International Journal of

Geo-Information

Article
GIS Based Procedural Modeling in 3D Urban Design
Ming Zhang 1,2 , Jielin Wu 1,2 , Yang Liu 1,2, *, Ji Zhang 1,2 and Guanyao Li 1,2

1 Guangzhou Urban Planning & Design Survey Research Institute, Guangzhou 510060, China
2 Guangdong Enterprise Key Laboratory for Urban Sensing, Monitoring and Early Warning,
Guangzhou 510060, China
* Correspondence: [email protected]

Abstract: Traditional urban design is time-consuming and laborious. We propose a computer-

generated architecture (CGA)-based workflow in this work, with the goal of allowing designers to
take advantage of a high level of automation. This workflow is based on procedural modeling. A
three-step CGA rule was applied to implement 3D urban procedural modeling, (1) parcel subdivision
and clustering, (2) building extrusion, and (3) texture mapping. Parcel subdivision and clustering is
the key step of layout modeling, giving the modeler flexibility to adjust the placement and size of the
inner building lots. Subsequently, a land-use-based combination of eight common building types
and layouts was used to generate various urban forms for different urban functional zones. Finally,
individual buildings were decorated by creating texture maps of a planar section of the building
facade or, alternatively, decomposing facades into sets of repeating elements and texture maps. We
employed the proposed workflow in the H-village urban redevelopment program and an air–rail
integration zone development program in Guangzhou. Three design proposals were generated for
each project. The results demonstrated that this workflow could generate multiple layout proposals
and alternative facade textures quickly and, therefore, address most of the collaborative issues with
its analysis functions, including a flexible adjustment mechanism and real-time visualization.

Keywords: procedural modeling; urban design; layout modeling; building extrusion; texture
Citation: Zhang, M.; Wu, J.; Liu, Y.; mapping; Guangzhou
Zhang, J.; Li, G. GIS Based Procedural
Modeling in 3D Urban Design. ISPRS
Int. J. Geo-Inf. 2022, 11, 531. https://
doi.org/10.3390/ijgi11100531 1. Introduction
Academic Editors: Wolfgang Kainz The main incentives for adopting generative design are to use computational capabili-
and Sisi Zlatanova ties to support human designers and automate parts of the design process [1]. However,
Received: 21 August 2022
it is less applied in urban design than in urban planning and architecture. Urban design
Accepted: 16 October 2022
“includes technical questions of urban functioning, economic issues of cost and benefit, aes-
Published: 19 October 2022
thetic issues of appearance, as well as social issues involving allocation and provision” [2],
which differentiates urban design from architecture design and urban planning. Urban de-
Publisher’s Note: MDPI stays neutral
sign generally involves multiple stakeholders, often representing conflicting requirements
with regard to jurisdictional claims in
and interests, thus intensifying the complexity of the design [3]. Hence, it is necessary for
published maps and institutional affil-
designers to come up with alternative design solutions, present in different formats, and
iations.
quickly respond to the pop-up requirements. Compared with urban planning, urban design
focuses more on design and user experience, and it operates at the feature and system level.
Bringing the designer’s concept to the non-design expert’s communicative level requires
Copyright: © 2022 by the authors.
significant application of communication media [4]. Thus, collaboration, communication,
Licensee MDPI, Basel, Switzerland. and visualization are at the heart of urban design.
This article is an open access article With advances in computer science and visualization, computer-assisted design has
distributed under the terms and been gradually replacing routine design with fully or semi-automated design procedures.
conditions of the Creative Commons However, studies on creative design issues remain elusive [5], as with collaborative issues.
Attribution (CC BY) license (https:// In urban design, the routine design issues are mainly related to the zoning (or detailed plan
creativecommons.org/licenses/by/ in China) regulation. That is to say, the design can only be created with certain functions
4.0/). and forms in the way that a detailed plan allows. The functions are defined by land use,

ISPRS Int. J. Geo-Inf. 2022, 11, 531. https://doi.org/10.3390/ijgi11100531 https://www.mdpi.com/journal/ijgi

ISPRS Int. J. Geo-Inf. 2022, 11, 531 2 of 20

while the forms are limited by parameters such as floor area ratio, density, and height in a
detailed plan. Since an individual function can be translated into several pattern languages,
which are common to all urban designers, it can theoretically be learned well by a computer,
making it possible to generate alternative patterns that follow the parameters. Even though
it is probably unrealistic to solve all the design issues with design automation, a collective
automation tool still offers extra support and advantages for human designers. The ideal
tool should consist of optimized, automated layout design methods, real-time visualization,
a mechanism to adapt the design to programmatic changes, and so on.
There have been some enlightening research works on constructing computer-assisted
frameworks and models to aid designers in urban design in recent years. These works
can be broadly divided into: (1) geometrical modeling, which focuses on algorithms that
produce intricate geometry quickly from a compact set of specifications (i.e., procedural
modeling) [6,7], and (2) behavioral modeling, the direction of which is to understand the
underlying socioeconomic, meteorological, and resource consumption/waste production
processes occurring within an urban space [8]. To achieve an optimized urban design, one
of the prerequisites is a thorough understanding of the underlying logic of the design which
relies on both geometrical modeling and behavioral modeling. It is especially true that
geometrical modeling plays a dominant role during the design stage [9–11]. The present
study puts forward a collective solution regarding procedural modeling and outlines its
application in projects to explore its future potential.
The main contribution of this study is the development of a complete interactive
workflow for semi-automated 3D urban design from the 2D representation of the urban area.
This workflow links the main urban planning parameters with the 3D urban design model
so that the urban design scheme can be changed on the fly according to the requirements
of urban planning and the ideas of urban designers. As a result, this workflow greatly
enhances collaboration amongst urban designers and significantly boosts their creativity.

2. Related Work
Our method builds upon previous work in procedural modeling [12–14]. Procedural
modeling is often used to create objects with a high degree of redundancy. It is developed
upon some production systems such as L-systems, shape grammars, and split grammars
which allow the creation of complex structures from small sets of inputs [15–17]. The L-
system was proposed by Lindermayer as a basis for geometric plant modeling [18]. Parish
and Müller [7] introduced L-systems to resemble the growth of streets. Shape grammars,
which define rules for the specification and transformation of 2D and 3D shapes, were
initially used for describing geometric shapes in artworks [19,20]. Wonka et al. [17] extended
the concept of shape grammars with split grammars by adding attributes as parameters
to the geometric shape itself. Split grammars provide an automated size-independent
approach to derive building models from a dataset of rules and attributes. Following
these novel approaches, Müller et al. [21] proposed a computer-generated architecture
(CGA) method for the generation of detailed 3D objects, particularly urban objects such
as buildings and roads. CGA has become one of the most powerful methodologies for
procedural modeling and urban design with a set of shape grammars, such as extrusion,
translation, scaling, and splitting.
A series of new methods have been further developed to improve procedural modeling
and urban design, aiming to provide an effective means for quick architecture creation.
These efforts have focused on the following four areas:
(1) Layout modeling refers to the procedural generation of parcels inside city blocks.
Several pieces of research aimed to synthesize new urban layouts by creating and/or
joining fragments of pre-existing examples [22–26]. For instance, Aliaga et al. [22] per-
formed both a structure-based synthesis and an image-based synthesis to create urban
layouts using example fragments from several real-world cities. Vanegas et al. [25]
presented skeleton-based subdivisions and oriented bounding box subdivisions to
generate spatial configurations of parcels with high similarity to those observed in
ISPRS Int. J. Geo-Inf. 2022, 11, 531 3 of 20

real-world cities. In contrast, our method generates urban layouts through a set of
procedural rules rather than reproduces or starts with existing urban layouts.
(2) Building modeling addressed the problem of generating 3D building models. Most
of the previous works focused on generating a compact, efficient, and reusable pro-
cedural representation to construct a new 3D architectural model that resembles the
original [27–31]. Aliaga et al. [27] proposed a method to construct a grammar from
photographs, enabling the rapid sketching of novel architectural structures in the
original style. Demir et al. [14] converted an architectural model into a split tree and
synthesized new geometric models with the extracted split tree. Compared to previ-
ous work, the 3D architectural model generated from our workflow is land-use-based.
That is to say, instead of reproducing a specific architectural style, we try to restore
the general building types of the different urban functional zone.
(3) Facade modeling applied a segmentation algorithm for facade reconstruction from
image data or LIDAR scanning results [32–37]. Van Gool et al. [36] discussed different
facade reconstruction algorithms and used one rule set for the reconstruction of differ-
ent kinds of buildings. Wan and Sharf [37] presented a method for finding the best
segmentation of facades through maximum likelihood formulation and then recon-
structing building facades from LIDAR scans using a grammar-based segmentation
algorithm. In this work, we seek to explore the procedural representation method of
facade modeling and texture mapping. The main difference from previous work is
that we try to implement a land-use-based texture mapping.
(4) Urban modeling explored an efficient way of urban scale modeling and
rendering [38–42]. Kuang [41] implemented highly memory-efficient modeling and
rendering of urban buildings by proposing a hierarchical grid-based modeling method
and a data structure called Non-Uniform Texture. Marvie et al. [42] introduced GPU
Shape Grammars for real-time generation, tuning, and rendering of procedural mod-
els. In this work, we applied the CGA grammars for large-scale procedural modeling
with the advantage of avoiding explicit storage of expanded geometry and delaying
the generation of buildings until the rendering stage.

