Research Progress in Spatiotemporal Dynamic Simulation of LUCC

Abstract

Land Use and Land Cover Change (LUCC) represents the interaction between human societies and the natural environment. Studies of LUCC simulation allow for the analysis of Land Use and Land Cover (LULC) patterns in a given region. Moreover, these studies enable the simulation of complex future LUCC scenarios by integrating multiple factors. Such studies can provide effective means for optimizing and making decisions about the future patterns of a region. This review conducted a literature search on geographic models and simulations in the Web of Science database. From the literature, we summarized the basic steps of spatiotemporal dynamic simulation of LUCC. The focus was on the current major models, analyzing their characteristics and limitations, and discussing their expanded applications in land use. This review reveals that current research still faces challenges such as data uncertainty, necessitating the advancement of more diverse data and new technologies. Future research can enhance the precision and applicability of studies by improving models and methods, integrating big data and multi-scale data, and employing multi-model coupling and various algorithmic experiments for comparison. This would support the advancement of land use spatiotemporal dynamic simulation research to higher levels.

1. Introduction

Land use is one of the primary means through which human activities impact the natural environment, and it serves as a reflection of the interaction between human activities and the natural environment. Land Use and Land Cover Change (LUCC) patterns induced by human activities can have an impact on environmental changes at regional and even global scales, contributing to a range of serious environmental issues [1,2,3,4] such as the greenhouse effect, water–heat balance, urban heat island effect, climate change [5,6], food security [7,8], and a decrease in biodiversity [9]. In Chinese urban and periurban contexts, LUCC is influenced by multiple processes and factors [10], especially the dominant role of human activities in urbanization [11]. Many government planning efforts, in China and elsewhere, are focused on the intersection of human activities and environmental issues, aiming to manage the development of land use patterns in ways that strive to minimize environmental degradation. Authorities often rely on modeling and simulation to support decision-making and planning, but a very wide range of methods and tools exist and not all are equally suitable for this purpose. Therefore, a systematic understanding of spatiotemporal dynamic simulation of LUCC in different applications holds significant importance in order to improve land use management and promote ecological protection.

Research on the spatiotemporal dynamics of land use change notably includes monitoring the spatial and temporal dynamics of land use change, assessing impacts of LUCC at global or regional scales [12,13], identifying LUCC drivers, and spatiotemporal dynamic LUCC simulation. Geographic information science (GIS) and remote sensing (RS) are utilized to monitor dynamics of land use change at multiple spatial and temporal scales. Approaches to assess LUCC impacts include geographically weighted regression (GWR) [14,15], Geodetector (a statistic that measures and attributes the phenomenon of within strata being more similar than between strata across all scales) [16], logistic regression (LR) [17], etc. With respect to spatiotemporal dynamic LUCC simulation, scholars have proposed different strategies that integrate geographical methods, mathematical methods, and computer technology with the aim of accurately obtaining information about historical and ongoing LUCC and simulating future LUCC patterns [18,19,20,21]. The approaches used by these scholars aimed to shed light on how future land use patterns may change, as well as to provide input for decisions on land use management. Therefore, spatiotemporal dynamic simulation serves not only as a means to understand past and present Land Use and Land Cover (LULC), but also as a crucial tool to support and guide future sustainable LULC.

Scholars have conducted comprehensive reviews in the field of LUCC spatiotemporal dynamic simulation, covering the application of specific models and their extensions [22,23,24], overviews of LUCC in particular regions [25], research methods for LUCC spatiotemporal dynamics [26], and summaries of LUCC driving forces [27]. Building on this foundation, the present review further integrates and analyzes the methodological framework, model applications, and driving force analyses of LUCC spatiotemporal dynamic simulation to provide a comprehensive discussion and summary of its development.

