Modeling land use with urban simulation

COURSE PROCEDURAL MODELING OF URBAN ENVIRONMENTS Modeling Land Use With Urban Simulation Benjamin Watson North Carolina State University 185 COURSE PROCEDURAL MODELING OF URBAN ENVIRONMENTS Modeling Land Use With Urban Simulation Benjamin Watson North Carolina State Univ. http://cde.ncsu.edu/watson/ and Learning Sciences, Northwestern University Architecture, Illinois Institute of Technology 186 Goals Automatic placement of buildings Not building generation (yet) City layouts should be Convincing, typical But completely novel Controlled automation Minimal effort, maximum effect Interruptable PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 187 Note that we are not trying to recreate existing cities, only perhaps some of the character of existing cities. E.g. “I want a city that is a mix of Paris and Tokyo”. Thus we are not using the widespread capture technology. The tradeoff between minimal effort and maximal effect is a delicate one. “I want it, but I want it for free. Mostly”. Interruptability is key. The ability to stop a sim, edit it slightly, and restart it is a basic element of control. 187 Previous: Urban Design OCEAN: very little automation, input heavy PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 188 Urban designers are interested in conceptual sketches embedded in current urban settings. This view was the results of hundreds of man-hours of work. 188 Previous: Urban Planning CommunityViz: input heavy, very small scale PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 189 Urban planners are interested in prediction. For high certainty, simulations require a great deal of input, and must generally operate at a very fine or coarse resolution. This is Ascutney, Vermont: a small New England town. 189 Previous: Urban Geography SprawlSim: input heavy, coarse resolution PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 190 Urban geographers are more interested in processes than predictions. They are trying to understand phenomena. This is Chicago IL, next to Lake Michigan. Urban geographic simulations are most relevant to our own, because they can work at appropriate levels of detail, and because they include significant automation. 190 Different Goals Non-existent places, minimizing Input requirements Prediction requirements User control Not accurate processes Instead, convincing output High resolution & large scale output Not just one or the other PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 191 We are not interested in existing places in this work. This means we don’t have to imitate what’s there, and we don’t require a great deal of input. We are also interesting in cities that match applied needs. Users should be able to control the sim, regardless of inaccuracies they produce. There are already large scale sims, and high resolution sims. There are very few that output entire cities at the resolutions we require (sub parcel/plot, approx 10m). 191 Approach Agent-based simulation Similar to flocking, particles Model urban development Environment: terrain and structures Agents: structure developers PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 192 We are using agent-based simulation, the same technology used for flocks, and particles. This approach has only very recently been taken up by the urban geography community. 192 Results: SimCity Layout populated by EA’s SimCity 3000: curvilinear road hierarchy, two urban cores PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 193 Our results are 2D maps of land use, meant to be populated with automatically generated structures. EA was kind enough to loan us SimCity, which we use to populate one of our maps here. With its limited gaming setting, SimCity can constrain its development a great deal: single lane roads only, rectangular parcels only, no structures on slopes. Despite these constraints our output adds realism. SimCity uses a very simple model, just complex enough to provide good play. It is not fully automated, requiring users to “paint” zones of use onto maps. That is the essential element of play: how good a planner are you? Our simulation can be completely automated. 193 Results: SimCity Layout populated by EA’s SimCity 3000: curvilinear road hierarchy, two urban cores PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 194 Note the curvilinear road hierarchy and dual urban cores generated by our sim. 194 Results: SimCity Zooming in on the cores: interesting, despite simplifications required by SimCity PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 195 195 Results: 3D output Rapid prototyped output: a neighborhood in Berlin, built by students PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 196 Our tool was not designed to imitate existing places, but we tested the control we provide by having architecture students recreate existing neighborhoods in Berlin… 196 Results: 3D output Madrid PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 197 …and Madrid. Students painted the primary roads here, and the sim filled in building and secondary road development. 197 Results: land use A 3x3 mile city: residential, commercial, industrial, park, roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 198 Here is a 2D vectorized map of our output. Colors indicate land use, with dark green indicating undeveloped land. Thick grey lines are primary roads, thinner lines secondary roads, and the thinnest lines are property boundaries. 198 Results: land use Artist paints a commercial zone, and various road layouts PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 199 In this example, the artist has directed the sim to place most commercial development around the bay, a somewhat unrealistic request. 199 Results: land use Artist paints a commercial zone, and various road layouts PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 200 Note the many road layouts that the artist has painted onto the terrain as input constraints. These can very in orientation, “tightness”, and spacing. 