Computers, Environment and Urban Systems 59 (2016) 2537

Contents lists available at ScienceDirect

Computers, Environment and Urban Systems

journal homepage: www.elsevier.com/locate/ceus

An improved LOD specication for 3D building models

Filip Biljecki , Hugo Ledoux, Jantien Stoter
3D Geoinformation Research Group, Delft University of Technology, Delft, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: The level of detail (LOD) concept of the OGC standard CityGML 2.0 is intended to differentiate multi-scale repre-
Received 19 December 2014 sentations of semantic 3D city models. The concept is in practice principally used to indicate the geometric detail
Received in revised form 16 February 2016 of a model, primarily of buildings. Despite the popularity and the general acceptance of this categorisation, we
Accepted 14 April 2016
argue in this paper that from a geometric point of view the ve LODs are insufcient and that their specication
Available online xxxx
is ambiguous.
Keywords:
We solve these shortcomings with a better denition of LODs and their renement. Hereby we present a rened
Level of detail set of 16 LODs focused on the grade of the exterior geometry of buildings, which provide a stricter specication
3D city modelling and allow less modelling freedom. This series is a result of an exhaustive research into currently available 3D city
3D GIS models, production workows, and capabilities of acquisition techniques. Our specication also includes two hy-
3D building models brid models that reect common acquisition practices. The new LODs are in line with the LODs of CityGML 2.0,
CityGML and are intended to supplement, rather than replace the geometric part of the current specication. While in
Scale our paper we focus on the geometric aspect of the models, our specication is compatible with different levels
of semantic granularity. Furthermore, the improved LODs can be considered format-agnostic.
Among other benets, the rened specication could be useful for companies for a better denition of their
product portfolios, and for researchers to specify data requirements when presenting use cases of 3D city models.
We support our rened LODs with experiments, proving their uniqueness by showing that each yields a different
result in a 3D spatial operation.
2016 Elsevier Ltd. All rights reserved.

1. Introduction This taxonomy has been developed in the German Special Interest
Group 3D (SIG 3D) initiative (Albert, Bachmann, & Hellmeier, 2003),
The level of detail (LOD) of a 3D city model is one of its most and has been further described in Grger and Plmer (2012). The ve
important characteristics. It denotes the adherence of the model to its LODs have become widely adopted by the stakeholders in the 3D GIS
real-world counterpart, and it has implications on its usability industry and they now also describe the grade and the design quality
(Biljecki, Ledoux, Stoter, & Zhao, 2014b). of a 3D city model, especially its geometric aspect (i.e. how much detail
The CityGML 2.0 standard from the Open Geospatial Consortium should be acquired?). They have gained importance also in the
(2012) denes ve LODs. The concept is intended for several thematic computer graphics (Verdie, Lafarge, & Alliez, 2015; Musialski et al.,
classes of objects but it is primarily focused on buildings, and the ve 2013), and BIM (Tolmer, Castaing, Diab, & Morand, 2013) communities
described instances increase in their geometric and semantic complex- when dealing with 3D building models.
ity (Fig. 1). LOD0 is a representation of footprints and optionally roof The LOD concept of CityGML is primarily intended to differentiate
edge polygons marking the transition from 2D to 3D GIS. LOD1 is a the grade of data resulting from different production workows, and
coarse prismatic model usually obtained by extruding an LOD0 model. they are driven by semantics as much as geometry. In the industry
LOD2 is a model with a simplied roof shape, and where the object's and research community they were accepted from the outlook on
parts can be modelled in multiple semantic classes (e.g. roof, wall). geometric richness, which was partly caused by the lack of applications
LOD3 is an architecturally detailed model with windows and doors, that require semantics. For instance, we have observed that while the
being considerably more complex than its preceding counterpart. LOD2 from the point of view of CityGML developers represents a
LOD4 completes an LOD3 by including indoor features (Kolbe, 2009). model with differentiated semantic surfaces, practitioners primarily
refer to models with roof shapes, even when not dealing with data
that is semantically structured.
Corresponding author at: 3D Geoinformation Research Group, Department of While the ve LODs generally provide a categorisation of the overall
Urbanism, Faculty of Architecture and the Built Environment, Delft University of
Technology, Julianalaan 134, 2628 BL Delft, The Netherlands.
level of abstraction, content, value, and usability of 3D city models, this
E-mail addresses: [email protected] (F. Biljecki), [email protected] (H. Ledoux), classication has several drawbacks and shortcomings as we show in
[email protected] (J. Stoter). Section 2. Since the specication is crucial among practitioners and

http://dx.doi.org/10.1016/j.compenvurbsys.2016.04.005
0198-9715/ 2016 Elsevier Ltd. All rights reserved.
26 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

Fig. 1. The ve LODs of CityGML 2.0. The geometric detail and the semantic complexity increase, ending with the LOD4 containing indoor features.

researchers for conveying the grade of a 3D city model and its adher- Bidarra, & Benes, 2014), conversion from BIM and computer graphics
ence to the real-world, in this paper we present a rened specication models (Donkers, Ledoux, Zhao, & Stoter, in press; Kumar, Saran, &
to solve such problems. It should be noticed that the topic of rening Kumar, in press), and generation from 2D drawings (Gimenez,
and improving the current specication of the LODs is currently under Hippolyte, Robert, Suard, & Zreik, 2015)); (2) the number of data pro-
consideration in the CityGML community for version 3.0 (Machl, ducers and national mapping agencies requesting 3D data is increasing
2013; Lwner & Grger, 2016), and we hope that our proposal will (Stoter et al., 2015), and without a ner specication data producers
help the discussions. However, our work is intended to be independent and users may resort to creating their own specications (e.g. see the
of any particular 3D format, and applicable to any format that can be series from Blom, 2011), which might increase the ambiguity; (3) the
used to store 3D building models, including ones such as COLLADA increase in quantity of data sets with non-homogenous LODs (Fan,
and OBJ. Zipf, Fu, & Neis, 2014; Touya & Reimer, 2015; Arroyo Ohori, Ledoux,
In Fig. 2 we give an example of the shortcomings of the current Biljecki, & Stoter, 2015a); and (4) use cases have different requirements
concept, from the point of view of the geometric detail. The gure when it comes to the complexity and quality of the data. Furthermore,
illustrates two LOD2 models: the model on the left has been acquired the number of 3D use cases is rapidly increasing (Biljecki, Stoter,
with two acquisition techniques, the walls are at their actual location Ledoux, Zlatanova, & ltekin, 2015b), for instance solar potential
and the roof overhangs are explicitly present. The representation in estimation (Freitas, Catita, Redweik, & Brito, 2015), studying the
the middle has been acquired with one technique (aerial photogram- thermal characteristics of the outdoor space (Maragkogiannis,
metry) where the walls are derived as projections from the roof outline Kolokotsa, Maravelakis, & Konstantaras, 2014), reghting simulations
(the third model will be introduced in another example in the following (Chen, Wu, Shen, & Chou, 2014), and advances in multi-scale navigation
section). This example illustrates how the CityGML LOD concept is (Hildebrandt & Timm, 2014). Each of these use cases may have different
ambiguous and that it falls short in dening the complexity of the requirements when it comes to the LOD of the models.
models: the two models are of the same LOD (LOD2) according to In this paper we improve the geometric aspect of the LOD specica-
CityGML while the rst one is more laborious to acquire and it may tion of 3D building models. We provide an extended and more informa-
bring better results in a spatial analysis (e.g. more accurate volume; tive series of 16 LODs that are compatible with the existing CityGML
see Biljecki, Ledoux, Stoter, & Vosselman, 2016). Hence, practitioners LODs. The rened taxonomy is a result of a research into currently avail-
would not consider them to be of equal value and usability. For able 3D city models and an investigation of the acquisition workows.
these reasons we argue in this paper that they should be consid- We review related work on this topic (Section 3), and for each LOD
ered as different LODs, and our specication differentiates such we give requirements and show an example (Section 4).
cases. We have generated a sample data set in 16 LODs and run them
This ambiguity is most evident in the production of the models. For through a few GIS operations to show that each LOD is unique from a
instance, in 3D generalisation where researchers produce multiple geometric point of view and may bring different results in a spatial anal-
geometric variants of LODs and discuss the ambiguity, among others ysis (Section 5).
see Guercke, Gtzelmann, Brenner, and Sester (2011), Fan and Meng In this paper we focus on the exterior of buildings (i.e. their
(2012), Stoter et al. (2011), Noskov and Doytsher (2014), and Deng et exterior shell in LOD03). The renement of the indoor and seman-
al. (2016). tics aspect of the specication can be considered as orthogonal topics
Solving the ambiguity is also important considering: (1) the increas- to this one. These topics are being tackled by other researchers who
ing number of acquisition techniques (e.g. the recently investigated decompose it into different levels of abstraction and integrate them
being drones (Nex & Remondino, 2013), radar (Zhu & Shahzad, 2014), into expanded LOD1, LOD2 and LOD3 models (for examples see the
handheld devices (Rosser, Morley, & Smith, 2015; Sirmacek & work of Boeters, Arroyo Ohori, Biljecki, and Zlatanova (2015) and
Lindenbergh, 2014), procedural modelling (Wonka, Wimmer, Lwner, Benner, Grger, and Hfele (2013)). While the semantic
Ribarsky, & Sillion, 2003; Mller, Wonka, Haegler, Ulmer, & van Gool, LOD and indoor LOD are out of scope of our paper, present work on
2006; Kelly & Wonka, 2011; Mller Arisona, Zhong, Huang, & Qin, these topics is compatible with our work because such specication
2013; Tsiliakou, Labropoulos, & Dimopoulou, 2014; Smelik, Tutenel, can be supplemented to ours. For instance, each of the newly rened

Fig. 2. Two variants of LOD2 and an LOD1 model exposing the shortcomings of the CityGML LOD concept, and why the computer graphics principles cannot be fully applied to GIS and 3D
city modelling.
 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 27

