ABSTRACT

As urban populations rapidly rise worldwide, it is increasingly necessary to determine the accurate distribution and configuration of the population in order to appropriate resources and services. Census-based methods for obtaining population counts are time consuming, labour intensive, and costly. Researchers have turned to remote sensing to estimate population from aerial and satellite datasets including lidar, which allows measures of building volume to be incorporated into population estimates. However, studies using lidar-derived building volumes have noted inconsistencies between population and building volume estimates in certain areas. In this article, we investigate this issue by incorporating both static and ambient population data into models using the US Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) database. To do this, we first develop a normalized home–work index to classify census blocks as primarily work-oriented, home-oriented, or mixed-use based on the LEHD data. We then employ ordinary least squares and geographically weighted regression (GWR) to explore the relationships between the different population groups (work, home, and mixed) and lidar-derived building volumes. We test these relationships across four diverse cities in Texas: Austin, Dallas, Houston, and San Antonio. Results suggest non-stationarity in the relationship between building volume and population with stronger, positive relationships in home-oriented and mixed-use blocks where the amount of building volume per person may be more consistent compared to work-oriented blocks. GWR models yielded high R² values (0.9), particularly in mixed-use areas, indicating the potential for predictive relationships.