200 Results: land use Zooming in… simulation takes place at sub-parcel level PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 201 Our sim happens at a sub-parcel level on a grid. Each grid element is 12m’ on a side. Obviously the grid itself will not produce good parcel boundaries. 201 Results: land use Zooming in… simulation takes place at sub-parcel level PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 202 Our vectorization is not adequate to smooth these. We are currently working on a good 2D smoothing post-process. 202 Results: land use Zooming in… simulation takes place at sub-parcel level PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 203 203 Results: land use A town, different clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 204 Here’s a smaller town with different clustering and primary road behavior. 204 Results: land use A town, different clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 205 205 Results: land use A town, different clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 206 206 Results: land use A town, different clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 207 207 Results: gridded Same town, on gridded agent environment; previous results were vectorized to produce continuous edges PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 208 Here, the first direct view of a corresponding simulation grid. All remaining results will be the sim grid itself. Here we had not yet improved our parcelling algorithm, so there are holes. 208 Results: growth Cities develop over time, starting shaped by artist constraints such as terrain, roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 209 One of the advantages of our growth modeling is that we generate history, which is a crucial element of the urban “feel”. Old structures can look different than new structures. 209 Results: growth Cities develop over time, starting shaped by artist constraints such as terrain, roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 210 210 Results: growth Cities develop over time, starting shaped by artist constraints such as terrain, roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 211 211 Results: growth Cities develop over time, starting shaped by artist constraints such as terrain, roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 212 Note that redevelopment and conversion of use can occur. (Compare sequential frames). 212 Results: growth History of growth can provide additional character/guidance for building placement PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 213 213 Results: density Density of use is also modeled: dense, sparse here a weak urban core PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 214 There is a great deal of internal data to our simulations. Here we look at density of use, with brighter areas indicating dense development. This is a very loosely clustered, sprawling city. 214 Input needed: None. Optional: terrain/roads/use, global or local constraints PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 215 Sims can run without *any* input, or can be guided by artist input constraints and parameters. Here, the artist has painted a simple geography, with high regions a brighter green/lighter blue. 215 Control: centrality Weak core: density smoothness constraint weak PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 216 Now we’ll demonstrate artist control. Artists can easily manipulate the centrality of urban use by changing one parameter. Here loose clustering... 216 Control: centrality Stronger core: stronger smoothness constraint PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 217 Tighter clustering… Note that these sequences are generated with the same input terrain, including a few roads. So you can see the subtle variations and randomness introduced by repeated runs of the sum on the same input (with one changed parameter). 217 Control: centrality Strong core: smoothness constraint strong PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 218 Tightest clustering. 218 Control: prevalence of each land use Little commercial use: land cover parameter smaller PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 219 Similarly, artists can control the amount of density and land coverage dedicated to each use by manipulating two parameters. Here, very little commercial use. 219 Control: prevalence of each land use More commercial use PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 220 More commercial use… 220 Control: prevalence of each land use Lots of commercial use PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 221 Still more. 221 Control: clustering Thinly distributed commercial use: importance of proximity weak PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 222 Land uses cluster together, with varying degrees of tightness. Artists can control tightness of clustering with a per usage type parameter. Here commercial clustering is weak. 222 Control: clustering More commercial clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 223 Now it’s stronger. 223 Control: clustering Lots of commercial clustering PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 224 And still stronger. Note the especially dense (bright red) development in the center, and that uses have largely abandoned the left side of the city. 224 Control: interruptability State of city: artist unhappy with commercial cluster top right PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 225 We mentioned interruptiblity. Artists can change and wipe out portions of the city as they like, then watch how the city adapts. 225 Control: interruptability Erase the cluster… PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 226 Here the artist levels the commercial concentration top right. The artist could then simply press “go”, and watch what ensues. 