3. Methods
This work proposed a series of CGA rules to provide a complete workflow of semi-
automated 3D urban design, including layout modeling, building modeling, and facade
modeling process. The project was conducted in ESRI CityEngine software which uses CGA
as the main scripting language to define rules of 3D content creation. The methodological
approach used in this project is summarized in Figure 1.

3.1. Data Preparation

A geographical information system (GIS) based dataset was collected for the construc-
tion of urban models, including satellite terrain imagery, road centerlines, and 2D zoning
plan polygons (Figure 2). Satellite terrain imagery provides basic geographic information
about the study area. All urban elements were placed on the terrain map to model actual
land-specific topography. Road centerlines, along with street width attributes, were used
for road creation among blocks.
Two-dimensional zoning plan polygons come with some key parameters, including
zoning usage, floor area ratio (FAR), building coverage (BC), maximum permitted height
(Hmax), and setback parameters. Zoning usage defines the land use type consisting
of built-up areas (residential, commercial, industrial institution, educational institution,
infrastructure, and utilities) and unbuilt areas (agriculture, forest, bare land). The FAR is
the ratio of a building’s total floor area to the size of land upon which it is built. BC defines
the maximum area that building footprints can cover on the plot surface. Hmax is the
maximum permitted height of the construction in meters. Setback refers to the minimum
distance between buildings and the lot line. These key parameters were used as inputs for
the generation of the 3D urban plan.
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR PEER REVIEW 4 of 24
ISPRS Int. J. Geo-Inf. 2022, 11, 531 4 of 20

Figure 1. The workflow of semi-automated procedure modeling on urban scale, including parcel
division, building extruding, and facade texture mapping process.

3.1. Data Preparation

A geographical information system (GIS) based dataset was collected for the con-
struction of urban models, including satellite terrain imagery, road centerlines, and 2D
zoning plan polygons (Figure 2). Satellite terrain imagery provides basic geographic in-
formation about the study area. All urban elements were placed on the terrain map to
Figure
Figure1.1.The
Theworkflow
workflowofofsemi-automated
semi-automated procedure modeling on
procedure modeling onurban
urbanscale,
scale,including
includingparcel
parcel
modelbuilding
division, actual land-specific
extruding, andtopography.
facade Road
texture centerlines,
mapping process. along with street width attrib-
division, building extruding, and facade texture mapping process.
utes, were used for road creation among blocks.
3.1. Data Preparation
A geographical information system (GIS) based dataset was collected for the con-
struction of urban models, including satellite terrain imagery, road centerlines, and 2D
zoning plan polygons (Figure 2). Satellite terrain imagery provides basic geographic in-
formation about the study area. All urban elements were placed on the terrain map to
model actual land-specific topography. Road centerlines, along with street width attrib-
utes, were used for road creation among blocks.

Figure 2. Dataset collected for the construction of urban models, including terrain with satellite im-
Figure 2. Dataset collected for the construction of urban models, including terrain with satellite
agery, road centerlines, and 2D zoning plan polygons.
imagery, road centerlines, and 2D zoning plan polygons.

3.2. Grammar Parsing and Derivation

CGA is a context-free grammar that can be written as

G = < T, N, ω, R > (1)

Figure 2. Dataset collected for the construction of urban models, including terrain with satellite im-
agery, road centerlines, and 2D zoning plan polygons.
ISPRS Int. J. Geo-Inf. 2022, 11, 531 5 of 20

n o
where G is the grammar, N is a set of non-terminals N = N1 · · · N| N | , representing the
n o
intermediate state of shape transformation, and T is the set of terminals T = T1 · · · T| N | ,
representing outputs of 3D models. ω is a starting axiom, and R is the collection of grammar
rules, where the paradigm can be defined as


h N i ::= O p (h N i|h T i){h N i|h T i} (2)

The output N is the result of transformation and combination of a series of grammar

rules from left to right. O is the parameterized operator, including 11 basic transformation
operations, i.e., extrude, split, rotate, comp, setback, insert, scale, transformation, L-shape,
U-shape, and O-shape.
CGA grammar parsing obtains the general rules and terminals for 3D urban design.
These rules describe the basic methods of: (1) parcel subdivision and clustering, (2) building
extrusion, and (3) texture mapping. The process of deriving a completely urban design
entails determining which production rules to apply and how many times to repeat them.
The details are elaborated on below.

3.3. Urban Procedural Modeling

3.3.1. Parcel Subdivision and Clustering
Parcel subdivision and clustering determine the layout of buildings within each
parcel. The subdivision of parcels to lots is carried out with one of the four algorithms:
recursive, offset, skeleton, or partition (Figure 3). The first three methods are provided by
the CityEngine. The recursive and offset subdivision algorithms create two types of lots:
with street access or located in the middle of a parcel without street access. The skeleton
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR subdivision
PEER REVIEWalgorithm can ensure that the created lots always have one side connected 6toofa 24
street (Figure 3A–C).

Figure 3. Parcel subdivision, clustering, and rotation ((A–C) showing subdivision results generated
Figure 3. Parcel subdivision, clustering, and rotation ((A–C) showing subdivision results generated
by recursive, offset, and skeleton algorithm; (D–F) showing subdivision results generated by parti-
by recursive, offset, and skeleton algorithm; (D–F) showing subdivision results generated by partition
tion algorithm with absolute values, specified partition number, and specified ratio; (G) showing
algorithm with absolute values, specified partition number, and specified ratio; (G) showing index of
index of inner lots, (H) showing inner lots clustered based on their index, (I) showing cluster lots
inner lots,as(H)
rotating showing inner lots clustered based on their index, (I) showing cluster lots rotating as
a whole).
a whole).
The partition algorithm proposed in this work enables subdivision of parcels into
regular grids with a series of CGA rules. Starting from the 2D zoning plan polygons, a
series of production rules 𝑁 determines layout generation by combining the splitting and
clustering operation. Firstly, 3 types of splitting methods are defined (Figure 3D–F). The
type 1 rule specifies lot size as absolute values. The remaining parts at the edge of parcel
remain the same (Figure 3D). The type 2 rule splits the geometry into a specified number
 ISPRS Int. J. Geo-Inf. 2022, 11, 531 6 of 20

The partition algorithm proposed in this work enables subdivision of parcels into
regular grids with a series of CGA rules. Starting from the 2D zoning plan polygons, a
series of production rules N determines layout generation by combining the splitting and
clustering operation. Firstly, 3 types of splitting methods are defined (Figure 3D–F). The
type 1 rule specifies lot size as absolute values. The remaining parts at the edge of parcel
remain the same (Figure 3D). The type 2 rule splits the geometry into a specified number of
parts. This method creates equal size shapes with adaptive size (Figure 3E). The type 3 rule
creates a repetitive pattern on the plot using the ratio (Figure 3F).
The clustering operation is then performed to create pre-defined layouts after splitting
the parcel into smaller inner lots. Splitting rules conducted in the parcel subdivision step
create an index system that can be used as the basis of the cluster operation. Starting
from a certain corner, each inner lot is assigned an identifier index as “X index, Y index”.
Clustering rules pick out specific lots based on their index and assign them as a group.
Normally, each group contains at most four lots. Defined clusters can be rotated and
transformed as a whole to adapt the general intended layout of the plot (Figure 3G–I).
Parcel subdivision, clustering, and rotation are crucial in terms of layout design quality
and diversity. Having the ability to choose different subdivision methods, define clusters,
and rotate clusters gives the modeler flexibility to adjust the placement and size of the inner
building lots.

