You are on page 1of 10

Ecological Modelling 417 (2020) 108924

Contents lists available at ScienceDirect

Ecological Modelling
journal homepage: www.elsevier.com/locate/ecolmodel

Land-use Simulation and Decision-Support system (LandSDS): Seamlessly T


integrating system dynamics, agent-based model, and cellular automata
Dongya Liua, Xinqi Zhenga,*, Hongbin Wangb
a
School of Information Engineering, China University of Geosciences, Beijing, China
b
Institute of Network Technology, Beijing University of Posts and Telecommunications, Beijing, China

A R T I C LE I N FO A B S T R A C T

Keywords: This article aims to construct a Land-use Simulation and Decision-Support system (LandSDS) to comprehensively
Land-use simulation explore the interference and impact of natural environment, individual choice, and national macro-policies on
Decision support land use and cover change. LandSDS takes full advantage of system dynamics (SD) model, agent-based model
System dynamics model (ABM), cellular automata (CA) model, and Geographic Information System (GIS): (1) On the one hand, SD model
Agent-based model
controls land-use area change on macro level. On the other hand, by passing the parameters values from SD
Cellular automata model
model to CA model, the iterations of CA model are controlled macroscopically, and at the same time, the
transition rules in the CA model are affected microscopically by SD model; (2) By simulating the complex
selection behaviors of different agents, ABM enforces or participates in the cells’ transformation behaviors in CA
model on the micro evolution scale; (3) Taking CA model as a breakthrough point, the micro-cell unit is used to
realize the data exchange and communication between SD-ABM-CA by integrating CA with SD and CA with
ABM; and (4) Using GIS’s powerful spatial computing and data processing capabilities, the simulation results are
further analyzed and processed to obtain more effective information. LandSDS can solve the top-down data
analysis and graphic visualization problems caused by target data or policy mutations, and can also handle the
data and graphic visualization problems caused by the temporal and spatial changes of cell units. The system
solves the analysis methods and technical problems about the top-down and bottom-up seamless integration of
the spatial and temporal models. Beijing urban expansion was taken as an example to demonstrate the avail-
ability of the system.

1. Introduction which can help people better understand and explore the land system’s
dynamic characteristics and laws. Among the existing research results,
Due to the rapid population growth, especially in developing there are many macro and micro models used to explore land-use
countries, urban expansion has become an increasingly serious problem change, for example, Markov model (Arsanjani et al., 2013; Gong et al.,
(He et al., 2006). In cities, the main factors driving land-use change are 2015), system dynamics (SD) model (Han et al., 2009), and cellular
derived from human activities. A lot of evidence has shown that human automata (CA) model (Singh et al., 2018; Vaz and Arsanjani, 2015).
behavior, especially excessive cultivation and urbanization, has caused These models have advantages in some respects, and can solve some
many problems, such as environmental problems, food security, and specific problems. And there are also some coupled models that in-
loss of biodiversity, which have become more and more serious (Fang tegrate some of these models (Clarke and Gaydos, 1998; Liu et al.,
et al., 2019; Hu et al., 2013; Zhang et al., 2015). How to achieve the 2015; Munshi et al., 2014). Among these coupled models, there mainly
harmonious development between human activities and urban sus- exist the following two problems: (1) The unidirectionality of the
tainability has become an important and severe problem. Therefore, it feedback mechanism. In the multi-dimensional integrated model, it is
is very necessary to construct a system for simulating land-use change loose coupling between top-down and bottom-up models. In fact, it is
to provide constructive opinions and suggestions for the government always the coupled model based on function and application, which
and explore urban growth and sustainable development issues. realizes the one-way communication from macro to micro. These cou-
Land use and cover change models are the main tools for studying pled models are unable to feedback the influence of micro-evolution
land-use problems, and are also the key content for land-use research, results on the macro change, and do not form the real-time and


Corresponding author.
E-mail address: zxqsd@126.com (X. Zheng).

https://doi.org/10.1016/j.ecolmodel.2019.108924
Received 18 December 2018; Received in revised form 10 December 2019; Accepted 17 December 2019
Available online 07 January 2020
0304-3800/ © 2019 Elsevier B.V. All rights reserved.
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Fig. 1. The LandSDS general structure.

