The Digital Twin of the City of Zurich for Urban Planning

Gerhard Schrotter, Christian Hürzeler•View original
ComprehensiveExpertharry voice
A city planner in Zurich pulls up a three-dimensional model of a neighborhood that doesn't exist yet. She rotates it, runs a summer heat simulation, and checks the shadow cast on a school courtyard at two in the afternoon. What she's looking at took years to build — fifty thousand buildings, centimeter-accurate terrain, and a governance framework that keeps all of it from going stale. This is what Gerhard Schrotter and Christian Hürzeler describe in their account of Zurich's digital twin. The details of how they built it, opened it, and what broke along the way are worth understanding carefully. Start with the pressure that drove the project. The Canton of Zurich projects an increase of two hundred eighty thousand people by 2040, and cantonal guidelines direct that eighty percent of that growth should happen in urban regions. Under those scenarios, Schrotter reports, around five hundred twenty thousand people could live within Zurich's city limits by 2040. The administration frames this as a dual imperative: capture the benefits of growth — economic dynamism and cultural density — while managing intensified competition for land and the need to densify existing residential areas without eroding what makes those quarters livable. These objectives are encoded in the city's planning document Strategies Zurich 2035, which identifies high-quality densification, noise reduction, and land-use conflict management as principal tasks. The digital twin sits inside the city council's "Digital City" initiative, one of six subprograms in that framework, linked explicitly to Smart City Zurich. The framing is institutional from the beginning: the twin is a spatial image intended to bridge municipal planning, citizen participation, and technical domains. It connects to Building Information Modelling and the construction sector. Components get updated at different intervals and, where the situation demands it, enriched with real-time sensor data. This is not a visualization project dressed up as infrastructure. It is designed as a decision-support system, and that design choice determines everything downstream. The technical foundation is organized around levels of detail, or LoD for short. LoD zero is the digital terrain model, LoD one is a block model of building volumes, and LoD two is a roof model with walls and detailed roof structures. The terrain base was derived from LiDAR imagery acquired in 2014 at a ground sampling resolution of fifty centimeters. Bridge areas received additional break lines; adjustments were made manually at open-water edges along Lake Zurich and the Limmat. Road geometries from the official cadastral survey were introduced as break lines, and an irregular triangular mesh was calculated with tolerances of one meter and ten centimeters. The mean height error for that terrain layer is thirty to forty centimeters. The block model, LoD one, is built from official cadastral floor plans with eave and gable heights determined by photogrammetry. Its mean planimetric position error falls within the official cadastral limits of ten to fifteen centimeters, and its mean height error is fifty centimeters, and it is continuously updated. The roof model, LoD two, was produced by semi-automatic stereo aerial photogrammetric evaluation. Together the system covers more than fifty thousand buildings at various levels of detail, in the Swiss reference system CH1903 plus. The standard underneath all of this is CityGML, which Schrotter uses as the conceptual data model, enabling semantic modeling of urban objects so that spatial entities carry not just geometry but meaning: what a building is, when it was recorded, how accurate the measurement was, and who is responsible for updating it. Metadata are captured in the Swiss catalogue Geocat and governed under the Federal Act on Geoinformation from 2007, which defines spatial data models and spatial metadata as legal constructs. That legal grounding matters because it assigns real obligations. The metadata is not optional documentation but a governed asset. Lifecycle management is where the methodology gets operationally serious. Drawing on Grieves's 2014 product lifecycle concepts applied to spatial data, the Zurich system assigns every modeled component — a building, a utility segment, a tree layer — lifecycle metadata: creation date, acquisition method, planned update interval, steward, and quality metrics. For utility pipelines and cables, owners have submitted two-dimensional cadastral data at least once a week since July 2013. The register operator's role is to verify completeness and timeliness; content responsibility stays with the owner. Utility data positional accuracy runs to plus or minus twenty centimeters. The terrain base, by contrast, dates to 2014 — a reminder that not everything updates on the same cycle, and tracking those differences explicitly is what makes the system trustworthy rather than misleading. Schrotter distinguishes two modeling modes. Predictive modeling builds digital prototypes before reality exists — using the twin as a sandbox to test scenarios before any concrete is poured. Reactive modeling links the digital object to live sensors so that real-time changes in the physical world propagate immediately into the digital one. Together these two modes cover the full range from long-term planning simulation to operational monitoring. Now, that data is useful internally. But in late 2018, Zurich made a choice that changed the scale of what the twin could do: it published the three-dimensional spatial datasets as Open Government Data. The terrain model, block model, and roof model went onto the Open Government Data platform, accompanied by a media conference on the twelfth of November 2018. Between that date and the twentieth of September 2019, the portal saw eleven thousand seven hundred twenty-two dataset downloads. The five three-dimensional spatial datasets alone accounted for roughly one-third of all downloads in that period. More broadly, Open Government Data across the city's portal was downloaded seven thousand four hundred twenty-two times, corresponding to about sixty-five percent of total downloads. That is not a symbolic number. It's evidence of external demand. The city has operated an "Open by Default" principle since 2012 — administrative data provided free of charge, machine-readable, under free license. Applying that principle to the three-dimensional models transformed them from internal assets into public infrastructure. Schrotter lists the application categories this enabled: start-ups around the Internet of Things, Smart City solutions, location-based gaming, and collaborative platforms. The city itself uses subsets for solar potential calculations, shadow and noise analysis, flood simulation, and submission to third parties for visualizing planned construction. Free access, the paper states, "simplifies the use enormously" and opens value-creation opportunities for private individuals and project partners. That framing ties the open-data decision back to the city's governance goals: transparency, participation, and economic value creation. It was an infrastructural and strategic move, not a transparency gesture. The applications built on that infrastructure are where the abstract architecture becomes concrete planning work. Start with urban climate. Zurich is currently averaging six heat days per year. Climate projections say that number will reach twenty-one heat days by 2060. The city used the digital twin — combining terrain, existing buildings, planned buildings, and tree locations — as input to meso- and micro-scale climate models. These simulations quantify the obstacle effect of planned buildings and compare it against current development, reporting impacts on temperature, wind, and cold-air volume flow as functions of building length, width, height, and position. The results feed directly into the city's Sectoral Planning on Heat Reduction, expected to be published in 2020, and are used as an additional basis for variant decisions on public-sector construction projects. Heat, shadow, noise, traffic — these can now be layered and weighed in the same decision process because the geometry is consistent across all of them. Solar potential and shadow analyses have also become routine planning tools. For the municipal development plan, planners combined three three-dimensional data levels in a single interactive web application: current building stock from the digital twin, maximum building capacity under current zoning from a parametric Esri CityEngine model, and target densification scenarios from urban design drafts. Visual comparison of those three levels in the browser allowed planners to derive quantitative statements — building utilization, floor space, numbers of inhabitants, and jobs — because the geometry was linked directly to planning parameters. That link between geometry and attribute data is what makes these visualizations analytically useful rather than just illustrative. For architectural competitions, two tools were tested that leverage the twin for jury evaluation. A web-based three-dimensional map hosts submitted competition models, allowing juries to view designs jointly, change perspectives and scales, compare entries synchronously, and toggle spatial data layers. The second tool uses Microsoft HoloLens Augmented Reality glasses to project competition designs and their surrounding built context as an interactive hologram in the judging room. Juries can simulate shadows, switch spatial data on and off, change scales, generate cross-sections, and at the extreme, visualize designs on site at a scale of one to one. The city is actively testing whether the Augmented Reality tool can replace physical scale models in the long term. For public participation, three approaches were tried. The city transformed parts of Zurich into a Minecraft world based on the digital twin to engage younger residents. A browser-based interactive participation tool built on Esri's JavaScript Application Programming Interface lets users draw buildings, green spaces, or trees within a planning area with a few mouse clicks and submit the result to city administration. Tests with users who had no