The Daily Zurich

Zurich news, every day

News

Zurich Tackles the Ghost Image Problem: How the City Compares to Amsterdam, Vienna and Singapore

As urban planning databases swell with duplicate and outdated building images, Zurich's cadastral office is testing automated detection tools — and the results put it ahead of some peers but behind others.

By Zurich News Desk · Published 4 July 2026, 8:28 pm

4 min read

Zurich Tackles the Ghost Image Problem: How the City Compares to Amsterdam, Vienna and Singapore
Photo: Photo by Magda Ehlers on Pexels

Zurich's cantonal building registry contains records for roughly 92,000 structures across the city and its surrounding municipalities. A growing share of those records, according to the Federal Statistical Office's annual geodata quality report published in March 2026, carry duplicate photographic entries — the same facade shot filed twice, sometimes under different parcel numbers, after decades of analogue-to-digital migrations and successive GIS platform upgrades. City planners, insurers and mortgage underwriters all draw on that registry. When the images don't match the parcels, decisions get made on bad data.

The problem surfaced publicly in late 2025 when the Amt für Städtebau, Zurich's urban development office headquartered on Lindenhofstrasse, flagged inconsistencies during a routine audit tied to the ongoing Wohnungsnot housing crisis response. The audit was trying to map vacant lots and underused commercial buildings in districts like Altstetten and Schwamendingen — exactly the areas the city is targeting for densification under its 2040 spatial strategy. Duplicate images were distorting the automated vacancy scores the office uses to prioritise rezoning applications.

What Zurich Is Actually Doing

The Amt für Städtebau began piloting a perceptual hashing tool in January 2026 in partnership with ETH Zurich's Institute of Cartography and Geoinformation, which sits on the Hönggerberg campus. Perceptual hashing converts images into compact numeric fingerprints and flags pairs that score above a similarity threshold — the current threshold is set at 96 percent — for human review. Since the pilot launched, the team has processed approximately 34,000 image entries covering Districts 9 and 12 and flagged just over 1,800 suspected duplicates, of which reviewers have confirmed and removed 1,340 as of last week. That is a duplication rate of roughly five percent in those two districts, which urban data specialists consider high for a registry of Zurich's administrative maturity.

The tool itself is not new. The same hashing approach has been deployed by Amsterdam's Basisregistratie Adressen en Gebouwen — the Dutch national address and building registry — since 2022, and by Vienna's Stadtentwicklung Wien geodata team since mid-2023. Both cities reported initial duplicate rates between three and seven percent after their first full scans, figures that broadly align with what Zurich is finding. Singapore's Urban Redevelopment Authority completed a city-wide image deduplication sweep in 2024 using a proprietary convolutional neural network rather than perceptual hashing, and publicly reported a post-clean duplicate rate of under 0.8 percent. That benchmark is what ETH's cartography team is now using as a long-term target for the Zurich project.

The Gap With Leading Cities — and Why It Matters Here

The divergence between Zurich and Singapore is partly a legacy infrastructure problem. Singapore's land registry was consolidated into a single digital platform in 2011, meaning fifteen years of images were captured under consistent metadata standards. Zurich, like Vienna and Amsterdam, went through multiple platform migrations — most recently from a legacy MapInfo system to a cloud-hosted ESRI environment between 2018 and 2021 — and each migration created new duplication risk as records were batch-imported rather than individually verified.

The stakes are not trivial. The Swiss mortgage market relies on cantonal building data for property valuations. UBS, which absorbed Credit Suisse's Swiss retail book in 2023 and now holds a dominant position in domestic mortgage origination, uses automated valuation models that pull directly from cantonal geodata feeds. Duplicate or misattributed images can push valuations in either direction. The Swiss Financial Market Supervisory Authority, FINMA, has not issued formal guidance on registry image quality as a systemic risk factor, but the topic was raised in a working group session documented in FINMA's 2025 annual report.

The Amt für Städtebau plans to extend the ETH pilot to all twelve city districts by the end of the third quarter of 2026, with a full-registry sweep targeting completion before January 2027. Residents and property owners in affected districts do not need to take any action — the deduplication process works on the backend registry and does not alter public-facing records visible through the cantonal geoportal at maps.zh.ch. Anyone who suspects their building entry is incorrectly linked to a duplicate image can file a correction request through the standard parcel objection form, which carries a processing fee of CHF 90 under the current cantonal fee schedule.

Topic:#News

How does this story make you feel?

Spread the word

See something wrong? Suggest a correction.

Have your say

Loading comments…

Sources

About this article

Published by The Daily Zurich

This article was produced by the The Daily Zurich editorial desk and covers news in Zurich. See our editorial standards for how we use AI.

The Daily Zurich brief

The day's Zurich news in a 2-minute read, every weekday morning. Free.

By subscribing you agree to receive emails from The Daily Zurich and accept our Privacy Policy. Unsubscribe anytime.

Daily brief

Enjoyed this? Wake up to Zurich news every morning.

Free, in your inbox before 7am. Weekdays.

By subscribing you agree to receive emails from The Daily Zurich and accept our Privacy Policy. Unsubscribe anytime.

More from The Daily Zurich

More in News

Enjoyed this story? Get tomorrow's briefing free.