226 Control: interruptability Add “honey”, incentives for development. PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 227 But the artist can provide incentives for different types of development, which we call “honey”. Here the artist paints some honey onto the city. Color indicates the sorts of development attracted by the honey. We’ve made the honey particularly strong here, for demonstration. But it can easily be made less sweet. 227 Control: interruptability Result: honey “sweetness” controllable PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 228 And appropriate development ensues. The artist does not have to carve out new roads and property boundaries by herself. 228 Developer agents Build structures in environment Property res, ind, com and park types Roads Primary, access extenders, access connectors PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 229 Agents build things. There are seven types of developers: residential, industrial, commercial and parks, as well as primary road, access road extenders, and access road connectors. 229 Property developers During each sim tick: Move toward valuable land Hill climbing w/ random drops Build on most valuable land seen recently Create parcel or increase density Keep development if profitable PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 230 This is the basic behavior of the four types of property developers. As noted earlier, property developers can convert uses if it’s profitable. We also don’t permit industrial to be converted directly to residential, and vice versa. 230 Residential value Near water Near residential Slightly above average elevation Far from industrial uses PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 231 Here are the major factors that residential developers use to value land, with a corresponding value field. 231 Commercial value Near market Res & com Near primary roads Near water Flat land (Away from parks) PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 232 And the major factors for commercial value, with a corresponding field. 232 Industrial value Flat land Near water Near industrial Near primary roads Far from residential Far from parks PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 233 Factors and field for industrial value. 233 Park value Near other parks Far from industrial, commercial Not valued by other uses Hilly terrain Near water Near residential PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 234 Factors for park value. 234 Calculating value Value = Constraints × (Importance•Terrain) + Honey Constraints: powerful limits on value Importance: relative value of terrain characteristics Terrain: attributes of local environment Honey: artist-provided development incentives PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 235 This is the high level math of calculating property value. Importance is a vector depending on developer type. It weighs terrain attributes. Terrain is a vector depending on location. It weighs elements such as proximity to roads, water, and usage types. 235 Terrain A vector with locally varying elements, e.g.: dw = 1 1 + dw eh = e (Ez − Ez − Eoffset )2 −128 Proximity to water Elevation PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 236 Here are two example terrain attributes: proximity to water and elevation. Value to water proximity falls with inverse square of distance. Residential folks like to have views and be slightly above average elevation. 236 Importance A vector of weights applied to each terrain element eh = elevation ev = elevation flatness epv = elevation variance efp = elevation at flood plain dw,dr = proximity to water, res di,dpk = proximity to industry, park dpr = proximity to primary roads dm = proximity to market Importance Usage Residential Commercial Industrial Park eh 0.3 ev e pv e fp dw 0.3 dr 0.4 0.15 di dpk dpr dm 0.2 0.5 0.1 0.1 0.15 0.3 0.1 0.4 .1 .1 0.4 0.1 PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 237 Here is a table with most of the terrain attributes in columns, and uses in rows. A few have been omitted for clarity. Importance weights are in rows. Weights sum to one, once off-screen attributes are included (just a few). For example, residential developers value above average elevation, proximity to water, and proximity to other residential development. 237 Constraints Proportions of land cover per use Proportions of population per use Density smoothness Per use proximity constraints Changing the smoothness constraint PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 238 Constraints are multiplicative factors that strongly limit value. We demonstrated the constraints earlier, here’s a reminder. 238 Road agents Access road extenders Ensure that developed land is reachable Access road connectors Ensure that access network is connected Primary roads Extend, connect primary network PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 239 There are three types of road agents. Currently we implement only two levels in a road hierarchy; more would increase realism, and be fairly simple. 239 Road constraints All paintable: Horizontal, vertical grid spacing Grid orientation Grid tightness Overall road density PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 240 Artists can control road layouts at a high level by painting them onto the input map. Remember that map with different griddings? Here it is again. Orientation, tightness, spacing and overall road density can be painted (pavement/sq m). Also roads can be directly painted by the artist. 240 Extenders Prospecting Look for sites away from network Check that grid constraints okay Try to build Attempting to build Shortest path back to road Avoid property lines Avoid elevation change Check constraints PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 241 Extenders navigate through a distance-from-road landscape. Road building does not always succeed: a completed segment may violate the road constraints. 