bidirectional communication between sub-models to achieve real in- spatial representation ability, and micro-individual behavior cannot be
teraction and integration; and (2) The lack of research on human-en- considered. Based on this, we introduced another dynamic model – CA
vironment mechanisms. The interaction, impact, and constraints be- model.
tween human beings and their environment are the important factors CA model is a “bottom-up” and microscopic model (White et al.,
that affect land-use change. In most single or coupled models, there is 1997). CA model determines the next state of cell based on the current
lack of human agents’ influence, which does not effectively consider the cell state, the neighborhood, the transition rule, and so on. Many stu-
interaction between human and land (He et al., 2005; Kazak, 2018; dies have shown that the CA model can achieve simulation studies on
Yang et al., 2012). Based on the above problems, we aim to construct a macro-scale by the simple evolution of each cell on micro-scale (Chen
unified basic model that simulates land-use change from macro to et al., 2002; White and Engelen, 1993). Its application in land-use
micro, multi-dimension, and multi-angle by screening out several dy- change and urban development is extensive and in-depth, which is also
namic models. a hot spot for CA research (Clarke and Gaydos, 1998; Lauf et al., 2012;
SD model combines system theory, cybernetics, and computer si- Wu et al., 2010). Meanwhile, CA model has spatial characters, and can
mulation, and can build a causal feedback loop among various factors be integrated with Geographic Information System (GIS) well. GIS can
to explore the interrelationships and influences, thereby gaining mac- capture, process, and calculate spatial data well, and have strong spatial
roscopic meaning and guidance. Since SD model can easily introduce storing, managing, and analyzing ability (Zhang and Huang, 2014). GIS
various macro-driving factors, it makes the scenarios simulation re- is widely used in the integration research with CA model (Clarke and
search simple, which is very useful for testing policy recommendations. Gaydos, 1998; Yeh and Li, 2001). But the CA model cannot simulate the
SD model has been widely used to solve complicated social and natural complex individual behavior, nor is it mobile. So we need to import
problems (Castellazzi et al., 2010; Guo et al., 2001; Liu et al., 2013). ABM to make the coupling model more perfect and comprehensive.
However, the most obvious shortcoming of SD model is that there is no In recent years, ABM has attracted the attention of more and more

2
D. Liu, et al. Ecological Modelling 417 (2020) 108924

system, the SD model is first run to obtain the values for the selected
indicator parameters, and then the results are fed back to the CA model.
Thirdly, according to different agents’ selection behaviors decided by
various socio-economic factors, the influence of ABM is introduced into
CA model. In CA model, according to the factors, such as the cell state,
and the influence of the neighborhood, the cell state at next moment is
calculated combined with the index parameters transmitted from SD
model and the influence from ABM. When the iterations in CA model
equals the number which is controlled by SD model, the index para-
meters calculated on the spatial-scale in CA model are transmitted to
the SD model to fine-tune the parameter values in SD equations, and the
simulation is pushed to the next moment. The loop is stopped until the
Fig. 2. Data interaction mechanism between SD model and CA model. simulation reaches the set time, and the simulation results are exported.
During the simulation process, the users can track the changes of any
elements and pause the simulation at any time. The users also can ex-
scholars. ABM is often employed to study social phenomena (Epstein,
port the simulation results of the spatial data and the tracked data for
1999; Liu et al., 2007; Tian et al., 2016). The ABM model is another
in-depth analysis and study under GIS or other statistical platforms at
“bottom-up” model, and is also microscopic. The biggest difference
any time.
with CA model is the mobility of ABM. In ABM, the whole system
In LandSDS, the data interaction and mutual influence between each
produces pattern changes in the macro-dimension according to the in-
sub-models are realized with the smallest grid unit as the breakthrough.
teraction between different microscopic individuals (Lambin et al.,
The overall relationship can be expressed as (modified from Liu et al.
2003). ABM can directly capture the interactive properties of nature
(2017)):
and human systems, as well as capture the complex system behaviors
brought by these interactivity. Unlike CA model, which focuses on Ft + 1 ∼ f (SDt + 1, ABMt , CAt ) (1)
transition rules, ABM is well-suited for representing the evolutionary
where Ft + 1 represents the land-use state at time t+ 1; SDt + 1 is the effect
process between human activities and natural environment. ABM can
of the SD model on the state transition at time t+ 1; ABMt denotes the
map real-world concepts and structures into the models (White and
effect of ABM on the state transition at time t ;
Engelen, 2000).
CAt represents the effect of CA model on the state transition at time t; f is
Based on the previous temporal-spatial dynamics method (TSDM)
the transition function; and ∼ indicates the proportional relationship.
(Liu et al., 2018, 2017) and the geographic system basic model (Zheng,
2012), this study presents a Land-use Simulation and Decision-Support
2.2. Integration between SD and CA
system (LandSDS), which contains a SD model, an ABM, a CA model,
and a GIS. By integrating the above four model, the system can build a
As shown in Fig. 2, the data bidirectional interaction mechanism
basic comprehensive model for land-use simulation. The model in-
between SD model and CA model is displayed. In the traditional in-
tegrate the four sub-models’ advantage and make up the weaknesses,
tegration model between SD model and CA model, the iterations of CA
which can simulate land-use change in multiple scales and multiple
model are generally controlled by the macro parameters in SD model,
perspectives and provide help for the decision-making. Section 2 mainly
which is only the data flow and impact form SD model to CA model. In
introduces the principle of LandSDS, the interaction between sub-
the interaction mechanism of LandSDS constructed in this paper, the
models, and how to build the three sub-models. Section 3 is about the
data bidirectional communication and influence between SD model and
data use in the paper. Section 4 provides the precision verification of
CA model are realized by selecting appropriate indicators. Firstly, the
the system based on the study area, and shows an in-depth discussion
SD model is run to get the values of the selected indicator. Then the
about the system. Section 5 presents conclusions.
parameters are passed to CA model, and the CA model is started to run.
In CA model, the iterations are controlled and the transition rules are
2. Methods affected on micro-scale according to the parameters which are passed
from SD model. When the evolution is push to the next time, the in-
2.1. LandSDS dicators’ values obtained from CA model are fed back to SD model to
adjust the parameters of dynamic equations. Bidirectional real-time
This paper constructed a LandSDS, which deeply integrates SD communication between the two sub-models is achieved through data
model, ABM, CA model, and GIS. In the LandSDS, the SD model controls transfer between SD model and CA model. Through the bidirectional
the distribution of each element on the macro amount according to its real-time communication mechanism, the trend of simulation results is
advantages in time-dimension and macro-dimension. At the same time, corrected in time, which is more in line with the reality that the gov-
it is convenient to set various reasonable scenarios simulation by ernment proposes some short-term goals to achieve long-term planning.
changing the values of some parameters in SD model. ABM influences
the spatial distribution of various elements on the micro-scale, ac- 2.3. Integration between ABM and CA
cording to the simulation for the complex dynamic behaviors of dif-
ferent agents. CA model uses its advantage in the spatial dimension to The CA model, which comprehensively considers the factors of
determine the micro-pattern by defining the transition rules for the cell. natural environment and neighborhood influence, can better simulate
Based on the above, the system make full use of the spatial computing the land-use change process, but does not consider the dominant role
power of GIS to make further analysis for the simulation results at the for controlling land-use change. And in real life, the dominant role,
same time. such as the government, usually play a decisive role in land-use change.
As shown in Fig. 1, before the system starts to run, the initial con- Based on this, we coupled CA model with ABM to introduce human-
ditions required in SD model, ABM, and CA model are first set, ac- land interactions into land-use change simulations.
cording to spatial data, statistical data, policy files, and other ancillary In this study, taking urban expansion as an example, agents are
data. Then, the appropriate indicator parameters (single or multiple) divided into two categories: resident agents and government agents.
are selected as the medium for interactive communication between SD The resident agents select the suitable land for living by considering the
model and CA model. After the setting for initial conditions of the surrounding environment, such as transportation convenience,