2. Data and Method

LUCC is a broad area of research within geography. This review focuses on studies directly related to “geographic models and simulations in LUCC” to clarify its methodological system and related research progress. The review conducted a search in July 2024 on geographic models and simulations in the field of LUCC. In the Web of Science “Core Collection” database, the search rules were set as “TS = ((“Land use change” OR “Land cover change” OR “LUCC”) AND (simulate* OR model*))”, and the search terms were in English. In China’s CNKI “CSSCI” and “CCSD” databases, the search rules were set as “(SU = Land use change OR SU = land use and land cover change OR SU = LUCC OR SU = Spatiotemporal changes in land use) AND (SU = simulate OR SU = predict)”, and the search terms were in Chinese. The literature publication period was up to 2024. The initial database search yielded 232 publications in the Web of Science “Core Collection” database, and 20 publications in China’s CNKI “CSSCI” and “CCSD” databases. Based on the “geographic models and simulations in LUCC”, after excluding articles in medicine, materials science and other unrelated fields, we judged whether the titles and abstracts of these articles were consistent with this topic. After excluding entries unrelated to the review topic, a total of 2118 English articles and 816 Chinese articles were obtained, including 2876 methodological articles and 58 review articles. Among the 2876 methodological articles, most of them were based on the application of existing models in different study areas, and only a small number of articles integrated, improved, and developed the models. Therefore, this literature review was constructed on 113 articles (see Figure 1). Figure 2 shows the time series of the number of articles published in Chinese/English from the first article in this field in 1994 to 2024. Before 2000, it was the initial stage of land use change simulation research, with relatively few studies on model simulation; entering the 21st century, with the development of geographic information technology and the use of multi-source remote sensing data, research on land use change simulation showed a significant upward trend; since 2013, the rapid development of technologies such as remote sensing cloud platforms and machine algorithms has greatly promoted research on land use change models. Globally, China is the country that has published the most articles on the theme of land use change simulation, with 699 related English articles, accounting for 33%; followed by the United States (519 articles, accounting for 24.5%).

3. Research Status of Spatiotemporal Dynamic Simulation of LUCC

3.1. Research Methods for Spatiotemporal Dynamic Simulation of LUCC

Liu et al. [26] categorized the research methods for LUCC spatiotemporal dynamic simulation into three parts: spatiotemporal process detection, analysis of spatiotemporal process driving mechanisms, and characterization and simulation of spatiotemporal processes. Drawing on the categorization of Liu et al. [26], we have organized the basic steps of LUCC spatiotemporal dynamic simulation research, as shown in Figure 3, both to clarify the elements and processes that are typically involved and to indicate for which parts of the process the various tools and approaches are employed. Later in the text, we provide an overview of the application of different models and their extensions, as well as an analysis of question on driving forces.

For spatiotemporal process detection (see Figure 3), sensing technology can be utilized to interpret long-term land use data from multiple or single remote sensing image datasets. Depending on research needs, land use data products across various spatial and temporal scales can also be employed. By applying land use-related parameters, such as the dynamic degree of single land use and land use transition matrices, the trends and patterns of LUCC can be quantitatively studied. This allows for a preliminary exploration of the causes and driving forces behind these changes. Furthermore, quantitative models (such as SD models and Markov models) can be used to set scenarios for future LULC patterns [28].

For driving mechanism analysis, spatial analysis methods can be employed to spatialize and quantify natural factors (such as climate, terrain, etc.) and human factors (such as population, economic development, etc.). Additionally, methods like GWR and LR models can be utilized to investigate the potential driving factors of LUCC. Policy factors that cannot be quantified (such as land use policies, urban planning) can be considered as constraints on land use type conversions.

On this basis, the simulation of spatiotemporal processes can be achieved by establishing the conversion rules and modelling the simulation models.

3.2. Spatiotemporal Dynamic Simulation Models

In the following, we will describe a number of common types of LUCC models, as well as providing some examples of specific models belonging to these broad families. Based on a review of research conducted in China and other countries in the period 1998 to 2020, Qiao et al. [28] found that the most popular models could be categorized in the following types: Markov model, System Dynamics model (SD), Cellular Automaton (CA), Slope, Land use, Exclusion, Urban extent, Transportation, Hillshade (SLEUTH model), Conversion of Land Use and its Effects at Small region extent (CLUE-S), Agent-Based Model (ABM), Patch-generating Land Use Simulation (PLUS), Future Land Use Simulation (FLUS), and Geographical Simulation and Optimization System (GeoSOS). Among them, Markov and SD, serving as quantitative models, calculate the demand for future LULC patterns using mathematical methods, while the other models facilitate spatial configurations of LULC patterns. Based on the research articles treating other models, we have supplemented the characteristics and limitations (based on their specific application scenarios and software design) identified by Qiao et al. [28] (see

Table 1. Commonly used spatiotemporal dynamic simulation models of land use.

