Intercensal building‑based population estimation

Urban planning faces challenges in monitoring population distribution amidst rapid urbanisation and climate change, underscoring the need for accurate small-scale population data. Despite this, census surveys are limited in their ability to provide timely and granular information, highlighting the need for innovative solutions. This paper proposes a method for intercensal population estimation, leveraging continuously available data sources to fill this gap between national census surveys. Our approach involves dasymetric mapping of historical census data to building footprints, enhanced with a range of relevant features, to train an extreme gradient boosting model and apply it to a contemporary building dataset to estimate the population. Germany serves as the case study in a data-rich scenario, offering access to numerous open data sources. We evaluate the estimated population on a 100 m grid level using official census data and examine errors statistically and spatially by applying a scale-robust metric called inverse hyperbolic sine error (IHSE). Our analysis reveals that the estimated population data meets or exceeds the acceptable quality threshold in 88% of the cases, with a mean relative error of less than 8%. While the model tends to underestimate in densely populated regions, the results demonstrate the potential for various practical applications.