Before modern Earth observation technologies, historical maps were the primary record of how cities changed over time. They offer a lens on the evolving identity of urban areas, but tapping into that record consistently and at fine detail has been difficult.
The new paper, posted on arXiv, addresses this challenge directly. It describes a deep learning approach for aligning historical maps, which enables urban change profiling. The work is positioned as a way to unlock long-term urban transformation data that predates satellite imagery.
While the abstract does not detail the specific architecture or results, it signals a methodological step forward. The significance lies in making historical cartography usable for quantitative analysis of city evolution, rather than leaving it as a static archive.