Fine-grained population mapping from coarse census counts and open geodata.

Area of Science:

• Geospatial analysis
• Demography
• Artificial intelligence

Background:

• Accurate fine-grained population maps are crucial for urban planning, environmental monitoring, public health, and humanitarian efforts.
• Existing census data is often aggregated over large areas and may not be up-to-date, limiting its utility for detailed spatial analysis.

Purpose of the Study:

• To introduce POMELO, a deep learning model designed to generate high-resolution population maps.
• To estimate population density at a [Formula: see text]m ground sampling distance using coarse census data and open geodata.
• To enable population estimation even in data-scarce regions lacking census counts.

Main Methods:

• Utilized a deep learning approach (POMELO) integrating coarse census data with open geodata.
• Developed a model capable of disaggregating aggregate census counts into fine-grained population maps.
• Enabled unconstrained population prediction by generalizing across countries, even without available census data.

Main Results:

• POMELO achieved strong agreement with detailed reference population counts in sub-Saharan African countries.
• Disaggregation of coarse census data yielded [Formula: see text] values between 85-89%.
• Unconstrained population prediction, in the absence of any census data, reached [Formula: see text] values of 48-69%.

Conclusions:

• The POMELO model provides a powerful tool for generating accurate, fine-grained population maps from limited or no census data.
• This approach significantly enhances the availability of detailed population data for critical applications in diverse geographical contexts.
• POMELO demonstrates the potential of deep learning to overcome data limitations in population mapping and analysis.