3.3.2. Building Extrusion

With parcel subdivision completed, various building types are extruded from pre-
defined layouts to generate different urban functional zone. All building extrusion rules are
assumed to be organized in the following manner: (1) a pre-defined layout generated from
nf. 2022, 11, x FOR PEER REVIEW 7 of 24
above step is selected, (2) a production rule with this non-terminal as predecessor is chosen,
(3) the non-terminal is replaced with the rule’s successor to create the shape transformation
result (e.g., a typical building of a selected zoning plan), and (4) the production process is
terminated if all non-terminals are substituted; otherwise, the process repeats from step 2.
This organization defines a hierarchy of production rules where the grammar termi-
This organization defines a hierarchy of production rules where the grammar terminals
nals consist of 8 common
consist ofbuilding
8 commontypes andtypes
building public
andfacilities (e.g., (e.g.,
public facilities parking lot and
parking park).
lot and park). These
These 8 building types are main hall with podium, multi-towers with podium, building
8 building types are main hall with podium, multi-towers with podium, building with
with connecting corridor,
connectingtower building,
corridor, O-shape
tower building, building,
O-shape L-shape
building, L-shapebuilding,
building, U-shape
U-shape building,
and I-shape building
building, and I-shape building (Figure 4). (Figure 4).

Figure 4. Building library containing

Figure 4. 8 common
Building library building
containing typesbuilding
8 common ((A). Main
typeshall
((A).with
Mainpodium,
hall with (B).
podium, (B).
Multi-towers with podium, (C). Building with connecting corridor, (D). Tower building, (E). (E).
Multi-towers with podium, (C). Building with connecting corridor, (D). Tower building, O- O-shape
shape building, (F). building,
L-shape (F).
building,
L-shape(G). U-shape
building, (G). building, and (H).
U-shape building, andI-shape building).
(H). I-shape building).

A land-use-based combination of building types and layout is proposed subse-

quently to generate various urban forms of different urban functional zones (Figure 5).
Urban form, referring to the size, shape, and configuration of an urban area or its parts, is
determined, at a broad scale, by the building type, street type, and their layout [43]. The
ISPRS Int. J. Geo-Inf. 2022, 11, 531 7 of 20

A land-use-based combination of building types and layout is proposed subsequently

to generate various urban forms of different urban functional zones (Figure 5). Urban form,
referring to the size, shape, and configuration of an urban area or its parts, is determined,
at a broad scale, by the building type, street type, and their layout [43]. The commercial
land use zone contains three typical urban forms. The retail park often includes 4 L-shape
buildings placed on a clustered parcel layout. A square, parking lot, or courtyard is
placed in the center without street access. Shopping malls are mainly high-rise buildings
(i.e., types I and II) in the street accessible parcels generated by recursive, offset, and
ISPRS Int. J. Geo-Inf. 2022, 11, x FORpartition algorithms. The pedestrian mall is characterized by dense buildings arranged
PEER REVIEW 8 of 24in
parallel partition parcels to form a narrow street space.

Figure 5. Various urban forms of different urban functional zone generated by land-use-based com-
Figure 5. Various urban forms of different urban functional zone generated by land-use-based
bination of building types and layout.
combination of building types and layout.
The residential land use zone mainly consists of panel-type and towel-type building
footprints (i.e., slab block and tower apartment). Both of them are arranged in parallel
partition parcels. The industrial land use zone is normally dominated by simple basic
building footprints (e.g., L-shaped, U-shaped, and I-shaped buildings). The educational
land use zone, on the other hand, is characterized by buildings with connecting corridors.
 ISPRS Int. J. Geo-Inf. 2022, 11, 531 8 of 20

The official land use zone consists of four typical urban forms. The U-shaped cluster
often includes two facing L-shaped buildings and two facing I-shaped buildings. It forms
a square enclosed on three sides. Similarly, an L-shaped cluster brings out an L-shaped
layout on a corner, and two I-shaped layouts on cross-sides, forming a square enclosed
on two sides. Office buildings with atriums are characterized by a series of O-shaped
buildings distributed in a compact way, and single office buildings are often arranged in
parallel partition parcels.
The residential land use zone mainly consists of panel-type and towel-type building
footprints (i.e., slab block and tower apartment). Both of them are arranged in parallel
partition parcels. The industrial land use zone is normally dominated by simple basic
building footprints (e.g., L-shaped, U-shaped, and I-shaped buildings). The educational
land use zone, on the other hand, is characterized by buildings with connecting corridors.
The building heights in each zone are determined using the FAR, building density,
Geo-Inf. 2022, 11, x FOR PEER REVIEW and floor height. They can be calculated as H = FAR ∗ f h /d, where H represents building
9 of 24
height, f h is floor height and d represents building density. By adding a random function,
the building height variation is created, while the overall FAR remains unchanged. The
setback parameter is used to fine-tune street configuration and design.
3.4. Texture Mapping
3.4. Textureconsists
A typical building Mappingof several floors, each floor is divided into various faces,
A typical
and each face consists of severalbuilding consists of
windows several floors,
surrounded by each
trimfloor
and iswall
divided into various
material. Our faces,
and this
algorithm exploits eachtypical
face consists of several
structure to addwindows
details on surrounded
extruded by trim and
building wall material.
footprint and Our
algorithm exploits this typical structure to add details on extruded building footprint
then map texture from image space to 3D architecture model space. In this process, texture
and then map texture from image space to 3D architecture model space. In this process,
maps and grammartexturerules
mapsareandcompiled together
grammar rules as buffer
are compiled objects
together as to support
buffer objectsefficient
to supportac-efficient
cess by fragment shaders. Each pixel that needs to be colored follows the decision
access by fragment shaders. Each pixel that needs to be colored follows the decision tree tree
constructed byconstructed
the grammar to derive
by the grammar thetotexture coordinates.
derive the texture coordinates.
For parsing, allForbuildings
parsing, are divided into
all buildings groundinto
are divided floor, intermediate
ground floors, and
floor, intermediate roofand roof
floors,
from bottom tofromtop bottom
(Figureto6).
topThere
(Figure 6).seven
are There are seven
types of types
roofs:offlat
roofs: flatflat
roof, roof, flat with
roof roof with
par-parapet,
single
apet, single slope slope
roof, roof, slope
double double slope
roof, roof,
hip hippyramid
roof, roof, pyramid
roof, roof, and stepped
and stepped roof.roof. Slope, hip
Slope,
and pyramid roofs are often observed on residential buildings,
hip and pyramid roofs are often observed on residential buildings, while other roof types while other roof types can
be observed on all kinds of buildings (Figure 7).
can be observed on all kinds of buildings (Figure 7).

Figure 6. Texture mapping of building, dividing a typical building to ground floor, intermediate
Figure 6. Texture mapping of building, dividing a typical building to ground floor, intermediate
floors, and roof from bottom to top; Each part consists of repeating pairs of elements; The detailed
floors, and roof from bottom to top; Each part consists of repeating pairs of elements; The detailed
facade synthesis is conducted by combing the grammar-based splitting with texture mapping.
facade synthesis is conducted by combing the grammar-based splitting with texture mapping.
ISPRS Int. J. Geo-Inf. 2022, 11, 531 Figure 6. Texture mapping of building, dividing a typical building to ground floor,9 of intermediate
20
floors, and roof from bottom to top; Each part consists of repeating pairs of elements; The detailed
facade synthesis is conducted by combing the grammar-based splitting with texture mapping.

Figure7.7.Seven
Figure Seventypes
typesofofroofs
roofs((A)
((A) Flat
Flat roof,
roof, (B)(B) Flat
Flat roof
roof with
with parapet,
parapet, (C) (C) Single
Single slopeslope
roof,roof, (D)
Double slope roof, (E) Hip roof, (F) Pyramid roof, and (G) Stepped roof).
(D) Double slope roof, (E) Hip roof, (F) Pyramid roof, and (G) Stepped roof).

