Urban digital twins compared: Singapore, Tokyo, Seoul and Rotterdam

A city announces a digital twin. The press image is always magnificent: every tower rendered, traffic flowing, weather moving across the screen. Then the practical questions begin. Does the model update? Can anybody outside city hall use it? Has a planner changed a real decision because of it? And what, exactly, is being twinned: buildings, pipes, people, or merely a very expensive map?

Singapore, Tokyo, Seoul and Rotterdam answer those questions differently. They are not four deployments of one product. They are four operating models hiding under one fashionable term: a national simulation environment, an open-data platform, a municipal virtual laboratory, and a planning model that takes the underground city as seriously as the skyline. Comparing them is more useful than ranking them, because the differences reveal what a city must decide before buying any software.

First, the definition problem

Our digital-twin glossary uses a demanding definition: a maintained virtual representation with a meaningful connection to the changing physical system. That does not require every lamp post to stream data every second. It does require more than geometry. A model can be visually exact and operationally dead; a simple asset twin can be visually plain and operationally vital.

This distinction matters because city programmes evolve in layers. Surveying teams first create terrain, buildings and roads. Agencies attach registers, planning rules and infrastructure records. Sensors or periodic surveys refresh selected layers. Simulation tools then ask “what if?” questions. Public viewers may expose only a fraction of the working environment. Calling the entire stack a twin is defensible, but it makes comparisons slippery.

The four cities below all clear a reasonable threshold. Each built more than a promotional rendering, each names practical uses, and each has an accountable public institution behind it — though one of the four, the most famous, has since quietly concluded. None publishes the kind of independent before-and-after evidence common in transport or energy projects. The evidence tells us what the platforms contain and what officials used them for. It rarely tells us how much faster, cheaper or safer the city became.

Singapore: the national simulation room

Virtual Singapore is the field's headline case — and, on closer inspection, a completed programme rather than a running one. Its boundary was unusually clean: one city-state, modelled as one national environment. The National Research Foundation, Singapore Land Authority and GovTech launched the programme in 2014 on Dassault Systèmes' 3DEXPERIENCity platform. The reported budget was about S$73 million, and the national 3D mapping beneath it was completed in 2022. There the story turns: the promised public phase never launched, and the programme's name has since disappeared from government pages.

Its strongest idea was not photorealism but shared context. Building geometry, terrain and semantic information could support solar-potential studies, flood modelling, accessibility analysis and crowd-flow scenarios without every agency rebuilding the island from scratch. Government documentation presented it as a collaborative platform for public agencies, researchers and businesses, although the environment itself ran on government machines behind Singapore's "internet separation", and access to the richest layers was always controlled.

That combination gives Singapore impressive breadth and an awkward evidence problem. The official use-case list is long, but a list of simulations is not a list of measured outcomes. We found no independent public evaluation that isolates how much Virtual Singapore shortened a planning process, reduced a project cost or prevented damage. The sharpest evidence is what Singapore built next: when the state needed twins that run something day to day, it commissioned new platforms — the sensor-fed Open Digital Platform in Punggol Digital District and the port's Maritime Digital Twin — while the world's most cited city model stayed a sealed simulation environment.

What another city can copy: establish one authoritative spatial base, assign durable institutional owners, and fund the unglamorous work of keeping data compatible — Singapore's mapping layer outlived the programme and is refreshed on a fixed cycle. What it should not copy blindly: national-scale scope, or the assumption that fame equals operation. A twin can be the most cited model in the world and still never acquire a documented day-to-day user.

Tokyo: make the model reusable

Tokyo's Digital Twin, begun in 2021, makes a different strategic bet. The Tokyo Metropolitan Government operates a free browser-based 3D Viewer and publishes data that others can reuse. In December 2024 it released high-precision aerial point clouds for the capital's 23 special wards, adding unusually rich raw material for disaster planning, urban development and research. The programme has kept shipping since: by mid-2026 the viewer layers more than 1,100 types of geospatial data over the base model, ward-level model updates were still arriving in 2026, and the strategy's stated goal remains a complete digital twin by 2030.

Tokyo therefore treats openness as part of the infrastructure, not an afterthought. The public viewer matters, but the reusable point cloud matters more: universities, companies and civic developers can build analyses without scraping a city-owned interface. This is a less cinematic story than a control room and potentially a more productive one.

The limitation is equally important. Openness does not automatically create operational synchronisation. Some layers are periodic surveys, some are administrative datasets, and some are hazard or planning overlays. Tokyo's own roadmap describes social implementation as a staged programme, not a finished omniscient replica. That honesty is healthy. A city-scale twin is an institution that keeps publishing and governing data, not a launch event.

What another city can copy: procure open formats and public access from the beginning. A twin that only one vendor can operate is a dependency disguised as an asset. The unresolved question: whether public reuse produces measurable planning improvements, rather than an impressive catalogue of possible applications.

