LandScan mosaic enables high-resolution gridded population estimates with explicit uncertainty

Area of Science:

• Geographic Information Systems (GIS)
• Machine Learning
• Population Dynamics

Background:

• Gridded population datasets are crucial for decision-making in environmental risk assessment, urban planning, and disaster preparedness.
• Traditional methods often lack uncertainty quantification, leading to potentially flawed decisions.
• Accurate population estimates are vital for effective policy and resource allocation.

Purpose of the Study:

• To introduce a probabilistic machine learning framework, LandScan Mosaic, that explicitly incorporates uncertainty into population modeling.
• To address the methodological gap of overlooking uncertainty in gridded population estimation.
• To provide a quantitative approach for incorporating confidence into structured decision-making processes.

Main Methods:

• Developed a probabilistic machine learning framework, LandScan Mosaic.
• Quantified uncertainty in building use types, floor counts, and occupancy rates using Monte Carlo simulations.
• Applied the framework to Iloilo City, Philippines, for flood risk assessment.

Main Results:

• Generated probability distributions of population counts instead of deterministic estimates.
• Demonstrated the framework's application in prioritizing areas affected by projected flooding.
• Showcased how probabilistic estimates support targeted interventions for economic and social risks.

Conclusions:

• The LandScan Mosaic framework advances population distribution modeling by explicitly accounting for data uncertainty.
• Probabilistic estimates enhance structured decision-making, particularly in disaster preparedness and risk assessment.
• Comparative analysis showed improved decision rankings when incorporating machine learning and uncertainty.