Although
Although ground
groundfloors andand
floors intermediate floorsfloors
intermediate normally have different
normally structures
have different structures
and textures, they can be decorated in two ways. The first method is the creation
and textures, they can be decorated in two ways. The first method is the creation of texture
of texture
maps representing each planar section of a building’s facade. The second is decomposition
maps representing each planar section of a building’s facade. The second is decomposition
of the facade by a set of repeating elements. The ground floor and intermediate floors
of the facade by a set of repeating elements. The ground floor and intermediate floors
consist of an ordered sequence of facades. Each facade can be subdivided into labeled
feature groups representing brick, trim, windows, and entries. The subdivision process
can be summarized as a facade-splitting schema, including the symbolic growth rule
and geometric properties. Reliably discovering the underlying symmetry in structure
and texture is crucial for facade analysis. For example, considering the face in Figure 6,
which can be subdivided into 10 columns, if A represents brick and B corresponds to a
window, then the pictured face has only 2 unique types of columns and can be written as
F = ABCDECDEBA. Considering that the repeating pairs of elements can be combined
and represented by borrowing the Kleene star notation from regular expression, the facade
splitting rule is rewritten as F = AB(CDE)∗ BA (Figure 6).
CityEngine provides a standard texture library that can be used to map appropriate
texture images to the matching building facade based on land used type and building
height. However, to generate a city block with specific local characteristics, a customized
texture library is required. Since the texture mapping process of building facade involves
the transformation from image space to building model space, strict constraints on texture
quality are proposed: (1) all texture images should be taken with orthorectified view since
oblique view often results in a skewed facade texture; (2) all texture images should be
regular shape since irregularly shaped polygons often cause geometric distortion; (3) all
texture images should be relatively clean since occluded elements often cause erroneous
texture mapping of the building model.
In this work, a rudimentary method proposed by Früh and Zakhor [44] has been
employed to perform local texture image acquisition. A series of post-processing techniques,
e.g., foreground removal, geometric correction, and color balancing, proposed by previous
researchers [45–47], was also applied to improve the quality of texture images. Thus, a land-
used-based texture library containing six types of facades (commercial, official, residential,
industrial, educational, infrastructure and utilities) and four types of basic facade elements,
including windows, walls, doors, and roofs, is constructed (Figure 8).
 In this work, a rudimentary method proposed by Früh and Zakhor [44] has been em-
ployed to perform local texture image acquisition. A series of post-processing techniques,
e.g., foreground removal, geometric correction, and color balancing, proposed by previ-
ous researchers [45–47], was also applied to improve the quality of texture images. Thus,
ISPRS Int. J. Geo-Inf. 2022, 11, 531 a land-used-based texture library containing six types of facades (commercial, official,10 ofres-
20
idential, industrial, educational, infrastructure and utilities) and four types of basic facade
elements, including windows, walls, doors, and roofs, is constructed (Figure 8).

Figure8.8.AAland-used-based
Figure land-used-basedtexture
texturelibrary.
library.((A)
((A)Six
Sixtypes
typesofofbuilding
buildingfacades,
facades,namely
namelycommercial,
commercial,
official, residential, industrial, educational, infrastructure, and utilities; (B–E) showing 4 types of
official, residential, industrial, educational, infrastructure, and utilities; (B–E) showing 4 types of
basic facade elements, namely windows, walls, doors, and roofs).
basic facade elements, namely windows, walls, doors, and roofs).

4. Case Study
To demonstrate the proposed urban procedural modeling described and explore future
potential, it was used to generate and optimize the urban design in two practical projects
in Guangzhou, a megacity in south China. Both projects are covered by a detailed control
plan, and each has its own characteristic. One is the urban renewal project of H-Village,
involving various stakeholders, namely the village manager, peasants, local community,
residents, developers, local government, and so on. In this project, our methodology
shows the capacity to coordinate and collaborate with multiple stakeholders by providing
a flexible urban design proposal that can be modified on the fly. Another is an urban design
strategy for the air–rail integration zone, which is planned as the major transportation hub
of Guangzhou. In this project, our methodology generates multiple urban design proposals
to cover possible urban development scenarios in order to fully consider the uncertainty
existing in the long-term development of the city.

4.1. Case 1: Urban Renewal Project of H-Village

4.1.1. Background
H-Village is located in an urban–rural fringe of Guangzhou, China, 11 km north of
the city center (Figure 9A). The project covers an area of 120 hectares. It can be seen
from the satellite map that there are rural communities, town communities, and industrial
zones on the site (Figure 9B). In 2018, the local government raised a conceptual plan
called Guangzhou Valley of Design to create a hub for design enterprises. After years of
negotiation, 90% of the existing residents agreed to the urban redevelopment proposal in
early 2021. The zoning adjustment begins soon according to the new development strategy
(Figure 10), as well as the demolition.
 the city center (Figure 9A). The project covers an area of 120 hectares. It can be seen from
the satellite map that there are rural communities, town communities, and industrial
zones on the site (Figure 9B). In 2018, the local government raised a conceptual plan called
Guangzhou Valley of Design to create a hub for design enterprises. After years of negoti-
ISPRS Int. J. Geo-Inf. 2022, 11, 531 ation, 90% of the existing residents agreed to the urban redevelopment proposal in 11 early
of 20
2021. The zoning adjustment begins soon according to the new development strategy (Fig-
ure 10), as well as the demolition.

Figure 9. Location and current situation of H-Village ((A) showing the location of H-Village, 11 km
Figure 9. Location and current situation of H-Village ((A) showing the location of H-Village, 11 km
north of the city center, and (B) showing the current situation of H-Village, consisting of rural, town,
ISPRS Int. J. Geo-Inf. 2022, 11, x FORnorth
PEER of
REVIEW
the city center, and (B) showing the current situation of H-Village, consisting of rural,12town,
of 24
and industrial communities, with some reserved buildings and landscape in the area).
and industrial communities, with some reserved buildings and landscape in the area).

Figure 10.The
Figure10. Thedetailed
detailedplan
planofofthe
theH-village
H-villageredevelopment
redevelopmentproject
projectafter
afterzoning
zoningadjustment
adjustment(the
(the
newlybuilt
newly built part
part consisting
consisting of 13 parcels ofof residential,
residential, 44parcels
parcelsofofeducational,
educational,and
and1515parcels
parcelsof
ofbusiness).
business).

In
Inthe
thelatest
latestdetailed
detailedplan,
plan,parts
partsofofthe
theexisting
existingfunctions
functionsarearereserved,
reserved,namely
namelythethe
town
towncommunity,
community,aahistorical
historicaltemple,
temple,an anancestorial
ancestorialhall,
hall,aavocational-technical
vocational-technical school,
school,
and
andaaresearch
researchinstitute,
institute,while
whilethe
therural
ruralcommunity
communitywill willbecome
becomeaamodern
moderncommunity
community
consisting
consistingofofapartments,
apartments,schools,
schools,parks
parksandandpublic
publicfacilities,
facilities,and
andthe
theindustrial
industrialzone
zonewill
will
be
bedeveloped
developedintointoaacreative
creativezone
zonetotooffer
offerjobs
jobsand
andhomes.
homes.The Thenewly-built
newly-builtpart,
part,namely
namely
the
theformer
formerrural
ruralcommunity
communityand andthe
theindustrial
industrialzone,
zone,mainly
mainlyconsists
consistsof
ofthirteen
thirteenparcels
parcelsofof
residential, four parcels of educational, and fifteen parcels of business (Figure
residential, four parcels of educational, and fifteen parcels of business (Figure 10). 10).

4.1.2. Restrictions and Requirements

In this case, restrictions and requirements are from different groups of stakeholders,
namely residents/peasants from rural communities, residents from town communities,
the owner and manager of the industrial zones, the local government authority, the de-
veloper, Guangzhou Metro, and enterprises which intend to settle in Table 1. There are
some interests all the stakeholders share, such as accessibility to amenities, appropriate
ISPRS Int. J. Geo-Inf. 2022, 11, 531 12 of 20

4.1.2. Restrictions and Requirements

In this case, restrictions and requirements are from different groups of stakeholders,
namely residents/peasants from rural communities, residents from town communities, the
owner and manager of the industrial zones, the local government authority, the developer,
Guangzhou Metro, and enterprises which intend to settle in Table 1. There are some
interests all the stakeholders share, such as accessibility to amenities, appropriate open
space, and energy saving. There are some special requirements as follows.