planning background found the tool easy and intuitive despite its three-dimensional interface. The third approach was commercial and remarkable in scale: Blindflug Studios used the city's building data to create a game called (re)format Z:, a dystopian cyberpunk reimagining of Zurich's Reformation history. Development took about seven months. The game launched in November 2017 and hit twenty-five thousand downloads within three weeks, reaching first place on the App Store's Top Free Puzzle Games list. By early 2019, it had reached one hundred thousand downloads and twenty-four thousand five hundred hours of total playing time worldwide. By the end of 2019, it was still generating about eighty thousand impressions per week in the App Store. That is reach no public engagement event achieves. Across all of these applications, the pattern is the same: the digital twin's payoff is visual, data-linked prototypes that enable scenario testing in actual decision processes. Geometry linked to planning parameters. Simulations linked to governance workflows. Open data linked to external application development. The infrastructure works as a platform precisely because the lifecycle management and metadata governance described earlier make the underlying data trustworthy enough to build on. But Schrotter is candid about where the platform is not yet complete. The central unsolved problem is subsurface modeling. Utility owners have been submitting two-dimensional pipeline and cable data weekly since 2013, and they already record three-dimensional coordinates internally. The problem is that "this information is lost at the interface to the utility cadaster." The three-dimensional coordinate data never makes it through. Schrotter says this gap will be closed to enable Building Information Modelling-capable utility cadastral datasets, but it has not been closed yet. Until it is, the digital twin has no reliable integrated underground layer. For a city managing aging utility infrastructure at the same time it is densifying above ground, that is a significant gap. Standards problems compound it. There are no jointly developed guidelines at cantonal and federal levels for modeling, recording, and updating three-dimensional geodata. CityGML offers extensive modeling capabilities, but Zurich does not use all of them. The practical question of how to use the standard sustainably and efficiently remains open. Building Information Modelling and Geographic Information Systems still operate in separate worlds: different levels of detail, different update cadences, no practical concept for where they converge, where they stay separate, and where they need to exchange data automatically. Schrotter puts the question bluntly — there are no practical concepts for how these two worlds meet. Data volume is a growing pressure. Mobile mapping is producing increasing volumes of point clouds. How to refine façades from those clouds, how to integrate point cloud vegetation models, and how to update and maintain the twin efficiently as data volumes grow are all open questions without settled answers. The paper cites the Digital Underground project and related modeling work by van Son and colleagues from 2018, and by Yan and colleagues from 2019, as promising directions — but describes them as evaluations of measurement methods, not operational solutions. What the Zurich implementation demonstrates, taken together, is a set of concrete prerequisites for this kind of system to work. Metadata and lifecycle governance are not optional: they are what makes simulation outputs trustworthy enough to put in front of a planning committee. The terrain model is from 2014. The block model is continuously updated. Those are different things, and knowing which is which — and encoding that difference in governed metadata — is what allows anyone using the system to know what they are actually looking at. Open data release functions as a force multiplier: eleven thousand seven hundred twenty-two downloads in less than a year, one-third of them from five three-dimensional datasets, is evidence that documented, accurate, open spatial data finds its users. Visualization has to be audience-specific — Augmented Reality for juries, browser tools for citizens, Minecraft for younger residents, and commercial games for everyone else. Institutional coordination has to span Geographic Information Systems, Building Information Modelling, utilities, and planning departments because no single department owns all the data the twin needs to be useful. The gaps Schrotter identifies — subsurface, Building Information Modelling and Geographic Information Systems convergence, shared standards — suggest that any city attempting to replicate Zurich's approach should treat these not as problems to solve after the system is built, but as design constraints to address from the beginning. The underground infrastructure is not a later add-on. The Building Information Modelling interface is not a future compatibility layer. These are architectural decisions, and deferring them creates exactly the data-flow gaps that Zurich is now working to close. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

