Abstract <p>Weather and climate modeling for urban areas requires a detailed description of the built environment. Moreover, the rapid transformation of urban environments requires regular updating of city-descriptive parameters. Despite the increasing number of urban climate models on different scales, there is still a lack of global and consistent datasets that characterize the relevant properties of urban areas, including the morphological properties of built environments. Today, the most reliable estimates of building morphological parameters required for weather and climate models are obtained based on aggregation of the vector datasets that represent the geometry of individual buildings, roads, etc. However, such datasets often lack information about building heights for a large fraction of buildings. Therefore, it is necessary to integrate available datasets and improve their completeness by adding necessary attributes and filling in missing data. This research aims to retrieve missing data on building heights using various sources and machine learning techniques on the example of St. Petersburg city. We used the novel Overture Maps buildings dataset, and derived a set of building height predictors based on these data, including: building function, morphological properties, characteristics of building surroundings. We also obtained the independent estimate of the building height derived from the high-resolution ArcticDEM digital surface model. Gradient boosting on decision trees, specifically CatBoost, was employed as the machine learning model.</p>