ORNL Releases LandScan Mosaic to Model Daily Population Movement Using Buildings.

Researchers at the Department of Energy Oak Ridge National Laboratory have released LandScan Mosaic, a next-generation global population distribution dataset that models human presence at the building level across a 24-hour cycle. Moving beyond traditional satellite-pixel methodologies, the dataset employs a building-centric framework integrated with machine learning to estimate structural characteristics, including height, floor count, and occupancy type. This approach generates high-resolution ambient population maps that track daily human movement between residential, commercial, and industrial zones, significantly improving spatial accuracy in regions with limited census data. LandScan Mosaic distinguishes itself by incorporating explicit uncertainty metrics, a first for a globally accessible population dataset. By quantifying confidence levels alongside population estimates, the model enables decision-makers to assess data reliability in real time. Daniel Adams, lead author and research scientist at ORNL, emphasized that pairing advanced computational models with transparent uncertainty measurements ensures trustworthy outputs for time-sensitive operations. The dataset is designed to support disaster response, humanitarian assistance, infrastructure development, and national security initiatives. The LandScan program, established in 1999, has long focused on ambient population modeling for emergency and risk assessment applications. Principal Investigator Marie Urban noted that the integration of high-performance computing and artificial intelligence marks a significant evolution in demographic tracking. The ORNL team combined expertise from computer science, geography, and data analytics to develop a system that balances cutting-edge algorithmic processing with operational reliability. Building on this foundation, ORNL is preparing to launch LandScan Mosaic Timeseries, a dataset that will provide annual population distributions from 1975 to 2024. Using a backcasting methodology, the historical archive will enable researchers to analyze urban growth, landscape transformation, and long-term settlement patterns at a previously unattainable temporal resolution. Andrew Zimmer, a geospatial scientist on the project, highlighted that the building-level framework applied to modern data is now being adapted for historical records, allowing comparative analysis across decades. Future iterations of the LandScan ecosystem will expand into demographic profiling. Researchers are developing algorithms that utilize building functions to infer population characteristics, such as age distribution. For instance, school occupancy patterns will be used to map daytime concentrations of younger demographics, while residential and industrial usage will inform broader population segmentation. This advancement aims to deliver dynamic, time-sensitive demographic insights without compromising spatial granularity. LandScan Mosaic is now available to researchers, government agencies, and international practitioners worldwide. The dataset represents a structural shift in how global mobility and urban density are quantified, offering a standardized, trust-driven resource for both immediate operational needs and long-term planning. By merging AI-driven spatial modeling with rigorous uncertainty analysis, ORNL has established a new benchmark for ambient population tracking that bridges real-time decision-making with historical demographic research.