A city planner in Zurich pulls up a three-dimensional model of a neighborhood that doesn't exist yet. She rotates it, runs a summer heat simulation, and checks the shadow cast on a school courtyard at two in the afternoon. What she's looking at took years to build — fifty thousand buildings, centimeter-accurate terrain, and a governance framework that keeps all of it from going stale. This is what Gerhard Schrotter and Christian Hürzeler describe in their account of Zurich's digital twin. The details of how they built it, opened it, and what broke along the way are worth understanding carefully. Start with the pressure that drove the project. The Canton of Zurich projects an increase of two hundred eighty thousand people by 2040, and cantonal guidelines direct that eighty percent of that growth should happen in urban regions. Under those scenarios, Schrotter reports, around five hundred twenty thousand people could live within Zurich's city limits by 2040. The administration frames this as a dual imperative: capture the benefits of growth — economic dynamism and cultural density — while managing intensified competition for land and the need to densify existing residential areas without eroding what makes those quarters livable. These objectives are encoded in the city's planning document Strategies Zurich 2035, which identifies high-quality densification, noise reduction, and land-use conflict management as principal tasks.

The digital twin sits inside the city council's "Digital City" initiative, one of six subprograms in that framework, linked explicitly to Smart City Zurich. The framing is institutional from the beginning: the twin is a spatial image intended to bridge municipal planning, citizen participation, and technical domains. It connects to Building Information Modelling and the construction sector. Components get updated at different intervals and, where the situation demands it, enriched with real-time sensor data. This is not a visualization project dressed up as infrastructure. It is designed as a decision-support system, and that design choice determines everything downstream. The technical foundation is organized around levels of detail, or LoD for short. LoD zero is the digital terrain model, LoD one is a block model of building volumes, and LoD two is a roof model with walls and detailed roof structures. The terrain base was derived from LiDAR imagery acquired in 2014 at a ground sampling resolution of fifty centimeters. Bridge areas received additional break lines; adjustments were made manually at open-water edges along Lake Zurich and the Limmat. Road geometries from the official cadastral survey were introduced as break lines, and an irregular triangular mesh was calculated with tolerances of one meter and ten centimeters. The mean height error for that terrain layer is thirty to forty centimeters.

The block model, LoD one, is built from official cadastral floor plans with eave and gable heights determined by photogrammetry. Its mean planimetric position error falls within the official cadastral limits of ten to fifteen centimeters, and its mean height error is fifty centimeters, and it is continuously updated. The roof model, LoD two, was produced by semi-automatic stereo aerial photogrammetric evaluation. Together the system covers more than fifty thousand buildings at various levels of detail, in the Swiss reference system CH1903 plus. The standard underneath all of this is CityGML, which Schrotter uses as the conceptual data model, enabling semantic modeling of urban objects so that spatial entities carry not just geometry but meaning: what a building is, when it was recorded, how accurate the measurement was, and who is responsible for updating it. Metadata are captured in the Swiss catalogue Geocat and governed under the Federal Act on Geoinformation from 2007, which defines spatial data models and spatial metadata as legal constructs. That legal grounding matters because it assigns real obligations. The metadata is not optional documentation but a governed asset.

Lifecycle management is where the methodology gets operationally serious. Drawing on Grieves's 2014 product lifecycle concepts applied to spatial data, the Zurich system assigns every modeled component — a building, a utility segment, a tree layer — lifecycle metadata: creation date, acquisition method, planned update interval, steward, and quality metrics. For utility pipelines and cables, owners have submitted two-dimensional cadastral data at least once a week since July 2013. The register operator's role is to verify completeness and timeliness; content responsibility stays with the owner. Utility data positional accuracy runs to plus or minus twenty centimeters. The terrain base, by contrast, dates to 2014 — a reminder that not everything updates on the same cycle, and tracking those differences explicitly is what makes the system trustworthy rather than misleading. Schrotter distinguishes two modeling modes. Predictive modeling builds digital prototypes before reality exists — using the twin as a sandbox to test scenarios before any concrete is poured. Reactive modeling links the digital object to live sensors so that real-time changes in the physical world propagate immediately into the digital one. Together these two modes cover the full range from long-term planning simulation to operational monitoring.

