Source-linked AI summary
Procedural modeling of urban land use
Thomas Lechner, Ben Watson, Uri Wilenski, Seth Tisue, Martin Felsen, Andy Moddrell, Pin Ren, Craig Brozefsky
TL;DR
Cities are difficult and costly to model at the detail demanded by digital production, while few tools support artists. The paper uses agent-based simulation to generate realistic, controllable urban land-use maps and automate building and road placement. The system produces artist-steerable cities with modern land-use patterns, but large worlds currently require several hours and the model omits important aspects of real urban development.
Problem
Cities are complex and challenging to model, while few tools help artists create richer digital urban content without matching production-cost increases.
Method
An agent-based simulation generates urban land-use maps from terrain and supports autonomous development, global controls, and local artist intervention.
Results
The system produces unique maps with modern land-use patterns, developmental history, road growth, and spatial relationships comparable through conditional proportions.
Takeaways & Limitations
The maps provide structured urban content that can be populated with 3D buildings and visualized in a city display engine.
Takeaways & Limitations
Largest nine-square-mile worlds take several hours to develop, making the system far from interactive, while the model omits mixed uses and richer transport hierarchy.
Abstract
from arXiv · showhide
Cities are important elements of content in digital productions, but their complexity and size make them very challenging to model. Few tools exist that can help artists with this work, even as rapid improvements in graphics hardware create demand for richer content without matching increases in production cost. We propose a method for procedurally generating realistic patterns of land use in cities, automating placement of buildings and roads for artists.
1 Simulating land use
The paper addresses the difficulty of modeling complex cities by proposing procedural land-use generation that automates building and road placement while preserving artist control.
- Cities are difficult to model because of their complexity and size, while few artist-oriented tools exist despite demand for richer digital content.
- The system procedurally generates five land-use types and outputs property lines, development age, density, and population from a geographical map.
- Artists can let development proceed autonomously, adjust global behavior, or guide and directly set local land use.
- Each generated city is unique while corresponding with modern land-use patterns.
2 Procedural modeling and motivation
The work extends procedural modeling toward realistic urban content for artists, combining automated land-use simulation with steerability and developmental history.
- Procedural modeling and motivation: Prior graphics research supported urban capture, walkthroughs, and procedural modeling, but procedural modeling of human artifacts remained relatively sparse.
- Procedural modeling and motivation: The project benefited from close collaboration with SimCity developers at Electronic Arts.
- Land use: The project contributes a system that simulates urban land-use distributions and automates building and structure placement for artists.
- Steerability: Agent-based simulation lets artists pause and alter development freely without needing changes to conform to an output-language construct.
- History: Continuous developmental history assigns different ages across the urban landscape, improving variety and realism.
3 Simulating urban development
The paper positions its artist-focused simulator between large-scale planning and geographic simulations, using agent-based development to produce controllable urban patterns.
- Procedural modeling and motivation: Planning and urban-design simulations commonly rely on GIS data and produce global statistics or abstract symbolic outputs rather than detailed building and street placement.
- Procedural modeling and motivation: Urban geographic simulations are realistic and carefully validated, but have generally operated at scales too large for detailed positioning of individual buildings and streets.
- Simulating urban development: Artists can increase commercial land use from left to right by manipulating land-cover constraints λ_t.
4 Patterns of urban development
The simulation models major urban land uses through developer preferences, terrain and transport relationships, and artist-controlled primary and tertiary road organization.
- Patterns of urban development: Residential, commercial, industrial, and road development comprise roughly 80% of urban land use, motivating the system’s primary focus.The stated proportions are 40%, 5%, 10%, and 25%, respectively.
- Patterns of urban development: Residential development favors clustered, small, low-density parcels near waterfronts, elevated areas, and parks, away from busy roads and industry.
- Patterns of urban development: Commercial development clusters near markets and transportation corridors, favors flat terrain, and uses larger, denser parcels than residential development.
- Patterns of urban development: Industrial development favors flat land and transport access, while its large parcels are less densely populated than commercial parcels.