Table 1. Restrictions and requirements of different groups of stakeholders in the H-Village project.

Group Restrictions and Requirements for the Urban Design

Proper ratio of the residential function
Rural Community
Keeping the cultural context
Keeping the existing community
Town Community
Control of height
Guangzhou Metro Sparing open space for Line 14 and stations
Identity of the site/Fantasy facade
Enterprises
Network of open space
Financial balance
Developer
Feasibility
Street view
Local Government
Future development
Authority
Satisfaction of residents

Most of the residents from the rural community will come back to settle down after
the redevelopment. They would argue for enough living places and care about the cultural
context, which leads to the proper ratio of the residential function and a proper design
solution for reserving the little pond, historic temple, and ancestral hall.
Since the town community will be reserved as before, residents from the town com-
munity require sufficient daylight and sunlight for the existing community when new
buildings rise from the ground in the south. Thus, designers should take the distance to
the existing community and the height of the new building into consideration.
According to the metro plan, Line 14 will pass through the site in the northwest
corner, and two stations of Line 14 will be located at the site. Guangzhou Metro, which is
responsible for the construction and operation of Line 16, requires that there should be open
space on the ground above Line 14 to provide enough space for near-future construction.
The stations could be attached to surrounding buildings or stand alone.
Enterprises intending to settle in are mainly in the field of design, including industrial
design, architecture design, graphic design, and fashion design. As future users, they may
prefer a fantasy facade to demonstrate innovative culture and continuing open space to
make chance encounters take place.
Both the developer and the local government authority are looking forward to the
comprehensive performance of the urban design proposal, as they are officially in charge of
the entire project and the future operation. However, they focus on different aspects. The
developer will concentrate on the financial balance and feasibility problem, while the local
government authority concerned with the street view, future development, and satisfaction
of the residents.

4.1.3. Multiple Proposals Generation

After preparing the data and conducting the first two steps in the CGA rules, some
alternative proposals were raised for the project. Three of them were chosen to focus on
how they correspond to the restrictions and requirements. The main design issues in the
H-Village project are to reserve certain parts in the site, to make the newly built parts work
with the reserved parts well, and to construct a new identity in the redevelopment project.
ISPRS Int. J. Geo-Inf. 2022, 11, 531 13 of 20

The site could be divided into reserved zones, zones adjacent to the reserved zones,
and newly built zones (Figure 11). The reserved zones are where the reserved constructions
or landscapes exist, while the other two zones are mainly newly built. The zones adjacent
to the reserved zones are influenced strongly by the reserved construction and the existing
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR PEER REVIEW community, mainly consisting of residential parcels and educational parcels. The newly14 of 24

built zones are the main component of the skyline and the road landscape.

Figure 11. Three alternative proposals in the H-Village project ((A–C) sharing the same reserved zones and the zones adjacent to the reserved zones but differing
in the newly built zones, mainly
Figure consisting
11. Three of business parcels
alternative and residential
proposals parcels,
in the especially Blocks
H-Village 1–9). ((A–C) sharing the same reserved
project
zones and the zones adjacent to the reserved zones but differing in the newly built zones, mainly
consisting of business parcels and residential parcels, especially Blocks 1–9).

All three proposals share similar reserved zones and the zones adjacent to the reserved
zones. The reserved parts are mainly in the west and southwest, including the pond, a
historic temple, an ancestral hall, the town community, the research institute, and the
vocational-technical school. They are identified and extruded to reflect their actual appear-
ance using the terrain with satellite imagery. The same pattern languages are applied in
zones adjacent to the reserved zones, as it is the optimal solution when considering the
orientation, FAR, proper open space, setback, and other limitations set by zoning.
To protect and continue the contexts shaped by the culture, some green spaces and
routes are spared for the residents/peasants from rural communities and the newcomers.
The residential parcel at the east of the historic temple employs the tower apartments to
avoid obstructing the view, while the other residential parcels utilize slab blocks to meet
the living requirements of residents/peasants (Figure 11).
The differences in the three proposals are the patterns in newly built zones, consisting
of business and residential parcels, which form the identity of the site and are mainly
decided by the design concept of street interfaces in every proposal. Specifically, these
parcels are divided into nine blocks, labeled from 1 to 9 clockwise (Block 1 along the
northern boundary, Block 2 at the northeast corner, Blocks 3–5 along the eastern boundary,
Block 6 at the southeast corner, Block 7 along the southern boundary, Block 8 north of Block
6, and Block 9 northeast of Block 8).
Plan A is the lowest density proposal in response to the residents’ demands for living
places and sunlight. It creates a lot of open space through the enclosure and semi-enclosure
forms. In this case, the blocks along the street are of intermediate height and have certain
patterns with complete street fronts. To be more specific, they are Block 1 with U-shaped
business buildings, Blocks 2–3 with O-shaped business buildings, Block 4 with the main
hall with a podium, Block 5 with residential building complex, Block 6 with single office
buildings, Block 7 with slab blocks, Block 8 with building clusters with a courtyard in the
center, and Block 9 with a compact layout.
Proposal C could be seen as a more identical solution, revealing the government’s
ambition. Almost every parcel along the main streets features tower buildings or tower
buildings with podiums, namely Blocks 2–4 and Blocks 6–7. Blocks 1 and 5 in proposal
ISPRS Int. J. Geo-Inf. 2022, 11, 531 14 of 20

C take their patterns from proposals A and B, respectively. There are many spare spaces
around the tower buildings, making it possible to implant a network of various open spaces.
Compared to proposals A and C, Plan B is a compromised option. There is a desire to
create an identity contrary to the reserved parts behind the road interface in proposal B, so
some tower buildings emerge in Block 6 to the south and Block 4 to the east. These tower
buildings are potential landmarks. The other parts along the street are moderate with some
varieties; for example, Block 1 applies a compact layout, and Block 2 combines U-shaped
and L-shaped business buildings.

4.1.4. Alternative Facade Proposals

Since enterprises are looking forward to a fantasy streetscape, there are several alterna-
tive facade proposals in the multiple proposals mentioned above (Figure 12). Colors, trans-
parency, and styles could indicate the functions of buildings, such as offices (Figure 12A,B),
schools (Figure 12C), and businesses (Figure 12D). There are three common textures for
modern office buildings (one in Figure 12A, two in Figure 12B). The first consists of white
walls and blue glass, in which the glass is fragmented following certain modules to express
an erect and exquisite image. The second is decorated by pale brown ceramic tiles and
ribbon windows. All the office rooms can receive equal light with the ribbon windows, so
the facade is applicable for research or design institutes. The third one applies glass curtain
walls to make the office building appear to be light and clear, which will be popular with
creative classes. In the typical facade for schools, the window height is much larger than
others to make natural light penetrate deeper into the building. Every classroom could
have a good light condition with three or four windows. Figure 12D shows a street view
with a business facade texture along the street. On the left is the ground floor of a business
building, where retail activity could take place, such as cafés and grocery stores, while on
the right is a vital texture with irregular segmentation. The same functions can also use
various different facade textures. There are several visual effects when applying certain
textures in the same building (Figure 12E–H). Figure 12E,F employs double slope roofs,
while Figure 12G,H utilizes flat roofs. Aside from the roofs, they vary in wall materials,
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR PEER REVIEW including brown bricks, brownish-red ceramic plate curtain walls, brown ceramic tiles,16and of 24

khaki bricks.

Figure 12. Different facade textures applied to the buildings ((A,B) showing textures for offices, (C) showing textures for schools, (D) showing textures for retail
space, (E–H) showing different visual effects when applied certain textures in the same building, (E) applying double slope roofs and brown bricks on walls, (F)
Figure 12. Different facade textures applied to the buildings ((A,B) showing textures for offices,
applying double slope roofs and brownish red ceramic plate curtain wall, (G) applying flat roofs and brown ceramic tiles on walls, (H) applying flat roofs and
khaki bricks on (C) showing textures for schools, (D) showing textures for retail space, (E–H) showing different
walls).

visual effects when applied certain textures in the same building, (E) applying double slope roofs and
brown bricks on walls, (F) applying double slope roofs and brownish red ceramic plate curtain wall,
(G) applying flat roofs and brown ceramic tiles on walls, (H) applying flat roofs and khaki bricks
on walls).

