Zurich's Digital Archives Are Full of Duplicate Images — And Officials Are Finally Talking About It

Zurich's public institutions are sitting on tens of thousands of duplicate digital images — redundant files clogging servers at everything from cantonal planning offices to the city's own Stadtarchiv on Neumarkt — and the people responsible for managing those systems say the problem has reached a point where it can no longer be quietly ignored.

The issue has moved from a technical nuisance to a budget and governance concern because several major digitisation pushes have converged at once. The aftermath of the UBS-Credit Suisse merger generated enormous volumes of duplicated compliance and property-valuation imagery that flowed into third-party data vendors. Separately, the city's ongoing Wohnungsnot documentation effort — tracking the chronic housing shortage across neighbourhoods like Aussersihl and Schwamendingen — has produced overlapping cadastral and street-level photo sets that municipal planners acknowledge are difficult to deduplicate without a shared standard.

What the Experts Are Saying

Researchers at ETH Zurich's Computer Vision Lab have been examining automated deduplication pipelines for the better part of two years. Their work, presented at an internal symposium in March 2026, found that naive hash-based deduplication catches only about 60 percent of visually identical images in large heterogeneous datasets — leaving a substantial tail of near-duplicates that require perceptual hashing or machine-learning classifiers to identify. The lab has not yet published a peer-reviewed figure for Zurich-specific datasets, but the methodological gap they describe maps directly onto what city archivists are encountering.

At the Stadtarchiv itself, staff have flagged that the 2024 expansion of the GeoDaten Zürich open-data portal introduced hundreds of georeferenced image layers with no cross-checking mechanism against existing holdings. Officials there have described the situation in internal working-group minutes — documents reviewed by The Daily Zurich — as requiring a city-wide image-metadata standard before any deduplication tool can be deployed effectively. The working group has met four times since January 2026 and has yet to produce a binding recommendation.

Cantonal-level voices are also weighing in. The Baudirektion Kanton Zürich, which oversees land-use planning data, has circulated a discussion paper arguing that the canton and the city should adopt a shared persistent identifier scheme for spatial imagery by the end of 2026. The paper sets a soft deadline of October for pilot institutions to sign up. Whether that timeline holds is an open question, given that budget negotiations for the 2027 cantonal cycle begin in September.

What Happens Next

The practical stakes are not trivial. Storage costs for unmanaged image repositories at Swiss public bodies have been rising alongside cloud-pricing shifts; industry benchmarks suggest that eliminating duplicate files in a mid-size municipal archive can cut active storage expenditure by between 20 and 35 percent, though those figures vary widely by institution and dataset type. For a city the size of Zurich, with its extensive urban-planning photography and building-permit imagery going back decades, even the lower end of that range would represent a meaningful annual saving.

Three concrete steps are now on the table. First, the Stadtarchiv working group is expected to circulate a draft metadata standard for public comment before the end of summer — a process that will likely involve the Zentralbibliothek Zürich on Zähringerplatz, which holds significant photographic collections of its own. Second, ETH Zurich's Computer Vision Lab is in preliminary talks with the city's GIS-Zentrum about a pilot deduplication run on a subset of the GeoDaten portal's imagery, potentially starting in autumn 2026. Third, the Swiss federal standards body, the Schweizerische Normen-Vereinigung, is reviewing whether existing ISO metadata standards adequately cover geospatial image provenance — a review that could have knock-on effects well beyond Zurich.

For now, the consensus among those working closest to the problem is that the technology is not the hard part. Coordinating a dozen institutions, each with its own procurement rules, data-protection obligations under the revised Datenschutzgesetz, and IT roadmaps, is where previous efforts have stalled. The October pilot deadline from the Baudirektion gives reformers something concrete to point to — but Zurich's direct-democracy culture means any binding framework will ultimately need a clear political sponsor willing to take it through the formal approval process.