DOE CSGF Practicum Profile: David Potere
Oak Ridge National Laboratory, 2005
"Tracking Wal-Marts to Make Better Maps"
An edited excerpt from DEIXIS (2006-2007), the DOE CSGF Annual
David Potere has lived in 18 homes in 29 years and visited five continents in six years, first as the son of an Air Force pilot and then as a Navy lieutenant. Despite that, “I’m always lost,” he jokes.
So it’s no wonder Potere’s path led to geography. He’s become part of a growing community of researchers who use high-performance computers and satellite imagery to make better maps – especially ones showing where people live. In doctoral degree research at Princeton University and his Department of Energy Computational Science Graduate Fellowship (DOE CSGF) practicum, Potere (pronounced poe-TEER) refined techniques to make global maps of human population more accurate.
With satellites transmitting complete images of Earth’s surface about every 48 hours, “We have so much information coming out of space right now – a terabyte (one trillion bytes) of data per day – we need high-performance computers to sort through this,” Potere says. “There’s a lot of speculation about what answers might be buried in the data, and people are just figuring out what questions to ask.”
For Potere, the questions have revolved around defining urban areas – and tracking Wal-Mart stores.
Satellite views of Durham, N.C. before and after construction of a Wal-Mart |
Potere’s work at Boston University and Princeton got the attention of Budhendra Bhaduri, leader of the Geographic Information Science and Technology (GIST) Group at DOE’s Oak Ridge National Laboratory (ORNL). He invited Potere to do his practicum with GIST in summer 2005.
GIST’s LandScan Global Population Project uses high-performance computers to calculate worldwide population distribution based on census figures, topography, transportation and other data. Computer programs estimate population even where no specific numbers are available.
Just as importantly, the project estimates population at a given time, such as on a highway through a desert. A census would show little or no population, even though the highway may have thousands of people on it at one time. Such estimates are important to calculate risks and plan for natural disasters, terrorist attacks and industrial accidents.
“LandScan is the finest population data that has ever been produced,” Bhaduri says, estimating population in each of hundreds of millions of cells of about one square kilometer.
But maps are only as good as the information they’re based on, Potere says. “If you want to build the kind of map we want – to talk about space and time, where has there been change and when has there been change … you need to have some sort of ground-truth” – accurate data on where and when a change occurred.
Potere and his fellow researchers needed a set of large, documented geographic and demographic changes to validate their mapping tools. They settled on Wal-Mart stores because of their large size and precise opening dates.
With Neal Feierabend, a student research assistant, and Edward Bright, a remote sensing and Geographic Information System (GIS) specialist who heads the LandScan Global Project, Potere translated the street addresses of more than 3,000 stores into map coordinates. Then the researchers tapped satellite images to precisely locate the coordinates of each.
The researchers next looked at “signatures” for selected stores. They examined data from NASA’s MODIS (Moderate Resolution Imaging Spectroradiometer) project – two satellites circling the earth over the poles to produce a complete image of its surface every one to two days. A “greenness index” of the images indicated how much was in vegetation.
Pre- and post-construction satellite views of a Wal-Mart distribution center in Apple Valley, Calif. |
The index rises and falls with the seasons, as plants grow greener in the spring and summer and recede in the fall and winter, Potere says. “What we were looking for is a depression in that signal, which would indicate a change in use, and we found it,” he says. “The greenness index still responds to the seasons, but it does so far less dramatically ” where a Wal-Mart has been built.
Such data can “train” high-performance computers to recognize land-use changes in satellite images, Potere says. “If I’m interested in new facilities in the forests of North Korea, I can’t go there, but I know what a forest looks like in Maine and I know what a Wal-Mart looks like in Maine,” he adds. Computers could scan MODIS images from anywhere on the planet and spot similar changes.
Bhaduri says computers also could spot urbanization, illegal logging, crop disease and other land-use changes.
Computer automation ties into another of Potere’s practicum projects. With Bhaduri and other GIST researchers including Anil Cheriyadat, he refined an algorithm that precisely delineates urban boundaries in high-resolution satellite images. Good estimates of a city’s boundaries are important because they affect how researchers distribute census counts.
The program Potere and his fellow researchers refined correlates gray levels in the photos with edges of geographic features. Urban areas typically have more edge features. The result is an image with urban areas that are precisely – and quickly – defined.
Spatial-temporal distribution of U.S. Wal-Marts 1964-2004 |
Potere earned a bachelor’s degree in American History at Harvard and did his Navy hitch before starting in the geography program at Boston University. His research there focused on forest loss along the Appalachian Trail, but after completing his master’s thesis, he found himself more interested in the human aspects of geography. In the fall of 2005, he moved to Princeton’s Office of Population Research to study demography.
His dissertation is rooted in a third ORNL-GIST project. Potere, ORNL student research assistant Karen McNeany, and Annemarie Schneider, assistant geography professor at the University of California, Santa Barbara, compared six independently produced global maps of urban land cover. Although they’re based on common sources, “There are order of magnitude differences between these products,” Potere says.
Potere’s doctoral advisor, Princeton Demography and Public Affairs Professor Burton Singer, says remote sensing technology like satellite images has “the potential to really change the game” in predicting things like the spread of diseases, particularly those borne by insects and parasites. Potere, he says, came to his program “with a much more sophisticated understanding of the remote sensing technology than any other student I’ve had.”