4.2. Case 2: Urban Design Strategy of Air–Rail Integration Zone

4.2.1. Background
The air–rail integration zone is adjacent to the railway station of the Beijing–Guangzhou
Railway, 28 km north of the city center and 11 km west of the airport (Figure 13A). The
 ronment. However, there are currently few rural communities and industrial zones along
the rivers (Figure 13B). With the integration plan of the airport and the railway station
proposed in 2020, an air–rail line is to pass through the site, leading to opportunities for
new development in the nearby area.
ISPRS Int. J. Geo-Inf. 2022, 11, 531 There are five clusters in the existing detailed plan. They are divided by rivers15 and
of 20
main roads (Figure 14). Clusters 1–3 are the core clusters along the air–rail line. Cluster 1
is next to the Railway Station with small blocks of business and commercial and residen-
tial areas. Cluster 2 is on an island with surrounding water. It applies a radiate road sys-
project covers
tem; thus, thereanis area of 1081
a visual focushectares, almost
area around 10 times
Y-Station. the area
Cluster 2 isofthe
H-Village. There are
most comprehen-
rivers running through the site, making it an outstanding place with a natural
sive cluster among the five, including business, business for innovation, administration,
environ-
ment. However, there are currently few rural communities and industrial
and commercial and residential areas. Cluster 3 is an innovation zone with large parcels
zones along
the rivers (Figure 13B). With the integration plan of the airport and the railway station
of business for innovation. Cluster 4 is north of Clusters 1–3, consisting of residential areas
proposed in 2020, an air–rail line is to pass through the site, leading to opportunities for
and other amenities, while Cluster 5 is south of Clusters 2–3, consisting of business, busi-
new development in the nearby area.
ness for innovation, and residential.

Figure 13.
Figure 13. Location
Location and
and current
currentsituation
situationofofair–rail
air–railintegration
integrationzone
zone((A)
((A)showing
showing
thethe location
location of of
site,
site, 28 km north of the city center and 11 km west of the airport, and (B) showing the current
28 km north of the city center and 11 km west of the airport, and (B) showing the current situation of situ-
ation of the site beside the Beijing–Guangzhou Railway, consisting of rural and industrial commu-
the site beside the Beijing–Guangzhou Railway, consisting of rural and industrial communities, with
nities, with the future air–rail line passing through).
the future air–rail line passing through).

There are five clusters in the existing detailed plan. They are divided by rivers and
main roads (Figure 14). Clusters 1–3 are the core clusters along the air–rail line. Cluster 1 is
next to the Railway Station with small blocks of business and commercial and residential
areas. Cluster 2 is on an island with surrounding water. It applies a radiate road system;
thus, there is a visual focus area around Y-Station. Cluster 2 is the most comprehensive
cluster among the five, including business, business for innovation, administration, and
commercial and residential areas. Cluster 3 is an innovation zone with large parcels of
business for innovation. Cluster 4 is north of Clusters 1–3, consisting of residential areas and
other amenities, while Cluster 5 is south of Clusters 2–3, consisting of business, business
for innovation, and residential.

4.2.2. Restrictions and Requirements

Rivers are the main natural elements in the case. To prevent flood disasters, it is
necessary to apply green buffers of a certain distance around every cluster. The rivers make
the site unique and provide an outstanding environment, making the area more suitable
for a landscape system.
The entire integration zone is located in the special customs supervision zone; therefore,
the design must meet the state’s technical standardized requirements of airport clearance
protection, namely 165 m. In that case, the height of the construction plus the ground
altitude is less than or equal to 165 m. Since the ground altitude varies in every parcel, it is
necessary to set the height control carefully (Table 2).
ISPRS Int. J. Geo-Inf. 2022, 11, 531 16 of 20
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR PEER REVIEW 18 of 24

Figure14.14.The
Figure Thedetailed plan
detailed of the
plan air–rail
of the integration
air–rail zonezone
integration project (five clusters
project near rivers
(five clusters near and main
rivers and
main cluster
roads; roads; cluster 1–3the
1–3 along along the air–rail
air–rail line, mainly
line, mainly consisting
consisting of business,
of business, business
business for innovation,
for innovation, and
and commercial
commercial and residential
and residential areas;areas;
clustercluster
4 north4 north of cluster
of cluster 1–3, consisting
1–3, consisting of residential
of residential areas
areas and
and amenities;
other other amenities; and cluster
and cluster 5 south5of south of 2–3,
cluster cluster 2–3, consisting
consisting of business,
of business, businessbusiness for innova-
for innovation and
tion and residential areas).
residential areas).

4.2.2.2.Restrictions
Table andrequirements
Restrictions and Requirementsin air–rail line project.
Rivers are the main natural elements in the case. To prevent flood disasters, it is nec-
essary toElements
apply green buffers of aRestrictions and Requirements
certain distance around every for cluster.
the Urban Design
The rivers make
the site unique and provide •an outstanding environment,
Green buffer along the riversmaking the area more suitable
for a landscape
Rivers system. • Proper road system and landscape system
The entire integration zone is located in the special customs supervision zone; there-
fore, the design
Special must meet the
Customs • state’s technical
Control standardized requirements of airport clear-
of height
ance Supervision
protection, Zone
namely 165 m. In that case, the height of the construction plus the ground
altitude is less than or equal•to 165 m. Since
Control the ground altitude varies in every parcel, it
of density
Air–Rail Line Development
is necessary to set the height control carefully (Table 2).

Table 2. Restrictions and requirements in air–rail line project.

In terms of air–rail line development, there should be appropriate density development
in each cluster. The station traffic is oneRestrictions and Requirements
of the most important for the Urban
factors influencing De-
the density,
Elements
especially in Clusters 1–3. When it comes to Clusters 4 and 5,sign the density should comply
with the facility capacity (Table 2). • Green buffer along the rivers
Rivers
• Proper road system and landscape system
4.2.3. Multiple Proposals Generation
Special Customs Supervision
When focusing • Control of height
Zoneon the main design issues, the parcels in Clusters 1–3 along the air–rail
line are the main
Air–Rail Linedifferences
Development in the three
• alternative
Control proposals since the green buffer and
of density
ISPRS Int. J. Geo-Inf. 2022, 11, 531 17 of 20

height control are the same, and the facility capability is stable in Clusters 4–5. The forms
of construction in Clusters 1–3 reveal different concepts to deal with the density issue and
landscape system (Figure 14).
We present three alternative proposals for the air–rail line project in Figure 15. Proposal
A (Figure 15A) is set for a high-traffic need situation in both the railway station and
Y-Station, so it utilizes a high-density development and centralized landscape system.
Specifically, tower buildings are the main form, no matter the business parcels in Cluster
1, residential parcels in Cluster 2, and some business for innovation parcels in Cluster 3,
maximizing the density to the limit of the FAR. Since Cluster 3 is beyond walking distance
from Y-Station, only the core district of Cluster 3 takes up the tower building form, while
the other parts consist of U-shaped and O-shaped buildings. The building complex that
stretches to the river and the surrounding parcels in Cluster 2 forms the focal point of the
ISPRS Int. J. Geo-Inf. 2022, 11, x FOR PEER REVIEW site, and the nonlinear architecture near the river in Cluster 1 acts as a landscape node, 20 of 24

together making the centralized landscape system.

Figure 15. Three alternative proposals in the air–rail line project (Proposal (A) employing a high-density development and centralized landscape system in a high
traffic need situation
Figure in both
15.theThree
railway station and Y-Station,
alternative Proposal (B)
proposals inemploying a low-density
the air–rail development
line project and decentralized
(Proposal landscape system
(A) employing ainhigh-
a low
traffic need situation in both the railway station and Y-Station, Proposal (C) employing a medium-density development and decentralized landscape system when
density
there is a high traffic need atdevelopment
the railway station and centralized
and a medium landscape
traffic need in Y-Station).system in a high traffic need situation in both the

railway station and Y-Station, Proposal (B) employing a low-density development and decentralized
landscape system in a low traffic need situation in both the railway station and Y-Station, Proposal
(C) employing a medium-density development and decentralized landscape system when there is a
high traffic need at the railway station and a medium traffic need in Y-Station).

