Dynamics of Land Use/Land Cover Considering Ecosystem Services for a Dense-Population Watershed Based on a Hybrid Dual-Subject Agent and Cellular Automaton Modeling Approach

Yutong Li; Yanpeng Cai; Qiang Fu; Xiaodong Zhang; Hang Wan; Zhifeng Yang

doi:10.1016/j.eng.2023.10.015

Engineering ›› 2024, Vol. 37 ›› Issue (6) :182 -195. DOI: 10.1016/j.eng.2023.10.015

Research

Article

Dynamics of Land Use/Land Cover Considering Ecosystem Services for a Dense-Population Watershed Based on a Hybrid Dual-Subject Agent and Cellular Automaton Modeling Approach

Author information +

History +

PDF (3607KB)

Abstract

Land use/land cover represents the interactive and comprehensive influences between human activities and natural conditions, leading to potential conflicts among natural and human-related issues as well as among stakeholders. This study introduced economic standards for farmers. A hybrid approach (CA-ABM) of cellular automaton (CA) and an agent-based model (ABM) was developed to effectively deal with social and land-use synergic issues to examine human-environment interactions and projections of land-use conversions for a humid basin in south China. Natural attributes and socioeconomic data were used to analyze land use/land cover and its drivers of change. The major modules of the CA-ABM are initialization, migration, assets, land suitability, and land-use change decisions. Empirical estimates of the factors influencing the urban land-use conversion probability were captured using parameters based on a spatial logistic regression (SLR) model. Simultaneously, multicriteria evaluation (MCE) and Markov models were introduced to obtain empirical estimates of the factors affecting the probability of ecological land conversion. An agent-based CA-SLR-MCE-Markov (ABCSMM) land-use conversion model was proposed to explore the impacts of policies on land-use conversion. This model can reproduce observed land-use patterns and provide links for forest transition and urban expansion to land-use decisions and ecosystem services. The results demonstrated land-use simulations under multi-policy scenarios, revealing the usefulness of the model for normative research on land-use management.

Graphical abstract

Keywords

Land use/land cover / Human-environment interactions / Agent-based model / Cellular automaton

Cite this article

Download citation ▾

Yutong Li, Yanpeng Cai, Qiang Fu, Xiaodong Zhang, Hang Wan, Zhifeng Yang. Dynamics of Land Use/Land Cover Considering Ecosystem Services for a Dense-Population Watershed Based on a Hybrid Dual-Subject Agent and Cellular Automaton Modeling Approach. Engineering, 2024, 37 (6) : 182-195 DOI:10.1016/j.eng.2023.10.015

登录浏览全文

4963

注册一个新账户忘记密码

1. Introduction

Land use/land cover (LULC) describes the activities, arrangements, and inputs of people natural land [1]. Many researchers have used it to reflect anthropogenic and natural features observed on Earth for a long time [2], [3], [4]. Land-use cover change (LUCC) is a continuous and dynamic long-term series of LULC transactions. They can replicate the changes and trends in human activity and the natural environment [5], [6], [7], [8]. However, LUCC is a labyrinthine process that consists of the impact between human individuals or parcels of land and the interplay between the surroundings and human beings [9], [10]. There are simultaneously many driving factors causing LUCC, which is regarded to be an issue [11], [12], [13], [14], [15]. This issue challenges the behaviors of stakeholders, including decision-makers and farmers. This study provides significant assistance in enhancing the efficiency of resource usage and sustainable development. In addition, it promotes coordinated regulation at multiple scales [16], [17], [18], [19]. Therefore, it is desirable to understand the impact and driving factors of the two-way interaction between human beings and their natural surroundings on LUCC.

Prediction and simulation have been the most broadly studied methods in previous studies on LULC. Most scholars believe that the prediction and simulation of LULC using time-series data under various scenarios is essential for future LULC management strategies [1], [20]. Many models have been developed to predict and simulate spatial transformations to simulate future LULC. A convenient approach, the cellular automaton (CA)-Markov model, has been proposed for temporal-spatial dynamic model of LUCC in complex systems [21], [22], [23], [24], [25]. The CA-Markov models are widely used [26], [27], [28]. The CA-Markov model has suitable spatial evaluation capacity [29]. Many researchers have successfully conducted LULC simulations using the CA-Markov model [30]. For instance, Permatasari et al. [9] and Naboureh et al. [31] predicted LULC using the CA-Markov model and announced land-use policies in China and Indonesia, respectively. However, there are limitations to the CA-Markov models, which include the inability to simulate human decision-making and correct land-use changes due to human disturbances.

The drawbacks of the CA-Markov model can be overcome using agent-based model (ABM). ABM is a simulation approach in which the behavior of agents or individuals is programmed through their interactions and selection principles [32], [33], [34]. At the same time, it can be used to examine household properties and income under the land-use change decision module. It can be used to predict LULC, the interactions between various agents or components [35], and the intrinsic properties of the land (such as elevation, slope, accessibility, soil type, and productivity). The ABM built by Walsh et al. [36] consisted of a population transfer module, household assets module, land suitability module, agricultural crop production module, and chemical fertilizer applications. Giri et al. [32] incorporated hydrological processes into the ABM to solve critical problems in which the spatial distribution of blue and green water varies with land-use changes. Zhao et al. [37] simulated the decision-making processes of stakeholders based on an agent-based model for land-use allocation. An agent-based model was constructed to comprehend the period of land degradation and famine in Democratic People’s Republic of Korea and investigate feasible strategies for reducing risk [38]. By combining a semi-distributed hydrologic model and an agent-based model, they evaluated the impact of projected changes in temperature and precipitation on streamflow while also considering the influence of these variables on land-use decisions [39]. The ABM framework has also been widely applied to space-clear land-use decisions that promote the scale of sections and compartments [40], [41], [42], [43] and has been used to link land-use policies to potential land-use patterns to avoid decision-making limitations. However, a disadvantage of ABM is that spatial evaluation cannot be performed. Therefore, this research will reveal the two-way interplay between human beings and the natural environment on the LUCC under the influence of human decision-making at the temporal-spatial scale, combining the CA-Markov and ABM.

