A hybrid remote sensing-based framework for high-resolution population mapping and reconstruction in Ecuador (2000–2024)

Global population distribution grids are characterised by low accuracy in fragmented urban areas and in regions dominated by extensive rural landscapes, a common situation in Ecuador and other Latin American countries. As a result, understanding population dynamics, exposure to natural hazards, and landscape change at the urban–rural interface remains limited. To address this gap, a hybrid framework is proposed to generate fine-resolution population grids at 250 m resolution by integrating deep learning (DL) methods, multi-source remote sensing (RS) products, and census data. The framework achieves high predictive performance, with R² ranging from 0.88 to 0.99 across spatial windows and time periods. Results indicate that Ecuador’s population continues to shift from dense urban cores towards peripheral areas, leading to the expansion of large polycentric urban agglomerations. Savannas and shrublands are increasingly populated, reflecting a maturing phase of urbanisation. At the same time, the influence of densification and distance to road network on land-cover (LC) change is diminishing. Finally, using the reconstructed population grids, the evolution of natural hazard (NH) relative risk associated with flooding, extreme rainfall, earthquakes, and volcanic hazards is evaluated, revealing expanding risk along polycentric corridors.