On the contrary, Proposal B (Figure 15B) aims to address the low traffic need situation
in both the railway station and Y-Station. More O-shaped and U-shaped buildings are
employed to adapt to a low-density development and decentralized landscape system. The
residential parcels in Clusters 1 and 2 take the form of low-rise slab blocks or O-shaped
buildings, and the business for innovation parcels in Cluster 3 utilize the low-rise U-shaped
buildings. When it comes to the landscape system, there are courtyards of various sizes
in Cluster 1; thus, every single building could face a semi-private green space beside the
river scene. Instead of the attractive building complex in Cluster 2 being the center of the
landscape system, there are landscape sequences from the riverside and green spaces beside
the station to the courtyard in O-shaped or U-shaped buildings, making the decentralized
landscape system.
Proposal C (Figure 15C) could be seen as an intermediate state between Proposal A
and B when there is a high traffic need at the railway station and a medium traffic need
at Y-Station simultaneously. In this case, Cluster 1 is the same as that in Proposal A, and
Cluster 3 is the same as that in Proposal B. Since the passenger flow from Y-Station is
expected to be lower than that in Proposal A and larger than that in Proposal B, in Cluster
2 there are fewer tower buildings than in Proposal A, but more than in Proposal B. As well
as Proposal B, Proposal C employs a decentralized landscape system, but the sequences are
simplified as two levels of the riverside and green spaces beside the station.
ISPRS Int. J. Geo-Inf. 2022, 11, 531 18 of 20

5. Discussion
Geometric modeling (GM), building information modeling (BIM), and procedural
modeling (PM) are the three main approaches to urban design modeling. Among them,
GM is the most widely accepted method by urban designers since it can realize the com-
plete expression of the designer’s ideas with a high degree of realism in urban modeling.
Application of GM, however, is labor-intensive and time-consuming. BIM is often used
in building design since it contains all the relevant information needed for the planning,
construction, and operation of a building, but it is not a suitable tool for urban design since
some key elements (e.g., roads and public facilities) are not included within BIM.
Compared to the aforementioned two methods, PM has two apparent advantages:
(1) PM is the most efficient modeling method of urban design. With the well-developed
CGA algorithm, designers can generate several square kilometers of urban design
models in a matter of seconds, which allows urban designers to generate design
models nearly real-time. Although the models produced by the PM method merely
achieve similar degrees of accuracy to those produced by the BIM or GM method, the
urban designer can easily use this method to complete the preliminary evaluation of
different schemes in the early stage of urban design.
(2) PM is capable of creating flexible geometries which can accommodate various bound-
ary conditions. Relying on this ability, an agile urban design working process is
applied, with the goal of allowing designers to take advantage of a high level of
automation while still maintaining a high degree of control over the output. There
are two working models: “generation” and “modification in the PM-based urban
design process. A designer can easily switch between the two working modes. The
generation model provides a fully automatic workflow that allows urban designers to
generate a large variety of urban building models with the constraints of a detailed
plan. The modification model gives urban designers options to modify selected model
data according to different urban design concepts and needs to handle information
that is not encoded into the automated processes.
In this PM-based method, an urban design process is divided into a series of steps,
allowing designers to intervene and modify the outputs of each step which enhances
collaboration between machines and humans. It ensures real-time communication between
conceptualization, design, and operation. Stakeholders’ engagement enables the improve-
ment of the quality of services and of wellness in the environment. The approach proposed
by this work can improves the automation level of urban design work significantly.
There, however, are two challenges that may limit its practical application: (1) The
general algorithm does not cover all special circumstances that need to be dealt with.
In order to effectively handle a variety of specific needs, researchers may need to add
more custom work beyond the general modeling process, which significantly increases
the complexity of algorithm development. In this circumstance, urban planners usually
prefer traditional methods since algorithm development can be more complicated and time-
consuming. (2) Quantitative assessment of urban planning indicators (e.g., land usability,
planned building volume, traffic accessibility, green space accessibility, traffic noise impact),
which would significantly reduce urban planners’ assessments of the feasibility of planning
options, is not covered in this approach. This will be implemented in future research to
better assist planners in realizing interactive urban design work.

6. Conclusions
The aim of this study was to explore a 3D urban procedural modeling solution for
urban designers to enhance collaboration among stakeholders with a computer-aided
design tool.
A series of CGA rules addressing the needs of 3D urban modeling is proposed. Com-
pared with the previous work, it involves a complete workflow of semi-automated design
procedures at an urban scale, including layout modeling, building modeling, and facade
modeling processes. This workflow was employed in the H-village urban redevelopment
ISPRS Int. J. Geo-Inf. 2022, 11, 531 19 of 20

program and the air–rail integration zone development program in Guangzhou. Three
preliminary design proposals were generated for each project. The result demonstrated
that this workflow could settle most of the collaborative issues with its analysis functions,
flexible adjustment mechanism, and real-time visualization. Meanwhile, creative design
issues can be dealt with well by generating multiple layout proposals and alternative facade
textures quickly.

Author Contributions: Conceptualization, Ming Zhang and Jielin Wu; methodology, Ming Zhang;
software, Ming Zhang; validation, Yang Liu, Ji Zhang, and Guanyao Li; formal analysis, Jielin
Wu; investigation, Ji Zhang; resources, Ji Zhang; data curation, Ming Zhang; writing—original
draft preparation, Ming Zhang and Jielin Wu; writing—review and editing, Yang Liu, Ji Zhang, and
Guanyao Li; visualization, Jielin Wu; supervision, Yang Liu; project administration, Yang Liu; funding
acquisition, Yang Liu. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Guangdong Enterprise Key Laboratory for Urban Sensing,
Monitoring and Early Warning (No.2020B121202019).
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Singh, V.; Gu, N. Towards an integrated generative design framework. Des. Stud. 2012, 33, 185–207. [CrossRef]
2. Batty, M.; Dodge, M.; Jiang, B.; Smith, A. GIS and Urban Design. In Geographical Information and Planning: European Perspectives;
Geertman, S., Openshaw, S., Stillwell, J., Eds.; Springer: Berlin/Heidelberg, Germany, 1998; 32p, in press.
3. Walmsley, K.; Villaggi, L. Generative Urban Design: A Collaboration between Autodesk Research and Van Wijnen. 2019.
Available online: https://www.autodesk.com/autodesk-university/article/Generative-Urban-Design-Collaboration-Between-
Autodesk-Research-and-Van-Wijnen-2019 (accessed on 1 May 2022).
4. Chowdhury, S. Design Thinking for Computer-Aided Co-Design in Architecture and Urban Design. In Augmented Reality and Its
Application; Cvetković, D., Ed.; IntechOpen: London, UK, 2021. [CrossRef]
5. Krish, S. A practical generative design method. Comput.-Aided Des. 2011, 43, 88–100. [CrossRef]
6. Aliaga, D.G. 3D design and modeling of smart cities from a computer graphics perspective. Int. Sch. Res. Not. 2012, 2012, 728913.
[CrossRef]
7. Parish, Y.I.H.; Müller, P. Procedural Modeling of Cities. In Proceedings of the 28th Annual Conference on Computer Graphics
and Interactive Techniques, Los Angeles, CA, USA; 2001; pp. 301–308. [CrossRef]
8. Batty, M. Cities and Complexity: Understanding Cities with Cellular Automata, Agent—Based Models, and Fractals; The MIT Press:
Cambridge, MA, USA, 2007.
9. Wu, Z.; Gan, W.; Zang, W.; Ma, C.; Zhou, J.; He, Z.; Zhou, X. Concept and Development of the City in Telligent Model (CIM). City
Plan. Rev. 2021, 45, 113–118.
10. Yang, J.; Shao, D.; Fu, X.; Min, H. Progressive optimization of urban design based on the digital map of physical environment:
Process exploration. City Plan. Rev. 2022, 64–80.
11. Yang, J.; Liu, Z.; Wang, Q.; Yao, L.; Li, T.; Hu, M.; Shi, Y.; Lu, J.; Zhang, J.; Cheng, Y.; et al. Research and Application of Key
Technologies of Urban Design Digital Platform. Constr. Sci. Technol. 2021, 117–120. [CrossRef]
12. Badwi, I.M.; Ellaithy, H.M.; Youssef, H.E. 3D-GIS Parametric modelling for virtual urban simulation using CityEngine. Ann. GIS
2022, 28, 325–341. [CrossRef]
13. Khayyal, H.K.; Zeidan, Z.M.; Beshr, A.A.A. Creation and spatial analysis of 3D city modeling based on GIS data. Civ. Eng. J. 2022,
8, 2676–6957. [CrossRef]
14. Pepe, M.; Costantino, D.; Alfio, V.S.; Vozza, G.; Cartellino, E. A novel method based on deep learning, GIS and geomatics software
for building a 3D city model from VHR satellite stereo imagery. ISPRS Int. J. Geo-Inf. 2021, 10, 697. [CrossRef]
15. Demir, I.; Aliaga, D.G.; Benes, B. Proceduralization for editing 3D architectural models. In Proceedings of the Fourth International
Conference on 3D Vision, Stanford, CA, USA, 25–28 October 2016; pp. 194–202. [CrossRef]
16. Marvie, J.-E.; Perret, J.; Bouatouch, K. The FL-system: A functional Lsystem for procedural geometric modeling. Vis. Comput.
2005, 21, 329–339. [CrossRef]
17. Wonka, P.; Wimmer, M.; Sillion, F.; Ribarsky, W. Instant Architecture. ACM Trans. Graph. 2003, 22, 669–677. [CrossRef]
18. Lindenmayer, A. Mathematical models for cellular interactions in development II. Simple and branching filaments with two-sided
inputs. J. Theor. Biol. 1968, 18, 300–315. [CrossRef]
19. Stiny, G.; Gips, J. Shape Grammars and the Generative Specification of Painting and Sculpture. In Information Processing 71;
Freiman, C.V., Ed.; North Holland: Amsterdam, The Netherlands, 1972; pp. 1460–1465.
20. Stiny, G.; Mitchell, W.J. The Palladian grammar. Environ. Plan. B Plan. Des. 1978, 5, 5–18. [CrossRef]
ISPRS Int. J. Geo-Inf. 2022, 11, 531 20 of 20

