Framework integrates mobile and remote sensing data for population maps

The research team from the Department of Geography at SUNY Buffalo developed an innovative framework that uses a combination of 34 models to map monthly population distributions at fine resolutions. By integrating mobile phone data, building area, and detailed residential classifications, they created highly accurate population maps.

The most successful model used ordinary least squares (OLS) regression, which incorporated mobile phone location data and a seven-class classification of buildings, such as single-family homes and mixed-use residential buildings. The model demonstrated high accuracy (R² = 0.82) and captured monthly population variations effectively.

This approach offers a practical and replicable method for urban planners and researchers to track population dynamics in detail. The study was published on August 23, 2024, in the Journal of Remote Sensing.

The framework leverages remote sensing orthoimage, GIS tax parcel data, and SafeGraph home panel data. Remote sensing data sources like LiDAR and Landsat 8 were used to enhance spatial detail by mapping building areas and vegetation cover. Through comparing different models, the research identified building area as a key variable in population distribution. Machine learning models were also tested to further improve accuracy in predicting population trends.

“This framework provides a novel solution to tracking urban population dynamics. By integrating mobile data with remote sensing, we can now create monthly population maps that are more accurate and timely, which is crucial for urban planning and disaster management,” said Le Wang, co-author of the study and professor at SUNY Buffalo’s Department of Geography.

The research employed a two-step hybrid method. First, mobile phone data were combined with population-related variables to update population estimates at the census block group (CBG) level. Then, a weighted layer was created using statistical models and machine learning techniques, refining the population data down to the census block (CB) level. Model validation used random sampling and showed high accuracy, with an R² value of 0.82.

This hybrid approach combining remote sensing and mobile phone data can be applied to track population changes in various cities. Future applications could extend the model to larger regions and integrate additional dynamic data sources, such as real-time traffic or public services data, to further improve prediction accuracy and scalability. This could be a valuable tool for city management, emergency response, and policy-making, providing more detailed and up-to-date population insights.

Provided by
TranSpread