The Community for Technology Leaders


Pages: pp. 6-9


Pam Frost Gorder

When William Gibson published his groundbreaking science-fiction novel Neuromancer in 1984, he painted a bleak future of concrete and technology, where an overpopulated American eastern seaboard had fused into one giant megalopolis—the Sprawl.

Gibson has often said that he wasn't trying to predict the future with Neuromancer. He did influence it, though, because whether the Sprawl could ever truly exist is a serious question confronted by scientists working in the interdisciplinary urban-planning realm.

Sleuth, a computer model developed by the United States Geological Survey (USGS), is helping settle that question. It divides maps into a grid of cellular automata—mathematical constructs such as cells or pixels—and charts the spread of urban land into rural.

In an early application, Sleuth simulated long-term urban growth along the eastern corridor. Discrete blobs of color representing Washington, DC, New York, and Boston in 1998 spread outward on the map until by 2100 the three had become one city.

Three Possible Futures

Claire Jantz, research associate and geographer at Woods Hole Research Center (WHRC), recently focused Sleuth on the Washington-Baltimore-Chesapeake Bay area for a more detailed, short-term study. In the March 2004 issue of the Environment and Planning B journal, she and colleagues from the University of Maryland, College Park, and the WHRC reported that developed land in that area could expand as much as 80 percent by 2030 if current land-use policies and growth trends remain in effect (see Figures 1 and 2).

Graphic: Developed land. Baltimore, Maryland, in 1986 (top) and 2000.

Figure 1   Developed land. Baltimore, Maryland, in 1986 (top) and 2000.

Graphic: Developed land. Washington, DC, in 1986 (top) and 2000.

Figure 2   Developed land. Washington, DC, in 1986 (top) and 2000.

The same study explored two other possible futures for the region: one in which forest and agricultural areas were more protected, and one in which growth boundaries limited development. In one scenario with especially strong ecological controls, expansion was limited to 20 percent. In another scenario with more moderate controls—a managed growth scenario—development reached only 30 percent.

To Jantz, what's interesting about the 80-percent growth scenario is that it showed many of the area's more remote rural counties transforming from agricultural into urban landscapes in only 30 years. Counties closer to Washington, DC, experienced a similar transition starting in the late 1960s, she points out.

"That dramatic shift occurred in roughly three decades, so it's not unrealistic that a similar transition could occur in other, more outlying counties," she says.

Chesapeake Bay residents are already aware of how these changes can affect water quality. As green areas are paved, water that would have soaked directly into the earth is redirected to the edges of urban areas, where it washes sediment and contaminants into the watershed.

Jantz considers the 30-year forecast to be only mid-range in terms of the time it allows the bay's communities to make decisions about urban planning. She wouldn't want to extend the model much further, though.

"We felt we were pushing the envelope with a 30-year horizon because of error propagation through time," she says. The factors that affect urbanization, such as local and regional economics, housing markets, and population trends, contain a certain amount of variability, and the resulting error adds up through successive iterations. Calibration is the key to minimizing errors and creating valid forecasts, she says.

One way scientists calibrate forecasts is to build "hindcasts" by fitting a model to data from the past. Jantz and her colleagues did that by linking commercial imagery from Space Imaging's IKONOS satellite with images from NASA's Landsat satellites to build a picture of land change in the region from 1986 to 2000. They then aligned the model with trends they saw taking shape.

Model Locally, Analyze Globally

Urban planning integrates several disciplines including cartography, geography, sociology, economics, statistics, and transportation engineering. Yet Keith Clarke, creator of Sleuth, says the model itself is remarkably simple.

The cellular automata change state—in this case, in varying degrees from rural to urban—depending on simple rules of interaction with neighbor cells, like pieces in the mathematical Game of Life (which was invented by British mathematician John Horton Conway in 1970 and is a good example of cellular automata). Six control parameters (slope, land use, exclusion, urban extent, transportation, and hillshade) give the model its name. Monte Carlo algorithms generate many possible patterns among the cells, and the results are averaged during calibration.

When Clarke developed Sleuth at USGS in 1992, he was attempting to provide portability, something that was lacking in urban models at the time. A model that worked in one geographic region wouldn't necessarily work in another, so Sleuth provides a generic framework in which users can plug in a new location's characteristics and get customized results. The software is free for download from Clarke's Project Gigalopolis Web site (

"I wanted a model that would work anywhere, but could 'learn' its local environment through a rigorous calibration phase," Clarke recalls. "This at least puts all urban areas—large and small, developed and developing—on the same scale. The problem is that the scale is typically five dimensional."

The five control parameters (six if land use is being analyzed) each can take on 101 values; with a large number of Monte Carlo iterations, Sleuth can produce billions of results. The main difficulty is crunching the numbers and reducing the data's dimensions so that the results are easy to see. "If the five dimensions can be reduced to three or two, we can look at the data on a map," he says.