Seoul: turn the city into a laboratory

Seoul belongs in this comparison on substance, not symmetry. Seoul's S-Map is one of the most concrete municipal implementations in the atlas. Drone-supported spatial collection began in 2018, and the full platform launched in April 2021. Official sources describe a model covering 605.23 km², roughly 600,000 structures and six underground-utility layers — figures from the 2021 launch that Seoul's official pages still carry today, a floor rather than a current census for a model refreshed since.

The distinctive feature is the Open Lab. Agencies and companies can use the model to test sunlight and wind, autonomous-vehicle control and delivery-robot routes. That moves S-Map from “view the city” toward “rehearse an intervention”. Seoul also has a plausible adjacent sensing layer: the city's S-DoT network measures environmental conditions across streets, mountains and riverbanks. The two are linked editorially in the atlas because together they show the geometry-and-observation stack a working twin needs, although official documentation does not claim that every S-DoT reading streams directly into every S-Map layer.

Seoul's published numbers are strongest on coverage, not impact. We know how much city is modelled and which simulations are offered. We do not have an independent result such as a percentage reduction in design time or avoided construction clashes. The platform is a convincing laboratory; the scientific paper on whether that laboratory changes the city is still missing.

What another city can copy: define a service around the model. “Open Lab” is clearer than “innovation ecosystem” because it identifies users and actions. What to resist: mistaking the number of buildings for the quality of decisions.

Rotterdam: the underground city counts too

Rotterdam's municipal twin begins with the condition that makes the city distinctive: water, soft ground and a crowded subsurface. The model is designed to bring above-ground geometry together with cables, pipes and foundations, supporting flood scenarios, energy-transition planning and emergency response, and the Open Urban Platform beneath it has been operational since the beginning of 2025. A new tower is only half a planning problem if nobody can see what runs below its footprint.

Rotterdam also clarifies why the phrase “the city's digital twin” can mislead. The municipality's model is separate from the Port Authority's IoT and hydro/meteo platform. The port system went live in 2019 with 44 sensors and about 1.2 million daily data points, combining tides, currents, salinity, wind and visibility for vessel planning. The authority describes this as a foundation for a future 4D port twin and autonomous navigation. Two twins can occupy one map while having different owners, update cycles and users.

The city's Reyeroord Living Lab Sensible Sensor adds a useful counterpoint. Instead of declaring the whole city intelligent, Rotterdam tests waste, groundwater, lighting and air-quality sensors with residents in one neighbourhood. It is not itself a city twin. It demonstrates the governance work that twin diagrams tend to omit: explaining what devices do, learning locally and deciding which data streams deserve to scale.

What another city can copy: start from a planning constraint that existing tools handle badly. In Rotterdam that is the relationship between surface, subsurface and water. What it should document better: actual decisions changed by the model and the cost of maintaining parallel municipal and port environments.

Four twins, four operating models

The comparison suggests three axes more useful than “maturity level”. First is purpose: shared national planning, public reuse, controlled simulation or infrastructure coordination. Second is openness: who can inspect data, download it and run an experiment? Third is coupling: how frequently and reliably does the virtual system follow the physical one?

A city can be strong on one axis and weak on another. Tokyo may expose more reusable data than a more sensor-rich closed system. Rotterdam's port can have tighter live coupling than its municipal planning twin while serving a much narrower geography. Singapore can coordinate a national base model without making every sensitive layer public. Those are governance choices, not software defects.

The outcome chart that does not exist

A rigorous comparison should show bars for planning time saved, construction clashes avoided, energy reduced or flood damage prevented. We cannot draw that chart. None of the four programmes publishes a comparable independent outcome series attributable to the twin itself. That absence is the most important finding in this article.

The field measures what is inside the model far better than what changed outside it.

This is partly a measurement problem. A twin informs decisions alongside engineers, regulations, consultations and other models. Attribution is hard. But difficulty is not an exemption. Cities can still record how often the platform was used, whether it changed a design, how long an analysis took before and after, which datasets went stale, and what maintenance cost each year. Without such operational evidence, vendors win the argument with screenshots.

What a city should ask before procurement

1. Which decision will change? “Improve planning” is not a use case. Testing flood routes for a named district is.
2. What is the update contract? List each layer, its owner, refresh interval and acceptable age.
3. Who can reuse the data? Public viewer, downloadable open data and internal operational access are different products.
4. Can the city switch vendors? The city must be able to export its geometry, metadata, semantic models and scenario logic in documented, reusable formats instead of leaving them trapped in a vendor-hosted platform.
5. What will be measured? Publish use, changed decisions, time saved, recurring cost and failed simulations, not only model coverage.
6. What must remain closed? Detailed utilities, security-sensitive assets and movement data need explicit access rules.

Verdict

There is no “best” urban digital twin in this group. Singapore offers the clearest lesson in institutional coordination — and, now that its programme has quietly ended, the clearest warning that fame is not operation. Tokyo offers the strongest open-data proposition. Seoul turns a city model into a named simulation service. Rotterdam shows why a twin should begin with an awkward local problem, and why one city may need separate twins rather than one universal control room.

The transferable product is not the 3D model. It is the operating agreement around it: purpose, ownership, update cycles, access and measurement. A city that cannot write those five things down does not yet need a twin. It needs a data inventory.