Models	Characteristics	Main Limitations
Cellular Automaton [30,31]	Changes cell states through local simple rules, generating macro-level LUCC results from the bottom up.	Transformation rules rely on empirical statistics or expert knowledge; challenges in incorporating social factors.
Dinamica EGO [32,33]	The variables in the model undergo changes as the model runs deeper.	Transformation rules rely on empirical statistics or expert knowledge.
FLUS Model [20]	Introduces adaptive inertia coefficients and a roulette competition mechanism to determine the final land use type.	Difficult to clearly reflect spatial differences in LUCC on different locations.
GeoSOS Model [18]	A comprehensive analysis platform that integrates CA, multi-agent, and bio-intelligence.	Limited in identifying non-linear transformation rules and spatial variability of conversion rules.
PLUS Model [19]	Combines the advantages of transition analysis and pattern analysis, proposing an approach that integrates CA model with patch generation simulation strategy.	Land use development is influenced by policies and future planning.
Spherical geographic automata [34,35,36]	The spatiotemporal dynamic simulation model at the global scale adopts regular hexagonal cells and Earth surface curvature to avoid spatial distortions caused by Earth projection.	The accessibility of data is constrained, and spatial scale is limited. Furthermore, as data spatial resolution increases, data volume grows, and modeling system complexity rises, higher-performance computers are required.
CLUE-S Model [21,37,38,39]	Enables spatial configuration of systematic LULC types.	Non-spatial modules require calculations with independent mathematical models; neglects the possibility of non-dominant land class conversions.
Agent-Based Model [40]	Simulates decisions and behaviors of heterogeneous agents, explaining the role of agents in LUCC.	Rules are defined based on case-specific situations, lacking universality; challenging to represent spatial behaviors of agents.
Land Change Modeler [41,42,43]	Simplifies the analysis process, and provides three methods to generate LULC transition probability maps; it can be integrated with biodiversity and greenhouse gas emission assessments.	The relationship between current LULC and explanatory factors was considered without taking into account the pattern of LULC during the calibration period.
SLEUTH Model [28]	Corrects different parameter combinations to achieve suitable results, replacing historical data, simulating LUCC.	Limited consideration of socioeconomic factors.

). Currently, for complex land change processes, it is challenging for independent models to acquire information on their changes and make simulation. Therefore, integrating diverse methods and coupling multiple models for improvement is a current trend in the development of spatiotemporal dynamic simulation models [29].

3.2.1. Research on the CA Model and Its Extended Models

CA is a dynamic model with the capability to simulate the spatiotemporal dynamics of complex systems. Its time, space, and states are all discretized [44]. The core of CA is the transformation rule. In the context of geographical simulation, CA typically employs “bottom-up” transformation rules, where the behavior of individual cells has a global impact on complex systems. Currently, most spatiotemporal dynamic simulation models are based on CA combined with other theoretical approaches, further exploring the transformation rules of CA and their potential applications.

The Markov model exhibits the characteristic of memorylessness, whereby the selection of the next state depends solely on the current state, independent of time or past states. By utilizing the Markov model, the effectively quantified LULC type transition states and transition rates can compensate for the shortcomings of CA in terms of time and quantity simulation [45,46]. Cunha et al. [47] utilized a CA–Markov model to simulate the future LULC changes in the native vegetation zone of the Paraguay river basin over the next approximately 80 years. Halmy et al. [48] utilized a CA–Markov coupling model to simulate the future LUCC in some parts of the desert in northwest Egypt, which can help the government management and protect wildlife. Azizi et al. [49] investigated the spatiotemporal dynamics of land use change and urban growth by coupling four methods: CA, Markov model, logistic regression, and analytic hierarchy process (AHP). The Remote Sensing Center of the Federal University of Minas Gerais developed the Dinamica EGO model [32], which is a dynamic stochastic model based on the CA model. The model’s parameters evolve as the model runs deeper, including the transition rules and spatial variables of the CA model. Initially used to simulate deforestation and regeneration processes in the Amazon rainforest and the comprehensive protection strategy of the Amazon basin [33], the model is now also employed for comparative urban risk assessment [50], multi-scenario LUCC modeling [33], and hydrological responses to urban expansion [51].

We divide CA’s extended models into the following two types:

(1)

CA integrated with artificial intelligence (AI).

The combination of machine learning and GIS spatial analysis can facilitate the simulation and emulation of complex LUCC [52]. Machine learning algorithms, such as Artificial Neural Networks (ANNs) [53,54], Multi-Layer Perceptron (MLP) [55,56], Long Short-Term Memory Networks (LSTMs) [57,58], Back Propagation neural networks (BPNNs) [59,60], Support Vector Machines (SVMs), and Random Forest (RF) [61], can capture the changing mechanisms of LUCC without relying on fixed functions. They effectively avoid subjectivity caused by human factors and, to some extent, reveal the regularity of change in complex systems. An ANN algorithm could be utilized to generate transition potential maps for each LULC category [62]; Tajbakhsh utilized it to identify variables influencing LUCC within a specified period of time, providing better insights into the causes and consequences of LUCC. Liu et al. [63] utilized Convolutional Neural Networks (CNNs) to extract land use patterns from past periods and combined natural and social spatial driving factors to establish a Long Short-Term Memory–Cellular Automaton model (LSTM-CA). Compared with the MLP-CA model, they found that the LSTM-CA could more thoroughly explore the intrinsic relationships in complex LUCC, leading to improved simulation accuracy. Introducing ANN, adaptive inertia coefficients, and the competitive mechanism of roulette into CA has elevated FLUS to a higher-precision predictive model [20]. Building upon the FLUS, introducing CA, agent-multi systems, swarm intelligence, and other theories and methods from complexity science and artificial intelligence, GeoSOS_V2.4 demonstrated certain advantages in simulating and optimizing complex resource environments.