3
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Fig. 3. The integrated simulation process be-


tween ABM and CA model: (a) is the initial
state of the cell; (b) is the cell state which is
determined by the CA model; (c) is the cell
state which is determined by the resident
agents; (d) is the cell transition state under the
joint action of CA model and resident agents;
(e) is the cells which are not allowed to transfer
to the certain state by the government agents;
(f) is the final cell transition state under the
combined action.

environmental suitability, and school convenience. Government agents as:


determine the land-use type for a particular plot through the macro-
St + 1 = f (St , Nt , Dt + 1, Bt ) (3)
planning. At the same time, according to the China’s condition, the
choices of government agents are absolutely authoritative in the deci- where Dt + 1 indicates the macro factors influence on cell state change at
sion for the land-use pattern. For example, if the government classifies a time t + 1; and Bt is the agent behaviors influence on cell state change
certain area as an ecological protection area, the land parcels in the at time t . The more detailed formulas can be found in (Liu et al. (2017)).
area will not be allowed to convert into other land-use types.
In the integration of ABM and CA model, the selection behaviors of 2.6. Agents’ behaviors rules in ABM
resident agents are determined according to the surrounding environ-
ment, human factors, and other factors. The behaviors influence the In this paper, the relationship of ABM can be expressed as:
transition rules in CA model to some extent to decide the land-use type Uk ∼ (Fresident , Fauthority ) (4)
for a certain plot. At the same time, combined with the macro-plan of
the government agents in a period of time, it is checked whether the where Uk represents the combined effect of the different agents on the
land-use type for a certain plot is allowed to determine the land-use land-use change; Fresident indicates the effect of the resident agents on the
type finally. Fig. 3 shows the integration mechanism between ABM and land-use change; and Fauthority indicates the effect of the government
CA model. agents on the land-use change.
In this study, the role of government mainly reflects in macro-
planning, which determines the land-use pattern based on a series of
2.4. The causal relationship in SD
factors. Given China’s conditions, the government planning has a high-
level mandatory role.
SD model explores the complex and non-linear effects between
The choices of resident agents are often influenced by the sur-
various factors by construction a causal relationship to obtain useful
rounding living environment. For example, people often choose to live
information at macro level (Theobald and Gross, 1994). In the process
in the places where the public facilities are relatively complete and the
of urban development, the land-use system is complex and variable, and
environment is friendly. After comprehensive consideration of various
is interfered by many factors such as human. Therefore, SD model is
factors, the choices of resident agents in this paper are mainly influ-
often used to explore complex urban expansion behaviors (Han et al.,
enced by the following factors: transportation convenience, facility
2009; Shen et al., 2009; Wang et al., 2019).
convenience, education convenience, hospital convenience, and en-
In this study, we construct a SD model to explore the impact of
vironmental friendliness. The relationship can be expressed as:
human activities on urban growth. Meanwhile, to simplify the calcu-
lation, we suppose that the land-use change is affected only by human Fresident = βtraffic X + βfacility X + βeducation X education
traffic facility
activities and ignore the natural choice. In SD model, there mainly
contains three subsystems, namely population subsystem, economy + βhospital Xhospital + βenvironment X environment + γ (5)
subsystem, and land-use subsystem (Fig. 4). The population increase
where βtraffic , βfacility , βeducation , βhospital , and βenvironment indicate the effec-
has led to the demand increase for built-up land. At the same time,
tiveness of transportation convenience, facility convenience, education
further investment in the land has also promoted the expansion of
convenience, hospital convenience, and environmental friendliness,
urban built-up land to a certain extent.
respectively; βtraffic + βfacility + βeducation + βhospital + βenvironment = 1; Xtraffic ,
Xfacility , Xeducation , Xhospital , and Xenvironment indicate the influence values of
2.5. Transition rules in CA transportation convenience, facility convenience, education con-
venience, hospital convenience, and environmental friendliness on
Transition rules is the core of CA model. The traditional CA tran- land-use change, respectively; and γ is the random interference value.
sition rules can be expressed as:

St + 1 = f (St , Nt ) 3. Study area and data


(2)

where St+1 is the cell state at time t + 1; St is the cell state at time t ; Nt Beijing is China’s political, economic, and cultural center, and is
represents the state of the neighbor cell at time t ; and f refers to the adjacent to Tianjin and Hebei Province. Beijing is one of the most
transfer function for cell state from time t to time t + 1. densely populated and oldest cities in the world, with a history of more
In this study, the impact of SD model and ABM is extended into the than 3000 years (Li et al., 2005), and it will host the Winter Olympic
CA transition rules. Therefore, the CA transition rules can be calculated Games in 2022. There are 16 districts under the jurisdiction of Beijing,

4
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Fig. 4. SD causality diagram for land use.

which are divided into four functional zones, namely capital functional Table 1
core zone, urban function development zone, urban development new Data used in the paper.
zone, and ecological conservation development zone. The total area of Data Data type Year
Beijing is 16410.54 km2. By 2016, the resident population of Beijing
reached 21.73 million, and the gross domestic product (GDP) reached Dependent variables Land-use map shape 1995, 2005,
2566.91 billion yuan (Beijing Municipal Statistics Bureau, 2017). The 2015
Socio-economic School features shape 2013
huge population makes Beijing so crowded. The road system in Beijing factors Market features shape 2013
is like a network, and there are six rings from the downtown area to the Hospital features shape 2013
suburbs. Similar to other big cities in developing countries, Beijing faces Restaurant shape 2013
many problems, such as environmental pollution and population features
Bank features shape 2013
overload (Chen et al., 2005).
Hotel features shape 2013
In this study, we selected the land-use data of Beijing for 1995, Road network shape 2013
2005, and 2015 to simulate the land use and cover change. The land- Park features shape 2013
use data came from Data Center for Resources and Environmental Gross Domestic Continuous 1995–2015
Sciences, Chinese Academy of Sciences (Liu et al., 2005). Through the Product variable
Population Continuous 1995–2015
visual interpretation of the Landsat data, the overall accuracy of in- variable
terpretation is over 94.3 % (Liu et al., 2014). The data is mainly divided
into six first-class categories: cropland, woodland, grassland, water
body, built-up land, and unused land (Liu et al., 2005). Because there logistic regression method was used to obtain the influence probability
are few unused land in Beijing, and most of them are located in of surrounding environment on the land-use type change. The formula
woodland, we reclassified the land-use types into four types: cropland, can be expressed as:
woodland and grassland, built-up land, and water body. According to
the availability and accessibility of the data, eight environmental ele- 1
φ=
ments were selected, as shown in Table 1, which were derived from the 1 + exp (−y ) (6)
Institute of Remote Sensing and Digital Earth, Chinese Academy of
Sciences. All of the spatial data was obtained based on the project y= β1 X1 + β2 X2 + ⋯+βn Xn + α (7)
supported by the Ministry of Land and Resources of China.
where φ represents the influence probability of the surrounding en-
vironment on the land-use type change; β1, β2 , ⋯, and βn are the re-
4. Results gression coefficients; X1, X2 , ⋯, and Xn represent the values of the
driving factors (e.g., the values of the distance to schools); and α is the
4.1. Simulation results regression constant.
According to the availability and acquirable of the data, the socio-
4.1.1. The construction of transition rules for CA model economic factors selected in this paper include distance to schools,
In CA model, the transition rules are the core content. The current distance to markets, distance to hospitals, distance to restaurants, dis-
mining methods for the transition rules are various, and the more tance to banks, distance to hotels, and distance to roads (Fig. 5).
commonly used are data mining (Li and Gar-On Yeh, 2004), ant colony In ArcGIS software, the values of these environmental elements
optimization (Liu et al., 2008, 2007; Yang et al., 2012), logistic re- were obtained through 4952 valid random points. Prior to the logistic
gression (Liu et al., 2015; Munshi et al., 2014), and so on. In this paper, regression, the values of these variables were normalized to exclude the

5
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Fig. 5. Distance to the social-economic factors.

effects of different scales. The normalization formula can be expressed 4.1.2. The construction of rules for ABM
as (Tian et al., 2011): For the government agents, in order to reduce the degree of inter-
ference for the simulation results, we only restricted the development of
⎧ Si − Min (Si ) built-up land by the macro planning data to a certain degree.
, Max (Si ) ≠ Min (Si )
Si′ = Max (Si ) − Min (Si ) In the rules of resident agents, we used the distance to roads to

