Using Virtual Cities to Plan Real Cities A Case Study Using the Virtual Terrain Project In Hangzhou, China Dr. Michael Flaxman Lecturer in Landscape Planning.

Using Virtual Cities to Plan Real Cities

A Case Study

Using the Virtual Terrain Project

In Hangzhou, China

Dr. Michael FlaxmanLecturer in Landscape PlanningHarvard Graduate School of Design

Ben DiscoeLead Developer, Virtual Terrain ProjectThe Open Planning Project (TOPP)

Cast of Characters

15 Graduate Students in Architecture, Landscape Architecture & PlanningOne Semester Interdisciplinary Studio (double course):Faculty: Carl Steinitz (Architecture & Landscape Architecture)

Mike Flaxman (Planning, GIS & Visualization)Ben Discoe, Lead Developer, VTP (not pictured)

Study Area

Location

SoutheasternChina

2 hours SW of Shanghai

Public Image of the Westlake and Hangzhou

Hangzhou 2000

One of the Most Prosperous Cities in China

6 Million People

Population Doubling in 20 Years

15X Auto Ownership Expected

Design Strategy• Conclusion of prior Harvard

study:

• To protect the West Lake, new development should be directed away from it and toward the river to the South.

• Build New City Center Along River

• Make this area more attractive to development than the edge of the West Lake

Basic Task: Propose & Evaluate Alternative Plans for Hangzhou 2020

Each Alternative Accommodates Projected Growth (Same Program)

Evaluate the Visual, Economic & Environmental Consequences

Present Results to the General Public in Form which Fosters Productive Discussion

Design Challenge: How to Develop and Evaluate Compelling Alternatives for Large Urban Areas – Where to place strategic public investments in transportation

infrastructure and civic buildings

– Where to allow various land uses and densities / heights

– Need to establish open space network planning

Problem Summary

Technical Challenge: How to Visualize Alternative Futures for Large Cities and Regions

–Efficiently generating models of current conditions–Visualizing multiple alternative plans in 3D–Visualizing detailed design in broader context–Workflow must facilitate “top down” design

Conventional Approaches to City Representation for Planning

Sketches -> maps Single-use zoning

representing classes of legally-permitted use

Building heights implicit in zoning

Problems with Conventional Representation

Nobody understands the maps except (perhaps) the planner Basically too abstract: all concerned prefer representation

with building massings to provide sense of scale Two dimensional representation of a truly 3D problem

Visualization Goals:

Visual representation of function, size and locations (not “realism” but “natural scene paradigm”)

Represent some individual buildings and neighborhood patterns

A good working model, not just a polished final rendering

Challenge #1: Number and Diversity of ElementsLandscape Components

Terrain

Buildings

Roads

Vegetation

Water

Lighting/Atmosphere

Any large landscape is likely to contain all of these components.

The relative proportion will be location-specific, but each component needs an efficient representation.

Challenge #2: InteractivityWork done in parallel by 20 students working on PCs (single high-end visualization system not practical)

Students and design critics need to be able to view work from any camera angle

Design revisions should require editing at the level of the critique (changing zoning rather than manually moving 3000 buildings)

Potential RealtimeCity Visualization Approaches

GIS Visualization Systems – ArcScene, ERDAS VirtualGIS

Game Engines VRML /GeoVRML Commercial flight simulation generators General conclusion

– Many technical solutions were available, but no single available package was suitable for interactive city design with real GIS/CAD

Why The Virtual Terrain Project? Supports most common GIS formats CLOD terrain simplification allows

realtime performance on normal PC hardware

Can generate buildings procedurally based on footprints

Integrates CAD models where available Cross Platform Open Source, so both free and extensible

VTP Foundations

Built atop three major libraries– Open Scene Graph

• High Performance, OpenGL-based• Reads Many 3D Formats: OpenFlight, 3DS

– wxWindows• Provides mature cross-platform GUI (Linux, Windows,

MacOS)

