Oak Ridge National Laboratory has launched LandScan Mosaic, a new population distribution dataset that uses AI to model building occupancy and movement patterns throughout the day.

Researchers at the Department of Energy's Oak Ridge National Laboratory (ORNL) have released LandScan Mosaic, a global population distribution dataset. The tool uses machine learning to estimate building characteristics and occupancy, providing high-resolution data on where people are located throughout a 24-hour period. This development aims to improve disaster response, humanitarian assistance, infrastructure planning, and national security missions by offering more precise location and temporal data.

Building-Centric Modeling Framework

Unlike traditional methods that rely primarily on satellite imagery pixels, LandScan Mosaic utilizes a building-centric approach. Researchers employed machine learning to estimate missing building attributes, such as height, floor count, and function, even in regions where detailed data is otherwise sparse. By integrating land-use information and standardized occupancy distributions, the model captures daily population movement between homes, workplaces, schools, and other activity spaces. Additionally, the dataset provides explicit measures of uncertainty, allowing users to assess the confidence level of specific population estimates. According to lead author Daniel Adams, this transparency is intended to assist decision-makers who must act before perfect information is available.

Historical Data and Future Development

The ORNL team is also developing LandScan Mosaic Timeseries (LSM-TS), which applies a 'backcasting' method to provide annual population distribution data spanning 50 years, from 1975 to 2024. Geospatial scientist Andrew Zimmer noted that this long-term look-back allows researchers to reconstruct historical populations and observe the growth of cities at a building-level resolution. Future enhancements for the LandScan program include the addition of demographic characteristics. Researchers plan to use building functions—such as schools—as a 'demographic fingerprint' to better estimate the distribution of specific age groups during daytime and nighttime hours.