⎩ 1, Max (Si ) = Min (Si ) (8) represent the transportation convenience, the distance to markets to
indicate the facility convenience, the distance to schools to represent
Si'
where is the value of the driving factor after the normalization; Si is the education convenience, the distance to hospitals to represent the
the value of the driving factor; Min (Si ) is the minimum for all values of hospital convenience, and the distance to parks to indicate the en-
the driving factor; and Max (Si ) is the maximum for all values of the vironmental friendliness (Fig. 5). The shorter distance, the better con-
driving factor. venience. Therefore, the convenience value is inversely proportional to
The normalized values were imported into the SPSS software for the distance value. At the same time, in order to prevent the denomi-
logistic regression analysis to obtain the regression coefficients for each nator from appearing 0, and to ensure that the result was still limited to
influence factor.

6
D. Liu, et al. Ecological Modelling 417 (2020) 108924

the range of 0∼1, we exponentially processed the denominator. The 5. Discussion


processed equation can be expressed as:
5.1. Comparison between LandSDS and traditional SD-CA coupled model
μtraffic
Xtraffic =
exp (droad ) (9) The performance of the LandSDS is compared with the traditional
SD-CA coupled model. In the usual integration models, the SD model is
μfacility only used to control the iterations of the CA model. In this paper, we
Xfacility =
exp (dmarket ) (10) used the common coupled method for SD and CA model (Han et al.,
2009; He et al., 2006) for simulation research as a comparative ex-
μeducation perimental group. It can be seen in Table 3 that the system which we
Xeducation =
exp (dschool ) (11) built in this paper has better overall accuracies, Kappa coefficients, and
Fuzzy Kappa coefficients than the common SD-CA coupled model. The
μhospital reason for the above results may be due to the data bidirectional ex-
Xhospital = change between SD model and CA model, and the intervention of dif-
exp (dhospital ) (12)
ferent agents’ selection behaviors decided by socio-economic factors.
Table 3 also shows that LandSDS can better simulate land-use change
μenvironment
Xenvironment = under long time series, which indicates that LandSDS can be used for
exp (dpark ) (13) future scenario simulation studies.
Of course, due to the availability of data, different experimental
where droad represents the distance to roads; dmarket represents the dis-
environments, and other reasons, it is very difficult to completely si-
tance to markets; dschool represents the distance to schools; dhospital re-
presents the distance to hospitals; dpark represents the distance to parks; mulate other researchers’ experiments for comparison. Therefore, in
and μtraffic , μfacility , μeducation , μhospital , and μenvironment are coefficients. In this paper, we used the same data to conduct the comparative experi-
this paper, to simplify the operation, μtraffic , μfacility , μeducation , μhospital , and ments based on different model integration principles in the same ex-
μenvironment are set to 1, respectively. perimental environment.

5.2. The policy recommendations for Beijing development


4.1.3. Precision verification
The test of the accuracy for the system is very necessary. We usually Due to the excessive concentration of resources, such as educational
compare the simulation result with the real data to obtain the simula- resources and medical resources, a large number of migrant workers
tion accuracy. The visual comparison method is a simple method for enter Beijing, which has brought tremendous ecological and land
evaluating the simulation result, which can visually show the difference supply pressure. From 1995 to 2005, Beijing’s built-up land increased
in spatial distribution between the simulation result and the real data. by more than 1200 km2, while the cropland decreased by more than
The comparison between the actual land-use maps and the simulation 1900 km2 (Table 4).
results for 2005 and 2015 in Beijing is shown in Fig. 6. It can be seen The government should take more effective measures to promote
from the visual comparison that the simulation results generally show Beijing green and harmonious sustainable development: (1) Control
the land-use distribution well, although the simulation effect on some population growth and ease urban expansion pressure. In the future
scattered land use is not very well. development plan, Beijing should refer to the development pattern of
The verification of the simulation results also needs to be calculated foreign satellite cities. Beijing need to transfer some resources and
from the perspective of quantification. This paper adopted the overall gradually evacuate population to reduce the expansion pace and im-
accuracy and Kappa coefficient to verify the accuracy of the system. prove ecological stability; and (2) Enhance linkages between regions
Kappa coefficient is widely used in accuracy evaluation, and is a stan- and promote regional coordinated development. With the rapid eco-
dard measurement method (Congalton, 1991). The calculation results nomic growth, uneven development between cities will lead to very
showed that the total accuracies of land-use for 2005 and 2015 were serious problems. The coordinated development of Beijing-Tianjin-
86.56 % and 90.40 %, and the Kappa coefficients were 0.7659 and Hebei and the construction of Xiongan New District are aimed to ease
0.8254, respectively. Meanwhile, we introduced the Fuzzy Kappa some non-capital function and promote regional harmonious develop-
coefficient to evaluate the system’s simulation effect. The Fuzzy Kappa ment. The government should strength Beijing economic leading role
coefficient takes into account the neighborhood effect of the grid and is for Tianjin and Hebei to gradually promote the overall development of
more intelligent, which can be used in combination with Kappa coef- the Beijing-Tianjin-Hebei urban agglomeration.
ficient to obtain better results (Dou et al., 2007; Visser and de Nijs,
2006). The Fuzzy Kappa coefficients of land-use change for 2005 and 6. Conclusion
2015 were 0.852 and 0.887, respectively. We also used areal simulated
error to further verify the system’s simulation results. Table 2 listed the This paper proposed a Land-use Simulation and Decision-Support
area error for the four land-use types in Beijing. The above four accu- system (LandSDS), which is actually an artificial intelligence temporal-
racy indicators indicate that the system has good simulation effect. Of spatial simulation model. LandSDS is a system that integrates macro-
course, the simulation accuracies of land-use change are affected by scopic SD model and microscopic ABM and CA model in cell unit.
many aspects, such as the quality of interpretation for remote sensing LandSDS also integrates GIS at the same time. The system can solve
image, which will have different degrees impact on simulation results. data analysis and graphic visualization problems from top to down
From 1995 to 2015, the population of Beijing increased from 12.51 caused by target data or policy mutations, and can also handle the data
million to 21.71 million, and the GDP increased from 150.77 billion and graphic visualization problems from bottom to up caused by tem-
yuan to 2368.57 billion yuan (Beijing Municipal Statistics Bureau, poral-spatial changes of cell units. The system solves the temporal-
2017). On the one hand, the increase of population has driven the rapid spatial analysis methods and technical problems about the top-down
economic growth, but on the other hand, it has promoted the sky- and bottom-up seamless integration, which plague researchers long. In
rocketing of urban built-up land area. It can be seen from Fig. 6 that the this system, the SD model controls the changes of various elements on a
rapid expansion of the central urban area in Beijing occupied a large macro scale, and decides the iterations of the entire simulation run at
amount of arable land and woodland and grassland, which is very un- the same time. ABM is used to simulate the complex selection behaviors
favorable for the sustainable development of Beijing. of different agents. CA model simulates the distribution of micro-cell