241 Connectors Prospecting Random walk of roads Choose random destination If shortest path too long, try building Attempting to build Trace path to destination Avoid roads, elevation change Check grid constraints, road length PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 242 Connectors navigate on the existing road network. If a random destination is too long on existing network, build a connecting segment. 242 Primary developers Prospecting Look for sites away from primaries Avoid other primary developers Build Building Toward nearest primary And toward or away from center Avoid water, prefer access roads PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 243 Primary road buildings navigate in a distance-from-primary road landscape. They prefer to upgrade existing access roads to primary roads. 243 Simulated vs. real world Simulated PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 244 Houston It’s easy to get road layouts and satellite imagery. Less easy to get parcel lines. Hard to get trustworthy land use information! Houston is completely unregulated: no zoning at all, zero. So it represents unfettered development activity. We have data for Chicago and Brooklyn, but this data has various flaws. Both these cities are almost completely gridded. Notice the concentration of commercial development on primary roads, with industrial development at the outskirts. 244 Simulated vs. real world Simulated PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 245 Ascutney, Vermont Ascutney is a small New England town. It has already been bypassed by a new highway. 245 Sim vs. real: configuration Conditional probability Given that use here is X, what is probability of having use Y nearby? other water r c i rd w o r = road res ind com PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 246 Beside just proportional areas of development by use, urban geographers use measures of spatial configuration. One important such measure is based on conditional probability. We summarize these probabilities into matrix rows, which we visualize here using a luminance mapping. Thus given that a location is developed for residential use, it’s very likely that residential use will be nearby. Road is also likely to be nearby. 246 Sim vs. real: configuration r c i rd w o r c i rd w o Simulation: Houston: radius 40’ radius 160’ radius 640’ PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 247 Here are the complete matrices for various radii. The bright diagonals indicate strong clustering by use. Roads are quite likely to be close to res, com and ind use, especially as the radius of the measured neighborhood increases. The two sets of matrices are strikingly similar. One exception is the greater proximity of water to industrial use in Houston; a geographic dependency. 247 Future work Near term Speed Mixed use Better parcels Deeper road hierarchy We are working on Higher level control History, culture, “character” PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 248 Speed: we have focused on correctness rather than speed to date. We use NetLogo, a language and development envt implemented on top of Java. A 3x3 mile city can take several hours to develop. We believe a port to C++ and simple optimizations would increase speed several fold. This is important for realizing interactive use. Mixed use: uses currently cannot share a parcel. This parcel sharing is quite common in urban centers. Better parcels. As we mentioned earlier, we must improve and “smooth” parcel boundaries in a post process. We will do this using 2D polygonal optimizations. Higher level control: We would like to describe high level character using sets of the parameters we have currently implemented, and permit artists to interact at that level, rather than just at the lower levels. For example, “Chinese” development, 18th century development, etc. History brings changing technology, and therefore changing patterns of development: thus the marked difference between urban cores and suburbs. We would like to implement mechanisms for changing parameter values over time. 248 References Urban planning Brail, R. and Klosterman, R. (eds.). 2001. Planning Support Systems. Redlands: ESRI Press. Urban design Testa, P., O'Reilly, U., Kangas, M. AND Kilian, A. 2000. MoSS: Morphogenetic Surface Structure — a software tool for design exploration. Proc. Digital Creativity Symp. (Univ. Greenwich), 71-80. Bettum, J. AND Hensel, M. 2000. Channeling systems: dynamic processes and digital time-based methods in urban design”, Contemporary Processes in Architecture, Architectural Design, 70, 3, 36-43. PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 249 249 References Urban geography Benenson, I. and Torrens, P. 2004. Geosimulation: automata-based modeling of urban phenomena. Chicester, England: John Wiley & Sons. Clarke, K., Hoppen, S. and Gaydos, L. 1997. A self-modifying cellular automata model of historical urbanization in the San Francisco Bay area. Environment and Planning B: Planning and Design, 24, 2, 247-261. Torrens, P. 2002. SprawlSim: modeling sprawling urban growth using automatabased models. In Parker, D., Berger, T., Manson, S. and McConnell, W. (eds.) Agent-Based Models of Land-Use/Land-Cover Change. Louvain-la-Neuve, Belgium: LUCC International Project Office, 69-76. Waddell, P. and Ulfarsson, G. 2004. Introduction to urban simulation: design and development of operational models. In Handbook in Transport, Volume 5: Transport Geography and Spatial Systems, Stopher, Button, Kingsley, Hensher eds. Pergamon Press, pages 203-236. PROCEDURAL MODELING OF URBAN ENVIRONMENTS Slide 250 250 COURSE PROCEDURAL MODELING OF URBAN ENVIRONMENTS Final Notes We will try to produce updated course notes. Please check the web pages of the speakers to get the latest version of these notes 251