Taking into account the previous investigation and the current trend towards LUCC projections, this paper discusses LULC under multi-policy scenarios (Fig. 1). This study aimed to develop an approach based on CA and a multi-agent-based model to investigate LUCC patterns and monitor dynamics under several policy scenarios. The subjects of the agent-based CA-spatial logistic regression (SLR)-multicriteria evaluation (MCE)-Markov (ABCSMM) model are divided into two categories: human agents, who make land-use decisions based on their expected utility in each period, and land cell agents, who obtain environmental attribute parameters that affect the expected utility of human agents. The results of this research will contribute to the interaction between humans and the environment and construct a novel scientific understanding developed by analyzing the spatiotemporal variations in LULC and their connection to ecosystem services (ESs). The following innovations in the research include: ① coupling the CA model with spatial self-organization and the agent-based model with emergence to explore human-environment interactions; ② calculating the suitability mapping module of urban land, ecological land, and constrained land; and ③ linking the land-use transition to the ES and farmers’ financial conditions. Furthermore, policymakers and researchers can use the results of this study for sustainable land-use management.

2. Methodology

The process of the ABCSMM model is divided into three steps. Fig. 2 shows the hierarchical structures of the three processes. The first phase included the selection of driving factors and constraints. The second phase involves the creation of a suitability map. The third stage of the ABCSMM is based on current land-use data-initializing agents and parameters. The agents of the ABCSMM are divided into two categories: human agents, who make land-use decisions according to their perceived expected utility in each period, and land grid cell agents, who obtain some environmental attribute parameters that affect the expected utility of human agents. These parameters include the probabilities of annual land-use conversion, attraction of land-use types to human agents, and initially expected utilities of human agents. These parameters were generated using previous procedures or were estimated based on agricultural surveys. Details of these parameters are presented in the following sections. The ABCSMM modifies the characteristics of both human and land cell agents, and subsequently classifies them according to their updated features. From 1999 to 2030, the ABCSMM simulated land-use patterns every year.

2.1. Driving forces and constraints

2.1.1. Drivers for LULC change

In general, changes in land use are affected by several factors. Previous studies have revealed that factors affect land-use changes [44], [45], [46]. Socioeconomic data included population, gross domestic product (GDP), and distance. The natural data contained elevation and slope data. All factor data must be normalized to eliminate the influence of the data dimension on subsequent simulation studies. The standardization process in this study adopted a fuzzy-set membership function divided into positive and negative directions. The positive fuzzy-set membership function is:

(1)

a_{ij}^{'} = \{\begin{matrix} 1 \\ \frac{a_{ij} - \min (a_{ij})}{\max (a_{ij}) - \min (a_{ij})} \\ 0 \end{matrix} \begin{matrix} a_{ij} > \max (a_{ij}) \\ \min (a_{ij}) \leq a_{ij} \leq \max (a_{ij}) \\ a_{ij} < \min (a_{ij}) \end{matrix}

The negative fuzzy-set membership function is:

(2)

a_{ij}^{'} = \{\begin{matrix} 1 \\ \frac{\max (a_{ij}) - a_{ij}}{\max (a_{ij}) - \min (a_{ij})} \\ 0 \end{matrix} \begin{matrix} a_{ij} > \max (a_{ij}) \\ \min (a_{ij}) \leq a_{ij} \leq \max (a_{ij}) \\ a_{ij} < \min (a_{ij}) \end{matrix}

where a_ij is the value of data j in data type i, a_ij′ is the value after normalization, max(a_ij) is the maximum data, and min(a_ij) is the minimum data. The factors must be standardized on a scale of 0 to 1.

2.1.2. Policy setting and scenario development

Using the ABCSMM model, different constraints were set to predict land-use types in 2030, which can be divided into six scenarios (S1: balancing development scenario; S2: grassland protection scenario; S3: ecological protection scenario; S4: forest protection scenario; S5: economic development scenario; S6: cultivated land protection scenario). S1 considers urban expansion and ecological protection as the standard scenarios. S2 prioritized protecting grasslands, whereas S4 and S6 prioritized protecting forests and croplands. The original grasslands remained unchanged in the S2 and other land types were converted according to certain rules. The S4 and S6 scenarios are similar in that they maintain the previous year’s forestland and cropland. S3 prioritizes the safety of three ecologically friendly land types: cropland, forestland, and grassland, which are beneficial for regional ecological development. The original ecological land remained unchanged, and other land types were converted according to certain rules. S5 prioritized urban development. The original urban land used for economic development remained unchanged, and other land types were converted according to certain rules. The details of the conversion rules are explained in the following section, forming an atlas of environmental suitability and constraining the transfer of land-use changes and probability.

2.2. Land-use suitability analysis and mapping

2.2.1. Spatial logistic regression

The primary objective of the SLR model is to establish a regression correlation between multiple predictors and binary variables, indicating the likelihood of an event taking place [47]. This study proposed a city expansion model to predict impervious expansion patterns using the SLR method. The SLR model is a fundamental algorithm that utilizes geospatial information systems, remote sensing images with flexible time intervals, and diverse environmental variables. The SLR model uses socioeconomic and environmental data as inputs, and generates urban and non-urban areas as outputs. The model was evaluated using receiver operating characteristic (ROC) [48]. By comparing a probabilistic image that describes the likelihood of an event occurring with a binary image that displays its location, the ROC method is an effective approach for assessing a model’s ability to predict event occurrence [49]. In model validation using ROC, a summary of the point coordinates on the curves used to calculate the ROC values, ROC curves and ROC values was reported. The ROC curve varies from positive to positive when the threshold changes between 0 and 1, and the ratio of false positive to negative classification [50]. The functional form below expresses the empirical relationships between the conversion probability of residential land and the influencing factors.

(3)

P (Y = 1 | x) = \frac{\exp (\sum \partial_{i} X_{i})}{1 + \exp (\sum \partial_{i} X_{i})}

where P(Y = 1|x) is the probability of the dependent variable, which is binary or dichotomous; x is independent variable; Y = 1 means a change in a cell of the raster map, transitioning from non-residential land use; X_i is independent variables including X₀ for the constant term; ∂_i is coefficients of variables (parameters). To estimate the coefficients, a linear logit transformation is applied to both sides of Eq. (4) in the logistic function, which considers the linear probability in a set of parameters with a probability ranging between 0 and 1 [32].

(4)

Y = logit (p) = \ln (\frac{P_{k}}{1 - P_{k}}) = \partial_{0} + \partial_{1} x_{1 k} + \partial_{2} x_{2 k} + \partial_{3} x_{3 k} + \dots + \partial_{12} x_{12 k}

where Y is the probability that the dependent variable is 1; p is the probability of the dependent variable; P_k is the predicted probability of the dependent variable of non-urban conversion into residential land use; ∂₀ is the intercept; ∂₁, …, ∂₁₂ are coefficients for variables; x_1k, …,x_8k reflect the distance to the nearest: hospital (X₁), park (X₂), railway (X₃), road (X₄), school (X₅), shop (X₆), station (X₇), and waterway (X₈); x_9k, …,x_12k reflect the elevation, slope, population, and GDP in year k, respectively. The symbol of the parameter represents the direction of influence of each interpretation variable on the conversion probability. The negative coefficient symbol indicates that the possibility of conversion is reduced by increasing the distance between the cell and adjacent externalities. In the model, distance to the city center expresses the degree of convenience to schools, stations, shops, and hospitals [51]. The proximity of roads indicates the level of accessibility to metropolitan and urban centers, stations, shops, and schools [52]. The distance to parks and streams is a relative measure of aesthetic amenities close to parks and streams. Areas with relatively large populations and high GDP attract more residents. Therefore, these factors are expected to positively affect the probability of residential land conversion [53].

