Publicado en: International Journal of Geographical Information ScienceVolume 32, 2018 – Issue 2, pp. 282-301 (ya publicado online el 26 Oct 2017)https://doi.org/10.1080/13658816.2017.1393543La revista está 46/146 en COMPUTER SCIENCE, INFORMATION SYSTEMS en el JCR (Q2)Abstract:Georeferenced 3D models are an increasingly common choice to store and display urban data in many application areas. CityGML is an open and standardized data model, and exchange format that provides common semantics for 3D city entities and their relations and one of the most common options for this kind of information.Currently, creating and maintaining CityGML models is costly and difficult. This is in part because both the creation of the geometries and the semantic annotation can be complex processes that require at least some manual work. In fact, many publicly available CityGML models have errors. This paper proposes a method to facilitate the regular maintenance of correct city models in CityGML. This method is based on the continuous deployment strategy and tools used in software development, but adapted to the problem of creating,maintaining and deploying CityGML models, even when several people are working on them at the same time. The method requires designing and implementing CityGML deployment pipelines. These pipelines are automatic implementations of the process of building, testing and deploying CityGML models. These pipelines must berun by the maintainers of the models when they make changes that are intended to be shared with others. The pipelines execute increasingly complex automatic tests in order to detect errors as soon as possible, and can even automate the deployment step, where the CityGML models are made available to their end users. In order to demonstrate the feasibility of this method, and as an example of its application, a CityGML deployment pipeline has been developed for an example scenario where three actors maintain the same city model. This scenario is representative of the kind of problems that this method intends to solve, and it is based on real work in progress. The main bene fits of this method are the automation of model testing, every change to the model is tested in a repeatable way; the automation of the model deployment,every change to the model can reach its end users as fast as possible; the systematic approach to integrating changes made by different people working together on the models, including the possibility of keeping parallel versions with a common core; anautomatic record of every change made to the models (who did what and when) and the possibility of undoing some of those changes at any time.