Now, that data is useful internally. But in late 2018, Zurich made a choice that changed the scale of what the twin could do: it published the three-dimensional spatial datasets as Open Government Data. The terrain model, block model, and roof model went onto the Open Government Data platform, accompanied by a media conference on the twelfth of November 2018. Between that date and the twentieth of September 2019, the portal saw eleven thousand seven hundred twenty-two dataset downloads. The five three-dimensional spatial datasets alone accounted for roughly one-third of all downloads in that period. More broadly, Open Government Data across the city's portal was downloaded seven thousand four hundred twenty-two times, corresponding to about sixty-five percent of total downloads. That is not a symbolic number. It's evidence of external demand. The city has operated an "Open by Default" principle since 2012 — administrative data provided free of charge, machine-readable, under free license. Applying that principle to the three-dimensional models transformed them from internal assets into public infrastructure. Schrotter lists the application categories this enabled: start-ups around the Internet of Things, Smart City solutions, location-based gaming, and collaborative platforms.

The city itself uses subsets for solar potential calculations, shadow and noise analysis, flood simulation, and submission to third parties for visualizing planned construction. Free access, the paper states, "simplifies the use enormously" and opens value-creation opportunities for private individuals and project partners. That framing ties the open-data decision back to the city's governance goals: transparency, participation, and economic value creation. It was an infrastructural and strategic move, not a transparency gesture. The applications built on that infrastructure are where the abstract architecture becomes concrete planning work. Start with urban climate. Zurich is currently averaging six heat days per year. Climate projections say that number will reach twenty-one heat days by 2060. The city used the digital twin — combining terrain, existing buildings, planned buildings, and tree locations — as input to meso- and micro-scale climate models. These simulations quantify the obstacle effect of planned buildings and compare it against current development, reporting impacts on temperature, wind, and cold-air volume flow as functions of building length, width, height, and position.

The results feed directly into the city's Sectoral Planning on Heat Reduction, expected to be published in 2020, and are used as an additional basis for variant decisions on public-sector construction projects. Heat, shadow, noise, traffic — these can now be layered and weighed in the same decision process because the geometry is consistent across all of them. Solar potential and shadow analyses have also become routine planning tools. For the municipal development plan, planners combined three three-dimensional data levels in a single interactive web application: current building stock from the digital twin, maximum building capacity under current zoning from a parametric Esri CityEngine model, and target densification scenarios from urban design drafts. Visual comparison of those three levels in the browser allowed planners to derive quantitative statements — building utilization, floor space, numbers of inhabitants, and jobs — because the geometry was linked directly to planning parameters. That link between geometry and attribute data is what makes these visualizations analytically useful rather than just illustrative.

For architectural competitions, two tools were tested that leverage the twin for jury evaluation. A web-based three-dimensional map hosts submitted competition models, allowing juries to view designs jointly, change perspectives and scales, compare entries synchronously, and toggle spatial data layers. The second tool uses Microsoft HoloLens Augmented Reality glasses to project competition designs and their surrounding built context as an interactive hologram in the judging room. Juries can simulate shadows, switch spatial data on and off, change scales, generate cross-sections, and at the extreme, visualize designs on site at a scale of one to one. The city is actively testing whether the Augmented Reality tool can replace physical scale models in the long term. For public participation, three approaches were tried. The city transformed parts of Zurich into a Minecraft world based on the digital twin to engage younger residents. A browser-based interactive participation tool built on Esri's JavaScript Application Programming Interface lets users draw buildings, green spaces, or trees within a planning area with a few mouse clicks and submit the result to city administration.