2.2.2. Multicriteria evaluation

Non- resident land-use types (cropland, forest, shrub, grassland, and barren) were mainly influenced by natural factors (elevation and slope). The main purpose of this section is to evaluate the Dongjiang Basin (DRB) area of an optimal spatial combination of non-resident land use based on optimal potential suitability using the MCE and fuzzy methods. The goal of MCE is to explore and evaluate various options based on multiple criteria and objectives [54]. This approach combined several criteria to produce a single score for each decision. Eq. (5) represents the mathematical model for evaluating land suitability using multiple criteria [55].

(5)

S = f (x_{1}, x_{2}, \dots, x_{n})

where S is the land suitability measure, x₁,x₂,…,x_n are standards that affect the suitability of the land. Multi-criteria analysis based on geographic information system (GIS) was performed using Boolean superposition, a weighted linear combination (WLC), and an ordered weighted average (OWA). The WLC compensation decision rule is a GIS-based decision-support tool [56].

2.2.3. CA-Markov chain

The CA-Markov model incorporates both the traditional Markov model and CA to create a time-series model. In the CA-Markov model, the Markov chain model is a discrete-time random model that uses a moving probability matrix to simulate LULC change probability [14], [57]. The probability of transitioning from one state (t₁) to another (t₂) was utilized to establish the transition trend among different LULC states [58], [59]. The mathematical expression of the Markov model is given in Eq. (6):

(6)

S (t + 1) = P_{ij} S (t)

where S(t + 1) is the status of the LULC at time (t + 1), P_ij is a transitional matrix (7):

(7)

P_{ij} = [\begin{matrix} p_{11} & \dots & p_{1 n} \\ \dots & \dots & \dots \\ p_{n 1} & \dots & p_{nn} \end{matrix}]

where pij is the conversion probability between pairs of land-use types, 0≤p_ij≤1, and $\sum_{j=1}^{n} p_{i j}=1$; i and j are the land uses, i, j = 1, 2, …, n. The matrix depicts the past and present LULC classes in rows and columns, respectively. Furthermore, the mathematical representation of the CA is shown in Eq. (8):

(8)

S (t, t + 1) = f (S (t), H)

where H = h × h.

The CA-Markov model uses both CA and the Markov model to forecast LULC. The CA-Markov model applies a standard filter with a size of 90 × 90 for prediction. One of the drawbacks of the Markov chain model is that it cannot offer a spatial distribution of LULC events but can provide an approximation of their scale [60], [61], [62]. CA can change and control complex spatial distribution processes and can simulate the ability to simulate space-time of powerful complex systems [63], [64]. The model utilizes the interplay of cell space, mesh size, cell neighborhood, and transition rules to create intricate patterns and accurately depict phenomena [65], [66], [67]. Kappa indices were employed to assess the consistency between the simulation and classification of the LULC status diagrams [59], [68], [69].

2.3. Agent-based two-way interaction analysis

2.3.1. Land cell agents

The ABCSMM land unit agent was defined as a 90 m × 90 m land grid cell. The land cell agent of the designated year has several attributes, including land-use type, land ownership, attraction to various human agents, neighborhood index, and sensing value for two ESs important to farmers: food provision (crops and livestock) and other ESs (non-food provisioning, supporting, regulating, and cultural services) [70].

The CA model includes the effect of neighboring land use. One method to improve the estimation of these proximity-effect parameters is to use spatial land-use change data for empirical parameterization. This practical method can be used to quantify the spatial externalities of adjacent land use and location features using the observed land-use transformation [71], [72]. The neighborhood index was introduced to represent the role of the neighborhood in this research. The following formula was used to calculate the neighborhood index:

(9)

σ_{ij}^{z} = \frac{\sum_{n \times n} con (s_{ij} = L_{landuse})}{n \times n - 1}

where L is land-use type; σ_ij^z is the neighborhood index in land-use type L; s_ij is a matrix; Z is the index changes during the CA iterations; n×n is the size of the neighborhood; con(∙) is the condition function.

Incorporating ESs into LUCC models provides a comprehensive evaluation of trade-offs, independence, or conflicts between the environment and human welfare [73], [74]. Therefore, it is important to promote technology and policy designs to ensure sustainability. We adopted the concept of ecosystem defined in the Millennium Ecosystem Assessment [75] and assumed that farmers had benefited from two kinds of bipartite ecosystems: food supply (ES_f) and all other ESs (ES₀). In this study, we utilized an ES valuation system that reflects a farmer’s perceived ES_f (food provisioning) and ES₀ (aggregate of all other ESs) for four distinct land uses (as shown in Table 1) [70]. They used normalized 1 to 10 scale evaluation systems to assign standard values to ES_f and ES₀. Attraction to various human agents and ownership of land units are connected between human agents and land unit agents. The formula for attraction and ownership is described in the chapter describing the human agents.

2.3.2. Dural decision-making agents

According to the census of agriculture, there is reason to assume that the landowners in the watershed are mainly farmland owners. The ABCSMM is programmed to update its expected utility by considering ES and other attributes of the land units provided by the land agent. Human agents assume the following three financial conditions. Simultaneously, the attraction force of the land cell agent to the human agent is given by

(10)

β = C_{1} N + C_{2} {ES}_{f} + C_{3} {ES}_{0} + C_{4} D_{density} + C_{5} D_{GDP} + C_{6} e^{- D 1} + \dots + C_{n} e^{- D n}

where C₁,…,C₅ are the coefficients; N is neighborhood index; ES_f is food provision ESs; ES₀ is non-food supply ESs; D_density is population density; D_GDP is GDP; D1,…, Dn are distances to a certain location.

The human agent of the designated year has attributes, such as land cell ownership, property size, and financial state. This study assumes that each farmer owns a square farm of one land cell of approximately 90 m × 90 m. One limitation of this version of the ABCSMM is that property sizes are assumed to remain unchanged over time. This limitation should be addressed in future studies. Farmers’ financial situations in a given year can be categorized into three independent states: financially feeling good (α), financially moderate stress (β), and financially significant stress (1 - α − β).