LODs can be assigned a semantic LOD depending on the achieved for the implementation. For instance, it is not stated whether dormers
spatio-semantic coherence. and other larger roof details should be acquired in LOD2. This may
lead to misunderstandings between stakeholders, and errors in the
2. Shortcomings of the current concept and difculties with utilisation of the models. For example, in solar potential estimations,
designing a specication which are most frequently carried out with LOD2 models (Biljecki,
Heuvelink, Ledoux, & Stoter, 2015a), it is important to have roof
The LOD concept used in CityGML 2.0 has been borrowed from superstructures since they cast shadows and they may reduce the area
computer graphics in which multiple representations of polygon available for the installation of photovoltaic panels. Hence there may
meshes are differentiated by their number of faces, and their simplica- be substantial differences between analyses carried out with LOD2
tion is performed by algorithms that reduce the number of faces while models with and without roof superstructures. For this reason there is
attempting to retain visual delity (Luebke et al., 2003; Clark, 1976). a need to differentiate between variants of LODs, and it is therefore
While the early purpose of the LOD in 3D GIS was to improve visualisa- important to provide a more expressive specication which diminishes
tion performance (e.g. see Coors & Flick, 1998), visualisation is now only errors caused by an ambiguous LOD specication.
one of the applications of 3D city models (Biljecki et al., 2015b). CityGML provides several conformance rules to test the validity of
Nowadays, LOD also implies the usability of the model and its adherence CityGML data, and there are other efforts such as Grger and Coors
to the real-world feature. When simplifying the 3D geometries (2011), Wagner et al. (2012), and Coors and Wagner (2015) to provide
(generalising), different approaches are used, since the goal is to also extended modelling guides and rules. However, these do not cover the
retain various geometric and semantic aspects that are not relevant geometric detail of the models. This drawback results in many valid
for visualisation, and to preserve the structure of the simplied building variants to be considered of the same LOD.
(Xie et al., 2012). However, it should be noted that in computer graphics The size (e.g. length or footprint) of real-world features and their
the rationale implies choosing the optimal LOD among multiple LODs, parts (e.g. a balcony of a building) that have to be acquired is designated
while in GIS practitioners almost always deal with a single model at a as one of the main differentiators of the LODs. However, this cannot be
certain LOD. Therefore, for the large part the LOD concept is used prior used as the general guideline to further specify LODs. For example, if an
to acquisition of a 3D model (e.g. to detail the procurement of data, LOD2 requires that certain building parts bigger than a threshold should
delivered in a single LOD), while in computer graphics they are on- be acquired, this cannot be applied to windows, because they are not
the-y derived from a ner model. intended to be acquired in LOD2, irrespective of their size. A second
Despite the historic relation, the design of an LOD specication in GIS example are overhangs (such as in Fig. 2). They may be required by a
is hampered by the fact that the LOD concept in 3D GIS is inherently stakeholder. However, in size they are smaller than other features
different from the one in computer graphics (Biljecki et al., 2014b). which may not be required at all (e.g. dormers), hence each group of
We clarify this with two further arguments related to Fig. 2. First, this related features should be treated separately. Finally, nowadays a
example exposes that the two models on the left have the same polygon signicant amount of models are constructed with a combination of
count, the foremost metric in computer graphics to distinguish two different data sources. The LOD concept does not consider the LOD of
representations. Second, besides the aforementioned LOD2 models, combined data, where some parts of buildings may be acquired in a
the model on the right is a geometry extruded from a ne footprint of ner or coarser detail than other parts.
the same building. This LOD1 model has a higher face count than the These shortcomings could be solved together by providing a general
LOD2 models. While the number of the primitives generally gives a list of features that should be acquired and the minimum size for each.
good impression about the geometric complexity of a 3D city model, it However, CityGML does not provide such. We provide these in our
cannot be considered as an unambiguous differentiator as it is the case specication described in Section 4.
in computer graphics (the only exception to this in GIS and in 3D city
modelling is terrain because of its usually triangular representation: 3. Related work
lower LOD means less triangles (De Berg & Dobrindt, 1998; Surez,
Trujillo, Santana, de la Calle, & Gmez-Deck, 2015)). The general LOD notion was examined in our earlier work (Biljecki
Consequently, unlike in computer graphics, the LODs in 3D GIS et al., 2014b). The concept is decomposed into six metrics: list of
cannot be ordered: the LOD1 model, intrinsically considered inferior features, their geometric complexity, dimensionality, appearance,
to an LOD2, may be accounted as more valuable than an LOD2 for spatio-semantic coherence, and attributes. We take into account the
certain scenarios when a ner footprint is more useful than the acquired rst three metrics when dening the geometric aspect of the LOD.
roof shape. An example of such use case is the computation of the net Stoter et al. (2011) recognise that CityGML lacks precise LOD deni-
internal area of a building, useful for energy estimations, real estate tions and allows ambiguity, and in a later research, Stoter, Vosselman,
valuation, and population counts (Kaden & Kolbe, 2014; Nouvel et al., Dahmen, Oude Elberink, and Ledoux (2014) argue that the specication
2015; Boeters et al., 2015; Lwin & Murayama, 2009). Hence it does not should be further dened by practitioners, depending on the intended
strictly hold that LOD(i + 1) N LOD i, i.e. the LODs are rather nominal, application of the 3D city model to be acquired. We agree with this
instead their ordinality rather depends on the use case and other reasoning, and think that our approach may help practitioners to do
aspects (Biljecki et al., 2014b). so in a standardised and justied way, while still leaving a signicant
Focusing on the CityGML 2.0 LOD concept, it might come to a degree of freedom to accommodate specic requirements of use cases.
surprise that this ubiquitous standard does not provide a strict specica- Due to the ambiguity and the differences of models that CityGML
tion for the ve LODs. It gives short narrative descriptions, with a table considers to be of the same LOD, He, Besuievsky, Tourre, Patow, and
(see Table 3 in Open Geospatial Consortium (2012)) that is considered Moreau (2012) and He, Moreau, and Martin (2013) refer to the CityGML
as a recommendation, and not a requirement. The description actually LODs as LOD groups, and further dene inter-level LODs within the LOD1
species the upper limit of each LOD, and not the minimal restriction group that vary in their geometric complexity. Besuievsky, Barroso,
for each, i.e. it restricts what can be a part of each representation. For Beckers, and Patow (2014)) have a similar approach for LOD3 buildings
instance, LOD2 cannot contain openings, but it is not stated that LOD3 where they create three variants of LOD3 buildings that are distinguished
must contain openings. by the size and type of features to be acquired. We have considered their
Hence, besides an insufcient number of LODs and their condensed granular LODs when designing our rened specication.
grouping, the main drawback of the concept in the current version of Borrmann, Flurl, Jubierre, Mundani and Rank (2014) provide an
CityGML is that it does not mandate what features and how detailed extended LOD specication for tunnels dening ve LODs to create
they should be acquired, and therefore it leaves ambiguity and freedom consistent multi-scale models and to use them for synchronous
28 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

engineering collaboration. The LODs are discerned primarily by the list and Lwner (2013) and Lwner et al. (2013) propose the orthogonal
of railway elements that are acquired. Chen (2013) does a similar decoupling of the exterior and indoor geometry, and a renement into
work for trees dening four LODs i.e. Level Of Tree-detail. The feasibil- multiple semantic LODs. The number of permutations, excluded by
ity of the acquisition of these representations has been conducted with some prohibited variants, is large enhancing the specication, since
different airborne laser scanning scenarios. they still t within the present CityGML LODs. For instance, a building
In the BIM community, Van Berlo and Bomhof (2014) have worked with a coarse exterior with no semantic structuring may include a ne
on the renement of the BIM LODs after analysing industrial practices interior, and such has a unique designation. The renement of the in-
and conducting a series of geometric tests. This is similar to our ap- door LODs is a current research topic, which is also in focus in
proach (Sections 4 and 5). Related to the BIM domain, Tolmer, Hagedorn, Trapp, Glander and Dllner (2009), Kemec, Zlatanova, and
Castaing, Morand and Diab (2014) propose additional LODs to allow Duzgun (2012), Billen, Zaki, Servires, Moreau, and Hallot (2012),
for a more transversal decomposition of data and objects organisation, Kang and Lee (2014), Kim, Yoo, and Li (2014), and Boeters et al. (2015).
and apply them to an urban motorway project.
Wate, Srivastav, Saran, and Murthy (2013) emphasise the 4. Rened levels of detail for buildings
importance of the relation of the acquisition technique to an LOD, and
give acquisition technique guidelines for each CityGML LOD. We provide a series that contains 16 LODs (4 rened LODs for each of
Vosselman and Dijkman (2001) show how the capabilities (resolution) the LOD03), which are shaped after a literature review and inventory
of acquisition techniques have a direct impact on the LOD of the of presently available models by nding their main relevant similarities
reconstructed 3D city model. In our work we have analysed acquisition and mutual aspects. A visual example of the rened LODs is shown in
workows, and we have taken them into account when designing the Fig. 3. We believe that these LODs allow for less ambiguity, and they
specication. aid practitioners to standardise their data with an improved denition
Coors (2003) distinguishes the LOD of the acquired geometry from of the complexity of the models. As mentioned before, this work is not
the presented model (view). Related to the visualisation aspect, intended to extend CityGML 2.0, it rather provides a supplementary
ltekin and Reichenbacher (2011) analyse balancing the cognitive specication that reects the current practices and that conforms to
and bandwidth aspects of multi-LOD data, emphasising the economical the current concept, and at the same time solves the ambiguities
aspect of each LOD. elaborated in Section 2.
Dllner (2005) expresses that in the current LOD approach it is
difcult to integrate buildings from different sources and of varying 4.1. Methodology
LOD. Furthermore, they discuss the models that can be considered of
an LOD between LOD2 and LOD3. Their observations are important for Besides investigating workows for producing 3D city models (e.g.
our work because we introduce two LODs that are designed to be Habib, Zhai, & Kim, 2010; Kedzierski & Fryskowska, 2014; Xiong,
acquired with a combination of different sources. Jancosek, Oude Elberink, & Vosselman, 2015), we have examined
Stadler and Kolbe (2007) propose several levels of spatio-semantic several categories of sources of data. For instance, national standards
coherence in 3D city models. Their work is one of the foundations in and guidelines (AdV, 2011, 2013; Stoter et al., 2014; Blaauboer et al.,
the semantic aspect of 3D city models. Benner, Geiger, Grger, Hfele, 2013; Chinese Ministry of Housing and Urban-Rural Development,

Fig. 3. Visual example of the rened LODs for a residential building.

 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 29

2010), examples of series and specications of 3D city models not relat- selection criteria is mostly based on their signicance and minimum
ed to CityGML (Kninger & Bartel, 1998; Batty et al., 2000; Kemec et al., size, and it depends on the object's class (Biljecki, 2007). This reasoning
2012; Schilcher, Roschlaub, & Guo, 1998; Thiemann, 2003; Demir & can be followed in 3D as well. However, it should be stated that the sig-
Baltsavias, 2012), usually in the eld of 3D generalisation (Zhao, Zhu, nicance and size are both fuzzy terms that also depend on the use case,
Du, Feng, & Zhang, 2012; Li, Sun, Yang, & Jiang, 2013), visualisation and cannot be strictly dened, as other concepts related to scale and
(Andujar, Brunet, Chica, & Navazo, 2010; Rau & Cheng, 2013) and 3D re- LOD.
construction (Huang, Brenner, & Sester, 2011; Verma, Kumar, & Hsu, The minimum size can be expressed as the minimum length and/or
2006). width of an object, and/or the minimum footprint area. For instance, a
Furthermore, we have examined examples of models that refer to requirement may state that dormers that are wider than 1 m, and/or
the CityGML LODs but do not seem to be stored in CityGML or have their footprint projected onto the roof is larger than 1 m2 should be
any other relation with the standard (Nex & Remondino, 2013; acquired. This can be applicable to both the size of a feature and its
Burochin et al., 2014; Qin, 2014; Becker, 2011; Kanuk, Gallay, & granularity, e.g. minimum size of a land cover area or its spikes.
Hoerka, 2015; Zhu & Hu, 2010). Finally, a number of publicly available In expressing the thresholds, it is important to dene both the 1D
data sets and specications from companies, tenders and local and 2D requirements. For instance, a chimney may be longer than a
governments have been examined as well (Novakovi, 2011; Frani, dormer (1.5 m N 1.0 m), but much smaller when considering its
Baci-Deprato, & Novakovi, 2009; Vande Velde, 2005; Blom, 2011; footprint on the roof (0.15 m 0.15 m 0.02 m2). A 3D requirement
Glasgow City Council, 2009; Vertex Modelling, 2013; NAVTEQ, 2011; (volume) will not be used because it is not applicable to all types of
Sanborn, 2014; CyberCity 3D, 2013). features (e.g. windows).
We have analysed the available models from different angles:
usability (their intended use cases), acquisition techniques (implying 4.3. General rules and principles of the specication
their cost and availability), and specication if available. The dozens of
variants of models have been grouped into original CityGML LODs to The main principles of our specication are:
which they correspond, and which we name LOD families. We have
found a few general aspects, and various spatio-semantic ambiguities 1. A model must adhere to all of the requirements to be considered of a
that surface in each LOD family, resulting in four groups of models specic LOD.
within each family. The LOD groups are concentrations of models and 2. With a few exceptions, the specication is designed in such a way
partially imply their correspondence to the real-world, acquisition that each LODx.i contains more detail over LODx.(i1), i.e. all
technique, accuracy, domain of applications, etc. For each of these LOD requirements of an LODx.i should also satisfy the requirements of
groups a common set of requirements has been established, resulting an LODx.(i 1). However, observe that an LODx.i is not necessarily
in a specication and rened set of LODs. more detailed than an LOD(x1).(i + 1). This is comparable to the
Note that in our specication we do not focus on the semantic aspect discussion related to Fig. 2.
and the texture. We address the amount of geometric detail that has to 3. The list of building elements to be acquired is composed from the
be acquired, by focusing on the list of elements of a building, and their most common occurrences of such features, for instance, windows
granularity. Non-geometric requirements can be supplemented to our and dormers. This list applies also to other elements of comparable
specication if required. function and size.
A few observations in the survey motivated specic choices. The rst 4. The selection criteria to model an object or not have been
matter which we have noticed in our survey is the large number of determined based on the minimum size of the building's element
unique specications and combinations of various aspects. It is not to be acquired. The minimum size is expressed as a distance, which
possible to regard each of the aspects while retaining a reasonable num- can be applied to the length, width, or height of a feature, and as a
ber of LODs. Hence, in this process we have balanced the scope of the projected footprint area. The footprint projection is not necessarily
specication and the number of the rened LODs, and we have taken on the ground. For instance, for windows the projection onto the
care that the models unambiguously correspond to a rened LOD. The walls is considered, and roof superstructures on the (inclined) roof.
goal is to provide a ner specication, but exible enough to still These requirements are specied separately for each feature's type
allow some freedom of modelling and not to result in a too large (e.g. chimney). If no footprint selection criteria is stated, only the
number of levels. This is benecial for use cases, since a large number length criteria applies. Such minimum size selection criteria may
of models have been acquired bearing in mind a specic use case, and cause disregarding some features that in certain settings could be im-
a strict specication would not be favourable towards such practices. portant for a particular application. If that is the case, users may still
Another observation that led to a specic choice was a small number opt for modelling smaller features beyond the threshold.
of outliers specications that are not in line with the common 5. The specication provides general requirements that leave space for
practices in 3D city modelling. They are rather designed for a specic an extra number of variants. An LODx model that is modelled ner
application and as such cannot be accommodated in a uniform speci- than it is required in LODx.i, but below the specication of
cation. For instance, Kemec et al. (2012), Ioannidis, Soile, and Potsiou LODx.(i + 1), is considered as LODx.i. For instance, if an LODx.i and
(2015), and Frommholz, Linkiewicz, Meissner, Dahlke, and Poznanska LODx.(i + 1) require all buildings parts larger than 4 m and 2 m to
(2015) dene an LOD2 model with generalised footprint and fenestra- be acquired, respectively, and a model contains a part 3 m long, it
tion; and the specication in Kartverket (2014) which denes LOD1 may be considered as LODx.i.
models with non-at top surfaces. Such models are not considered, 6. The specication denes that sizeable building parts, extensions and
since their inclusion would compromise the simplicity of our concept, annexes such as garages and alcoves, may be acquired and treated dis-
but it would also not be in line with the standard CityGML LODs. tinctly. This should be distinguished from the cadastral point of view.
However, the rationale of our specication can still be used to dene Such objects are still part of the building, but their geometry is ac-
such customised LODs in addition to our series. quired in a way that these features are perceptually distinguishable.

