Adaptive weighted least squares (AWLS): A new vector-based model to improve urban population estimation at small-area scale using morphology and attractiveness criteria

Abstract

Fine-resolution urban population mapping is vital for many applications, including urban planning and disaster management. However, these data are rarely available. Despite the well-established correlation between urban population distribution and the physical parameters of residential areas and urban mobility, there needs to be a comprehensive model that effectively utilizes these relationships. This article proposed a novel model for population estimation, adaptive weighted least squares (AWLS), based on the correlations between urban morphology (e.g., physical parameters and shape of residential areas) and attractiveness (e.g., points of interest) and weighing them in the least square regression. This hierarchical model first uses the AWLS method to account for local correlations. Then, it disaggregates the population into lower-level spatial units specifically city blocks and parcels. The efficacy of the method is demonstrated in three neighborhoods in Tehran, Iran, with differences and similarities in terms of morphology and attractiveness. This method significantly improved population estimation accuracy, outperforming common global and local estimation models (11 and 8 in Neighborhood A, 14 and 13 in Neighborhood B, and 5 and 5 in Neighborhood C). This model successfully disaggregated the census tract (CT) population into city block and parcel levels, providing valuable data for crisis management.