1. Introduction
Land use/cover change (LUCC) has been generally considered a main driving force of global ecosystem and climate change [
1]. Urbanization is the most typical form of LUCC, and has a significant impact on biological diversity and ecosystem services [
2]. Nowadays, about 55% of the world’s population lives in cities with this rate expected to reach 68% by 2050 [
3]. Therefore, the study of urban LUCC is of great importance to understand and grasp global LUCC. The process of urbanization influences the flow of material, energy, and information, and affects the structure and function of ecosystems [
4]. Therefore, on the one hand, the socio-economic level may be significantly improved; on the other hand, it leads to loss of farmland, fragmentation of habitats, and increases in the heat island effect [
5,
6,
7]. However, these problems can be alleviated through reasonable urban space planning and efficient utilization of urban resources [
8]. Therefore, urban space planning is considered an effective tool/means to improve urban sustainable development.
The precise simulation of urban space evolution and analysis of the leading factors are the most important bases and essential prerequisites for urban space planning. Therefore, many studies of urban space evolution simulation and leading factors analysis have been carried out, and some typical simulation models have been constructed, such as the econometric statistics model (ES) [
9], the system dynamics model (SD) [
10], the cellular automata model (CA) [
11], the multi-agent system model (MAS) [
12], etc. Each of these models has its own advantages and disadvantages. In ES models, mathematical statistics methods are used to simulate the variations in scale of urban space, such as the logistic regression model [
13], Kuznets curve model [
14], and panel econometric model [
15]. These ES models are easy to construct and use, but they cannot simulate the dynamic process and variation in urban spatial distribution. In SD models, the process of urban space evolution is expressed by simulating the interactions of urban elements, e.g., MEPLAN model [
16], Dortmund model [
17] and LILT model [
18]. However, these SD models also cannot describe variation in urban spatial distribution. The ES and SD models are considered “top-down” models.
To improve the ability to simulate urban spatial variation, the CA model was first introduced by Chapin and Weiss in 1968 [
19]. From the late 1990s to the early 21st century, the CA model entered a high-speed development period, and saw widespread use in urban space planning [
20,
21]. However, it cannot well represent macro-scale political, economic and cultural driving forces that influence urban spatial variations [
22]. Therefore, some improved methods have been proposed, for example taking the macro factors as constraint conditions in the CA model [
23]. With the development of artificial intelligence (AI) technology, the study and application of the MAS model became popular over the past two decades. MAS defines a set of agents living in a common environment, with all agents coming to a joint decision on urban space use type within this system. The MAS model can provide a more powerful tool for simulating the multi-level decision-making processing in urban space evolution than the other existing methods [
24,
25]. However, the MAS model has to be used in conjunction with CA for considering the effects of neighborhood space use type on urban space evolution [
26]. Both CA and MAS are considered “bottom-up” models.
Table 1 shows the comparison of the characteristics of the four kinds of models.
From
Table 1, it is apparent that the simulation ability of the MAS model is stronger than that of other models. However, there are still some defects in the current MAS model. For example, the interaction between macro and micro factors is rarely considered, despite having a significant effect on the simulation precision of the model. Macro factors (e.g., the natural condition, government policy) often restrict the decision-making behaviors of micro factors (a resident, farmer, entrepreneur or environmentalist), particularly in developing countries [
27]. The choice preferences of micro factors may also have an important influence on the decision-making behaviors of macro factors [
28]. If the interaction and mutual influence between macro and micro factors are neglected, the simulation precision of the model will inevitably decrease. Moreover, the simulation objects are usually unit plots of different types of land use (e.g., urban construction land, cultivated land, woodland, grassland, water bodies, etc.) in current models. If the types of land use are different from the categories of urban planning (e.g., traffic planning, garden planning, land planning) for the same unit plot at the same time, it will lead to land use conflicts.
