In large, rapidly growing urban environments, mapping populations at small spatial scales is often a significant challenge. The highly heterogeneous nature of urban land use means that population density can vary by orders of magnitude between adjacent areas and change dynamically over very short timescales. In many resource-poor settings, existing demographic data from national censuses may be outdated or incomplete, causing services to be distributed inequitably and people to be left out of critical decision-making processes. 

To address these data gaps, we are testing the ability of AI-derived geospatial datasets and hybrid expert-informed GeoAI modelling frameworks to construct reliable small area urban population estimates. This project explores whether novel AI-derived geospatial datasets canto enhance the accuracy of traditional spatial demographic modelling. By fusing multimodal geospatial data with advanced machine learning and automated machine learning (AutoML), the framework aims to reveal human presence and movement at unprecedented scales while remaining transparent and interpretable for users. 

The project is being conducted in close collaboration with local researchers and national statistical offices across several countries, including Bangladesh and Thailand. This participatory approach the resulting tools and datasets to be tailored to local policy needs, such as urban health analytics, disease surveillance, and census planning. 

Through this work, we seek to provide an empirical evidence base for the utility of AI-derived data and to release open-access, dynamic population data products. These outputs will support more equitable and data-driven decision-making, ensuring that marginalized communities can be more accurately represented. 

Image: Derived from Bangkok, Thailand, Marcelo  Benzuca (Pexels), 2024.