Abstract
Traditional gridded population datasets often focus on residential locations—where people sleep. However, addressing global human security challenges like disaster response and infrastructure planning requires an "ambient" population perspective: an estimate of average human presence over a 24-hour period. LandScan High Definition (LSHD) provides this at a high spatial resolution of 3 arcseconds (~90m). By combining top-down administrative data with bottom-up building-level modeling, LSHD offers a scalable workflow for mapping human activity worldwide.
Introduction: The Need for Ambient Data
Most population maps are static, based on census data that ties individuals to their homes. While effective for long-term planning, these residential models fail during "unwarned" events such as flash floods or industrial accidents occurring during the workday. In such scenarios, a business district that appears empty on a residential map may actually contain thousands of workers and students. LSHD addresses this gap by modeling the movement and presence of people across residential, commercial, and industrial spaces.
The LSHD Methodology
The core of the LSHD approach is a hybrid top-down and bottom-up redistribution model. The process begins with authoritative subnational population totals (top-down) which are then allocated to specific built-up areas based on high-resolution occupancy estimates (bottom-up).
To manage this at scale, researchers utilize a vector unit called a MOLECULE (Model Leverageable Cut-up Labeled Entity). These features are created by intersecting building footprints with administrative boundaries and land-use data. Each MOLECULE is assigned an occupancy weight derived from Oak Ridge National Laboratory’s (ORNL) Population Density Tables (PDT), which use Bayesian machine learning to estimate how many people occupy different facility types (e.g., schools vs. factories) per unit of floor area.
Data Integration and Scalability
LSHD is a "data-hungry" framework that fuses disparate global datasets. Because data quality varies by region, the model employs a hierarchical preference system for building footprints and land-use attributes.
The workflow prioritizes high-fidelity 2D footprints (from ORNL, Microsoft, or Google) and 3D building morphology. Where these are unavailable, "failsafe" layers like the Global Human Settlement Layer (GHSL) are used to ensure no built-up areas are missed. This entire stack is organized into a 1° resolution tileset, enabling a repeatable and efficient global workflow.
Validation and Real-World Application
The utility of LSHD was demonstrated through a case study in the Philippines. By comparing LSHD with residential-only models (such as WorldPop) in Iloilo City, researchers found that LSHD identified significantly higher population counts in moderate-to-high flood risk zones. This discrepancy occurs because commercial and academic hubs—often located in vulnerable areas—are largely invisible to residential models.
Furthermore, the model was validated against subnational CO2 emissions data. Ambient population estimates showed a significantly stronger correlation with emissions ($\rho=0.761$) than residential estimates ($\rho=0.682$), confirming that LSHD accurately captures the footprint of human economic and social activity.
Limitations and Future Outlook
While LSHD represents a major leap in global population modeling, some limitations remain. In regions where building height data is missing, the model currently defaults to a two-floor assumption. Additionally, many buildings still lack specific land-use tags, defaulting to "residential" by necessity.
Future work focuses on "MapSpace," a machine learning tool designed to predict building use and height based on morphology. The team also aims to automate the detection of informal settlements and incorporate more granular demographic characteristics, such as age and income, into the global ambient grid.
Summary updated