(2)

Integrated models of CA.

The combination of economic models, ecological models, and Cellular Automaton (CA) models can offer insights into the rules of CA, enabling the exploration of the mechanisms behind LUCC from various perspectives. Additionally, it can reflect the impacts of changes in land use patterns on both human society and the natural environment. Chen et al. [64] integrated a dynamic global vegetation model with a CA model at the raster level, creating the coupled model (LPJ-FLUS). Compared to traditional CA models, this coupled model exhibits significant potential for analyzing environmental impacts under extreme climate conditions in the future. Liu [65] coupled a CA–Markov model with the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model, investigating the relationship between LUCC and changes in carbon storage on Hainan Island. They simulated the response of carbon storage to LUCC under different future scenarios.

PLUS v1.04 is an innovative future LUCC simulation model that integrates a land expansion analysis strategy (LEAS) and a CA model based on multi-type random patch seeds (CARS) [19]. Compared with other models, PLUS can simulate future land changes by interpreting deep-level relationships between land uses, analyzing land change strategies, and thereby improving simulation accuracy. While the PLUS model is frequently applied in domestic research, there is limited secondary development and extended application based on this model. PLUS has been used to simulate seven different land use types in the study area, addressing the limitations of the CA model in rule mining and asserting that this method demonstrates higher accuracy for short-term LUCC [66]. Additionally, some researchers conducted studies on the impact of LUCC on carbon storage in the study area by coupling the InVEST model with the PLUS model [67].

To address the data analysis and management challenges brought about by the era of big Earth data in recent decades, the Spatial Analysis and Modeling Laboratory of the Department of Geography at Simon Fraser University proposed a new generation of Earth simulation model: spherical geographic automata (SGA) [68]. SGA combines the capability of Cellular Automata (CA) to simulate complex spatiotemporal dynamic processes with a novel spherical geographic spatial model, aiming to tackle geographical simulation at a global scale. This innovative spherical spatial model accommodates Earth surface curvature and utilizes regular hexagons as fundamental units, thus avoiding shape distortions caused by Earth projection and providing uniform spatial neighborhoods for CA transition rules. SGA, developed relatively late and less applied compared to other models, is currently primarily utilized for scenario simulations of global urban land use change [34]. Additionally, it can integrate multi-criteria analysis to enhance adaptive analysis for simulating global urban land use change [34]. Furthermore, SGA can also be employed for simulating deforestation, thereby aiding decision-making for the sustainable development of global forests [36].

3.2.2. Research on the CLUE Model and Its Extended Model

CLUE is a dynamic model proposed by Veldkamp et al. [21] to simulate LU conversion and its impacts. Building upon this, the team led by Verburg [39] further developed it into a global LUCC simulation model tailored for small-scale study areas, known as CLUE-S (Conversion of Land Use and its Effects at Small regional extent). CLUE-S consists of two modules: the first module is a non-spatial land demand module, which uses independent quantity models such as the Markov model and System Dynamics (SD) to calculate land demand based on the quantities of different land use types. Another one is a spatial allocation module with unique advantages in global land use type spatial configuration. Combining the Markov model to simulate land demand with Genetic Algorithm (GA) and the CLUE-S model, Wang et al. ultimately obtained an optimized land use pattern for controlling non-point source pollution [69]. Dyna-CLUE, an enhanced version of CLUE-S [38], combines top-down land use demand allocation with grid units and integrates it with the bottom-up land use conversion process. Dyna-CLUE was used to scenario simulate future LULC in the Munneru river basin, India [70]. Through six different user-defined scenarios, they analyzed LULC change patterns for the years 2030, 2050, and 2080, providing valuable decision support for understanding human activities’ impact on natural resources and promoting sustainable land and water resource management. CLUMondo, developed by van Asselen and Verburg [37], offers an innovative method to global-scale LULC modeling. It treats the land system as the modeling object, driven by macro factors such as food production, livestock development, and capital investment, analyzing the dynamic simulation of land use system competition based on driver analysis. The simulation made by this model, including the relationship between location factors and location adaptability during the simulation period, remain constant, but this spatial relationship may change with new technologies and social changes.

