Classifying urban areas into residential, non-residential and mixed-use proportions using building footprints and geospatial models

Abstract High-resolution building footprints have transformed urban mapping, yet functional use information remains scarce in rapidly urbanising, data-limited settings. This study presents a geospatial framework for estimating the proportions of residential, non-residential, and mixed-use areas, demonstrated in Lagos, Nigeria, with methodological components suitable for adaptation in other data-scarce cities. Using over 180,000 ground-truth building samples and 68 geospatial covariates, we apply Random Forest and Bayesian Hierarchical models to characterise urban function. Both models perform strongly (residential r = 0.85, 0.84; non-residential r = 0.72, 0.69), while the Bayesian model provides enhanced uncertainty quantification. The resulting 1-km² gridded functional surface captures Lagos’s urban structure, including dense residential districts, commercial corridors, and mixed-use transition zones. This study provides a method for producing a semantically enriched representation of urban function in an African megacity, offering a transferable framework for advancing urban analytics, population modelling, and sustainable development planning.