– GDAL (Geographic Data Abstraction Library)• Reads/writes most GIS formats• Deals with Map Projections

1. Establish data meaning & representation options• Using “Builder” for 2D GIS/Image Data• “Culture Editor” for 3D and proxies

2. View/Edit Using Runtime “Enviro”

VTP Visualization Process

Production Pipeline 1: Base Data

Current Conditions– Base data from 1m satellite scene– Existing building footprints from 1:1000 CAD– Elevation from 1:10K contours + spot elevations

Production Pipeline 2: Buildings

Developed Building Typology Developed custom ArcView

scripts to place template building footprints with attributes

Generic buildings derived procedurally from footprints in Builder

Custom buildings/bridges developed in CAD / surfaced in 3DS. Insert locations specified in GIS.

Production Pipeline 3: Terrain & Ground Cover

Grading plans developed in GIS using ArcView Tin

Surface cover changes specified in GIS

Exported from GIS as an image file (BMP)

Production Pipeline 4: Vegetation

– Planting plans done by landscape architects in GIS (placing stems + plant ID attribute)

– Street trees generated procedurally in GIS based on road type

Results: Process Efficient: Four plans in four weeks

– “a good plan is a done plan”

– Students with no prior 3D visualization experience were able to productively use the tools

Effective: Visualizations influenced design choices– Building densities of first round proposals “looked like Cleveland” so adjusted upward

– Visualization of greenspace and street trees had large impact in public reviews in China

Results of “GIS Templates to 3D” Strategy Weaknesses

– No good tools for building footprint generation/alignment

– Generating grading plans using ArcView very tedious

– Some design best done in 3D.

Strengths–Efficient for large areas–Good, cleanly-attributed data–GIS intermediate products useful

Results: VTP Strengths Supports all major landscape elements:

– billboard vegetation, CLOD terrain

– Procedural & CAD buildings

GIS integration facilitated top-down design and revisions

Simple navigation interfacePowerful real-time visualization technologies for

work on standard commodity hardware

Results: VTP Weaknesses(for use in Interactive Planning)

– Constraints on terrain size and ground plane resolution

– Multiple file conversions required for some design changes (no auto-update/build)

– Lack of integration of feature types (particularly roads & terrain)

– GIS integration was “one way” (GIS->Vis) rather than “round trip” (no Vis->GIS)

General Conclusions about City & Regional Visualizations for Planning Use• Chosen technique must efficiently represent common objects

types• GIS/database essential for effective real-world visualization

creation/management• Procedural modeling greatly facilitates both creation and

updates• Inter-object 3d topology/constraints very challenging

Next Steps Beyond aesthetics - visualizing consequences– Economic, Biodiversity, Water Quality

Scenario Comparison Interface– Visualize differences using crossfade or split screen

Internet integration– Easy data acquisition from web map servers– Facilitate sharing of building models, ecosystems, etc.

Better pattern generation tools

For More Information… On the Hangzhou Project:

– http://www.gsd.harvard.edu/~gsd98mf2 The Virtual Terrain Project

– http://vterrain.org/

Future Directions Interface Enhancements

– Incorporate Viewpoint and Animation Path Enhancements

– Ecosystem/vegetation import and setup– Improve dynamic scene updating in Enviro

Future Directions 2 Better integration with GIS

– Make VTP Client of OpenGIS Web Map Servers– Facilitate GIS <-> VTP Ground plane transfer– Allow edits in 3D, saving to GIS

Improvements to Procedural Generation– Dynamic modification of buildings based on attribute changes (currently only on import)– Semi-random placement with offset from linear features (buildings along road, vegetation

along street, etc.)– Stratified random vegetation placement relative to terrain attributes

Future Directions 3 Efficiency and Quality Enhancements

– Support “Impostors” for buildings– Support tiled/paged terrain with textures– Higher-quality rendering modes for non-realtime use (or

geomorphed with steady camera) Focus on Community and Environmental Planning