3.2.3. Research on the ABM Model

Agent-Based Modeling (ABM) is originally developed as a computational simulation model for studying complex systems. The interaction among the “agents” within the model can induce macroscopic changes in the system. For example, the behavior of “agents” representing land parcels can manifest in the conversion of agricultural land to urban development. Therefore, it provides a new perspective for spatiotemporal dynamic simulation of LUCC. Similar to the “bottom-up” modeling approach of CA, ABM is a model based on translation rules, where multiple independent “agents” generate changes that influence the global outcome. It better reflects the complex human–land relationships and has gradually been applied in LUCC spatiotemporal dynamic simulation [71]. Ligtenberg et al. [40] coupled ABM with CA models to establish a land use scenario simulation model. Hosseinali et al. [72] simulated the land development process under the influence of individual decisions by government and developers. Mirzahossein et al. [73] utilized ABM to simulate the trend of urban residential land expansion. Li et al. [74] proposed a new ABM model to simulate the growth of urban residential land in Nanjing city. As the ABM model relies more on the independent behavior of “agents” to influence global features, simulating complex LUCC may require consideration of a large number of “agents”, necessitating support from high-performance computing equipment. Therefore, existing studies often focus on simulation for single land types.

3.2.4. Research on Land Change Modeler

Land Change Modeler (LCM) is a land planning and decision-making module within the TerrSet 2020 software (AKA IDRISI). Similar to the design philosophy of other models, LCM is an empirical model based on the relationship between LUCC patterns and influencing factors. What sets LCM apart is its consideration that certain influencing factors change along with alterations in LULC. These factors, referred to as “dynamic variables,” enable the staged calculation of LUCC transition potential [42]. LCM effectively integrates functions such as LUCC analysis, LUCC potential modeling, LUCC simulation, and ecological environment modeling, making it a multifunctional tool aimed at ecological sustainability research. It has been applied in dynamic LULC monitoring, LUCC simulation, urban growth, and more. Rajit et al. [43] utilized LCM to investigate the impact of human activities on LUCC within a wildlife sanctuary, simulating the future LULC patterns of the sanctuary and recommending regulatory measures to mitigate human-induced disturbances. Jatin et al. [41] utilized LCM to simulate future LULC and further integrated it with the SWAT model to evaluate watershed hydrological responses to LUCC.

4. The Main Issues in the Existing Research

This review summarizes the following three questions based on the retrieved review articles and method articles related to model application.

4.1. Methodological Issues for Spatiotemporal Dynamic Simulation of LUCC

While existing studies on LUCC simulation and model applications can to some extent reveal the external connections between LUCC and driving factors, this method still faces limitations when dealing with complex real-world scenarios that need to be overcome. The objectives that these studies are concerned with are real geographical entities and phenomena after digitization. In practical applications, the processing of these data is influenced by the subjective viewpoints and preconceived assumptions of researchers [75]. Although satellite image classification has been highly automated, the selection of training data in machine learning algorithms still requires the intervention of artificial visual image interpretation. Some prior knowledge gained from manual interpretation provides targeted support for training samples, aiding image classification and analysis [76]. This leads to a situation where certain features expressed in images of specific landforms or vegetation regions may lack “universality” in large-scale LULC data or products. Moreover, during the LUCC simulation process, setting the weights for various land use transformations often relies on expert judgment, which also introduces subjective biases. Data generated purely through computational models and empirical methods often lack realism and applicability.

Moreover, because the data generated through computer models often lack a certain degree of authenticity and applicability, it is also essential to consider the role of human agency. Over the past few millennia, humans have impacted three-quarters of the Earth’s land surface [77], primarily through land use and modification. Therefore, human factors are indispensable in the study of spatiotemporal dynamics of LUCC, with particular attention needed for government-led policies and regulations. Effective policies and regulations can constrain and mitigate disorderly land expansion and change to some extent.

By integrating computer-aided analysis with human factors, we can gain a more comprehensive understanding of the complexity of LUCC. Computers can provide objective, data-driven analysis, while human factors emphasize the importance of human agency and policy factors. This combined approach enables us to more effectively address the challenges in LUCC research, promoting land use planning and management towards more scientific and sustainable development.

4.2. Model Issues

The mathematical methods employed in LUCC simulation models typically require input data to have identical row and column information and the same spatial resolution. However, the available data are sourced from various origins, covering different spatial and temporal scales, including remote sensing data and population distribution data [78]. This disparity makes it challenging to achieve spatial and temporal scale uniformity between LUCC data and explanatory variables. Such diversity presents challenges for constructing LUCC models, as different explanatory variables exhibit different behaviors across various spatiotemporal scales, and the same explanatory variable may behave differently at different scales. Transforming explanatory variables across scales can also result in cumulative errors.