To solve these problems, a macro–micro joint decision model is proposed in this study for improving the simulation ability of urban space evolution. In this model, all simulation objects are unified into production, living and ecological space for realizing “multiple planning in one”. Compared with current models, the proposed model has two main characteristics. One is that the interaction and mutual influence between the macro and micro factors are fully considered; the other is that “Production–Living–Ecological” (PLE) space is taken as the simulation object, replacing exclusive types of land use. For validation of the proposed model, remote sensing images, geographic information and socio-economic data from Xuzhou, China between 2000–2020 were collected and tested. The results show that the simulation precision of CA model was 87.14% (Kappa coefficient), with an increase to 89.63% when using CA + MAS model. An additional 2.68% improvement of simulation precision was achieved by using the CA + MAS + Correlation model. Moreover, the result of factor weight determination indicated that the micro factors affected the evolution of production and living space more than the macro factors. However, the macro factors had more influence on the evolution of ecological space evolution than the micro factors. Therefore, we should pay more attention to the micro factors to realize the orderly development of living and production spaces. For example, some active policies should be formulated to strengthen the ideological guidance for micro individuals (e.g., residents, farmers, entrepreneurs), helping them establish scientific views of urban space utilization. Meanwhile, macro factors should be paid more attention to ensure the sustainable development of ecological space. For example, ecological space planning should closely link with the local environment and natural conditions to improve the urban ecological carrying capacity.
5. Discussions
The scientific planning of urban space is an important way to realize urban sustainable development. The precise simulation and understanding of the leading factors in urban space evolution are essential prerequisites for urban space planning. However, there are still some defects in existing models. For example, the econometric statistics model and the system dynamics model have a strong ability to predict the size of urban space evolution, but they are unable to simulate the variation in urban spatial distribution [
36]. For another example, the cellular automata model is a key milestone in the development of urban spatial simulation technology, which can simulate not only the variation in urban space size but also the variation in urban spatial distribution, but it is unable to simulate the effects of macro-scale factors (e.g., policy, economy, culture) on urban space evolution, and nor can it simulate the decision-making behaviors of different urban agents (e.g., resident, entrepreneur, environmentalist) [
37]. With the development of artificial intelligence (AI) technology, the study of a joint model of MAS and CA has become a hot spot for urban space evolution simulation. This type of model can simulate the influences of macro factors (e.g., natural environment, geographical location, policy) and micro factors (e.g., different urban agents) on urban space evolution [
38]. Therefore, the joint model of MAS and CA has a stronger ability to simulate urban space than other models.
However, the current CA + MAS model does not consider the correlations between macro and micro factors. It is assumed that the effects of macro and micro factors are completely independent, which is inconsistent with the facts, as macro factors (e.g., the natural condition, government policy) often restrict the decision-making behaviors of micro factors (e.g., residents, entrepreneurs, environmentalists). Meanwhile, the preferences of micro factors also have an important influence on the decision-making behaviors of macro factors. Therefore, if the interaction between and mutual influence of macro and micro factors are neglected, the simulation precision of CA + MAS model will decrease. In addition, the simulation objects of current models are usually unit plots with different types of land use (e.g., woodland, grassland, water body). Therefore, the simulation result ascribes a specific type of land use to each unit plot. However, the type of land use may be different from the urban planning for a given unit plot (e.g., traffic planning, garden planning, land planning) [
39]. Therefore, if multiple plans are implemented in the same place at the same time, it will lead to difficulties in decision-making and implementation (e.g., the conflict areas) [
40].
To alleviate the above problems, a macro–micro joint decision model is proposed based on the CA + MAS model in this study to improve the simulation of urban space evolution. The simulation objects are unified into production, living and ecological space for the convenience of unified implementation of multiple plans. For validation of the proposed model, remote sensing images, geographic information and socio-economic data from Xuzhou, China between 2000 and 2020 were collected and tested. The results proved that the simulation and prediction precisions of the proposed model were better than those of current models (e.g., CA, CA + MAS) for urban space evolution simulation, particularly for the simulation of living space evolution.
It is very important for urban planning and sustainable development strategies that the precision of PLE evolution simulation models be improved, because urban space evolution simulation models are often used to compare the implementation effects of different urban planning schemes. Therefore, it is beneficial to select the optimal planning scheme for realizing sustainable urban development, if the model simulation precision is high. Otherwise, this can result in faulty government decision-making and the waste of urban resources if the model simulation precision is poor. A high-precision simulation model helps urban planners discover the problems associated with different plans in advance, and is also an important tool to analyze the effects of different factors on the evolution of urban space. Therefore, it is very useful to grasp the rules governing urban space evolution and operation mechanisms.