4.2. Selection criteria for objects to be acquired 4.4. Rened LOD specication for the geometry

The selection of objects (e.g. building elements such as dormers) to The specication is given in Table 1, and a visual representation of
be mapped is an important part of the LOD specication. This is also the models is provided in Fig. 3. In this section each LOD family is rened
analogous to 2D maps (Touya & Brando-Escobar, 2013), where the with four LODs that are described.
30 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

Table 1
Specication of the rened levels of detail tting the current CityGML 2.0 LODs.

Requirements Rened levels of detail

0.0 0.1 0.2 0.3 1.0 1.1 1.2 1.3 2.0 2.1 2.2 2.3 3.0 3.1 3.2 3.3

Individual buildings
Large building parts (N4 m, 10 m2)
Small building parts, recesses and extensions (N2 m, 2 m2)
Top surfacea S M S S S M
Explicit roof overhangs (if N0.2 m)
Roof superstructuresb
(larger than 2 m, 2 m2)
Other roof details
(e.g. chimneys N1 m)
Openings (c) (N 1 m, 1 m2) R W
Balconies (N1 m)
Embrasures, other faade and roof details, and smaller windows
(N0.2 m)
a
Applicable only to LOD0.y and LOD1.y: S single top surface; M multiple top surfaces if the difference in height of the extruded building elements is signicant (larger than 2 m).
b
It includes dormers and features of comparable size and importance (e.g. very large chimneys).
c
R only openings on roofs; W only openings on walls. In R, openings on dormers are not required.

Because LOD1 is in essence an extrusion of the LOD0 model (or its (2007) and Mayer (2005) all generalise LOD1 models creating a coarser
generalised product), both are examined and redened in conjunction. LOD1 model. Agugiaro (2014) generates two variants of block models
from footprints: one from a cadastral source, and one from a topograph-
4.4.1. LOD0 and LOD1 families ic map. Therefore, we have identied the following relevant aspects in
The coarsest volumetric representation that the standard contains is the LOD0 and LOD1 families:
the LOD1 model, a generalised model which is only described as the
well-known blocks model comprising prismatic buildings with at The models may represent individual or aggregated buildings
roof structures (Open Geospatial Consortium, 2012). Block models (buildings that in reality are not adjacent and between which there
have also been described in patent language in the US Patent is a gap, but are close enough to be modelled as one entity, at a smaller
Application by Guskov and Brewington (2015), as a set of extruded scale). Some specications enforce this, for instance, the 3D standard
polygons (right prisms) that comprise a volume which is dened by a of The Netherlands requires individual buildings prohibiting their
base height from which extrusion begins, and an extrusion distance. aggregation (Stoter et al., 2014; Blaauboer et al., 2013).
LOD0 is briey described as a representation by 2.5D horizontal Since 2D footprints are extruded to a uniform height, and the
polygons with footprint level height and optionally roof level height resolution of the footprint directly implies the LOD of the 3D model,
(Grger & Plmer, 2012). they are the focus of LOD0 and LOD1. However, their complexity
LOD1 models are usually derived with extrusion to a uniform height may considerably deviate (Yang, Zhang, Ma, Xie, & Liu, 2011), from
(Buyuksalih, Isikdag, & Zlatanova, 2013; Ledoux & Meijers, 2011; Over, coarse to ne footprints as seen in Prandi et al. (2013) and Ellul and
Schilling, Neubauer, & Zipf, 2010; Ordnance Survey, 2014; SwissTopo, Altenbuchner (2013). This is especially the case in generalisation
2010; Arroyo Ohori, Ledoux, & Stoter, 2015b; Sargent, Holland, & from ner LODs (Forberg, 2007; Anders, 2005).
Harding, 2015), and generalisation from ner LODs (Baig & Besides the footprint, LOD0 models may contain a roof-edge surface.
Abdul-Rahman, 2013; Meng & Forberg, 2007), for instance, as a Multiplicity of top surfaces: LOD1 models are not necessarily extruded
bounding box of an LOD2 (Diakit, Damiand, & Van Maercke, 2014; to a uniform height, which is a common misconception about the pro-
El-Mekawy, stman, & Shahzad, 2011). As a consequence, there are duction of the LOD1 model, since the number of top surfaces are not
only horizontal and vertical surfaces, and no projection to the xy plane restricted by the standard. We have encountered several instances
of two horizontal surfaces can overlap (not counting the ground of enhanced LOD1 block models with differentiated roof tops (Emem
surface). & Batuk, 2004; Commandeur, 2012; Oude Elberink, Stoter, Ledoux, &
Although LOD1 models are the coarsest volumetric representation, Commandeur, 2013; Dllner, Kolbe, Liecke, Sgouros, & Teichmann,
they can be derived from very accurate and detailed footprints (Van 2006; Ellul & Altenbuchner, 2013; He et al., 2013; Hfele, 2011;
den Brink et al., 2013a; Kolbe et al., 2015) (see also Fig. 2). Sanborn, 2014) that include multiple at surfaces instead of a single
LOD1 models provide a relatively high information content and surface for the top, to differentiate terraced houses, large roof
usability comparing to their geometric detail (Henn, Rmer, Grger, & constructions, etc. Related to extrusion, sizeable parts of buildings
Plmer, 2012; Hoerka & Zlocha, 2012). For instance, they may be (e.g. veranda, carport, garage, and alcove) may be modelled separately
used for shadowing simulations (Strzalka, Alam, Duminil, Coors, & with a different value of the height, resulting in multiple top surfaces,
Eicker, 2012; Alam, Coors, & Zlatanova, 2013; Li, Zhang, & Davey, even if they belong to the same footprint. This may be to a degree
2015), estimation of noise pollution (Stoter, de Kluijver, & Kurakula, incompatible with the traditional LOD1 notion, however, their
2008), energy demand estimation (Strzalka, Bogdahn, Coors, & Eicker, occurrences warrant a separate LOD, and not all LOD1 models are
2011; Bahu, Koch, Kremers, & Murshed, 2013), simulating oods derived with extrusion.
(Varduhn, Mundani, & Rank, 2015), analysing wind comfort (Amorim,
Valente, Pimentel, Miranda, & Borrego, 2012), and visualisation
(Gesquire & Manin, 2012). We dene four LODs in each LOD family with the minimum
While the LOD1 model is the simplest volumetric 3D city model, it requirements: LOD0.0 and LOD1.0 are the coarsest models: they
may be modelled in multiple ways. Gtzelmann, Guercke, Brenner, require all buildings larger than 6 m to be acquired, and buildings may
and Sester (2009), Glander and Dllner (2009), Meng and Forberg be aggregated. These are the only two instances in our specication in
 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 31