Fig. 3 shows the financial condition of a farmer, perceptions of ES_f and ES₀ in a given year, and all possible land-use decision processes for farmers [70]. It is dependent on ES₀ and ES_f to determine whether farmers with good economic conditions will expand their arable land. As shown in Fig. 3, when wealthy farmers perceive ES_f > ES₀, they have the probability of converting forest land into cropland (P_FoAg), the probabilities of the conversion of grassland to cropland (P_GrAg), the probabilities of the conversion of barrenland to cropland (P_BaAg), and the probabilities of the conversion of shrubland to cropland (P_ShAg) to primarily expand croplands from different land-use types. However, if ES_f < ES₀, then the farmer has the highest possibility of turning some of his croplands into grasslands or shrublands. Finally, when ES_f = ES₀, the farmers were most likely to maintain their current farming practices. Farmers who face moderate fiscal pressure control their farming practices. Hence, croplands did not change during that year. However, forests, grasses, shrubs, water, and barren areas grow through natural vegetation. When faced with significant financial stress, a shortage of funds causes farmers to reduce their agricultural activities. This discards the farmland. Among these abandoned croplands, the previous farmland was either grassland, shrub, or barren. The ABCSMM was implemented such that P_AgBa represents the simulated conversion probability in which abandoned farmland transforms into barren land, and P_AgGr and P_AgSh represent the simulated transition probability from cropland to grass/shrub, respectively. These land-use transition probabilities are parameters that must be calibrated in this model.

The environmental layer refers to the spatial environment of a land-use system, which can directly affect the land-use mode or change the land-use type by acting on the layer of dual decision-making agents. It is a comprehensive expression of land-use status and the driving force of intelligent agents. The probability of land-use transformation in the ABM is shown in Fig. 3. Influencing the natural transformation of land without human interference can align better with the expected utility of human agents. For example, when the land-use type is cropland, the farmer agent is present, the farmer’s financial type is Financially Feel Good, and the land ES_f > ES₀ within a 3 × 3 area centered on the land, land-use type as forestland has a P_FoAg probability of being converted to cropland, land-use type as grassland has a P_GrAg probability of being converted to cropland, land-use type as barrenland has a P_BaAg probability of being converted to cropland, and land-use type as shrubland has a P_ShAg probability of being converted to cropland.

3. Application

3.1. Overview of the studying area

The research area is the DRB (Fig. 4), a branch of the Pearl River in southern China [76]. The vegetation coverage was relatively high in the DRB. Mountains and hills dominate the terrain of the middle and upper reaches. The downstream area, with gentle terrain, high population density, and rapid economic development, is economically developed in the Guangdong Province. There are many vegetation types in the DRB with a wide coverage area [77]. The total area of forest and grassland accounted for 82.25% of the total area, of which the forest type was dominated by evergreen broad-leaved forests, accounting for 51.6%. Based on the spatial distribution characteristics, the districts and counties with a high proportion of forestland were mainly concentrated in the middle and upper reaches of the watershed. The downstream area has low vegetation cover [78]. In 2020, the annual GDP of the DRB was approximately 1656.996 billion CNY, and the GDP of the primary, secondary, and tertiary industries accounted for roughly 3.46%, 49.86%, and 46.68%, respectively. There were apparent spatial differences in the contribution rates of different industries in the upper, middle, and lower reaches of the study area. The terrain significantly affects the upstream region, and the industrial economy is relatively underdeveloped. Tertiary industries have a high level of development. The primary land-use type in the region is forest, and since the 1980s, the urban area has expanded annually as the pace of urbanization has accelerated [79], [80], [81], [82], [83], [84]. Therefore, model research is required to evaluate the potential impacts of human-environmental interactions.

3.2. Status of LULC

The simulation of future DRB LULC was based on baseline LULC data (1990-2019) for the catchment area as described by Yang et al. [50]. Nine major LULC categories (cropland, forest, shrub, grassland, water, snow/ice, barren, impervious, and wetland) were identified at depths of 30 m × 30 m. Approximately 19 million land-cell agents are available in the DRB. The seven types of land-use patterns in any given year were cropland (class code = 1), forestland (class code = 2), shrub (class code = 3), grassland (class code = 4), water bodies (class code = 5), barren (class code = 6), and impervious (class code = 7). Yang et al. [50] resampled these seven land-use types. The ABCSMM displays the land-use pattern using the color scheme shown in Fig. 5. The ice/snow type was not observed in the DRB.

3.2.1. Land-use structure and changes

In general, changes in land use are affected by several factors. In this study, we considered the central driving factor data, including socioeconomic (Fig. 6, Fig. 7) and natural data (Fig. 8).

Table 2 presents the periodic land-use changes in the DRB (i.e., 1990, 2000, 2010, and 2019) and the tabulation of these changes resulting from the spatial analysis. According to the table, land-use changed in 1990, 2000, 2010, and 2019. Land use underwent significant transformation, with certain land types experiencing increased usage, whereas others declined in usage. From 1990 to 2019, grassland decreased from 0.21% to 0.07%, impervious increased from 0.38% to 2.51%, and shrubs decreased from 0.04% to 0.01%. Table 2 indicates that forest covered most of the DRB. The cropland area decreased from 17.86% to 15.71% from 1990 to 2010, and then increased to 16.24% in 2019.

The single land dynamic degree was used to study the changes within a certain period of specific land-use types within the area:

(11)

K_{i} = \frac{U_{bi} - U_{ai}}{U_{ai}} \times \frac{1}{t} \times 100 %

where K_i is the dynamic degree of land type i during the study period; U_ai,U_bi are the study period encompassing an analysis of the number of land types at the beginning and end of the period, respectively; t is the study period. When the period of T is set as a year, the value of K_i is the annual change rate of the land-use type in the study area.

The integrated conversion rate of the land-use type effectively integrates the change speed information of various land-use types in the unit, considering the total land area and structural factors of the unit. This value V reflects the overall LUCC rate in the study area.

(12)

V = \frac{1}{t} \sqrt{\sum_{i = 1}^{m} [{(\frac{A_{i}^{j + 1} - A_{i}^{j}}{A_{i}^{j}})}^{2} \frac{A_{i}^{i}}{S}}]

where A_i^j is the area of the i-th type of land in the j-th period, A_i^j+1 is the area of the i-th type of land in the (j + 1)-th period, A_iⁱ is the area of the i-th type of land in the i-th period, S is the total size of the study area, m is the number of land-use types, t is the study period.

According to Eqs. (1), (2), the rate of land-use change during period 1 (1990-2000), period 2 (2000-2010), and period 3 (2010-2019) was calculated, as shown in Table 3.

