A Novel Framework for Mapping Updated Fine-resolution Populations with Remote Sensing and Mobile Phone Data

This paper presents a new framework for producing monthly population maps at the census block level, which are crucial for population-related research and emergency response. Existing population products are outdated (e.g., decennial) and at coarse spatial resolution (e.g., national and global), as they rely on data that is collected and processed with a long lag time. The proposed framework is based on a comprehensive comparison of 34 models that use different methods (housing units, ordinary least squares, and machine learning), variables (social-economic, building, and vegetation), and classifications (7 and 2 classes). We employed the remote sensing Orthoimage, GIS tax parcel data, and SafeGraph home panel data to acquire the necessary variables that can reflect the spatial-temporal dynamics of the census block level populations. The best-performing model uses ordinary least squares with 3 kinds of information: the number of mobile phones, building area, and 7 class classifications (Single family, Two family, Three family, Mix family, Mix commercial family, Apartment, and Non-residential house). The model has a high accuracy (R² = 0.82) and can capture the monthly variations of population at the census block level. The framework is easy to implement and replicate by stakeholders, as it uses intuitive methods and readily available datasets. It can also reveal the detailed population patterns of cities over time, which can inform urban planning decisions.