Let's set the scene.
It's 3 o'clock on a Tuesday in a sprawling cityscape, teeming with life and energy. Professionals work on their computers in towering high rises. Children play on playgrounds as they wait for their parents to pick them up from school. Pet owners jog with their animal companions along the sidewalks. Buses pick up passengers and transport them around the city. A grandmother stops by a farmers' market and picks up a new vegetable to try with dinner.
But then, suddenly, an earthquake shakes the city, causing immense destruction. First responders rush onto the scene and must determine where to begin in their emergency response protocol. They may know roughly how many people are in the city, but where are all of those people now, at 3 o'clock on a Tuesday? With so many moving pieces in this scenario, time is of the essence, and knowing where the majority of the city's residents are in that moment is essential to saving as many lives as possible.
Daniel Adams, an associate research scientist at the Department of Energy's Oak Ridge National Laboratory and member of the LandScan team, recognizes the importance of having this information readily accessible for first responders. Stories such as these are realities that Adams and the Geospatial Science and Human Security Division are preparing for, as they conduct research and gather information that can be used to inform and strengthen rescue efforts and provide data for future population modeling research.
For Adams, working in data science means being a storyteller. He must parse through a large amount of information to determine which elements are most important, paring down the data to result in the most efficient and accurate data set possible. He then figures out how to use the data to communicate the "so what?" to the research community, much in the same way that an author may whittle down their narrative ideas and characters to reveal a story to their readers.
"You really have to be able to tell a story," Adams said, "because storytelling is just as important for data scientists as the actual analysis of the data is."
Adams and his fellow LandScan team members create population models based on patterns of human activity and the way in which humans interact with their built environment. ORNL's population modeling project LandScan - a now publicly accessible data set - provides a modeled estimation of global populations that includes a local area fine-tuning to enable authorities to respond to natural and human-caused crises. The team aims to understand the distribution of people across both natural landscapes - such as parks, greenspaces and open land - as well as built landscapes - like houses and other buildings.
In the LandScan program, the team is now moving toward a more temporally specific representation of populations that are modeled based on time-dependent human activity patterns. Using both models and real-world data allows the team to estimate human activity in a particular area, providing more accurate population representations for use by first responders in emergency response scenarios.
One of Adams's roles on the LandScan program is classifying buildings, facilities, and other designated spaces based on their usage types, such as residential versus non-residential buildings. Because not every building serves the same purpose, not every building will hold the same number of people at any given time. Using ORNL's advanced computing systems, artificial intelligence, and machine-learning capabilities, the team has developed a machine learning model to classify buildings based on morphological aspects - such as the building's architectural features - as well as the features of the buildings around it.
Data from Adams's research supports improvements in the likely occupancy and activity levels within residential or nonresidential buildings. From this data, the team can construct a population model that predicts the number of people within a facility over a 24-hour period. For example, it can be assumed that there would be more people in office buildings, schools, or other non-residential areas during the day, and it could also be assumed that more people would be in residential areas at night. In the event of an emergency, first responders could prioritize certain locations over others depending upon the predicted number of people at each location, resulting in more effective rescue missions.
However, Adams's contribution shows only a fragment of the overall "story" of LandScan. When Adams's contributions are combined with those of his colleagues, a clearer image begins to emerge - the world's most accurate estimate of human population in a given area. Adams's work supports an essential component, and improvement, to modeling the world's population.
UT-Battelle manages ORNL for the Department of Energy's Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science. - Lena Shoemaker
