Modeling population density guided by land use-cover change model: a case study of Bogotá

Population densities provide valuable spatial information to identify populations at risk, quantify mobility, and improve our understanding of future urban settlements. Advancements in machine learning algorithms open up new horizons to face these challenges. This research proposes a supervised machine learning approach, Random Forest, for population density appraisal in a large and dense developing city. We studied Bogotá, where functional integration with neighboring municipalities exists, although they have different governments and uncoordinated urban development plans. As a starting point, we use simulated residential land-use patterns, classified according to socioeconomic levels, from a cellular automata-based model. We estimate population density with reliable land-use change models and nine simple representations of the urban structure, such as land values and the distance to urban amenities. Therefore, combining a cellular automata model with a classification model, considering both continuous and categorical variables, demonstrates this methodology’s potential and promises a reliable assessment of population density. Finally, we present a trip generation model integrated with densities and spatial location. A comprehensive results discussion suggests this study’s importance in urban planning and the accuracy of the proposed methodology to support decision-making processes and policy evaluation.