Tests with users who had no planning background found the tool easy and intuitive despite its three-dimensional interface. The third approach was commercial and remarkable in scale: Blindflug Studios used the city's building data to create a game called (re)format Z:, a dystopian cyberpunk reimagining of Zurich's Reformation history. Development took about seven months. The game launched in November 2017 and hit twenty-five thousand downloads within three weeks, reaching first place on the App Store's Top Free Puzzle Games list. By early 2019, it had reached one hundred thousand downloads and twenty-four thousand five hundred hours of total playing time worldwide. By the end of 2019, it was still generating about eighty thousand impressions per week in the App Store. That is reach no public engagement event achieves. Across all of these applications, the pattern is the same: the digital twin's payoff is visual, data-linked prototypes that enable scenario testing in actual decision processes. Geometry linked to planning parameters. Simulations linked to governance workflows. Open data linked to external application development. The infrastructure works as a platform precisely because the lifecycle management and metadata governance described earlier make the underlying data trustworthy enough to build on.

But Schrotter is candid about where the platform is not yet complete. The central unsolved problem is subsurface modeling. Utility owners have been submitting two-dimensional pipeline and cable data weekly since 2013, and they already record three-dimensional coordinates internally. The problem is that "this information is lost at the interface to the utility cadaster." The three-dimensional coordinate data never makes it through. Schrotter says this gap will be closed to enable Building Information Modelling-capable utility cadastral datasets, but it has not been closed yet. Until it is, the digital twin has no reliable integrated underground layer. For a city managing aging utility infrastructure at the same time it is densifying above ground, that is a significant gap. Standards problems compound it. There are no jointly developed guidelines at cantonal and federal levels for modeling, recording, and updating three-dimensional geodata. CityGML offers extensive modeling capabilities, but Zurich does not use all of them.

The practical question of how to use the standard sustainably and efficiently remains open. Building Information Modelling and Geographic Information Systems still operate in separate worlds: different levels of detail, different update cadences, no practical concept for where they converge, where they stay separate, and where they need to exchange data automatically. Schrotter puts the question bluntly — there are no practical concepts for how these two worlds meet. Data volume is a growing pressure. Mobile mapping is producing increasing volumes of point clouds. How to refine façades from those clouds, how to integrate point cloud vegetation models, and how to update and maintain the twin efficiently as data volumes grow are all open questions without settled answers. The paper cites the Digital Underground project and related modeling work by van Son and colleagues from 2018, and by Yan and colleagues from 2019, as promising directions — but describes them as evaluations of measurement methods, not operational solutions. What the Zurich implementation demonstrates, taken together, is a set of concrete prerequisites for this kind of system to work. Metadata and lifecycle governance are not optional: they are what makes simulation outputs trustworthy enough to put in front of a planning committee. The terrain model is from 2014.

The block model is continuously updated. Those are different things, and knowing which is which — and encoding that difference in governed metadata — is what allows anyone using the system to know what they are actually looking at. Open data release functions as a force multiplier: eleven thousand seven hundred twenty-two downloads in less than a year, one-third of them from five three-dimensional datasets, is evidence that documented, accurate, open spatial data finds its users. Visualization has to be audience-specific — Augmented Reality for juries, browser tools for citizens, Minecraft for younger residents, and commercial games for everyone else. Institutional coordination has to span Geographic Information Systems, Building Information Modelling, utilities, and planning departments because no single department owns all the data the twin needs to be useful. The gaps Schrotter identifies — subsurface, Building Information Modelling and Geographic Information Systems convergence, shared standards — suggest that any city attempting to replicate Zurich's approach should treat these not as problems to solve after the system is built, but as design constraints to address from the beginning. The underground infrastructure is not a later add-on. The Building Information Modelling interface is not a future compatibility layer. These are architectural decisions, and deferring them creates exactly the data-flow gaps that Zurich is now working to close.

This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

More in Engineering →

Related lectures