Abstract. Automatic update of 3D city models has become a crucial operation in the context of urban digital twins. It commonly relies on a reconstruction from ALS point clouds. However, frequent reconstructions due to a dynamic urban environment are resource-intensive and lack change information about the scene. In this paper, we present a novel framework which aims to combine an instance change detection approach based on a distance-computing algorithm, geometric features and thresholding with a building reconstruction algorithm to ensure an efficient geometric, semantic and thematic (change labeling) update of an existing 3D city model. This approach comprises three stages spanning from data preparation to the integration of change results in the updated model. First, we prepare our input data. Next, we assess the changes between the two epochs. This process involves two stages. New and lost buildings are extracted in the first, and the changed and unchanged in the other. The results of the change detection are evaluated using standard evaluation metrics. The evaluation results are encouraging considering the various sources of errors. Finally, unchanged buildings are kept in the model, while the changed and new ones are reconstructed using Geoflow3D. The final model is semantically augmented using a change attribute. Since the 3D city model undergoes an update rather than complete reconstruction, the tracking of both geometric and semantic changes of some buildings can be made possible through a versioning system. The change information can be leveraged in multiple applications like 3D cadastre, urban inventory, urban planning…