which neighbouring buildings may be aggregated in a single geometric et al., 2013), national mapping agencies (Kartverket, 2014; Stoter et al.,
entity. In LOD0.1 and LOD1.1 buildings must be individually modelled 2015), and in municipal data sets (Dllner et al., 2006).
and all large building parts shall be acquired. LOD0.2 and LOD 1.2 have We have grouped the occurrences of LOD2 models into four LOD2
the requirement that smaller building parts and extensions should be variants based on the aforementioned aspects:
acquired (e.g. alcoves), and are extruded to a single height. This addition
may result in more accurate spatial analyses, such as line of sight LOD2.0 is a coarse model with standard roof structures, and potential-
(Yaagoubi, Yarmani, Kamel, & Khemiri, 2015). In addition, LOD0.2 re- ly including large building parts such as garages (larger than 4 m and
quires the roof-edge polygon to be acquired as well. We have found 10 m2).
that the LOD1.2 is the most common LOD1 model in practice. LOD0.3 LOD2.1 is similar to LOD2.0 with the difference that it requires smaller
and LOD1.3 also require the same features to be acquired, but it allows building parts and extensions such as alcoves, large wall indentations,
multiple top surfaces if their differences are higher than a threshold and external ues (larger than 2 m and 2 m2) to be acquired as well. In
(e.g. 2 m). For instance, a large recess in a wall might have its height sep- comparison to the coarser counterpart, modelling such features in this
arately modelled and may be individually extruded. This approach can LOD could benet use cases such as estimation of the energy demand
benet use cases such as estimating the internal area. Such modelling because the wall area is mapped more accurately.
practice may be considered counter-intuitive from the extrusion point LOD2.2 follows the requirements of LOD2.0 and LOD2.1, with the
of view, where a footprint polygon is extruded to a single height. addition of roof superstructures (larger than 2 m and 2 m2) to be
Howeverdue to a substantial number of such models, the increased acquired. This applies mostly to dormers, but also to other signicant-
accuracy they may bring to spatial analyses, and the fact that not all ly sized roof structures such as very large chimney constructions. Be-
LOD1 models are derived by extrusionintroducing such LOD is neces- cause the roof is mapped in more detail, this LOD can be advantageous
sary. Finally, LOD1.3 cannot contain multiple horizontal surfaces at the for the estimation of the insolation of roofs (Biljecki et al., 2015a).
same planar coordinate. LOD2.3 requires explicitly modelled roof overhangs if they are longer
than 0.2 m, therefore the roof edge and the footprints are always at
their actual location, which has advantages for use cases that require
4.4.2. LOD2 family the volume of the building.
LOD2 is a more detailed model than LOD1, in which individual build-
ings are mandated, and are modelled as simple structures containing
standard and simplied roof structures. They are usually derived from 4.4.3. LOD3 family
point clouds or photogrammetry, and their combination with building LOD3 adds openings (i.e. windows, doors), balconies, more detailed
footprints (Alexander, Smith-Voysey, Jarvis, & Tansey, 2009; Haala & roof structures (e.g. chimneys and antennas), and mandatory roof over-
Brenner, 1999; Haala & Kada, 2010). They can also be derived with hangs. This enhancement benets some applications, for instance, open-
generalisation from an LOD3 (Mao, Harrie, & Ban, 2012). ings are important for estimating heat losses (Lee et al., 2013),
The models provide a relatively favourable relation between the costs luminance mapping and glare analysis (Saran, Wate, Srivastav, &
of acquisition and usability. Acquisition-wise, they can be automatically Krishna Murthy, 2015), planning energy-efcient retrots (Previtali
derived from point clouds (Suveg & Vosselman, 2004; Brenner, 2005; et al., 2014), and for accounting the area available on vertical walls for
Kada & McKinley, 2009). Usability-wise, they can be used in a wider solar panel installation (Catita, Redweik, Pereira, & Brito, 2014). LOD3
range of applications than LOD1, such as the estimation of the solar poten- models are also appreciated in visualisation (Garnett & Freeburn, 2014).
tial of rooftops (Fath et al., 2015; Biljecki et al., 2015a), or as an improve- The acquisition of LOD3 models is a laborious process (Buhur, Ross,
ment in accuracy over LOD1 (e.g. in energy demand estimation Kaden & Byksalih, & Baz, 2009), hence they are in practice of limited availabil-
Kolbe, 2014). ity, and are usually restricted to smaller areas. They are normally de-
Stoter et al. (2014) recognise the ambiguity of roof overhangs in LOD2, rived from terrestrial laser scanning (El Meouche, Rezoug, Hijazi, &
i.e. whether they should be explicitly modelled or not, as overhangs may Maes, 2013; Akmalia, Setan, Majid, Suwardhi, & Chong, 2014), very
add value to certain use cases. The CityGML standard allows overhangs in dense airborne laser scanning (Truong-Hong & Laefer, 2015), their com-
LOD2 if known, but it does not require them (see again Fig. 2). This results bination (Kedzierski & Fryskowska, 2014), from the conversion and
in various LOD2 models with and without overhangs (for examples see generalisation from architecturally detailed models such as BIM
Benner et al. (2010); SwissTopo (2014); Steinhage et al. (2010); Oude (Donkers et al., in press; Geiger, Benner, & Haefele, 2015; Isikdag &
Elberink (2010); Leszek (2015); and Fan and Meng (2012); Van den Zlatanova, 2009; de Laat & van Berlo, 2011) and CAD (Lewis & Squin,
Brink et al. (2013b); Kada and McKinley (2009); Schilling et al. (2012); 1998; Huang, Lo, Zhi, & Yuen, 2008), from architectural plans (Yin,
Prechtel (2014); Schwalbe et al. (2005); Frstner (1999); Henn et al. Wonka, & Razdan, 2009), ground imagery (Xiao et al., 2009), and with
(2013), respectively). Most of the models from national mapping agen- procedural modelling (Goetz, 2013; Smelik et al., 2014; Martinovi,
cies do not have explicitly modelled roof overhangs (Aringer & Knopp, Riemenschneider, & van Gool, 2015). Recent research in the
Roschlaub, 2014; Brasebin, Perret, Mustire, & Weber, 2012). LOD2 acquisition of LOD3 models is focused to automatisation, especially
models with differentiated roof overhangs cost more to acquire since automatic detection of windows and other faade details (Becker,
they generally require a combination of two acquisition techniques (air- 2009, 2011; Van Gool and Martinovi, 2013).
borne and terrestrial). When roof overhangs are not available, the walls LOD3 models are signicantly more detailed than LOD2 models, and
are usually obtained as projections from the roof edges to the ground, in- less ambiguity is present from the 3D GIS point of view (while it would
herently increasing the volume of the building (Biljecki et al., 2016). be possible to nitpick among different variants of LOD3, they will hardly
A second ambiguity of LOD2 buildings are building installations such make any difference to spatial analyses). The only differentiation
as dormers, and chimneys, which are allowed in LOD2, but they are between LOD3 models we have detected is the minimum size of
rarely found in LOD2 models in practice. LOD2 models that contain features that are acquired, especially whether the embrasures of open-
building installations only include large features which considerably ings are modelled or not. For instance, (Besuievsky et al., 2014) create
protrude the wall or roof structure (Grger & Coors, 2011; Ben Fekih multiple variants of LOD3 models, one with at windows, one with
Fradj & Lwner, 2012; Vosselman & Dijkman, 2001). The German Feder- the embrasures, and a third with minor faade details. The ner two
al 3D building modelling guideline (AdV, 2013) explicitly states have the roughly the same size of linear features rendering their
that dormers and other objects of similar size should not be acquired. difference negligible from the GIS standpoint, hence, we merge them.
On the other hand, we have encountered LOD2 models which have dor- We dene: LOD3.2, an architecturally detailed model that contains
mers and other roof superstructures modelled, in academia (Buyuksalih features of size larger than 1.0 m, and LOD3.3one that contains
32 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

features of size larger than 0.2 m, including embrasures of windows (i.e. Table 2
making windows 3D), awnings and similar features of comparable size. Results of the experiments with the models acquired in the rened levels of detail. The size
of the test data set is 100 buildings. The geometric results of the area and volume compu-
The latter is benecial over the former for high-quality visualisation and tations are expressed as absolute deviations in percentages from the results of the compu-
virtual reality applications (e.g. Portals, Lerma, & Navarro, 2010), and it tations on the LOD3.3 model.
is usually a product of the conversion from BIM and architectural
Family Characteristic Rened level of detail
models.
However, we have encountered a number of models that cannot be LOD0 LOD0.0 LOD0.1f LOD0.2e,f LOD0.3e,f
Trianglesb 0.32 2.16 4.72 4.54
fully accommodated in the traditional LODs, but are common in the
Total surface area dev. 17.94 85.82 71.62 71.62
acquisition workows and which technically belong to LOD3. For in- WallSurface area dev. a a a a

stance, Frani et al. (2009) and Novakovi (2011) create a model from Volume dev.c a a a a

an aerial survey with roof details ner than in LOD2.3, but with other Memoryd [kB] 16 98 152 156
features of lesser detail comparable to LOD2. We denote such models LOD1 LOD1.0 LOD1.1e,f LOD1.2e,f LOD1.3e,f
Trianglesb 1.92 12.64 13.44 13.44
of mixed LODs (e.g. different LOD for aerial and terrestrial features) as Total surface area dev. 214.49 7.45 7.59 7.29
hybrid LODs. Two such LODs that reect the acquisition workows WallSurface area dev. a a a a

have been dened, and to attempt to accommodate models of specic Volume dev.c 784.40 23.36 23.43 22.94
congurations. Because they both contain openings, we add them to Memoryd [kB] 45 283 291 291
LOD2 LOD2.0f LOD2.1f LOD2.2f LOD2.3
the LOD3 group.
Trianglesb 13.70 14.50 19.30 26.18
LOD3.0 is a model where roof structures are mapped in ner detail Total surface area dev. 5.74 5.80 6.04 3.72
than LOD2.2, but other features such as walls are acquired as LOD2.2. WallSurface area dev. 17.41 17.47 17.90 7.51
Roofs may include roof windows. Windows of dormers do not have to Volume dev.c 21.45 21.48 24.02 1.77
be acquired. Memoryd [kB] 528 545 686 661
LOD3 LOD3.0f LOD3.1 LOD3.2 LOD3.3
LOD3.1 is its terrestrial counterpart, dened for terrestrial acquisi-
Trianglesb 26.22 302.58 317.50 596.30
tion techniques, such as mobile mapping systems (Kaartinen, Hyypp, Total surface area dev. 6.07 3.72 3.45 0.00
Kukko, Jaakkola, & Hyypp, 2012). Since these techniques operate WallSurface area dev. 17.95 5.77 5.41 0.00
from the ground, roof features may be out of reach. Therefore this Volume dev.c 21.55 1.84 0.70 0.00
Memoryd [kB] 808 4242 4554 11,775
instance requires all features below roof to be acquired with the
a
LOD3.2 grade, and the roof as LOD2.3, since overhangs may be explicitly Computation not possible due to dimensionality and/or insufcient semantic level.
b
modelled. This LOD is advantageous for use cases that require only the Average number of triangles per building. The values of the aggregated LOD0.0 and
LOD1.0 are low due to one block model covering multiple buildings.
wall surface to be modelled in ner detail, such as estimating the solar c
Volume of the solid representation. The deviation of LOD1.0 is high due to aggregation.
potential of faades or for pedestrian navigation. d
For all buildings in the data set, without compression.
e
The geometric reference of half of the roof was selected to represent the top of the
LOD0 and LOD1 models, which is common in the reconstruction of buildings from point
5. Proving the specication with geometric experiments clouds.
f
For these representations the geometric reference for the walls are projections from
roof edges, as it is common in airborne acquisition techniques.
For the experiments, we have generated a CityGML data set of 100
buildings in the 16 LODs we propose. The data has been generated
procedurally with a CityGML engine developed by Biljecki, Ledoux, The used semantic levels in these representations are the standard
and Stoter (2014a), and it has been converted to the OBJ le format to ones in CityGML (e.g. see thematically differentiated surfaces for each
broaden their usability. The conversion was done with the tool LOD in Fig. 1).
CityGML2OBJs (Biljecki & Arroyo Ohori, 2015), and in the process the The results are given in Table 2 and in Fig. 4. They prove numerically
models were triangulated with Triangle (Shewchuk, 1996). that each of the representations is different from the others, and that it
We have computed the following quantitative properties of the can be considered as a unique and standalone LOD, not only from the
models: (1) average triangle count per building, akin to the computer geometric aspect, but also from the pointview of a spatial analysis.
graphics complexity metric; (2) total surface area; (3) area of the From these results also the magnitude of the deviations within the
WallSurface; (4) volume of the corresponding solids; and (5) size in same LOD family are evident (e.g. the volume the difference between
memory for each of the representations. two models within LOD2 may be 24%). The rened specication
The results (2), (3), and (4) have been calculated as a sum for all presented in this paper decreases such differences.
buildings in the data set. Notice that these values are commonly used These experiments show the grouped results for all buildings in the
as a base in several spatial analyses. For instance, they are used in data set, and are sufcient for the aim of this section. In future work on
urban planning (Ahmed & Sekar, 2015), in the volumetric visibility this matter we plan to conduct a detailed error propagation analysis for
analysis of urban environments (Fisher-Gewirtzman, Shashkov, & each of the representations (e.g. root mean square error and distribu-
Doytsher, 2013), energy estimations (Nouvel, Schulte, Eicker, tions of the deviations), and to relate it to particular use cases, such as
Pietruschka, & Coors, 2013; Eicker, Nouvel, Duminil, & Coors, 2014), energy demand estimation.
predicting thermal comfort (Chwieduk, 2009), predicting cooling
requirements of buildings (Perez, Kmpf, & Scartezzini, 2013), in 6. Conclusions and future work
thermal simulations involving computational uid dynamics (Hsieh,
Aramaki, & Hanaki, 2011; Maragkogiannis et al., 2014), in urban design The current LOD categorisation of CityGML has two shortcomings:
evaluation (Yang, Putra, & Li, 2007), for calculating development
1. lack of a precise specication of each LOD; and that
densities (Meinel, Hecht, & Herold, 2009), and in investigating the
2. the current ve LODs are too generic and therefore they are not
urban heat island effect (van der Hoeven & Wandl, 2015).
always capable to separate one LOD from the other (i.e. two
The values are given in percentages of the absolute deviation from
signicantly different levels of abstraction may still be considered
the results of the computations on LOD3.3, as the nest LOD in this
as the same LOD as per the current specication).
series. The geometric computations were conducted with MeshLab1
(Cignoni, Corsini, & Ranzuglia, 2008). The rened LODs that we have introduced are a result of a litera-
ture review, analysing acquisition workows, and examining public-
1
A tool developed with the support of the 3D-CoForm project. ly available specications. The specication is compliant with the
 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 33

Fig. 4. Graphical depiction of the results of the experiments. The newly dened 16 levels of detail are examined from ve numerical aspects. The values are normalised according to the
average value in the corresponding LOD family, and it proves that each LOD is unique. Furthermore, these values expose the fundamental problem with the CityGML LODs: due to their
insufcient number, the relative differences of the models within each family may be signicant, e.g. for the total surface area the difference between two models within LOD3 may be 6%,
which may cause discrepancies in spatial analyses. Our paper renes the LODs to diminish the ambiguity and alleviate potential misunderstandings between stakeholders.

