In the last two decades there has been a steady rise in the gathering and use of 3D geo-information. A common way to store and use 3D data is by using 3D city models. In 3D city models, geo-information can be stored at different levels of detail. CityGML, the most commonly used data model and encoding for 3D city models, uses five levels of detail in order to model increasing geometric and semantic complexity. These levels of detail may be interpreted as a model quality measure, and as a guideline for users that need data for a certain application. CityGML consists of several thematic modules, each with their own level of detail specification. Some of these modules have a more further developed level of detail specification than others. Recently, several authors have proposed improvements to the Transportation module. This has led to proposals for various changes, especially concerning road data. However, these proposed changes have not been encoded yet. The main critiques are the lack of a level of detail specification for linear representations of roads, no ability to model networks, no representation of intersections and general ambiguity in the level of detail specification. Many road data use cases might potentially benefit from improvements on these points. In this thesis I attempt to improve the current level of detail specification of roads in the CityGML data model. The improvements are encoded in CityJSON, a JSON encoding of this data model. I assess the shortcomings in the current CityGML transportation module. After, a road data needs analysis is performed on three use cases: transport modelling, navigation and road maintenance. The data needs are compared to modelling approaches of other road data standards. This has resulted in several encoded improvements. A topological structure has been added to CityJSON. This includes the addition of two new modelling classes: Nodes and Edges. This structure is general such that it can be used by other thematic modules as well. Moreover, the level of detail specification for roads has been further developed to include both the linear representations and less ambiguous areal representations. This includes a prescription on how to model intersections and roundabouts at different levels of detail. Finally, I provide a structure which enables one to link the linear and areal road data together. This link is made at the scale level of the object, which data providers may choose themselves. This way one object can be modelled in two representation types without needing a one-to-one mapping between linear road segments and areal triangulated surfaces. These concepts are then tested by creating a CityJSON road data file for all new levels of detail.

There are several 3D city models available openly, worldwide. These models are used in various applications, from which many expects a homogeneous Level of Detail (LoD). Validating the accuracy of the LoD of a model requires the inference its LoD class and its conformance to the real-world object. This process quickly becomes infeasible for large models when done manually. Yet there is no automatic method for LoD inference and validation. Therefore the thesis proposes a method to automatically infer the geometric LoD (LoD0-2.3) in 3D city models.

A central aspect of this work is the use of machine-learning to classify building models based on their LoD. It follows the assumption that a process is possible where a classifier trained in a synthetic 3D city model containing all LoD classes, and applied in real city model. Therefore ten geometry measures (features) are computed from the objects and tested with six classification algorithms. The six experiments the transferability of a classifier from the synthetic city model to the real one, multi-class (LoD0-2.3) and binary (LoD2 or not) classification, and the effect of LoD class imbalance by introducing various amounts of LoD1 objects into the LoD2 model. Furthermore, by using a point cloud as ground truth, this explored the possibility of validating the inferred LoD classes.

The results indicate that the classifier is not transferable to the real data set when trained on the synthetic city model, which is probably due to the significant difference in object shapes between the two models. Binary classification outperforms the multi-class case and it is favourable for LoD validation where the main question is whether the model conforms the stated LoD or not. Finally, class-imbalance can reduce the classification with as much as 20%.

3D city models are now common planning and analysis tools. Urban vegetation as a feature in these models, however, is neglected overshadowed by the focus on buildings, so its inclusion in 3D city models is often symbolic. On the other hand, urban vegetation improves the comfort and social well being of a city’s inhabitants and is a resource for sustainable urban growth and an environmentally friendly resource for mitigating the negative effects of climate change, e.g., frequent heat waves, floods from storm downpours and extended dry periods. Trees also mitigate Urban Heat Island (UHI) effects. Urban vegetation’s ecosystem services (ecoservices) as mitigation functions of pollution and negative effects of climate change have propelled research, studies, applications, and simulations that need data and 3D models of existing vegetation e.g., for spatial simulations, to assess the canopy cooling impact to surroundings or to identify areas prone to UHI. In view of these needs, urban vegetation in 3D city models is underrepresented. Guidelines for modelling vegetation are already provided in CityGML, the 3D city modelling standard, but they are insufficient for today’s needs because it has vagueness and focuses mostly on built infrastructures.

In this research, 14 Single Vegetation Object (SVO) Levels of Details (LOD) and four root LODs are proposed. They target to meet different adherence requirements and scales. Their formulation is based on LOD specification approaches, and on a needs analysis that identified the vegetation models and data most commonly required in applications in the urban environment. Vegetation LOD description approaches include semantic 3D modelling standards, industry, and common practices of municipality users which are also GIS data providers.

Current vegetation LOD descriptions approaches fall into two different groups based on the geometry they adopted for their specifications: implicit or explicit, and no one approach fulfills identified needs. Acquisition techniques and demand in IT resources have influenced the definitions of vegetation LODs, the adoption of one geometry type or the other, and the wide use (or not) of certain LODs.

Refined SVO LODs specifications of this research combine the strengths of each group with descriptions that cover beyond geometric specifications. Refined LODs incorporate implicit components, underground representations and reconstruction LODs not defined by any approach. With them, most datasets can be represented by at least one LOD, and modelers can tell what LOD is possible to implement based on the data they already have. For acquisition, it is possible to tell what data is required for a particular LOD, and which LOD can be used to obtain data needed for an application. The broad spectrum of refined LOD allows them to meet different requirements.

A shadow analysis case study was done with implementations of refined SVO LOD specifications. Acquisition from aerial and mobile LiDAR data was done in a workflow that brought the 0D tree inventory of the municipality of Rotterdam to 3D models using mainstream and open source tools. The case study confirmed a quantitative impact in shadow duration and extent by each LOD indicating that each is independently differentiated. Volumetric and non-volumetric models had different shadow over and underestimation impacts. The study further highlighted the properties of the crown of the real-world object that help in choosing a LOD and gave insights offered by lower LOD models.

The implementation of the assorted LODs revealed that while much research has been done in acquiring vegetation parameters from LiDAR data, the many options, methods and algorithms are scattered necessitating a unifying process or tool.