As populations continue to grow, LUCC becomes more pronounced in areas with frequent human activity, particularly under the mandates of national and local government policies and regulations [79,80]. Current models, though capable of incorporating some quantitative data related to population or economy, primarily use policy or conceptual drivers that are more often used to constrain the expansion or contraction of LULC patches in these regions. There is a lack of systematic theoretical basis for selecting data scales, and achieving uniformity in spatiotemporal scales while ensuring data precision remains a challenge. The factors influencing LUCC encompass various natural and human dimensions, which exhibit complex interactions when further detailed. The core of LUCC simulation models lies in the conversion rules between different land types. Relying solely on predefined conversion rules and traditional single models is insufficient to meet the demands of complex LUCC simulation research. In the calibration process of existing models, the datasets for explanatory variables are often fixed; the variables used in the first calibration step are also employed in subsequent simulation stages. This approach may overlook changes that have occurred over the years, such as newly constructed roads and subways that could affect land use types.

To address these issues, future LUCC models need more flexible and dynamic methods capable of comprehensively handling data from different sources and scales, while accommodating dynamic changes in explanatory variables. This requires breakthroughs in data integration and processing technologies, as well as innovative theoretical methods to manage the complexity and diversity of LUCC research. By introducing more advanced computational methods and considering a broader range of variables, we can develop more accurate and reliable LUCC simulation models, providing scientific foundations for land use planning and management.

4.3. Issues in the Driver Analysis

Complex LUCC is a long-term process with a high degree of spatial heterogeneity, often influenced by natural or human factors. Analyzing the driving forces can thoroughly explore the internal mechanisms causing LUCC, provide data and theoretical support for studying the future development of land and internal model conversion rules, offer a scientific basis for government decision-making and regional planning, and also be a core aspect of researching the spatiotemporal dynamics of LULC. In the analysis of driving forces, major methods include principal component analysis (PCA), partial least squares regression (PLSR), geographically weighted regression (GWR), logistic regression (LR), analytic hierarchy process (AHP), grey relational analysis (GRA), etc. These methods can explore the driving forces behind LUCC and guide future studies on simulating LULC patterns.

Table 2. Commonly used methods for driving force analysis.

Method	Features
Principal Component Analysis (PCA) [81]	Facilitates a more understandable and usable outcome of the dataset.
Partial Least Squares Regression (PLSR) [82]	Suitable for complex samples and situations with multiple variable correlations.
Geographically Weighted Regression (GWR) [14]	Explores spatial variations and related driving factors at a specific scale.
Logistic Regression (LR) [17]	Highly efficient, requires minimal computational resources, and offers calibrated prediction probabilities without feature scaling.
Analytic Hierarchy Process (AHP) [83]	Treats the research object as a system, making decisions through decomposition, comparison, judgment, and synthesis.
Grey Relational Analysis [84]	Applicable to varied sample sizes and irregular patterns, regardless of whether the samples exhibit regularities.

lists commonly used methods for driving force analysis.

For the case of the spatiotemporal changes in LULC, the observed results are often discrete and discontinuous. LR analysis, capable of handling both continuous and categorical variables, can be utilized to further explore the systematic transition rules of CA [17]. However, there is a spatially non-stationary relationship between the driving factors and LULC. Utilizing ordinary logistic regression to determine the transition rules in CA model may overlook this aspect. Thus, some scholars have applied the GWR to calculate the transition probabilities of land use types, introducing spatial non-stationarity to the transition rules of CA models [14,15,85]. For other models, there are similar coupling applications. PLUS also can combine multiple linear regression and the Markov model, to explore the spatiotemporal characteristics of landscape pattern changes and Landscape Ecological Risk (LER) in the study area from 2000 to 2020 under the Shared Socioeconomic Pathways (SSPs) proposed by the Intergovernmental Panel on Climate Change (IPCC) [86]. Utilizing autologistic regression analysis, which could reflect the neighborhood abundance factor of land self-organization processes to improve the spatial allocation module of the CLUE-S model, could enhance the model’s simulation accuracy [87].

LUCC is the result of the interaction between natural and human factors in a region [86]. In current research, the driving factors influencing LUCC are summarized into natural and human factors [88,89,90,91], which can be further detailed into different driving factors, as shown in

Table 3. Common factors of LUCC.

Type	Driving Factor	Examples
Natural Factors	Topography	Elevation, Slope, Aspect, etc.
	Climate	Temperature, Growing Degree Days, Precipitation, Sunshine, Wind, etc.
	Soil Properties	Soil pH, Soil Salinity, Soil Organic Matter, Soil Thickness, etc.
Human Factors	Development	GDP, Technological Development, etc.
	Population	Population Distribution, Age Structure, Education Level, etc.
	Locational Conditions	Urban Centers, Roads, Rivers, Transportation Facilities, etc.
	Institutions, Policies	Government Reports, National and International Ecosystem-related Plans

.