The comprehensive land-use conversion rate was 0.15% in period 1, 0.09% in period 2, and 0.03% in period 3. The total land-use conversion rate showed a downward trend. From the single land dynamic degree, this was mainly due to the stabilization of various changes. In period 2, the barren land change rate was most notable, with a negative change in period 1 and a large positive change in period 2. The land-use type of impervious land shows positive growth but a decreasing rate. The cropland decreased slowly in the first stage, decreased rapidly in period 2, and then increased in period 3.

3.2.2. Analysis of land use degree

The degree of land use reflects the extent and depth of the land. It represents the natural attributes of the land and reflects the overall role of humans and natural elements. Based on the natural balance of land under the influence of social factors, this study proposes a classification standard for the degree of land use, divides the land use of the research area into four grades, and assigns a graded index (i.e., L₁ = barren; L₂ = forestland, shrub, grassland, and water; L₃ = cropland; and L₄ = impervious). To evaluate the level of integrated land use and its tendency to change within the research area, we proposed a land-use degree index and a land-use degree change model.

(13)

D_{j} = 100 \sum_{i = 1}^{n} {(A}_{i} C_{i})

where D_j is the comprehensive index of land-use degree, A_i is the grading index of the i-level land-use degree classification, C_i is the percentage of area occupied in i-level land-use degree, n is the number of grades of land-use degree.

(14)

R_{i} = (L_{b} - L_{a}) / L_{a}

where R_i is the change index of land-use degree; L_a, L_b are the comprehensive indices of the land-use degree in period a and b, respectively.

According to Eqs. (3), (4), the comprehensive index and change index of land-use degree from 1990 to 2019 are shown in Table 4.

4. Results analysis

4.1. Validation and comparing goodness-of-fit

To predict LULC, the model evaluation parameters for annual land cover calibration in the DRB are presented in Table 5.

This study proposes an urban expansion model using GIS and SLR based on various social and environmental factors. The integration of SLR with GIS is crucial for model urban changes owing to the spatial characteristics of several input variables. The ROC index was used to determine the performance of SLR in the research domain (Table 6). We constructed a map displaying the probability of urban expansion that was used to predict future urban patterns. An ROC value of 93% indicated that the probability diagram was valid. We conclude that the SLR is spatially clear and can be used to further understand the power to promote the growth and formation of urban spatial structures.

An assessment of the actual and simulated LULC in the DRB was conducted in 2019. Minor discrepancies between LULC classes are shown (Fig. 9). Accordingly, the essential and fake maps of 2019 displayed sound similarities in water body cover. The area coverage of the two maps showed that the consistency of all LULC categories was acceptable, and the rate of difference was less than 10%. For model verification, a consistency kappa index comparison was performed between the actual and simulated LULC maps for 2019. The overall kappa value is 98.93%, indicating a strong match between the two map categories. This verification process evaluated the placement of the two maps (prediction and practice) based on the number of pixels and pixel positions per land-use coverage category.

As presented in Table 7, this was calculated from the difference between the 2019 classification and the simulated LULC. This difference underlies an underestimation of approximately 0.75% and 0.03% for forestland and impervious land, respectively. However, the model overestimated LULC categories, such as cropland (0.76%) and water (0.03%). The fitting optimality estimation indicates that the model can be applied on the DRB to simulate LULC changes.

4.2. Land-use type suitability evaluation

Suitability mapping (Fig. 10) was completed using a logistic regression equation, multicriteria evaluation, and Markov and driving force grid data. Appropriate values range from 0 to 1; a value close to 1 indicates that the suitability for a specific land use is high, and a value close to 0 indicates that the compatibility of land use is low. The land-use suitability for forestland was higher in the area, with the highest value of 1. The suitability value of the impervious surfaces is higher downstream. A suitably high value for the water area was mainly concentrated in the Dongjiang River, with a maximum value of about 0.75. Cropland is primarily distributed at an elevation of 0-250 m and a slope of 0°-8°, where the terrain plays a vital role in increasing cropland. The suitability of grasslands for the entire basin was low, and the highest suitability was 0.01.

4.3. Policy scenarios simulation

Table 8 shows the land areas and percentage change rates for the different LULC types in 2030. The LULC in the balancing development scenario (S1) indicated that the area of 19.25% was covered by cropland, 75.97% by forestland, 2.18% by water, 2.53% by impervious land, 0.01% by shrubs, and approximately 0.00% by barren land. In the grassland protection scenario (S2), the LULC indicated that the area of 19.22% was covered by cropland, 75.91% by forest, 2.18% by water, 2.62% by impervious land, and almost the same as shrub and barren land in S1. Cropland, forestland, grassland, and impervious land categories cover approximately 7 318 874, 28 870 800, 23 838, and 954 406 hm² in the ecological protection scenario (S3) and 7 312 184, 28 872 100, 22 270, and 961 587 hm² in the forest protection scenario (S4), respectively. The LULC in the economic development scenario (S5) indicated that 19.23% was covered by cropland, 75.90% by forestland, and 2.62% by impervious land. In the cultivated land protection scenario (S6), the LULC indicated 7 309 094 hm² area of cropland, 28 845 500 hm² of forestland, 21 833 hm² of grassland, and 991 841 hm² impervious land. In contrast, the shrubland and barren categories were consistent across the six scenarios, accounting for 0.01% and almost 0.00% of the catchment area, respectively.

The changes in the LULC percentage graph for the six scenarios (S2-S1, S3-S1, S4-S1, S5-S1, and S6-S1) for 2030 are shown in Fig. 11. During S2-S1, cropland decreased by 6360.00 hm², forestland decreased by 1200.00 hm², grassland increased by 1044.90 hm², and impervious land increased by 35 868.00 hm². While during S3-S1, a decrease was observed in the impervious land by 5588.00 hm². Cropland experienced an increase of 10 800.00 hm², forestland by 22 500.00 hm², shrub by 133.65 hm², grassland by 1688.00 hm², water by 400.00 hm², and barren by 36.45 hm². During S4-S1, an increase was observed in cropland, forest, shrub, grassland, water, and impervious areas of 4110.00, 23 800.00, 109.35, 121.50, 182.00, and 1593.00 hm², respectively. The barren areas decreased by 60.75 hm². At the same time, in the S6-S1 barren decreased by 48.60 hm², and cropland, shrub, water, and impervious increased by 1020.00, 72.90, 207.00, and 31 847.00 hm². However, forestlands and grasslands decreased by 2800.00 and 315.80 hm², respectively. During S5-S1, the impervious area increased substantially by 36 792.00 hm², and shrub, water, and barren increased by 97.20, 291.00, and 85.05 hm², repectively. Cropland, forestland, and grassland decreased by 3270.00, 3900.00, and 145.70 hm², respectively.

