, their use has often been limited to more qualitative analyses at the site or block scale.
Fig. 4. Sample street index and map sheets from the City of Victoria, 1891. https://doi.org/10.58066/12DC-5G44
Fig. 5. Zoom in map of Victoria Planing Mill, Gas Works and Albion Iron Works. Victoria Harbour, City of Victoria, 1891. https://doi.org/10.58066/12DC-5G44
Today, these maps offer a distinctive and previously unavailable high resolution way to understand how cities shifted and changed. But digital scans are raster imagery, not vector files, so getting meaningful data from them requires georectifying each image and then manually digitizing it into point, line, and polygon features with descriptive attributes.
Fig. 6. Block and building level details, both written and graphical from City of Victoria, 1891. https://doi.org/10.58066/12DC-5G44
Hypothesis and Outcome
We hypothesized that fire insurance plans could be georeferenced solely by running the maps through a vision-language model (VLM) tuned for optical character recognition (OCR) and cross-referencing the resulting text transcriptions with modern street data. This flips the usual order of things: instead of georectifying a map before its features can be digitized, the digitized text and shapes are what provide the necessary reference points to georectify. This turned out to be correct in the case of fire insurance plans (see Fig. 7), resulting in a geotagging tool, which we presented at iPres 2026.
Fig. 7. 1912 Vancouver Goad Fire insurance map, unaltered by our pipeline.
With text on the maps transcribed with optical character recognition (OCR), historical street names are compared to modern street names. With just a few matches, we can fit the modern coordinates to the historical maps (see Fig. 8), and georectify the features on the map (see Fig. 9), as well as geotag the map itself (see Fig. 10). So far, this is successful for 18th-19th century fire insurance plans across North America, averaging 5 minutes per page with human assistance.
Fig. 8. 1912 Vancouver Goad Fire insurance map, with modern streets overlay. Blue streets indicate where historical and modern street names matched, red are streets that are inferred.
Fig. 9. Buildings from 1912 Vancouver Goad Fire insurance map, visualized in arcGIS with building material features identified by our pipeline.
Fig. 10. Full map page oriented in geojson.io, open source GIS platform.
We are continuing to develop this work and will write soon with more updates! If you have any questions on how to use this, or about the project in general, we welcome you to get in touch with us.
Thank you to the Kula Academy and TIB Open Science Lab for supporting the presentation of this work at iPres 2026. This work is supported by a SSHRC Doctoral Fellowship and Mitacs Accelerate Grant.
Figures 1-6.
Goad, C. E. (1891). Victoria, British Columbia [Fire Insurance Plans] [Map]. University of Victoria Libraries. https://doi.org/10.58066/12DC-5G44
Figures 7-9.
City of Vancouver. (n.d.). Goad’s fire insurance map 1912 [Data set]. Vancouver Open Data Portal. Retrieved September 10, 2026. https://opendata.vancouver.ca/explore/dataset/goads-fire-insurance-map-1912-georectified/table/