7
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Fig. 6. The simulation of Beijing land-use change from 1995 to 2015.

state by changing the state of each cell on a micro scale. GIS is used for data, the better overall simulation accuracies, Kappa coefficients, and
visual display and further in-depth spatial analysis of the simulation Fuzzy Kappa coefficients were obtained. The system was also compared
results. In the whole system, each sub-module communicates and with the common coupled SD-CA model, and the compared results
transfers data through the micro-individual cell unit, and realizes the showed that LandSDS was superior to the traditional coupled model.
mutual influence and restriction between sub-models. Through the interaction between SD model and CA model and the in-
Beijing was used as a case to verify the effectiveness of the LandSDS. troduction of agents’ complex behaviors, the simulation results are
The results showed that the integrated SD-ABM-CA system could si- more realistic.
mulate the rapidly growing regions well. By comparing with the real The influencing factors of urban land-use change are complex and

Table 2
Areal simulated error for 2005 and 2015 in Beijing.
Cropland Woodland and grassland Built-up land Water body

2
2005 Actual Area (km ) 3528.50 10001.63 2590.38 283.94
Simulated Area (km2) 3536.5 9423.25 2992.25 454.0625
Error 0.0023 0.0578 0.1551 0.5991
2015 Actual Area (km2) 2603.06 10497.31 2991.25 313.25
Simulated Area (km2) 2680.19 9834.50 3605.81 283.94
Error 0.0296 0.0631 0.2055 0.0936