existing LOD concept in the well established standard CityGML, investigate how to automatically validate whether a data set is
hence, while improving the shortcomings of the current concept, modelled according to a certain LOD, and we intend to conduct detailed
the rened specication does not damage the commonly used ve experiments of the performance of each LOD in a particular use case,
standard LODs it is completely compatible with it. It is possible aiding the practitioners to choose the optimal LOD for a use case when
to determine the LOD of an existing data set, and to store in the doc- also considering the costs of the acquisition. Experiments carried out
umentation or metadata of the model (since CityGML 2.0 does not in Section 5 are the rst step towards such research.
support storing such information). However, we foresee that our
specication may be integrated as a user dened prole in the Acknowledgements
upcoming CityGML 3.0, where a generic LOD framework is planned
(Lwner & Grger, 2016). We gratefully acknowledge the comments of the members of the
We have covered the vast majority of cases we have found in EuroSDR 3D Special Interest Group, members of the OGC CityGML
practice, and while further differentiations are possible, we believe Standard Working Group, members of the OGC CityGML Quality Inter-
that it is not benecial to dene more than 16 LODs. The extended operability Experiment, and those of the anonymous reviewers. We
specication is simple and it is intended to be of special interest to the appreciate the input given by companies specialised in the acquisition
data producers, addressing their critic of ambiguity that the current of 3D city models.
LODs present. Most importantly, not much modelling freedom is This research is supported by the Dutch Technology Foundation
allowed anymore, diminishing potential misunderstandings between STW, which is part of The Netherlands Organisation for Scientic
3D stakeholders and potential errors in the usability of the models. Research (NWO), and which is partly funded by the Ministry of
As much as the current LODs of CityGML 2.0 are used outside the Economic Affairs (project code: 11300).
CityGML format, these improved LODs may also be considered as
independent of CityGML, and applicable to any other 3D format or
References
context. For instance, they may become an important factor in
contracting the data acquisition, and as a more precise paradigm to AdV (2011). Produktstandard fr 3D-Gebudemodelle. Arbeitsgemeinschaft der
Vermessungsverwaltungen der Lnder der Bundesrepublik Deutschland (Working Com-
express and benchmark the capabilities of a 3D city model reconstruc-
mittee of the Surveying Authorities of the States of the Federal Republic of Germany)
tion technique. Furthermore, practitioners can nd them useful to (Available online at http://www.adv-online.de, (last accessed on 12 February 2016)).
better dene their product portfolios, and national mapping agencies AdV (2013). Modellierungsbeispiele fr 3D-Gebudemodelle. Arbeitsgemeinschaft der
to implement them in their specications. The classication has already Vermessungsverwaltungen der Lnder der Bundesrepublik Deutschland (Working Com-
mittee of the Surveying Authorities of the States of the Federal Republic of Germany)
been adopted by the EuroSDR 3D Special Interest Group (SIG) as a base (Available online at http://www.adv-online.de, (last accessed on 12 February 2016)).
for a forthcoming European standard for national mapping in 3D. Agugiaro, G. (2014). From sub-optimal datasets to a CityGML-compliant 3D city model:
Finally, researchers can nd them useful to unambiguously express Experiences from Trento, Italy. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, XL-4, 713.
the level of detail of models they are analysing, primarily in work on Ahmed, F. C., & Sekar, S. P. (2015). Using three-dimensional volumetric analysis in
generalisation and 3D use cases. everyday urban planning processes. Applied Spatial Analysis and Policy, 8, 393408.
We have reinforced our solution by running the data sets through a Akmalia, R., Setan, H., Majid, Z., Suwardhi, D., & Chong, A. (2014). TLS for generating
multi-LOD of 3D building model. IOP conference series: Earth and environmental sci-
few 3D GIS operations, which yield different results for each, showing ence. Proceedings of the 8th International Symposium of the Digital Earth (ISDE8), Vol.
that the variants may be considered as standalone LODs. In addition, 18. (pp. 18).
the signicant difference between different LODs of the same LOD Alam, N., Coors, V., & Zlatanova, S. (2013). Detecting shadow for direct radiation using
CityGML models for photovoltaic potentiality analysis. In C. Ellul, S. Zlatanova, M.
family shows how important it is to further differentiate LODs
Rumor, & R. Laurini (Eds.), Urban and regional data management (pp. 191196).
and to specify these differentiations. While the new LODs are de- London, UK: CRC Press.
ned in a more precise way, allowing signicantly less ambiguity, Albert, J., Bachmann, M., & Hellmeier, A. (2003). Zielgruppen und Anwendungen fr
Digitale Stadtmodelle und Digitale Gelndemodell. Technical report. SIG3D (Available
they still permit a degree of exibility allowing additional require-
online at http://www.ikg.uni-bonn.de/leadmin/sig3d/pdf/Tabelle_Anwendungen_
ments driven by use cases, following the reasoning of Stoter et al. Zielgruppen.pdf, last accessed on 12 February 2016).
(2014). Alexander, C., Smith-Voysey, S., Jarvis, C., & Tansey, K. (2009). Integrating building foot-
We have covered buildings as the most important and most prints and LiDAR elevation data to classify roof structures and visualise buildings.
Computers, Environment and Urban Systems, 33, 285292.
modelled features of the urban environment. For future work, we plan Amorim, J. H., Valente, J., Pimentel, C., Miranda, A. I., & Borrego, C. (2012). Detailed model-
to work on the other thematic modules. Furthermore, we plan to ling of the wind comfort in a city avenue at the pedestrian level. In T. Leduc, G.
34 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

Moreau, & R. Billen (Eds.), Usage, usability, and utility of 3D city models European Brasebin, M., Perret, J., Mustire, S., & Weber, C. (2012). Measuring the impact of 3D data
COST Action TU0801 (pp. 16). Nantes, France: EDP Sciences (03008). geometric modeling on spatial analysis: Illustration with Skyview factor. In T. Leduc,
Anders, K. H. (2005). Level of detail generation of 3D building groups by aggregation and G. Moreau, & R. Billen (Eds.), Usage, usability, and utility of 3D city models European
typication. Proceedings of the 22nd International Cartographic Conference: Mapping COST Action TU0801 (pp. 116). Nantes, France: EDP Sciences (02001).
approaches into a changing world, La Corua, Spain. Brenner, C. (2005). Building reconstruction from images and laser scanning. International
Andujar, C., Brunet, P., Chica, A., & Navazo, I. (2010). Visualization of large-scale urban Journal of Applied Earth Observation and Geoinformation, 6, 187198.
models through multi-level relief impostors. Computer Graphics Forum, 29, 24562468. van den Brink, L., Stoter, J., & Zlatanova, S. (2013a). Establishing a national standard for 3D
Aringer, K., & Roschlaub, R. (2014). Bavarian 3D building model and update concept based topographic data compliant to CityGML. International Journal of Geographical
on LiDAR, image matching and cadastre information. Innovations in 3D geo- Information Science, 27, 92113.
information sciences (pp. 143157). Springer International Publishing. van den Brink, L., Stoter, J., & Zlatanova, S. (2013b). UML-based approach to developing a
Arroyo Ohori, K., Ledoux, H., & Stoter, J. (2015b). A dimension-independent extrusion al- CityGML application domain extension. Transactions in GIS, 17, 920942.
gorithm using generalised maps. International Journal of Geographical Information Buhur, S., Ross, L., Byksalih, G., & Baz, I. (2009). 3D city modelling for planning activities,
Science, 29, 11661186. case study: Haydarpasa train station, Haydarpasa port and surrounding backside
Arroyo Ohori, K., Ledoux, H., Biljecki, F., & Stoter, J. (2015a). Modeling a 3D city model and zones, Istanbul. The International Archives of the Photogrammetry, Remote Sensing
its levels of detail as a true 4D model. ISPRS International Journal of Geo-Information, 4, and Spatial Information Sciences, XXXVIII-1-4-7/W5, 16.
10551075. Burochin, J. P., Vallet, B., Brdif, M., Mallet, C., Brosset, T., & Paparoditis, N. (2014).
Bahu, J. M., Koch, A., Kremers, E., & Murshed, S. M. (2013). Towards a 3D spatial urban Detecting blind building faades from highly overlapping wide angle aerial imagery.
energy modelling approach. ISPRS Ann. Photogramm. Remote Sensing and Spatial ISPRS Journal of Photogrammetry and Remote Sensing, 96, 193209.
Information Sciences II-2/W1 (pp. 3341). Buyuksalih, I., Isikdag, U., & Zlatanova, S. (2013). Exploring the processes of generating LOD
Baig, S. U., & Abdul-Rahman, A. (2013). Generalization of buildings within the framework (02) CityGML models in greater municipality of Istanbul. The International Archives of
of CityGML. Geo-spatial Information Science, 16, 247255. the Photogrammetry, Remote Sensing and Spatial Information Sciences, XL-2/W2, 1924.
Batty, M., Chapman, D., Evans, S., Haklay, M., Kueppers, S., Shiode, N., ... Torrens, P. M. (2000). Catita, C., Redweik, P., Pereira, J., & Brito, M. C. (2014). Extending solar potential analysis in
Visualizing the city: communicating urban design to planners and decision-makers. buildings to vertical facades. Computers & Geosciences, 66, 112.
Technical report paper 26. London, United Kingdom: University College London. Chen, M. (2013). Comparison of 3D tree parameters. (Master's thesis) Wageningen, The
Becker, S. (2009). Generation and application of rules for quality dependent faade recon- Netherlands: Wageningen University and Research Centre.
struction. ISPRS Journal of Photogrammetry and Remote Sensing, 64, 640653. Chen, L. C., Wu, C. H., Shen, T. S., & Chou, C. C. (2014). The application of geometric
Becker, S. (2011). Towards complete LOD3 models Automatic interpretation of building network models and building information models in geospatial environments for
structures. In D. Fritsch (Ed.), Proceedings of the 53rd photogrammetric week 11 re-ghting simulations. Computers, Environment and Urban Systems, 45, 112.
(pp. 3956). Chinese Ministry of Housing and Urban-Rural Development (2010). Technical specica-
Ben Fekih Fradj, N., & Lwner, M. O. (2012). Abschtzung des nutzbaren Dachchenanteils tion for three dimensional city modeling. PRC Industry Standard 82765. China Building
fr Solarenergie mit CityGML-Gebudemodellen und Luftbildern. In M. O. Lwner, F. Industry Press.
Hillen, & R. Wohlfahrt (Eds.), Geoinformatik 2012 Mobilitt und Umwelt. Chwieduk, D. A. (2009). Recommendation on modelling of solar energy incident on a
Konferenzband zur Tagung Geoinformatik, Braunschweig, Germany. (pp. 171177). building envelope. Renewable Energy, 34, 736741.
Benner, J., Geiger, A., Grger, G., Hfele, K. H., & Lwner, M. O. (2013). Enhanced LOD con- Cignoni, P., Corsini, M., & Ranzuglia, G. (2008). MeshLab: an open-source 3D mesh
cepts for virtual 3D city models. ISPRS Annals of the Photogrammetry, Remote Sensing processing system. Ercim News, 73, 4546.
and Spatial Information Sciences II-2/W1 (pp. 5161). Clark, J. H. (1976). Hierarchical geometric models for visible surface algorithms.
Benner, J., Geiger, A., Hfele, K.H., 2010. Concept for building licensing based on standard- Communications of the ACM, 19, 547554.
ized 3d geo information. The International Archives of the Photogrammetry, Remote ltekin, A., & Reichenbacher, T. (2011). High quality geographic services and bandwidth
Sensing and Spatial Information Sciences, 912 (XXXVIII-4/W15). limitations. Future Internet, 3, 379396.
van Berlo, L. A. H. M., & Bomhof, F. (2014). Creating the Dutch national BIM levels of Commandeur, T. (2012). Footprint decomposition combined with point cloud segmentation
development. 2014 International Conference on Computing in Civil and Building for producing valid 3D models. (Master's thesis) Delft University of Technology.
Engineering (pp. 129136). Orlando, FL, United States: American Society of Coors, V. (2003). 3D-GIS in networking environments. Computers, Environment and Urban
Civil Engineers. Systems, 27, 345357.
Besuievsky, G., Barroso, S., Beckers, B., & Patow, G. (2014). A congurable LoD for Coors, V., & Flick, S. (1998). Integrating levels of detail in a web-based 3D-GIS. In R.
procedural urban models intended for daylight simulation. In G. Besuievsky, & V. Laurini, K. Makki, & N. Pissinou (Eds.), Proceedings of the 6th ACM international
Tourre (Eds.), Eurographics workshop on urban data modelling and visualisation symposium on advances in geographic information systems (pp. 4045). Washington,
(pp. 1924). Strasbourg, France: The Eurographics Association. DC, USA: ACM.
Biljecki, Z. (2007). Concept and implementation of Croatian Topographic Information Coors, V., & Wagner, D. (2015). CityGML quality interoperability experiment des OGC.
SystemCROTIS. (Ph.D. thesis) Institut fr Photogrammetrie und Fernerkundung, DGPF Tagungsband. Publikationen der Deutschen Gesellschaft fr Photogrammetrie,
Technische Universitt Wien. Fernerkundung und Geoinformation e.V., Vol. 24. (pp. 288295).
Biljecki, F., & Arroyo Ohori, K. (2015). Automatic semantic-preserving conversion be- CyberCity 3D (2013). 3D modeling. (URL:) http://www.cybercity3d.com/
tween OBJ and CityGML. Eurographics workshop on urban data modelling and visuali- De Berg, M., & Dobrindt, K. T. G. (1998). On levels of detail in terrains. Graphical Models
sation 2015, Delft, Netherlands (pp. 2530). and Image Processing, 60, 112.
Biljecki, F., Heuvelink, G. B. M., Ledoux, H., & Stoter, J. (2015a). Propagation of positional Demir, N., & Baltsavias, E. P. (2012). Automated modeling of 3D building roofs using
error in 3D GIS: estimation of the solar irradiation of building roofs. International image and LiDAR data. ISPRS Annals of the Photogrammetry, Remote Sensing and
Journal of Geographical Information Science, 29, 22692294. Spatial Information Sciences, I-4, 3540.
Biljecki, F., Ledoux, H., & Stoter, J. (2014a). Error propagation in the computation of Deng, Y., Cheng, J. C. P., & Anumba, C. (2016). Mapping between BIM and 3D GIS in differ-
volumes in 3D city models with the Monte Carlo method. ISPRS Annals of the ent levels of detail using schema mediation and instance comparison. Automation in
Photogrammetry, Remote Sensing and Spatial Information Sciences II-2 Construction, 67, 121.
(pp. 3139). Diakit, A. A., Damiand, G., & Van Maercke, D. (2014). Topological reconstruction of com-
Biljecki, F., Ledoux, H., Stoter, J., & Vosselman, G. (2016). The variants of an LOD of a 3D plex 3D buildings and automatic extraction of levels of detail. In G. Besuievsky, & V.
building model and their inuence on spatial analyses. ISPRS Journal of Tourre (Eds.), Eurographics workshop on urban data modelling and visualisation, Stras-
Photogrammetry and Remote Sensing, 116, 4254. bourg, France (pp. 2530).
Biljecki, F., Ledoux, H., Stoter, J., & Zhao, J. (2014b). Formalisation of the level of detail in Dllner, J. (2005). Continuous level-of-detail modeling of buildings in 3D city
3D city modelling. Computers, Environment and Urban Systems, 48, 115. models. In C. Shahabi, & O. Boucelma (Eds.), GIS '05 Proceedings of the 13th annual
Biljecki, F., Stoter, J., Ledoux, H., Zlatanova, S., & ltekin, A. (2015b). Applications of 3D ACM international workshop on geographic information systems, Bremen, Germany
city models: State of the art review. ISPRS International Journal of Geo-Information, 4, (pp. 173181).
28422889. Dllner, J., Kolbe, T. H., Liecke, F., Sgouros, T., & Teichmann, K. (2006). The virtual 3D city
Billen, R., Zaki, C., Servires, M., Moreau, G., & Hallot, P. (2012). Developing an ontology of model of Berlin Managing, integrating, and communicating complex urban
space: Application to 3D city modeling. In T. Leduc, G. Moreau, & R. Billen (Eds.), information. Proceedings of 25th Urban Data Management Symposium (UDMS 2006),
Usage, usability, and utility of 3D city models European COST action TU0801. Nantes, Aalborg, Denmark (pp. 112).
France: EDP Sciences. Donkers, S., Ledoux, H., Zhao, J., Stoter, J., 2016. Automatic conversion of IFC datasets to
Blaauboer, J., Goos, J., Ledoux, H., Penninga, F., Reuvers, M., Stoter, J., & Vosselman, G. geometrically and semantically correct CityGML LOD3 buildings. Transactions in GIS
(2013). Technical specications for the reconstruction of 3D IMGeo CityGML data. (in press), http://dx.doi.org/10.1111/tgis.12162.
Technical report. Apeldoorn, The Netherlands: Kadaster (Version 2.02. Available on- Eicker, U., Nouvel, R., Duminil, E., & Coors, V. (2014). Assessing passive and active solar
line at http://www.geonovum.nl/sites/default/les/3DFinalReport_2013_2.02_0pdf. energy resources in cities using 3D city models. Energy Procedia, 57, 896905.
Last accessed on 12 February 2016). El Meouche, R., Rezoug, M., Hijazi, I., & Maes, D. (2013). Automatic reconstruction of 3D
Blom, ASA (2011). Blom3D whitepaper for Blom partners, clients and developers. Techni- building models from terrestrial laser scanner data. ISPRS Annals of the
cal report (Available online at http://blomasa.com/ftp/products/bis/Blom3D Photogrammetry, Remote Sensing and Spatial Information Sciences, II-4-W1, 712.
(accessed on 12 February 2016. Oslo, Norway)). Ellul, C., & Altenbuchner, J. (2013). LOD 1 VS. LOD 2 Preliminary investigations into
Boeters, R., Arroyo Ohori, K., Biljecki, F., & Zlatanova, S. (2015). Automatically enhancing differences in mobile rendering performance. ISPRS Annals of the Photogrammetry,
CityGML LOD2 models with a corresponding indoor geometry. International Journal Remote Sensing and Spatial Information Sciences, II-2/W1, 129138.
of Geographical Information Science, 29, 22482268. El-Mekawy, M., stman, A., & Shahzad, K. (2011). Towards interoperating CityGML and
Borrmann, A., Flurl, M., Jubierre, J. R., Mundani, R. P., & Rank, E. (2014). Synchronous col- IFC building models: A unied model based approach. In T. H. Kolbe, G. Knig, & C.
laborative tunnel design based on consistency-preserving multi-scale models. Nagel (Eds.), Advances in 3D geo-information sciences (pp. 7393). Berlin, Heidelberg:
Advanced Engineering Informatics, 28, 499517. Springer Berlin Heidelberg.
 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 35