- Patterns of urban development: Parks generally appear later in neglected regions and are closer to residential than industrial areas.
- Patterns of urban development: Roads form a hierarchy from freeways to primary and tertiary roads, with networks organized in grid or organic patterns.
- Patterns of urban development: The system currently models primary and tertiary roads and lets artists organize them organically or into grids.
5 Procedural modeling of land use
The system models urban land use through agent-based developers that prospect, build, and evaluate sites while artists control global and local behavior. It produces land-use maps with parcels, roads, development age, density, and interactive editing controls.
- Simulation environment: NetLogo implements distinct developer agents in a patch-based world representing a 1.5-mile-square environment at 40-foot patch resolution.A 200x200 patch world supports one-patch-wide tertiary roads.
- Input and interaction: Artists provide terrain and adjust global proportions, density clustering, land-use clustering, and local road-layout behavior through a painting interface.They can also remove development, add desired development, and apply local incentives or prohibitions during simulation.
- Output and control: The system outputs rasterized land-use maps containing residential, commercial, industrial, and park parcels plus primary and tertiary road networks.Parcels record population, age, value, and density; honey can swap adjacent lower-value parcels to strengthen local incentives despite global-use constraints.
- Property developers: Property developers repeatedly prospect, build or modify sites, and check profitability using local and global site selection.Prospecting favors valuable nearby sites but can relocate globally among the most valuable fifth of developable sites.
- Property developers: Developers form parcels by expanding from road-adjacent patches, while parks use population, existing park coverage, site value, and city size constraints.Residential, commercial, and industrial parcels expand perpendicular to roads before widening along the block.
6 Results
The system produces controllable urban land-use maps that support building selection and automated development, while matching characteristic spatial relationships observed in Houston. Results also demonstrate artist-guided and autonomous development, although numerical validation targets typical urban patterns rather than reproducing a specific city.
- Generated maps: The system generates vectorized urban maps with varied gridded and organic road layouts, including artist-guided commercial development near the bay.Vectorization applies polygon- and spline-based post-processing to raster parcel and road outlines.
- Generated maps: Initial terrain, water, mountains, and roads were painted in less than 10 minutes, after which development proceeded completely automatically.The simulation automatically grew primary roads and concentrated development along primary roads and shorelines.
- Validation: Numerical validation is difficult because the method targets typical urban patterns that feel real rather than reproducing any existing city.The evaluation therefore addresses composition and spatial configuration instead of exact city replication.
- Spatial validation: The conditional-probability comparison with Houston shows self-clustering of uses, road proximity to most uses, and weaker spatial patterns at larger scales.Figure 16 visualizes these relationships with grayscale matrices across multiple radii.
- Application: The resulting 2D maps are designed to be populated with 3D building and structure models for digital productions.A SimCity 3000 visualization illustrates this intended application.
- Application: The system's maps contain enough information to support meaningful choices among SimCity's building library and add ungridded higher-level structure.Comparison with SimCity required simplifying the output because that engine could not display primary roads, structures on inclines, or non-rectangular parcels easily.
7 Limitations and future work
The simulation remains limited by development speed, incomplete representation of urban structure, and insufficient support for historical or cultural dynamics. Future work targets faster execution, more regular and mixed parceling, deeper transport hierarchies, and time-varying urban behavior.
- 7 Limitations and future work: Nine-square-mile worlds take several hours to develop, making the simulation far from interactive despite a current focus on correctness and robustness.The authors identify implementation in a high-performance language such as C++ as an obvious first optimization step.
- 7 Limitations and future work: The simulation does not capture all urban development, including regular parceling, mixed uses within parcels, or rich multi-level transport hierarchies.The authors suggest larger-unit parceling, mixed-type development mechanisms, and adding freeways to address these gaps.
- 7 Limitations and future work: Higher-level historical or cultural control, such as representing old towns or Chinatowns, would require simulating urban dynamics across changing transport eras.The paper gives the transition from horses to trains to cars as an example of the required temporal dynamics.