However, there are still some limitations of the proposed model in this study. (i) The urban space is expressed as a regular grid (30 m × 30 m) in the proposed model, but actual urban space is an irregular polygon. This led to inconsistencies between the simulation result and the RS imaging-observed results, as well as a decrease in simulation precision [
41]. (ii) In theory, the smaller the area of unit space, the higher the precision of urban space simulation. However, the computational burden increases at exponential levels with decreasing areas of unit space. Therefore, the selection of optimal geographic unit scale for simulation of urban space evolution remains an unsolved problem. For larger areas (e.g., an urban agglomeration or economic zone), an adaptive theory should be applied to define the grid scale. Regions with rapid land use change could be defined using a small grid (e.g., 30 m × 30 m), while regions with slow land use change could be simulated using a large grid (e.g., 300 m × 300 m). On the one hand, the differentiating grids could improve simulation accuracy; on the other hand, they could ensure operational efficiency. (iii) The estimated precision of the factor initial weight is strongly related to the sample selection when using a statistical method. Therefore, the selection of representative and diverse samples is a key problem. In this study, an even sampling strategy was adopted to select the experimental samples. However, this was not an optimal solution. In addition, the variation in urban population size and spatial distribution is an important factor that affects urban space evolution [
42]. In this study, the information on urban population size was used to predict the variation in living space size, but the data of population spatial distribution is not used. Therefore, urban space evolution simulation models should be further investigated considering the variations in population spatial distribution in the future.
6. Conclusions
The accurate simulation of urban space evolution and understanding of the leading factors are key issues to improve the sustainability of urban development. In this study, a macro–micro joint decision model was constructed based on the CA + MAS model in order to improve the ability to simulate urban space. A method of factor weight determination was proposed to analyze the effects of different factors on urban space evolution. For validation of the proposed model and method, experimental data (e.g., RS data, GIS data and socio-economic data) were collected and tested. The results proved that the proposed model and method were valid and reliable, and could improve the simulation and prediction of urban space evolution. The main conclusions of this study are as follows:
(1) The simulation precision of urban space evolution from the CA + MAS model was better than that from the CA model, because the decision-making behaviors of different urban agents (e.g., resident, entrepreneur, environmentalist) were considered in the CA + MAS model and not the CA model. Moreover, if the interactions and influences of macro and micro factors were considered (e.g., CA + MAS + Correlation model), the simulation precision of the CA + MAS model could be further improved. In this study, the simulation precisions (Kappa coefficient) of urban space evolution from the CA, CA + MAS and CA + MAS + Correlation models were 87.14%, 89.63% and 92.31%, respectively. The improvement in living space simulation precision was the most significant when using the CA + MAS+ Correlation model, compared with simulation of other types of spaces.
(2) The prediction precisions of CA, CA + MAS and CA + MAS + Correlation model were worse than their simulation precisions, as the factor weights in prediction models were replaced with those from the simulation models, disregarding changes that occurred during the prediction period. Therefore, errors were inevitably introduced. However, the prediction precision of the CA + MAS + Correlation model remained better than that of the CA and CA + MAS models, and the improvement in prediction precision was larger than that in simulation precision using the CA + MAS + Correlation model, compared to the CA and CA + MAS models. It was proved that the CA + MAS + Correlation model had a stronger ability to predict PLE space evolution than the CA and CA + MAS models.
(3) According to the results of factor weight determination, it was determined that the effects of micro factors on the evolution of living and production space were greater than those of macro factors. This indicated that the influences of desires and behaviors of human beings on the evolution of living and production space are increasing, correlating with the development of science and technology. The decision-making behaviors of city residents played the most important role in urban living space development. In the evolution of production space, rural residents and entrepreneurs had more influence than the other factors. By comparison, the effects of macro factors on the evolution of ecological space were more significant than those of micro factors, where the three most important factors were the location condition, neighborhood influence and natural condition. This means that the location and natural condition have more influence on the evolution of ecological space than the other factors.
According to the above analysis and conclusions, some policy implications are proposed to improve the sustainability of urban development. At present, the micro factors (e.g., city and rural residents and entrepreneurs) are the leading factors in the evolution of living and production space. Therefore, active policies should be formulated to strengthen the ideological guidance for these micro individuals, help them establish scientific views of urban space utilization, and realize the ordered development of living and production space. In addition, macro factors (e.g., location condition, natural environment) have the most important influence on the evolution of ecological space. Therefore, urban ecological space planning should closely link with the local environmental and natural conditions, to improve urban ecological carrying capacity and realize urban sustainable development.