Emem, O., & Batuk, F. (2004). Generating precise and accurate 3D CITY MODELS USING Henn, A., Rmer, C., Grger, G., & Plmer, L. (2012). Automatic classication of building
PHOTOGRAMMETRIC DATA. The International Archives of the Photogrammetry, types in 3D city models. GeoInformatica, 16, 281306.
Remote Sensing and Spatial Information Sciences, XXXV(B4), 431436. Hildebrandt, D., & Timm, R. (2014). An assisting, constrained 3D navigation technique for
Fan, H., & Meng, L. (2012). A three-step approach of simplifying 3D buildings multiscale virtual 3D city models. GeoInformatica, 18, 537567.
modeled by CityGML. International Journal of Geographical Information Science, van der Hoeven, F., & Wandl, A. (2015). Hotterdam: How space is making Rotterdam warmer,
26, 10911107. how this affects the health of its inhabitants, and what can be done about it. Delft, The
Fan, H., Zipf, A., Fu, Q., & Neis, P. (2014). Quality assessment for building footprints data on Netherlands: TU Delft.
OpenStreetMap. International Journal of Geographical Information Science, 28, Hoerka, J., & Zlocha, M. (2012). A new 3-D solar radiation model for 3-D city models.
700719. Transactions in GIS, 16, 681690.
Fath, K., Stengel, J., Sprenger, W., Wilson, H. R., Schultmann, F., & Kuhn, T. E. (2015). A meth- Hsieh, C. M., Aramaki, T., & Hanaki, K. (2011). Managing heat rejected from air condition-
od for predicting the economic potential of (building-integrated) photovoltaics in urban ing systems to save energy and improve the microclimates of residential buildings.
areas based on hourly radiance simulations. Solar Energy, 116, 357370. Computers, Environment and Urban Systems, 35, 358367.
Fisher-Gewirtzman, D., Shashkov, A., & Doytsher, Y. (2013). Voxel based volumetric visibility Huang, H., Brenner, C., & Sester, M. (2011). 3D building roof reconstruction from point
analysis of urban environments. Survey Review, 45, 451461. clouds via generative models. In D. Agrawal, I. Cruz, C. S. Jensen, E. Ofek, & E. Tanin
Forberg, A. (2007). Generalization of 3D building data based on a scale-space approach. (Eds.), Proceedings of the 19th ACM SIGSPATIAL International Conference on Advances
ISPRS Journal of Photogrammetry and Remote Sensing, 62, 104111. in Geographic Information Systems (pp. 1624). Chicago, IL, United States: ACM Press.
Frstner, W. (1999). 3D-city models: Automatic and semiautomatic acquisition methods. Huang, H. C., Lo, S. M., Zhi, G. S., & Yuen, R. K. K. (2008). Graph theory-based approach for
In D. Fritsch, & R. H. Spiller (Eds.), Proceedings of the 47th Photogrammetric Week '99, automatic recognition of CAD data. Engineering Applications of Articial Intelligence,
Stuttgart, Germany (pp. 291303). 21, 10731079.
Frani, S., Bai-Deprato, I., & Novakovi, I. (2009). 3D model and a scale model of the City Ioannidis, C., Verykokou, S., Soile, S., & Potsiou, C. (2015). 5D multi-purpose land information
of Zagreb. The International Archives of the Photogrammetry, Remote Sensing and system. Eurographics Workshop on Urban Data Modelling and Visualisation, Delft,
Spatial Information Sciences, XXXVIII-2/W11, 17. Netherlands (pp. 1924).
Freitas, S., Catita, C., Redweik, P., & Brito, M. C. (2015). Modelling solar potential in the Isikdag, U., & Zlatanova, S. (2009). Towards dening a framework for automatic genera-
urban environment: State-of-the-art review. Renewable and Sustainable Energy tion of buildings in CityGML using building information models. 3D Geo-Information
Reviews, 41, 915931. Sciences (pp. 7996). Heidelberg: Springer Berlin.
Frommholz, D., Linkiewicz, M., Meissner, H., Dahlke, D., & Poznanska, A. (2015). Kaartinen, H., Hyypp, J., Kukko, A., Jaakkola, A., & Hyypp, H. (2012). Benchmarking the
Extracting semantically annotated 3D building models with textures from oblique performance of mobile laser scanning systems using a permanent test eld. Sensors,
aerial imagery. The International Archives of the Photogrammetry, Remote Sensing and 12, 1281412835.
Spatial Information Sciences, XL-3/W2, 5358. Kada, M., & McKinley, L. (2009). 3D building reconstruction from LiDAR based on a cell
Garnett, R., & Freeburn, J. T. (2014). Visual acceptance of library-generated CityGML LOD3 decomposition approach. The International Archives of the Photogrammetry, Remote
building models. Cartographica: The International Journal for Geographic Information Sensing and Spatial Information Sciences, XXXVIII-3(W4), 4752.
and Geovisualization, 49, 218224. Kaden, R., & Kolbe, T. H. (2014). Simulation-based total energy demand estimation of
Geiger, A., Benner, J., & Haefele, K. H. (2015). Generalization of 3D IFC building models. 3D buildings using semantic 3D city models. International Journal of 3-D Information
Geoinformation Science (pp. 1935). Dubai, UAE: Springer International Publishing. Modeling, 3, 3553.
Gesquire, G., & Manin, A. (2012). 3D visualization of urban data based on CityGML with Kang, H. Y., & Lee, J. (2014). A study on the LOD (level of detail) model for applications
WebGL. International Journal of 3-D Information Modeling, 1, 115. based on indoor space data. Journal of the Korean Society of Surveying, Geodesy,
Gimenez, L., Hippolyte, J. L., Robert, S., Suard, F., & Zreik, K. (2015). Review: Reconstruc- Photogrammetry and Cartography, 32, 143151.
tion of 3D building information models from 2D scanned plans. Journal of Building Kauk, J., Gallay, M., & Hoerka, J. (2015). Generating time series of virtual 3-D city
Engineering, 2, 2435. models using a retrospective approach. Landscape and Urban Planning, 139, 4053.
Glander, T., & Dllner, J. (2009). Abstract representations for interactive visualization of Kartverket (2014). SOSI Del 3 Produktspesikasjon for Felles KartdataBase (FKB). Product
virtual 3D city models. Computers, Environment and Urban Systems, 33, 375387. specication by Kartverket (The Norwegian Mapping Authority). (Available online
Glasgow City Council (2009). Urban model specication. Glasgow City Council. at) http://kartverket.no/globalassets/standard/sosi-kap3-produktspesikasjoner/
Development and regeneration services. (Available online at) https://www. fkb-4.5/0-generelldel-2014-03-01.pdf (last accessed on 12 February 2016)
glasgow.gov.uk/urbanmodel (last accessed on 12 February 2016) Kedzierski, M., & Fryskowska, A. (2014). Terrestrial and aerial laser scanning data integra-
Goetz, M. (2013). Towards generating highly detailed 3D CityGML models from tion using wavelet analysis for the purpose of 3D building modeling. Sensors, 14,
OpenStreetMap. International Journal of Geographical Information Science, 27, 1207012092.
845865. Kelly, T., & Wonka, P. (2011). Interactive architectural modeling with procedural
van Gool, L., & Martinovi, A. (2013). Towards semantic city models. In D. Fritsch (Ed.), extrusions. ACM Transactions on Graphics, 30, 115.
Proceedings of the 54th Photogrammetric Week 13, Stuttgart, Germany (pp. 217232). Kemec, S., Zlatanova, S., & Duzgun, S. (2012). A new LoD denition hierarchy for 3D city
Gtzelmann, T., Guercke, R., Brenner, C., & Sester, M. (2009). Terrain-dependent aggregation models used for natural disaster risk communication tool. In T. Bandrova, & C. Knig
of 3D city models. The International Archives of the Photogrammetry, Remote Sensing and (Eds.), The 1st International Conference on Information Science and Engineering
Spatial Information Sciences, XXXVIII-2(W11), 5. (ICISE2009), Albena, Bulgaria (pp. 95104).
Grger, G., & Coors, V. (2011). Modeling guide for 3D objects. SIG 3D. (Available online Kim, J. S., Yoo, S. J., & Li, K. J. (2014). Integrating IndoorGML and CityGML for indoor space.
at) http://en.wiki.quality.sig3d.org (last accessed on 12 February 2016) In D. Pfoser, & K. J. Li (Eds.), Web and wireless geographical information systems. Lecture
Grger, G., & Plmer, L. (2012). CityGML Interoperable semantic 3D city models. ISPRS notes in computer science proceedings of the 13th International Symposium (W2GIS
Journal of Photogrammetry and Remote Sensing, 71, 1233. 2014) (pp. 184196).
Guercke, R., Gtzelmann, T., Brenner, C., & Sester, M. (2011). Aggregation of LoD 1 build- Kolbe, T. H. (2009). Representing and exchanging 3D city models with CityGML. In S.
ing models as an optimization problem. ISPRS Journal of Photogrammetry and Remote Zlatanova, & J. Lee (Eds.), 3D geo-information sciences (pp. 1531). Heidelberg: Springer
Sensing, 66, 209222. Berlin.
Guskov, I., & Brewington, B. (2015). United States Patent Application US20150187130A1: Au- Kolbe, T. H., Burger, B., & Cantzler, B. (2015). CityGML goes to broadway. Photogrammetric
tomatic generation of 2.5D extruded polygons from full 3D models. (Available online at) Week 15, Stuttgart, Germany (pp. 343356).
http://www.google.com/patents/US20150187130 (last accessed on 12 February 2016) Kninger, A., & Bartel, S. (1998). 3D-GIS for urban purposes. GeoInformatica, 2, 79103.
Haala, N., & Brenner, C. (1999). Virtual city models from laser altimeter and 2D map data. Kumar, K., Saran, S., Kumar, A.S., 2016. CityGML based Interoperability for the transforma-
Photogrammetric Engineering and Remote Sensing, 65, 787795. tion of 3D data models. Transactions in GIS (in press).
Haala, N., & Kada, M. (2010). An update on automatic 3D building reconstruction. ISPRS de Laat, R., & van Berlo, L. (2011). Integration of BIM and GIS: The development of the
Journal of Photogrammetry and Remote Sensing, 65, 570580. CityGML GeoBIM extension. Advances in 3D geo-information sciences (pp. 211225).
Habib, A. F., Zhai, R., & Kim, C. (2010). Generation of complex polyhedral building models Heidelberg: Springer Berlin.
by integrating stereo-aerial imagery and Lidar data. Photogrammetric Engineering and Ledoux, H., & Meijers, M. (2011). Topologically consistent 3D city models obtained by ex-
Remote Sensing, 76, 609623. trusion. International Journal of Geographical Information Science, 25, 557574.
Hfele, K. H. (2011). CityGML model of the FJK-Haus. Dataset and specication. Karlsruhe In- Lee, D., Pietrzyk, P., Donkers, S., Liem, V., van Oostveen, J., Montazeri, S., ... Verbree, E. (2013).
stitute of Technology (Published online at http://www.iai.kit.edu/www-extern-kit/ Modeling and observation of heat losses from buildings: The impact of geometric detail
leadmin/download/download-vrsys/FJK-Haus-Lod1-LoD4-V2.pdf, last accessed on on 3D heat ux modeling. Proceedings of the 33rd European Association of Remote Sensing
12 February 2016). Laboratories (EARSeL) Symposium, Matera, Italy (pp. 353372).
Hagedorn, B., Trapp, M., Glander, T., & Dllner, J. (2009). Towards an indoor level-of-detail Leszek, K. (2015). Environmental and urban spatial analysis based on a 3D city model.
model for route visualization. 10th International Conference on Mobile Data Manage- Computational science and its applications ICCSA 2015 (pp. 633645). Springer Inter-
ment: Systems, Services and Middleware (pp. 692697). Taipei, Taiwan: IEEE. national Publishing.
He, S., Besuievsky, G., Tourre, V., Patow, G., & Moreau, G. (2012). All range and heteroge- Lewis, R., & Squin, C. (1998). Generation of 3D building models from 2D architectural
neous multi-scale 3D city models. In T. Leduc, G. Moreau, & R. Billen (Eds.), Usage, usabil- plans. Computer-Aided Design, 30, 765779.
ity, and utility of 3D city models European COST action TU0801. Nantes, France: EDP Li, Q., Sun, X., Yang, B., & Jiang, S. (2013). Geometric structure simplication of 3D building
Sciences. models. ISPRS Journal of Photogrammetry and Remote Sensing, 84, 100113.
He, S., Moreau, G., & Martin, J. Y. (2013). Footprint-based generalization of 3D building Li, Z., Zhang, Z., & Davey, K. (2015). Estimating geographical PV potential using LiDAR data
groups at medium level of detail for multi-scale urban visualization. International for buildings in downtown San Francisco. Transactions in GIS, 19, 930963.
Journal on Advances in Software, 5, 378388. Lwner, M. O., & Grger, G. (2016). Evaluation criteria for recent LoD proposals for CityGML
Henn, A., Grger, G., Stroh, V., & Plmer, L. (2013). Model driven reconstruction of roofs from buildings. Photogrammetrie - Fernerkundung - Geoinformation 2016 (pp. 3143).
sparse LIDAR point clouds. ISPRS Journal of Photogrammetry and Remote Sensing, 76, Lwner, M. O., Benner, J., Grger, G., & Hfele, K. H. (2013). New concepts for structuring
1729. 3D city models An extended level of detail concept for CityGML buildings. In B.
36 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537