5. Discussions and concluding remarks

The ABCSMM model developed in this study couples a CA model with spatial self-organization and a multi-agent model with emergence to calculate the suitability of urban and ecological land. It also incorporates ESs and farmers’ financial status to address the interactions between humans and the environment. The model was used to simulate and predict the spatial patterns of future land use under six policy scenarios: balancing development scenario (S1), grassland protection scenario (S2), ecological protection scenario (S3), forest protection scenario (S4), economic development scenario (S5), and cultivated land protection scenario (S6).

The main results are as follows: ① The ROC index was used to define the accuracy of the SLR simulation, and a value of 93% indicates that SLR has an advantage in spatial relationship performance and can be applied to urban spatial structure prediction. ② The consistency kappa index was used to compare the actual land-use cover in 2019 with the simulated land-use cover, and the difference ratio was less than 10%. The overall kappa index was 98.93%, indicating a high degree of match between the two maps. ③ The land suitability for forests in the region was relatively high. The suitability for urban land was higher near the lower reaches of the DRB. High suitability values for the water areas were mainly concentrated in the middle and lower reaches of the DRB. The suitability for grassland is relatively low, and arable land is mainly distributed in areas with an elevation of 0-250 m and a slope of 0°-8°. ④ The categories of shrub and wasteland are consistent in six situations, accounting for approximately 0.01% of the watershed area. The area occupied by forests will exceed 75% in any scenario in 2030. ⑤ Compared to the S1, the area of shrubs increased in all other scenarios, with the largest absolute change in urban land area and the smallest in wasteland area. Due to the past policy of returning farmland to forests, cultivated land has shown a downward trend, whereas forest land has shown an upward trend. However, in recent years, a moderate reversion from afforestation to farmland has occurred in response to the national call to ensure national food security.

The model developed in this study performed well in simulating spatiotemporal land-use changes in the DRB. However, this study only considered limited socioeconomic and natural environmental driving factors; factors such as climate variability, natural disasters, and soil conditions were not considered. At the same time, in the CA simulation, many land-use types remained unchanged, which may have caused the kappa value to inflate. In future research, more accurate indicators, such as the figure-of-merit (FoM), should be used to evaluate the simulation accuracy and analyze the accuracy of land-use change simulation based on CA. Finally, correlations were identified among the driving factors selected by the authors. Analyzing these using SLR may lead to unstable simulations and erroneous results. In future research, more precise methods should be developed to characterize the relationships between multiple driving factors. Future research on land-use change simulation should leverage the advantages of big data technology and promote the development of more refined and diversified research directions for land-use change simulation. Combined with practical issues in the field of the ecological environment, we need to explore the feedback mechanism between land-use change and its ecological and environmental effects. The research perspective should gradually shift from exploring the impact of human activities on land-use change to their interactions. Ultimately, this will promote the coordinated development of human-land relations and enhance our understanding of the process of land-use change.

Acknowledgments

This research was supported by the Program for Guangdong Introducing Innovative and Entrepreneurial Teams (2021ZT090543), the National Natural Science Foundation of China (U20A20117), and the Key-Area Research and Development Program of Guangdong Province (2020B1111380003). The authors would also like to extend the appreciation to the editors and the anonymous reviewers for their efforts in helping improve the quality of this paper.

Compliance with ethics guidelines

Yutong Li, Yanpeng Cai, Qiang Fu, Xiaodong Zhang, Hang Wan, and Zhifeng Yang declare that they have no conflict of interest or financial conflicts to disclose.

References

Publishing order | Descend order by publishing year | Descend order by cited within

[1]	T. Gashaw, T. Tulu, M. Argaw, A.W. Worqlul. Evaluation and prediction of land use/land cover changes in the Andassa watershed, Blue Nile Basin, Ethiopia. Environ Syst Res, 6 (1) (2017), p. 17.

[2]	H.A. Kaul, I. Sopan. Land use land cover classification and change detection using high resolution temporal satellite data. J Environ, 1 (4) (2012), pp. 146-152.

[3]	H. Wang, X. Liu, C. Zhao, Y. Chang, Y. Liu, F. Zang. Spatial- temporal pattern analysis of landscape ecological risk assessment based on land use/land cover change in Baishuijiang National nature reserve in Gansu Province, China. Ecol Indic, 124 (2021), p. 107454.

[4]	J. Yang, J. Xu, Y. Zhou, D. Zhai, H. Chen, Q. Li, et al. Paddy rice phenological mapping throughout 30-years satellite images in the Honghe Hani Rice Terraces. Remote Sens, 15 (9) (2023), p. 2398.

[5]	H.W. Zheng, G.Q. Shen, H. Wang, J. Hong. Simulating land use change in urban renewal areas: a case study in Hong Kong. Habitat Int, 46 (2015), pp. 23-34.

[6]	E. Yirsaw, W. Wu, X. Shi, H. Temesgen, B. Bekele. Land use/land cover change modeling and the prediction of subsequent changes in ecosystem service values in a Coastal Area of China, the Su-Xi-Chang Region. Sustainability, 9 (7) (2017), p. 1204.

[7]	Batunacun C. Nendel Y. Hu T. Lakes. Land-use change and land degradation on the Mongolian Plateau from 1975 to 2015—a case study from Xilingol, China. Land Degrad Dev, 29 (6) (2018), pp. 1595-1606.

[8]	M. Minta, K. Kibret, P. Thorne, T. Nigussie, L. Nigatu. Land use and land cover dynamics in Dendi-Jeldu hilly-mountainous areas in the central Ethiopian highlands. Geoderma, 314 (2018), pp. 27-36.

[9]	R.J. Permatasari, A. Damayanti, T.L. Indra, M. Dimyati. Prediction of land cover changes in Penajam Paser Utara Regency using cellular automata and Markov model. IOP Conf Ser Earth Environ Sci, 623 (1) (2021), p. 012005.

[10]	D.F. Ren, A.H. Cao, F.Y. Wang. Response and multi-scenario prediction of carbon storage and habitat quality to land use in Liaoning Province, China. Sustainability, 15 (5) (2023), pp. 1-23.

[11]	K. Mohan, P.K. Rai, V.N. Mishra. Prediction of land use changes based on land change modeler (LCM) using remote sensing: a case study of Muzaffarpur (Bihar), India. J Geogr Inst Jovan Cvijic SASA, 64 (1) (2014), pp. 111-127.

[12]	S.K. Singh, S. Mustak, P.K. Srivastava, S. Szabó, T. Islam. Predicting spatial and decadal LULC changes through cellular automata Markov Chain Models using earth observation datasets and geo-information. Environ Processes, 2 (1) (2015), pp. 61-78.