Maps are critical for urban planning. Different scenarios' images, such as the three produced in the Chesapeake Bay study, give people the global perspective they need to make decisions.

"A model is just a model," Clarke says. "Many people can't understand the details, but they want to use models that are credible and useful. We found that scenarios are the usual way that people other than modelers interact with models and data. People might want to understand the basic premise on which models are built, but they really want simple choices to make planning decisions."

Sleuth images make the extent of possible urbanization easy to understand, but what else can they tell communities?

"I think the real question here is related to scale, pattern, and process," Jantz says. The different processes that drive urban development patterns operate at different scales. Local economics is one factor; topography is a larger issue. Then there is the smallest-scale process: an individual's decision to move, which depends on any number of personal factors, from the need for more space to the desire to escape bothersome neighbors.

"Even with highly detailed data sets, a model like Sleuth would not be able to capture all local-scale factors that would result in the development of a specific parcel of land. Rather, it captures patterns that result from these local-scale processes at work across the landscape," Jantz says. "While Sleuth is able to simulate development patterns, it's difficult to use this modeling framework to make hypotheses about the underlying processes creating the patterns."

One area of the country that imposes strict urban-growth policies is Portland, Oregon. Sonny Conder, a senior planner for Portland's Metropolitan Service District, praises the level of detail on Sleuth's maps. But Jantz's study only makes him ask more questions.

"The results show that imposing limits on growth curbs urbanization, but where does the growth go? Does it increase densities in already urbanized areas, and where? Does real estate prices change? How about traffic volumes, mode of transportation, and travel times?" Conder asks. "While Sleuth represents a strong step in the right direction and provides a platform for informing regions of the consequences of unbridled urban expansion, it does little to inform regions of what can be done about it."

He uses a different model to link a regional econometric model, a transportation model, and residential and nonresidential real estate models in conjunction with Geographic Information Systems data. The model, called MetroScope, simulates actual market response to urban factors. "Our approach attempts to explicitly represent the operation of supply and demand in the marketplace," he says.

Ethan Seltzer, director of the School of Urban Studies and Planning at Portland State University, agrees that models like Sleuth are extremely useful. "That said, it's important to keep in mind that the fundamental requirement for models like this is excellent data, coupled with ongoing calibration and revision," he says. "Although advances in software and computing power have been exceptional, the utility of these models is still limited by the extent of our understanding of communities and the built-up and natural environment. Clearly, there is much to do in those realms."

Computing Issues

Clarke, now a research cartographer and professor of geography at the University of California, Santa Barbara, says he also designed Sleuth for portability to different computing platforms. "I've used almost every [computing] configuration that you can imagine. Right now, we're using everything from Sparc stations and PCs to Blue Horizon at the San Diego Supercomputing Center," he says.

Jantz's group did their work on the Beowulf PC Cluster at USGS's Rocky Mountain Mapping Center in Denver. Their model of one small section of Chesapeake Bay measured roughly 4,000 by 4,000 cellular automata, and required more than 2 Gbytes of RAM per machine on the cluster. Until recently, motherboards that could hold that much memory weren't available off the shelf, but Jantz thinks the model would benefit more from developing a computationally efficient calibration process than a boost in computing power.

"The most widely used calibration method for Sleuth is brute force, which means that the model runs multiple simulations using every possible combination of growth-parameter values," she says. "This is computationally intensive and also produces a large amount of data that must be interpreted by the user."

Clarke is working on the problem. He hosts the Sleuth user forum ( whose members have figured out how to use a Linux emulator to make the Sleuth code run on Windows machines. That technique has allowed his group to test calibration methods on PCs. Meanwhile, they're continuing their work on USGS's Beowulf and Blue Horizon. "We'll take cycles wherever we can find them," he says.

One of his students, Noah Goldstein, has also written a macro that does the calibration using a genetic algorithm. "It may buy us a huge speed-up," Clarke says.

The Future Is Now

The NASA Land Cover and Land Use Change Program, the Chesapeake Bay Foundation, and the National Center for Smart Growth at the University of Maryland all funded Jantz's previous study. Now the US Environmental Protection Agency has awarded her group funding to construct a 30-year forecast of the entire Chesapeake Bay watershed—an area seven times larger, covering parts of six states and the District of Columbia.

"We hope to couple Sleuth to other models that address trends in regional economics and population growth," Jantz explains. "If successful, this would be research at the forefront of urban modeling, producing fine-scale forecasts over a large area, using methods grounded in urban theory and economics, and incorporating satellite remote sensing products."

Woods Hole Research Center plans to post Sleuth-related work to its Web site (

About the Authors

Pam Frost Gorder is a freelance science writer based in Columbus, Ohio.
59 ms
(Ver 3.x)