Natural factors such as topography, climate, and soil properties are crucial in inducing LUCC [92]. The undulation and slope of the terrain directly impact the availability and manner of land utilization. In mountainous regions with significant elevation differences and steep slopes, the limited stable land is more suitable for land cover types like forests, grasslands, or bare land. In contrast, flat plains are more conducive to agricultural production, leading to a predominant use for cultivation and farming. Additionally, climate conditions, such as precipitation and temperature, play a vital role in the growth and development of different vegetation and crops, thereby influencing the land use patterns. For instance, in arid regions where water resources are scarce, agricultural activities are relatively limited, and more land is designated for grasslands or other land cover types suitable for arid environments [93]. Climate conditions also impact vegetation distribution; for instance, in tropical rainforest regions, a humid climate fosters lush vegetation growth, resulting in expansive rainforests. The texture and fertility of the soil determine its cultivability and utility. Fertile soils are suitable for crop cultivation, thus being predominantly used for agricultural purposes, while infertile soils hinder agricultural development and may be allocated for forest or grassland use.

The level of economic prosperity in a country or region often influences land use patterns. Higher levels of economic development are typically associated with larger-scale industrialization and urbanization processes, leading to increased demand for urban land for constructing factories, residential areas, commercial centers, etc. [94,95]. This can result in the encroachment on agricultural facilities and ecological land, such as forests [96], posing threats to food security [97] and biodiversity. Advanced technologies and engineering methods can alter land use patterns. For example, land reclamation and irrigation projects make previously unsuitable land for agriculture more suitable, with fields becoming more level and irrigation facilities more sophisticated. The development of transportation technologies like cars and high-speed trains can also change land use patterns by increasing the demand for stations, parking lots, etc. Changes in population size and distribution have profound effects on land use. Areas with high population density often require more residential and commercial areas and infrastructure, accelerating the urbanization process. On the other hand, the growing population may necessitate the cultivation of grasslands and undeveloped land to expand farmland or enhance agricultural productivity to meet food demands [98]. Geographical location and spatial conditions, such as areas close to transportation hubs, are more suitable for developing industries, logistics, and commercial centers. Government institutions and policies are the most direct factors influencing current LUCC. Government management of land planning and allocation, policy direction for agriculture, industry, and urban development, all directly impact land use patterns.

5. Discussion

5.1. Data Scales and Models

High spatial resolution remote sensing imagery interpretation data or land use products offer significant advantages in acquiring land use information, particularly in areas with fragmented land use distribution. This type of imagery can provide more detailed land use information, especially in urban built-up areas or systems with pronounced spatial heterogeneity, where its superiority is more pronounced. Additionally, when combined with more accurate digital elevation models and reliable population distribution data, it facilitates a thorough exploration of the factors influencing LUCC. In global-scale spatiotemporal dynamic simulation studies of land use, low spatial resolution data, while practical, falls short in capturing changes in detail, especially in scenarios requiring high-precision information, such as urban built-up areas. For different scale study areas, the driving factors of land use change, such as soil acidity, climate, temperature, etc., exhibit significant variations. In large-scale regional studies, these factors have a relatively pronounced impact on land use changes. Conversely, in small-scale regional studies, the internal differences in these factors are not as apparent and may not be the primary considerations for research. Therefore, considering the trade-off between practicality and accuracy, the choice of data resolution still needs to be comprehensively considered for different scales of land use research.

However, remote sensing imagery data are often affected by meteorological factors such as clouds, fog, and aerosols, as well as sensor malfunctions, resulting in data loss and inconsistent spatiotemporal information. This poses challenges for long time series land use studies, necessitating the use of advanced data processing and correction techniques to ensure data quality and continuity. The growth of vegetation and crops significantly influences land use patterns. Vegetation exhibits strong reflectance during the growing season, and different vegetation types show distinct spectral characteristics in different seasons. Agricultural activities, such as planting and harvesting, also impact the information in remote sensing images during specific seasons. Therefore, when conducting LUCC analysis, seasonal changes’ impact on remote sensing data must be thoroughly considered to obtain accurate and reliable research results.

To address challenges like precise identification of driving forces, spatiotemporal scale issues, and multi-scale adaptation of models, diversification, and coupling of multiple models have become mainstream trends in this field [99,100,101]. This approach allows models to complement each other, considering multiple influencing factors comprehensively, and gaining a comprehensive understanding of the internal mechanisms driving land use changes. Sun et al. utilized Generative Adversarial Networks (GANs) [102] to calculate the future states of pixels according to transformation rules. According to their research, this not only overcomes the subjectivity of traditional model transformation rules but also combines spatial planning data with artificial intelligence models. It achieves a fusion of subjective policy factors with objective data generated through unsupervised learning, opening up new perspectives for land use simulation research.