Murgante, S. Misra, M. Carlini, C. M. Torre, H. Q. Nguyen, D. Taniar, B. O. Apduhan, & O. Sanborn (2014). 3D cities. Product brochure. (Available online at) http://www.sanborn.
Gervasi (Eds.), Spatial information theory. Cognitive and computational foundations of com/3d-cities/ (last accessed on 12 February 2016)
geographic information science (pp. 466480). Heidelberg: Springer Berlin. Saran, S., Wate, P., Srivastav, S. K., & Krishna Murthy, Y. V. N. (2015). CityGML at semantic
Luebke, D., Reddy, M., Cohen, J. D., Varshney, A., Watson, B., & Huebner, R. (2003). Level of level for urban energy conservation strategies. Annals of GIS, 21, 2741.
detail for 3D graphics. San Francisco: Morgan Kaufmann Pub. Sargent, I., Holland, D., & Harding, J. (2015). The building blocks of user-focused 3D city
Lwin, K., & Murayama, Y. (2009). A GIS approach to estimation of building population for models. ISPRS International Journal of Geo-Information, 4, 28902904.
micro-spatial analysis. Transactions in GIS, 13, 401414. Schilcher, M., Roschlaub, R., & Guo, Z. (1998). Vom 2D-GIS zum 3D-Stadtmodell durch
Machl, T. (2013). Minutes of the International OGC, SIG 3D and TUM workshop on require- Kombination von GIS-, CAD- und Animationstechniken. Tagungsband ACS 98,
ments for CityGML 3.0. International OGC, SIG 3D and TUM workshop on requirements for Fachseminar Geoinformationssysteme, Frankfurt, Germany (pp. 12).
CityGML 3.0 (pp. 128). Munich, Germany: Technische Universitt Munchen. Schilling, A., Hagedorn, B., & Coors, V. (2012). Final report of the OGC 3D portrayal
Mao, B., Harrie, L., & Ban, Y. (2012). Detection and typication of linear structures for dy- interoperability experiment. Technical report 12075. Open Geospatial Consortium.
namic visualization of 3D city models. Computers, Environment and Urban Systems, 36, Schwalbe, E., Maas, H. G., & Seidel, F. (2005). 3D building model generation from airborne
233244. laser scanner data using 2D GIS data and orthogonal point cloud projections. The
Maragkogiannis, K., Kolokotsa, D., Maravelakis, E., & Konstantaras, A. (2014). Combining International Archives of the Photogrammetry, Remote Sensing and Spatial Information
terrestrial laser scanning and computational uid dynamics for the study of the Sciences, XXXVI(3/W19), 209214.
urban thermal environment. Sustainable Cities and Society, 13, 207216. Shewchuk, J. R. (1996). Triangle: Engineering a 2D quality mesh generator and Delaunay
Martinovi, A., Knopp, J., Riemenschneider, H., & van Gool, L. (2015). 3D all the way: Se- triangulator. Applied computational geometry towards geometric engineering
mantic segmentation of urban scenes from start to end in 3D. CVPR 2015: The 28th (pp. 203222). Heidelberg: Springer Berlin.
IEEE Conference on Computer Vision and Pattern Recognition (pp. 44564465). Boston, Sirmacek, B., & Lindenbergh, R. (2014). Accuracy assessment of building point clouds
United States: IEEE. automatically generated from iphone images. The International Archives of the
Mayer, H. (2005). Scale-spaces for generalization of 3D buildings. International Journal of Photogrammetry, Remote Sensing and Spatial Information Sciences, XL-5, 547552.
Geographical Information Science, 19, 975997. Smelik, R. M., Tutenel, T., Bidarra, R., & Benes, B. (2014). A survey on procedural modelling
Meinel, G., Hecht, R., & Herold, H. (2009). Analyzing building stock using topographic for virtual worlds. Computer Graphics Forum, 33, 3150.
maps and GIS. Building Research and Information, 37, 468482. Stadler, A., & Kolbe, T. H. (2007). Spatio-semantic coherence in the integration of 3D city
Meng, L., & Forberg, A. (2007). 3D building generalisation. In W. Mackaness, A. Ruas, & T. models. The International Archives of the Photogrammetry, Remote Sensing and Spatial
Sarjakoski (Eds.), Challenges in the portrayal of geographic information: Issues of general- Information Sciences, XXXVI-2(C43), 8.
isation and multi scale representation (pp. 211232). Amsterdam, The Netherlands: Steinhage, V., Behley, J., Meisel, S., & Cremers, A. B. (2010). Automated updating and
Elsevier Science. maintenance of 3D city models. The International Archives of the Photogrammetry,
Mller Arisona, S., Zhong, C., Huang, X., & Qin, H. (2013). Increasing detail of 3D models Remote Sensing and Spatial Information Sciences, XXXVIII(48-2/W9), 16.
through combined photogrammetric and procedural modelling. Geo-spatial Informa- Stoter, J., de Kluijver, H., & Kurakula, V. (2008). 3D noise mapping in urban areas.
tion Science, 16, 4553. International Journal of Geographical Information Science, 22, 907924.
Mller, P., Wonka, P., Haegler, S., Ulmer, A., & van Gool, L. (2006). Procedural modeling of Stoter, J., Roensdorf, C., Home, R., Capstick, D., Streilein, A., Kellenberger, T., ... Ilves, R.
buildings. ACM Transactions on Graphics, 25, 614623. (2015). 3D modelling with national coverage: Bridging the gap between research
Musialski, P., Wonka, P., Aliaga, D. G., Wimmer, M., van Gool, L., & Purgathofer, W. and practice. Advances in 3D geo-information sciences (pp. 207225). Cham,
(2013). A survey of urban reconstruction. Computer Graphics Forum, 32, Switzerland: Springer International Publishing.
146177. Stoter, J., Vosselman, G., Dahmen, C., Oude Elberink, S., & Ledoux, H. (2014). CityGML
NAVTEQ (2011). 3D landmarks A product guide for developers. NAVTEQ. implementation specications for a countrywide 3D data set: The case of The
Nex, F., & Remondino, F. (2013). UAV for 3D mapping applications: a review. Applied Netherlands. Photogrammetric Engineering and Remote Sensing, 80, 1321.
Geomatics, 6, 115. Stoter, J., Vosselman, G., Goos, J., Zlatanova, S., Verbree, E., Klooster, R., & Reuvers, M.
Noskov, A., & Doytsher, Y. (2014). Preparing simplied 3D scenes of multiple LODs of (2011). Towards a national 3D spatial data infrastructure: Case of The Netherlands.
buildings in urban areas based on a raster approach and information theory. Thematic Photogrammetrie - Fernerkundung - Geoinformation, 2011, 405420.
cartography for the society (pp. 221236). Springer. Strzalka, A., Alam, N., Duminil, E., Coors, V., & Eicker, U. (2012). Large scale integration of
Nouvel, R., Mastrucci, A., Leopold, U., Baume, O., Coors, V., & Eicker, U. (2015). Combining photovoltaics in cities. Applied Energy, 93, 413421.
GIS-based statistical and engineering urban heat consumption models: Towards a Strzalka, A., Bogdahn, J., Coors, V., & Eicker, U. (2011). 3D city modeling for urban scale
new framework for multi-scale policy support. Energy and Buildings, 107, 204212. heating energy demand forecasting. HVAC&R Research, 17, 526539.
Nouvel, R., Schulte, C., Eicker, U., Pietruschka, D., & Coors, V. (2013). CityGML-based 3D Surez, J. P., Trujillo, A., Santana, J. M., de la Calle, M., & Gmez-Deck, D. (2015). An
city model for energy diagnostics and urban energy policy support. Proceedings of efcient terrain level of detail implementation for mobile devices and performance
BS2013: 13th Conference of International Building Performance Simulation Association, study. Computers, Environment and Urban Systems, 52, 2133.
Chambry, France (pp. 218225). Suveg, I., & Vosselman, G. (2004). Reconstruction of 3D building models from aerial
Novakovi, I. (2011). 3D model of Zagreb. GIM International, 25, 2529. images and maps. ISPRS Journal of Photogrammetry and Remote Sensing, 58,
Open Geospatial Consortium (2012). OGC city geography markup language (CityGML) 202224.
encoding standard 2.0.0. Technical Report. SwissTopo (2010). swissBUILDINGS3D 1.0. Vereinfachte 3D-Gebude der Schweiz. Prod-
Ordnance Survey (2014). OS MasterMap topography layer Building height attribute. uct brochure and documentation.
Getting started guide (1.0 ed.). SwissTopo (2014). Produktinformation swissBUILDINGS3D 2.0. Product brochure and
Oude Elberink, S. (2010). Acquisition of 3D topography: automated 3D road and building documentation.
reconstruction using airborne laser scanner data and topographic maps. (Ph.D. thesis) Thiemann, F. (2003). 3D-Gebude-Generalisierung. In W. G. Koch (Ed.), Theorie 2003 -
Enschede, The Netherlands: ITC, University of Twente. Vortrge der Dresdner Sommerschule fr Kartographie (pp. 5261) (ikg.uni-hanno-
Oude Elberink, S., Stoter, J., Ledoux, H., & Commandeur, T. (2013). Generation and dissem- ver.de, Dresden, Germany).
ination of a national virtual 3D city and landscape model for The Netherlands. Tolmer, C. E., Castaing, C., Diab, Y., & Morand, D. (2013). CityGML and IFC: Going further
Photogrammetric Engineering and Remote Sensing, 79, 147158. than LOD. Digital Heritage International Congress DigitalHeritage (pp. 645648). Mar-
Over, M., Schilling, A., Neubauer, S., & Zipf, A. (2010). Generating web-based 3D city models seille, France: IEEE.
from OpenStreetMap: The current situation in Germany. Computers, Environment and Tolmer, C. E., Castaing, C., Morand, D., & Diab, Y. (2014). Structuration des informations
Urban Systems, 34, 496507. pour les projets d'infrastructures. Proposition de niveaux complmentaires aux
Perez, D., Kmpf, J. H., & Scartezzini, J. L. (2013). Urban area energy ow microsimulation Level Of Detail et Level Of Development. Proceedings of the Confrence SCAN14,
for planning support: A calibration and verication study. International Journal on Luxembourg (pp. 110) (Luxembourg).
Advances in Systems and Measurements, 6, 260271. Touya, G., & Brando-Escobar, C. (2013). Detecting level-of-detail inconsistencies in
Portals, C., Lerma, J. L., & Navarro, S. (2010). Augmented reality and photogrammetry: A volunteered geographic information data sets. Cartographica: The International
synergy to visualize physical and virtual city environments. ISPRS Journal of Journal for Geographic Information and Geovisualization, 48, 134143.
Photogrammetry and Remote Sensing, 65, 134142. Touya, G., & Reimer, A. (2015). Inferring the scale of OpenStreetMap features. In J. J.
Prandi, F., De Amicis, R., PIffer, S., Soave, M., Cadzow, S., Boix, E. G., & D'Hondt, E. (2013). Arsanjani, A. Zipf, P. Mooney, & M. Helbich (Eds.), OpenStreetMap in GIScience
Using CityGML to deploy smart-city services for urban ecosystems. The International (pp. 8199). Springer International Publishing.
Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XL-4/ Truong-Hong, L., & Laefer, D. F. (2015). Quantitative evaluation strategies for urban
W1, 8792. 3D model generation from remote sensing data. Computers and Graphics, 49,
Prechtel, N. (2014). On strategies and automation in upgrading 2D to 3D landscape 8291.
representations. Cartography and Geographic Information Science, 42, 244258. Tsiliakou, E., Labropoulos, T., & Dimopoulou, E. (2014). Procedural modeling in 3D GIS en-
Previtali, M., Barazzetti, L., Brumana, R., Cuca, B., Oreni, D., Roncoroni, F., & Scaioni, M. vironment. International Journal of 3-D Information Modeling, 3, 1734.
(2014). Automatic faade modelling using point cloud data for energy-efcient Vande Velde, L. (2005). Tele Atlas 3D navigable maps. In G. Grger, & T. H. Kolbe (Eds.),
retrotting. Applied Geomatics, 6, 95113. EuroSDR International Workshop on Next Generation 3D City Models (pp. 4750).
Qin, R. (2014). Change detection on LOD 2 building models with very high resolution Varduhn, V., Mundani, R. P., & Rank, E. (2015). Multi-resolution models: Recent progress
spaceborne stereo imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 96, in coupling 3D geometry to environmental numerical simulation. 3D geoinformation
179192. science (pp. 5569). Springer International Publishing.
Rau, J. Y., & Cheng, C. K. (2013). A cost-effective strategy for multi-scale photo-realistic Verdie, Y., Lafarge, F., & Alliez, P. (2015). LOD generation for urban scenes. ACM
building modeling and web-based 3-D GIS applications in real estate. Computers, Transactions on Graphics, 34, 114.
Environment and Urban Systems, 38, 3544. Verma, V., Kumar, R., & Hsu, S. (2006). 3D building detection and modeling from aerial
Rosser, J., Morley, J., & Smith, G. (2015). Modelling of building interiors with mobile LIDAR data. Proceedings of the IEEE Computer Society Conference on Computer Vision
phone sensor data. ISPRS International Journal of Geo-Information, 4, 9891012. and Pattern Recognition (CVPR06) (pp. 22132220). IEEE.
 F. Biljecki et al. / Computers, Environment and Urban Systems 59 (2016) 2537 37

