We've added 63 new sheets to the catalog from the Service Géographique de l'Indochine — the French colonial mapping bureau that produced the most systematic topographic survey of northern Vietnam before WWII. The collection covers the Red River delta and surrounding provinces, with most sheets surveyed between 1903 and 1927 at roughly 1:100,000 scale.
What's in the batch
The sheets span the densely populated lowland north: Hà Nội itself (sheet 20, 1903), the port of Hải Phòng (31, 1904), and the major delta cities — Nam Định (55), Thái Bình (56), Ninh Bình (59), Phát Diệm (69), Thanh Hóa (75). Inland coverage extends to Sơn Tây (12), Bắc Ninh (10), Hưng Yên (42), and the Quảng Yên coastal zone (25, 32). A handful of named sheets sit outside the numbered grid: Hà châu, Nhà nam, Bảo Lộc, Cẩm Lý, and a cover plate that we've kept in the archive for completeness.
For anyone tracing genealogy or urban history in northern Vietnam, this is the layer beneath everything: the administrative boundaries, village names, and road networks that the post-1954 state inherited and reshaped. Place names are in pre-reform romanization (no diacritics), which is itself a useful diff against modern toponymy.
Pipeline notes
This is the first batch processed entirely through the consolidated bulk-upload pipeline. Each sheet now flows in one pass: tile_map.sh writes a DZI pyramid to R2 under iiif.maparchive.vn/iiif/<uuid>, the maps row is inserted with the sheet number preserved in extra_metadata.sheet_number, and the thumbnail URL is derived from the smallest pyramid level the worker advertises in info.json. No separate backfill step.
The previous version of the script split parsing across two passes — upload first, then run a clean-up to strip leading sheet numbers and fetch thumbnails. That worked but hid a latent bug: a set -e + pipefail shell trap silently aborted the cleanup on the first row whose name didn't have a leading number, which meant later uploads weren't being cleaned. Folding everything into the upload script eliminates the second pass and the failure mode at once.
What's next for these sheets
None of the new maps are georeferenced yet — they're in draft status until the corners are placed. The sheets are gridded enough that GCP propagation should work well: once we manually georeference one or two anchors per band, the rest can inherit corners arithmetically the same way the L7014 series did. After that, the OCR pipeline can scout the sheets for place names, and the toponym layer for northern Vietnam starts to materialize.
If you can read the older romanization confidently and want to help validate place-name extractions, this is exactly the kind of contributor work we're set up to support. Browse the catalog to see the new sheets.