5.2. Challenges and Prospects of Spatiotemporal Dynamic Simulation

The widespread application of high-resolution satellite remote sensing data, unmanned aerial vehicle (UAV) technology, and a range of advanced ground measurement tools has provided multi-source data and more accurate spatial information for the spatiotemporal dynamic simulation of LUCC. The popularity of these technologies offers researchers more comprehensive and detailed information on LUCC. With the continuous upgrade of measuring instruments, other data related to Earth observation, including physical and chemical parameters, will also achieve higher precision, providing robust data support for LUCC spatiotemporal dynamics and other research fields. However, integrating these multi-scale data into a high-computational-efficiency, large-capacity database remains a significant challenge. This process requires addressing issues such as spatiotemporal scale consistency, format compatibility, data processing complexity, and data storage and management technical difficulties. Additionally, maintaining the accuracy and consistency of data during integration is a crucial problem that researchers must overcome.

Artificial intelligence is continually aiding scientific research, opening up new paradigms. In recent years, large AI models like Generative Pre-trained Transformer-4 (GPT-4), Bard, and Wenxin have achieved remarkable success in large-scale data processing, profoundly impacting LUCC spatiotemporal dynamic simulations and geographic research as a whole. Large language models, represented by GPT-4, are not only expected to enhance the progress of scientific research but also assist scientists in proposing more innovative hypotheses, guiding experimental design, and rapidly interpreting large-scale data [103]. The introduction of this technology promises to provide unique insights for scientific exploration, surpassing the limitations of traditional methods and bringing disruptive development to research fields. In land use research, we can analyze and process vast datasets to gain a deeper understanding of the impact of human activities on the Earth system. Through these models, we can more accurately simulate future land use changes, thereby proposing more scientific sustainable development strategies. The emergence of these large language models accelerates the acquisition of new knowledge and experiences, marking a transition from traditional rule-based models to more adaptive and flexible approaches, injecting new ideas and technical means into land use research. This revolutionary change not only enhances the efficiency of data processing and analysis but also provides scientists with unprecedented tools to explore and understand complex geographical phenomena more deeply, promoting the development of land use research and the entire field of geography. Based on past trends, machine learning methods often extrapolate under assumptions of a “status quo” [104], which poses a challenge for LUCC, influenced by complex human activities.

LUCC is both the result of various interacting factors and a factor influencing these variables. While deeply exploring future LUCC, it is also necessary to focus on the multi-level responses these changes might trigger, such as Ecological Security Patterns (ESPs) (which consist of ecological nodes, ecological corridors, and ecological sources) [105], biodiversity [106], and ecological risks [107,108,109,110]. Specifically, research on ESPs can help us identify and protect important ecological areas, ensuring the stability and functionality of ecosystems. Biodiversity studies contribute to the maintenance of various species and their habitats, preventing ecosystem degradation. Ecological risk assessments can help us identify and simulate potential ecological crises, providing a basis for formulating preventive and response measures. By integrating these multi-level response studies, it may be possible to construct a more comprehensive integrated LUCC model, encompassing functions such as digital image processing, remote sensing image interpretation, multi-method driving force analysis, LUCC simulation, and ecological response, and offering researchers the capability for secondary development. This comprehensive research approach helps advance LUCC spatiotemporal dynamic simulation research to a more thorough and in-depth level, adapting to the continually changing natural and human factors.

6. Conclusions

The development of spatiotemporal dynamic simulation models for land use is progressing in the direction of positive multi-model coupling, with new models continually emerging and undergoing improvements. These computer models, by integrating the advantages of various theoretical approaches such as mathematical methods and geographic theories, further explore the complex causes of changes in human–environment systems. This approach, to some extent, addresses the limitations of single models and contributes to a more comprehensive simulation of land use patterns under various scenarios in the future. It provides crucial theoretical support for regional planning and development.

The existing models have already found extensive applications in areas such as urban expansion, delineation of future urban boundaries, multiple-scenario simulations of future land use, and ecological effects [61,111,112,113]. They play a significant role in decision support and ecological assessments. As a core aspect of research in this direction, in-depth exploration of the driving forces behind LUCC becomes particularly crucial, considering the influence of numerous factors on the development of complex human–environment systems. This research is essential for achieving more realistic simulations of future land use patterns.

However, current research faces challenges such as data uncertainty, making it necessary to advance new technologies through diverse data sources. In future studies, improvements are expected to be made upon existing models and methods, integrating big data and multi-scale data, conducting experimental comparisons with multi-model coupling and different algorithms, and continuously refining research methods. The aim is to provide more applicable research approaches for specific regions and propel spatiotemporal dynamic simulation studies of land use to a higher level.