21. Müller, P.; Wonka, P.; Haegler, S.; Ulmer, A.; Gool, L.V. Procedural modeling of buildings. ACM Trans. Graph. 2006, 25, 614–623.
[CrossRef]
22. Aliaga, D.G.; Beneš, B.; Vanegas, C.A.; Andrysco, N. Interactive reconfiguration of urban layouts. IEEE Comput. Graph. Appl.
2008, 28, 38–47. [CrossRef]
23. Chen, G.; Esch, G.; Wonka, P.; Müller, P.; Zhang, E. Interactive procedural street modeling. ACM TOG 2008, 27, 103. [CrossRef]
24. Nishida, G.; Garcia-Dorado, I.; Aliaga, D.G. Example-driven procedural urban roads. Comput. Graph. Forum 2015, 35, 1–13.
[CrossRef]
25. Vanegas, C.A.; Kelly, T.; Weber, B.; Halatsch, J.; Aliaga, D.G.; Müller, P. Procedural generation of parcels in urban modeling.
Comput. Graph. Forum 2012, 31, 681–690. [CrossRef]
26. Yang, Y.; Wang, J.; Vouga, E.; Wonka, P. Urban pattern: Layout design by hierarchical domain splitting. ACM TOG 2013, 32, 1–12.
[CrossRef]
27. Aliaga, D.G.; Rosen, P.A.; Bekins, D.R. Style grammars for interactive visualization of architecture. IEEE Tran Actions Vis. Comput.
Graph. 2007, 13, 786–797. [CrossRef]
28. Bekins, D.; Aliaga, D.G. Build-by-Number: Rearranging the real world to visualize novel architectural spaces. In Proceedings of
the IEEE Visualization, Minneapolis, MN, USA, 23–28 October 2005; pp. 143–150.
29. Hohmann, B.; Havemann, S.; Krispel, U.; Fellner, D. A GML shape grammar for semantically enriched 3D building models.
Comput. Graph. 2010, 34, 322–334. [CrossRef]
30. Nishida, G.; Garcia-Dorado, I.; Aliaga, D.G.; Benes, B.; Bousseau, A. Interactive sketching of urban procedural models. ACM
Trans. Graph. 2016, 35, 130. [CrossRef]
31. Vanegas, C.A.; Allaga, D.G.; Benes, B. Building reconstruction using manhattan-world grammars. In Proceedings of the 2010
IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Francisco, CA, USA, 13–18 June 2010;
pp. 358–365.
32. Alegre, F.; Dallaert, F. A Probabilistic Approach to the Semantic Interpretation of Building Facade. 2004. Available online:
https://smartech.gatech.edu/handle/1853/4483 (accessed on 1 May 2022).
33. Cohen, A.; Schwing, A.G.; Pollefeys, M. Eicient Structured Parsing of Facades Using Dynamic Programming. In Proceedings of
the 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 23–28 June 2014; pp. 3206–3213.
34. Müller, P.; Zeng, G.; Wonka, P.; Gooll, V. Image-based procedural modeling of facades. ACM Trans. Graph. 2007, 26, 85. [CrossRef]
35. Teboul, O.; Kokkinos, I.; Simon, L.; Koutsourakis, P.; Paragios, N. Parsing Facades with shape grammars and reinforcement
learning. IEEE Trans. Pattern Anal. Mach. Intell. 2013, 35, 1744–1756. [CrossRef] [PubMed]
36. Van Gool, L.; Zeng, G.; van den Borre, F.; Müller, P. Towards mass-produced building models. Int. Arch. Photogramm. Remote Sens.
Spat. Inf. Sci. 2007, 36, 209–221.
37. Wan, G.; Sharf, A. Grammar-based 3D facade segmentation and reconstruction. Comput. Graph. 2012, 36, 216–223. [CrossRef]
38. Demir, I.; Aliaga, D.G.; Benes, B. Proceduralization of building at city scale. In Proceedings of the Second International Conference
on 3D Vision, Tokyo, Japan, 8–11 December 2014; pp. 456–463. [CrossRef]
39. Vanegas, C.A.; Garcia-doradoi, I.; Aliaga, D.G.; Beneš, B.; Waddell, P. Inverse design of urban procedural models. ACM Trans.
Graph. 2012, 31, 168. [CrossRef]
40. Weber, B.; Müller, P.; Wonka, P.; Gross, M. Interactive geometric simulation of 4D cities. Comput. Graph. Forum 2009, 28, 481–492.
[CrossRef]
41. Kuang, Z.; Chan, B.; Yu, Y.; Wang, W. A compact random-access representation for urban modeling and rendering. ACM Trans.
Graph. 2013, 32, 1–12. [CrossRef]
42. Marvie, J.; Buron, C.; Gautron, P.; Hirtzlin, P.; Sourimant, G. GPU Shape Grammrs. Comput. Graph. Forum 2012, 31, 2087–2095.
[CrossRef]
43. Živković, J. Urban Form and Function; Climate Action: London, UK, 2019; ISBN 978-3-319-71063-1.
44. Früh, C.; Zakhor, A. An automated method for large-scale, ground-based city model acquisition. Int. J. Comput. Vision 2004, 60,
5–24. [CrossRef]
45. Korah, T.; Rasmussen, C. Analysis of building textures for reconstructing partially occluded facades. In Computer Vision—ECCV
2008, Proceedings of the 10th European Conference on Computer Vision, Marseille, France, 12–18 October 2008; Springer:
Berlin/Heidelberg, Germany, 2008. Part I.
46. Tan, Y.K.A.; Kwoh, L.K.; Ong, S.H. Large scale texture mapping of building facades. Int. Arch. Photogramm. Remote Sens. Spat. Inf.
Sci. 2008, XXXVII Pt B5, 1–6.
47. Teller, S.; Antone, M.; Bodnar, Z.; Bosse, M.; Coorg, S.; Jethwa, M.; Master, N. Calibrated, registered images of an extended urban
area. Int. J. Comput. Vis. 2003, 53, 93–107. [CrossRef]