Vertex Modelling (2013). Wide Area Models. (URL:) http://www.vertexmodelling.co.uk/ Yaagoubi, R., Yarmani, M., Kamel, A., & Khemiri, W. (2015). HybVOR: A Voronoi-based 3D
site/about/wide-area-models/ GIS approach for camera surveillance network placement. ISPRS International Journal
Vosselman, G., & Dijkman, S. (2001). 3D building model reconstruction from point clouds of Geo-Information, 4, 754782.
and ground plans. The International Archives of the Photogrammetry, Remote Sensing Yang, P. P., Putra, S. Y., & Li, W. (2007). Viewsphere: A GIS-based 3D visibility analysis for
and Spatial Information Sciences, XXXIV(3/W4), 3744. urban design evaluation. Environment and Planning B: Planning and Design, 34, 971.
Wagner, D., Wewetzer, M., Bogdahn, J., Alam, N., Pries, M., & Coors, V. (2012). Geometric- Yang, L., Zhang, L., Ma, J., Xie, J., & Liu, L. (2011). Interactive visualization of multi-
semantical consistency validation of CityGML models. Progress and new trends in 3D resolution urban building models considering spatial cognition. International Journal
geoinformation sciences (pp. 171192). Berlin, Heidelberg: Springer Berlin Heidelberg. of Geographical Information Science, 25.
Wate, P., Srivastav, S. K., Saran, S., & Murthy, Y. V. N. K. (2013). Formulation of hierarchical Yin, X., Wonka, P., & Razdan, A. (2009). Generating 3D building models from architectural
framework for 3D-GIS data acquisition techniques in context of Level-of-Detail (LoD). drawings: A survey. IEEE Computer Graphics and Applications, 29, 2030.
Proceedings of the 2013 IEEE Second International Conference on Image Information Zhao, J., Zhu, Q., Du, Z., Feng, T., & Zhang, Y. (2012). Mathematical morphology-based
Processing (ICIIP-2013), Shimla, India (pp. 154159). generalization of complex 3D building models incorporating semantic relationships.
Wonka, P., Wimmer, M., Ribarsky, W., & Sillion, F. (2003). Instant architecture. ACM ISPRS Journal of Photogrammetry and Remote Sensing, 68, 95111.
Transactions on Graphics, 22, 669677. Zhu, Q., & Hu, M. Y. (2010). Semantics-based 3D dynamic hierarchical house property
Xiao, J., Fang, T., Zhao, P., Lhuillier, M., Quan, L., Xiao, J., ... Quan, L. (2009). Image-based model. International Journal of Geographical Information Science, 24, 165188.
street-side city modeling. ACM Transactions on Graphics, 28, 114. Zhu, X. X., & Shahzad, M. (2014). Facade reconstruction using multiview spaceborne
Xie, J., Zhang, L., Li, J., Wang, H., & Yang, L. (2012). Automatic simplication and visualiza- TomoSAR point clouds. IEEE Transactions on Geoscience and Remote Sensing, 52,
tion of 3D urban building models. International Journal of Applied Earth Observation 35413552.
and Geoinformation, 18, 222231.
Xiong, B., Jancosek, M., Oude Elberink, S., & Vosselman, G. (2015). Flexible building
primitives for 3D building modeling. ISPRS Journal of Photogrammetry and Remote
Sensing, 101, 275290.