[13]	S.K. Singh, P.B. Laari, S. Mustak, P.K. Srivastava, S. Szabó. Modelling of land use land cover change using earth observation data-sets of Tons River Basin, Madhya Pradesh, India. Geocarto Int, 33 (11) (2018), pp. 1202-1222.

[14]	S. Tavangar, H. Moradi, A.M. Bavani, M. Gholamalifard. A futuristic survey of the effects of LU/LC change on stream flow by CA-Markov model: a case of the Nekarood watershed, Iran. Geocarto Int, 36 (10) (2021), pp. 1100-1116.

[15]	C. Ning, R. Subedi, L. Hao. Land use/cover change, fragmentation, and driving factors in Nepal in the last 25 years. Sustainability, 15 (8) (2023), p. 6957.

[16]	M. Kindu, T. Schneider, D. Teketay, T. Knoke. Land use/land cover change analysis using object-based classification approach in Munessa-Shashemene landscape of the Ethiopian highlands. Remote Sens, 5 (5) (2013), pp. 2411-2435.

[17]	S.D. Dayamba, H. Djoudi, M. Zida, L. Sawadogo, L. Verchot. Biodiversity and carbon stocks in different land use types in the Sudanian Zone of Burkina Faso, West Africa. Agric Ecosyst Environ, 216 (2016), pp. 61-72.

[18]	A. Alam, M.S. Bhat, M. Maheen. Using Landsat satellite data for assessing the land use and land cover change in Kashmir valley. GeoJournal, 85 (6) (2020), pp. 1529-1543.

[19]	Y. Xu, Y. Chen, Y. Ren, Z. Tang, X. Yang, Y. Zhang. Attribution of streamflow changes considering spatial contributions and driver interactions based on hydrological modeling. Water Resour Manage, 37 (5) (2023), pp. 1859-1877.

[20]	M. Witjes, L. Parente, C.J. van Diemen, T. Hengl, M. Landa, L. Brodský, et al. A spatiotemporal ensemble machine learning framework for generating land use/land cover time-series maps for Europe (2000-2019) based on LUCAS, CORINE and GLAD Landsat. PeerJ, 10 (2022), p. e13573.

[21]	H. Memarian, S.K. Balasundram, J.B. Talib, C.T.B. Sung, A.M. Sood, K. Abbaspour. Validation of CA-Markov for simulation of land use and cover change in the Langat Basin, Malaysia. J Geogr Inf Syst, 4 (6) (2012), pp. 542-554.

[22]	C. Hyandye, L.W. Martz. A Markovian and cellular automata land-use change predictive model of the Usangu Catchment. Int J Remote Sens, 38 (1) (2017), pp. 64-81.

[23]	Y. Lu, P. Wu, X. Ma, X. Li. Detection and prediction of land use/land cover change using spatiotemporal data fusion and the Cellular Automata-Markov model. Environ Monit Assess, 191 (2) (2019), p. 68.

[24]	Z. Zhang, G. Hörmann, J. Huang, N. Fohrer. A random forest-based CA-Markov model to examine the dynamics of land use/cover change aided with remote sensing and GIS. Remote Sens, 15 (8) (2023), p. 2128.

[25]	L. Liu, S. Yu, H. Zhang, Y. Wang, C. Liang. Analysis of land use change drivers and simulation of different future scenarios: taking Shanxi Province of China as an example. Int J Environ Res Public Health, 20 (2) (2023), p. 1626.

[26]	M.T.U. Rahman, F. Tabassum, M. Rasheduzzaman, H. Saba, L. Sarkar, J. Ferdous, et al. Temporal dynamics of land use/land cover change and its prediction using CA-ANN model for southwestern coastal Bangladesh. Environ Monit Assess, 189 (11) (2017), p. 565.

[27]	M.G. Munthali, S. Mustak, A. Adeola, J. Botai, S.K. Singh, N. Davis. Modelling land use and land cover dynamics of Dedza district of Malawi using hybrid Cellular Automata and Markov model. Remote Sens Appl Soc Environ, 17 (2020), p. 100276.

[28]	S. Sibanda, F. Ahmed. Modelling historic and future land use/land cover changes and their impact on wetland area in Shashe sub-catchment, Zimbabwe. Model Earth Syst Environ, 7 (1) (2021), pp. 57-70.

[29]	M. Mwabumba, B.K. Yadav, M.J. Rwiza, I. Larbi, S. Twisa. Analysis of land use and land-cover pattern to monitor dynamics of Ngorongoro world heritage site (Tanzania) using hybrid cellular automata-Markov model. Curr Res Environ Sustainability, 4 (2022), p. 100126.

[30]	B. Matlhodi, P.K. Kenabatho, B.P. Parida, J.G. Maphanyane. Analysis of the future land use land cover changes in the Gaborone dam catchment using CA-Markov model: implications on water resources. Remote Sens, 13 (13) (2021), p. 2427.

[31]	A. Naboureh, M.H.R. Moghaddam, B. Feizizadeh, T. Blaschke. An integrated object-based image analysis and CA-Markov model approach for modeling land use/land cover trends in the Sarab plain. Arabian J Geosci, 10 (12) (2017), p. 259.

[32]	S. Giri, N.N. Arbab, R.G. Lathrop. Water security assessment of current and future scenarios through an integrated modeling framework in the Neshanic River Watershed. J Hydrol, 563 (2018), pp. 1025-1041.

[33]	A. Gausen, W. Luk, C. Guo. Using agent-based modelling to evaluate the impact of algorithmic curation on social media. J Data Inf Qual, 15 (1) (2023), pp. 1-24.

[34]	B.J. Sattler, J. Friesen, A. Tundis, P.F. Pelz. Modeling and validation of residential water demand in agent-based models: a systematic literature review. Water, 15 (3) (2023), p. 579.

[35]	Hunter E. A hybrid agent-based and equation based epidemiological model for the spread of infectious diseases [dissertation]. Dublin: Technological University Dublin; 2020.

[36]	S.J. Walsh, G.P. Malanson, B. Entwisle, R.R. Rindfuss, P.J. Mucha, B.W. Heumann, et al. Design of an agent-based model to examine population-environment interactions in Nang Rong District, Thailand. Appl Geogr, 39 (2013), pp. 183-198.

[37]	X. Zhao, X. Ma, W. Tang, D. Liu. An adaptive agent-based optimization model for spatial planning: a case study of Anyue County, China. Sustain Cities Soc, 51 (2019), p. 101733.