8
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Table 3 long-term urban growth prediction for San Francisco and Washington/Baltimore. Int.
Comparison for simulation accuracies. J. Geogr. Inf. Sci. 12, 699–714.
Congalton, R.G., 1991. A review of assessing the accuracy of classifications of remotely
Start year 1995 2005 sensed data. Remote Sens. Environ. 37, 35–46.
Dou, W., Ren, Y., Wu, Q., Ruan, S., Chen, Y., Bloyet, D., Constans, J.-M., 2007. Fuzzy
Simulation year 2005 2015 2015 kappa for the agreement measure of fuzzy classifications. Neurocomputing 70,
726–734.
LandSDS Overall accuracy 86.56 % 83.53 % 90.40 % Epstein, J.M., 1999. Agent‐based computational models and generative social science.
Kappa coefficient 0.7659 0.7078 0.8254 Complexity 4, 41–60.
Fuzzy Kappa 0.852 0.794 0.887 Fang, C., Cui, X., Li, G., Bao, C., Wang, Z., Ma, H., Sun, S., Liu, H., Luo, K., Ren, Y., 2019.
Modeling regional sustainable development scenarios using the Urbanization and
coefficient
Eco-environment Coupler: case study of Beijing-Tianjin-Hebei urban agglomeration,
Common SD-CA coupled Overall accuracy 81.57 % 75.65 % 83.82 %
China. Sci. Total Environ. 689, 820–830.
model Kappa coefficient 0.6792 0.5775 0.7143
Gong, W., Yuan, L., Fan, W., Stott, P., 2015. Analysis and simulation of land use spatial
Fuzzy Kappa 0.782 0.682 0.796 pattern in Harbin prefecture based on trajectories and cellular automata—Markov
coefficient modelling. Int. J. Appl. Earth Obs. Geoinf. 34, 207–216.
Guo, H., Liu, L., Huang, G., Fuller, G., Zou, R., Yin, Y., 2001. A system dynamics approach
for regional environmental planning and management: a study for the Lake Erhai
Table 4 Basin. J. Environ. Manage. 61, 93–111.
Han, J., Hayashi, Y., Cao, X., Imura, H., 2009. Application of an integrated system dy-
Land-use areal changes for Beijing from 1995 to 2015 (km2). namics and cellular automata model for urban growth assessment: a case study of
Cropland Woodland and Built-up Water body Shanghai, China. Landsc. Urban Plan. 91, 133–141.
He, C., Okada, N., Zhang, Q., Shi, P., Zhang, J., 2006. Modeling urban expansion sce-
grassland land
narios by coupling cellular automata model and system dynamic model in Beijing,
China. Appl. Geogr. 26, 323–345.
1995 4538.94 9635.69 1777.44 454.00
He, C., Shi, P., Chen, J., Li, X., Pan, Y., Li, J., Li, Y., Li, J., 2005. Developing land use
2005 3528.50 10001.63 2590.38 283.94
scenario dynamics model by the integration of system dynamics model and cellular
2015 2603.06 10497.31 2991.25 313.25 automata model. Sci. China Ser. D Earth Sci. 48, 1979–1989.
Changed Area −1935.88 861.63 1213.81 −140.75 Hu, Y., Zheng, Y., Zheng, X., 2013. Simulation of land-use scenarios for Beijing using
CLUE-S and Markov composite models. Chin. Geogr. Sci. 23, 92–100.
Kazak, J., 2018. The use of a decision support system for sustainable urbanization and
changeable. In this study, in order to highlight the feasibility and op- thermal comfort in adaptation to climate change actions-the case of the wroclaw
larger urban zone (Poland). Sustainability 10.
erability of the integration method, many simplifying work was carried Lambin, E.F., Geist, H.J., Lepers, E., 2003. Dynamics of land-use and land-cover change in
out in the construction of the system, such as the simplification of the tropical regions. Annu. Rev. Environ. Resour. 28, 205–241.
natural, economic and policy indicators. In the future research, these Lauf, S., Haase, D., Hostert, P., Lakes, T., Kleinschmit, B., 2012. Uncovering land-use
dynamics driven by human decision-making–a combined model approach using
issues should be studied in depth to make the simulation results more cellular automata and system dynamics. Environ. Model. Softw. 27, 71–82.
realistic and efficient. Meanwhile, based on China’s basic national Li, F., Wang, R., Paulussen, J., Liu, X., 2005. Comprehensive concept planning of urban
conditions, we can see the effects of urban changes in a short period of greening based on ecological principles: a case study in Beijing, China. Landsc. Urban
Plan. 72, 325–336.
time. However, for some developed or underdeveloped countries, the Li, X., Gar-On Yeh, A., 2004. Data mining of cellular automata’s transition rules. Int. J.
temporal-spatial changes are slow. Therefore, collection a long series Geogr. Inf. Sci. 18, 723–744.
for data to calculate the relevant indicators should be considered to Liu, D., Zheng, X., Wang, H., Zhang, C., Li, J., Lv, Y., 2018. Interoperable scenario si-
mulation of land-use policy for Beijing–Tianjin–Hebei region, China. Land Use Policy
ensure the basic needs of the system for running. 75, 155–165.
Liu, D., Zheng, X., Zhang, C., Wang, H., 2017. A new temporal–spatial dynamics method
Declaration of Competing Interest of simulating land-use change. Ecol. Modell. 350, 1–10.
Liu, J., Kuang, W., Zhang, Z., Xu, X., et al., 2014. Spatiaotemporal characteristics, pat-
terns and causes of land-use changes in China since the late 1980s. J. Geogr. Sci. 24
We promise that there is no conflict of interest exiting in the sub- (2), 195–210.
mission of this manuscript, and manuscript is approved by all authors Liu, J., Liu, M., Tian, H., Zhuang, D., Zhang, Z., Zhang, W., Tang, X., Deng, X., 2005.
for publication. I would like to declare on behalf of my co-authors that Spatial and temporal patterns of China’s cropland during 1990–2000: an analysis
based on Landsat TM data. Remote Sens. Environ. 98, 442–456.
the work described was original research that has not been published Liu, X., Li, X., Liu, L., He, J., Ai, B., 2008. A bottom‐up approach to discover transition
previously, and not under consideration for publication elsewhere, in rules of cellular automata using ant intelligence. Int. J. Geogr. Inf. Sci. 22,
whole or in part. All the authors listed have approved the manuscript 1247–1269.
Liu, X., Li, X., Yeh, A.G.-O., He, J., Tao, J., 2007. Discovery of transition rules for geo-
that is enclosed. graphical cellular automata by using ant colony optimization. Sci. China Ser. D Earth
Sci. 50, 1578–1588.
Acknowledgements Liu, X., Ou, J., Li, X., Ai, B., 2013. Combining system dynamics and hybrid particle swarm
optimization for land use allocation. Ecol. Modell. 257, 11–24.
Liu, Y., Dai, L., Xiong, H., 2015. Simulation of urban expansion patterns by integrating
This research was supported by the Fundamental Research Funds auto-logistic regression, Markov chain and cellular automata models. J. Environ.
for the Central Universities. We thank Institute of Remote Sensing and Plan. Manag. 58, 1113–1136.
Munshi, T., Zuidgeest, M., Brussel, M., van Maarseveen, M., 2014. Logistic regression and
Digital Earth, Chinese Academy of Sciences for providing the data used cellular automata-based modelling of retail, commercial and residential development
in this study. in the city of Ahmedabad, India. Cities 39, 68–86.
Shen, Q., Chen, Q., Tang, B.-s., Yeung, S., Hu, Y., Cheung, G., 2009. A system dynamics
model for the sustainable land use planning and development. Habitat Int. 33, 15–25.
References
Singh, S.K., Laari, P.B., Mustak, S., Srivastava, P.K., Szabó, S., 2018. Modelling of land use
land cover change using earth observation data-sets of Tons River Basin, Madhya
Arsanjani, J.J., Helbich, M., Kainz, W., Boloorani, A.D., 2013. Integration of logistic re- Pradesh, India. Geocarto Int. 33, 1202–1222.
gression, Markov chain and cellular automata models to simulate urban expansion. Theobald, D.M., Gross, M.D., 1994. EML: a modeling environment for exploring land-
Int. J. Appl. Earth Obs. Geoinf. 21, 265–275. scape dynamics. Comput. Environ. Urban Syst. 18, 193–204.
Beijing Municipal Statistics Bureau, 2017. Beijing Statistical Yearbook. China Statistic Tian, G., Ma, B., Xu, X., Liu, X., Xu, L., Liu, X., Xiao, L., Kong, L., 2016. Simulation of
Press (in Chinese). urban expansion and encroachment using cellular automata and multi-agent system
Castellazzi, M., Matthews, J., Angevin, F., Sausse, C., Wood, G., Burgess, P., Brown, I., model—a case study of Tianjin metropolitan region, China. Ecol. Indic. 70, 439–450.
Conrad, K., Perry, J., 2010. Simulation scenarios of spatio-temporal arrangement of Tian, G., Ouyang, Y., Quan, Q., Wu, J., 2011. Simulating spatiotemporal dynamics of
crops at the landscape scale. Environ. Model. Softw. 25, 1881–1889. urbanization with multi-agent systems—a case study of the Phoenix metropolitan
Chen, H., Ganesan, S., Jia, B., 2005. Environmental challenges of post-reform housing region, USA. Ecol. Modell. 222, 1129–1138.
development in Beijing. Habitat Int. 29, 571–589. Vaz, E., Arsanjani, J.J., 2015. Predicting urban growth of the Greater Toronto Area-
Chen, J., Gong, P., He, C., Luo, W., Tamura, M., Shi, P., 2002. Assessment of the urban coupling a Markov cellular automata with document meta-analysis. J. Environ.
development plan of Beijing by using a CA-based urban growth model. Photogramm. Inform. 25, 7.
Eng. Remote Sens. 68, 1063–1072. Visser, H., de Nijs, T., 2006. The map comparison kit. Environ. Model. Softw. 21,
Clarke, K.C., Gaydos, L.J., 1998. Loose-coupling a cellular automaton model and GIS: 346–358.

