Interpretable deep learning for consistent large-scale urban population estimation using Earth observation data

Sugandha Doda, Matthias Kahl, Kim Ouan, Ivica Obadic, Yuanyuan Wang, Hannes Taubenböck, Xiao Xiang Zhu

Abstract: Highlights•Proposed population estimation approach combining deep learning and explainable AI.•Using only open geo data, model could predict population, even without census data.•The new population maps show higher accuracy than GHS-POP data in European cities.•Explainable AI indicates land use data as most important indicator of population.•Proposed method assessed in a disparate continent which is missing in most studies.