Researchers at the Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL) have released LandScan Mosaic, a next-generation global population distribution dataset that estimates where people are by modeling how buildings are used and occupied throughout the day. Built on the foundation of ORNL's widely used LandScan Global, one of the world's most accurate population distribution datasets , Mosaic's enhancements will improve disaster response, humanitarian assistance, infrastructure planning and national security missions worldwide.
LandScan Mosaic's research innovations include:
- AI-driven ability to estimate building characteristics where data is sparse, expanding consistent, high-resolution population modeling to previously underrepresented areas.
- Uncertainty quantification measurements that provide users with greater confidence in population estimates.
- Improved temporal and location specificity compared to LandScan Global, so users have a better idea of where people are, no matter the time of day.
Buildings provide a clearer picture of population
While traditional approaches rely largely on pixels from satellite imagery to estimate populations across landscapes, LandScan Mosaic introduces a new building-centric modeling framework. Researchers implemented machine learning techniques to estimate missing building characteristics - including height, floor count, function and use type -enabling consistent global application even in regions where detailed building information is sparse.
By bringing in additional information about buildings and land use, along with standardized occupancy distributions, the model represents how people occupy residential, commercial, industrial and other structures around the world. The result is a high-resolution representation of ambient population - the average number of people present in a location over a 24-hour period - that captures daily movement between homes, workplaces, schools and other activity spaces.
LandScan Mosaic also provides something unavailable in previous global population datasets: explicit measures of uncertainty that help users understand confidence in the underlying population estimates.
Daniel Adams, R&D scientist at ORNL and lead author of a Nature Scientific Reports paper describing LandScan Mosaic methodology, said, to the team's knowledge, this is the first globally available population dataset release with these characteristics.
"Decision makers often need to act before perfect information is available," Adams said. "By pairing population estimates with transparent measures of uncertainty, LandScan Mosaic helps users understand not only where people are likely located, but also how confident they can be in those estimates."
As they've incorporated AI into the LandScan program, the team has remained focused on ensuring users can trust the data to make decisions in rapidly changing situations such as disaster response as well as more stable conditions such as infrastructure planning.
"We want to power everything we're doing, using state-of-the-art methodologies, but still preserve that trust piece," Adams said. "This is a really important project for showcasing how AI and decision analysis can really be blended together in a trustworthy manner."
Established in 1999, the LandScan program was created to address the need for more accurate population distribution data for emergency response, risk assessment and national security applications. LandScan Global pioneered the concept of ambient population, modeling the full activity space of people over a 24-hour period rather than relying solely on residential locations.
Marie Urban, principal investigator for the LandScan program, said decades of population modeling expertise combined with cutting-edge high-performance computing and AI technologies uniquely position ORNL to perform this work.
"The multidisciplinary approach we take allows us to bring together computer scientists, data scientists, geographers, people with expertise from different backgrounds … We have a lot of great research staff bringing a lot of different skill sets in to help us take the program into another dimension," Urban said.
Looking back to advance population dynamics technology
To go forward, Urban said the team is looking backward. They are using historical population and building development data to understand population and landscape change over time. This information can help predict future populations and long-term impact of events.
The next release, called LandScan Mosaic Timeseries (LSM-TS), is a dataset containing annual population distribution for the 50-year period from 1975 to 2024. It uses LandScan Mosaic as the input and uses a 'back-casting' method to produce historical population distributions.
Andrew Zimmer, geospatial scientist at ORNL, said LSM-TS's unprecedented 50-year look back in time enables researchers to reconstruct historical populations and reflect the growth of many cities, providing the same building-level modeling for these historic datasets as it does for modern-era data.
"That isn't possible with a lot of other datasets that are maybe available for 20 years, where a lot of those cities are already established," Zimmer said. "The building level framework and the daytime and nighttime data allow opportunities to look at both temporal granularity within the day and how that may have changed over time but also look at those settlements and identify those as well."
Zimmer said the team is also working to enhance LandScan Mosaic's capabilities by adding demographic population characteristics. He explained that building-level data can contribute to this effort as well.
"Buildings are used to identify almost a demographic fingerprint," he said. "For example, there is a certain age group that attends schools. So, when we're building a demographic dataset, we can use the function of buildings at a fine spatial resolution to inform the distribution of those younger populations to certain locations and capture daytime and nighttime dynamics."
LandScan Mosaic is available to researchers, practitioners and decision-makers worldwide.
UT-Battelle manages ORNL for DOE'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 .