9
D. Liu, et al. Ecological Modelling 417 (2020) 108924

Wang, T., Kazak, J., Han, Q., de Vries, B., 2019. A framework for path-dependent in- Yang, X., Zheng, X.-Q., Lv, L.-N., 2012. A spatiotemporal model of land use change based
dustrial land transition analysis using vector data. Eur. Plan. Stud. 27, 1391–1412. on ant colony optimization, Markov chain and cellular automata. Ecol. Modell. 233,
White, R., Engelen, G., 1993. Cellular automata and fractal urban form: a cellular mod- 11–19.
elling approach to the evolution of urban land-use patterns. Environ. Plan. A Yeh, A.G.-O., Li, X., 2001. A constrained CA model for the simulation and planning of
1175–1199. sustainable urban forms by using GIS. Environ. Plann. B Plann. Des. 28, 733–753.
White, R., Engelen, G., 2000. High-resolution integrated modelling of the spatial dy- Zhang, H., Jin, X., Wang, L., Zhou, Y., Shu, B., 2015. Multi-agent based modeling of
namics of urban and regional systems. Comput. Environ. Urban Syst. 24, 383–400. spatiotemporal dynamical urban growth in developing countries: simulating future
White, R., Engelen, G., Uljee, I., 1997. The use of constrained cellular automata for high- scenarios of Lianyungang city, China. Stoch. Environ. Res. Risk Assess. 29, 63–78.
resolution modelling of urban land-use dynamics. Environ. Plann. B Plann. Des. 24, Zhang, W., Huang, B., 2014. Land use optimization for a rapidly urbanizing city with
323–343. regard to local climate change: shenzhen as a case study. J. Urban Plan. Dev. 141,
Wu, D., Liu, J., Wang, S., Wang, R., 2010. Simulating urban expansion by coupling a 05014007.
stochastic cellular automata model and socioeconomic indicators. Stoch. Environ. Zheng, X., 2012. On the geographic system simulation basic model. Chin. J. Nat. 34,
Res. Risk Assess. 24, 235–245. 143–149.

10

You might also like