[38]	Y. An, A. Park. Developing an agent-based model to mitigate famine risk in North Korea: insights from the “Artificial North Korean Collective Farm” model. Land, 12 (4) (2023), p. 735.

[39]	D. Dziubanski, K.J. Franz. Projecting hydrologic change under land use and climate scenarios in an agricultural watershed using agent-based modeling. Front Water, 5 (2023), p. 1020080.

[40]	Crooks AT. The repast simulation/modelling system for geospatial simulation [dissertation]. London: University College London; 2007.

[41]	KM. Johnston. Agent analyst:agent-based modeling in ArcGIS, Esri Press, Redlands (2024). In press.

[42]	H. Mirzahossein, V. Noferesti, X. Jin. Residential development simulation based on learning by agent-based model. TeMA J Land Use Mobility Environ, 15 (2) (2022), pp. 193-207.

[43]	G. Ravaioli, T. Domingos, R.F.M. Teixeira. A framework for data-driven agent-based modelling of agricultural land use. Land, 12 (4) (2023), p. 756.

[44]	S. Abolhasani, M. Taleai. Assessing the effect of temporal dynamics on urban growth simulation: towards an asynchronous cellular automata. Trans GIS, 24 (2) (2020), pp. 332-354.

[45]	J. Hao, Q. Lin, T. Wu, J. Chen, W. Li, X. Wu, et al. Spatial-temporal and driving factors of land use/cover change in Mongolia from 1990 to 2021. Remote Sens, 15 (7) (2023), p. 1813.

[46]	W. Azemeraw, M. Matebie. Landslide susceptibility mapping using information value and logistic regression models in Goncha Siso Eneses area, northwestern Ethiopia. SN Appl Sci, 2 (2020), pp. 1-19.

[47]	I.M. Perez, A. Airola, P.J. Boström, I. Jambor, T. Pahikkala. Tournament leave-pair-out cross-validation for receiver operating characteristic analysis. Stat Methods Med Res, 28 (10,11) (2018), pp. 2975-2991.

[48]	R.G. Pontius Jr. Statistical methods to partition effects of quantity and location during comparison of categorical maps at multiple resolutions. Photogramm Eng Remote Sens, 68 (10) (2002), pp. 1041-1049.

[49]	B.C. Pijanowski, A. Tayyebi, M.R. Delavar, M.J. Yazdanpanah. Urban expansion simulation using geospatial information system and artificial neural networks. Int J Environ Res, 3 (4) (2009), pp. 493-502.

[50]	J. Yang, X. Huang. The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019. Earth Syst Sci Data, 13 (8) (2021), pp. 3907-3925.

[51]	C. Llorca, N. Kuehnel, R. Moeckel. Agent-based integrated land use/transport models: a study on scale factors and transport model simulation intervals. Procedia Comput Sci, 170 (2020), pp. 733-738.

[52]	C.G.C. Coelho, C.G. Ralha. MASE-EGTI: an agent-based simulator for environmental land change. Environ Modell Software, 147 (2022), p. 105252.

[53]	E.G. Irwin, P.W. Jeanty, M.D. Partridge. Amenity values versus land constraints: the spatial effects of natural landscape features on housing values. Land Econ, 90 (1) (2014), pp. 61-78.

[54]	O.N. Mensour, B. El Ghazzani, B. Hlimi, A. Ihlal. A geographical information system-based multi-criteria method for the evaluation of solar farms locations: a case study in Souss-Massa area, southern Morocco. Energy, 182 (2019), pp. 900-919.

[55]	T. Everest, A. Sungur, H. Özcan. Determination of agricultural land suitability with a multiple-criteria decision-making method in Northwestern Turkey. Int J Environ Sci Technol, 18 (5) (2021), pp. 1073-1088.

[56]	J. Malczewski. Integrating multicriteria analysis and geographic information systems: the ordered weighted averaging (OWA) approach. Int J Environ Technol Manage, 6 (1,2) (2006), pp. 7-19.

[57]	Mishra M, Mishra KK, Subudhi AP, Phil M. Urban sprawl mapping and land use change analysis using remote sensing and GIS (case study of Bhubaneswar city, Orissa). In: Proceeding of Geospatial World Forum; 2018 Jan 15-19; Hyderabad, India; 2018.

[58]	L. Wu, P. Shi, H. Gao. State estimation and sliding-mode control of Markovian jump singular systems. IEEE Trans Autom Control, 55 (5) (2010), pp. 1213-1219.

[59]	H.M. Mosammam, J.T. Nia, H. Khani, A. Teymouri, M. Kazemi. Monitoring land use change and measuring urban sprawl based on its spatial forms: the case of Qom City. Egypt J Remote Sens Space Sci, 20 (1) (2017), pp. 103-116.

[60]	X. Yang, X.Q. Zheng, L.N. Lv. A spatiotemporal model of land use change based on ant colony optimization, Markov chain, and cellular automata. Ecol Modell, 233 (2012), pp. 11-19.

[61]	A. Maviza, F. Ahmed. Analysis of past and future multi-temporal land use and land cover changes in the semi-arid Upper-Mzingwane sub-catchment in the Matabeleland south province of Zimbabwe. Int J Remote Sens, 41 (14) (2020), pp. 5206-5227.

[62]	J. Lin, X. Li, Y. Wen, P. He. Modeling urban land-use changes using a landscape-driven patch-based cellular automaton (LP-CA). Cities, 132 (2023), p. 103906.

[63]	D. Guan, H. Li, T. Inohae, W. Su, T. Nagaie, K. Hokao. Modeling urban land use change by the integration of cellular automaton and Markov model. Ecol Modell, 222 (20-22) (2011), pp. 3761-3772.

[64]	X. Yang, X.Q. Zheng, R. Chen. A land use change model: integrating landscape pattern indexes and Markov-CA. Ecol Modell, 283 (2014), pp. 1-7.

[65]	H. Keshtkar, W. Voigt. A spatiotemporal analysis of landscape change using an integrated Markov chain and cellular automata models. Model Earth Syst Environ, 2 (1) (2016), p. 10.

[66]	X. Liang, X. Liu, D. Li, H. Zhao, G. Chen. Urban growth simulation by incorporating planning policies into a CA-based future land-use simulation model. Int J Geogr Inf Sci, 32 (11) (2018), pp. 2294-2316.

[67]	Z. Liu, P.H. Verburg, J. Wu, C. He. Understanding land system change through scenario-based simulations: a case study from the drylands in northern China. Environ Manage, 59 (3) (2017), pp. 440-454.

[68]	N.Q. Omar, M.S.S. Ahamad, W.M.A.W. Hussin, N. Samat, S.Z.B. Ahmad. Markov CA, multi regression, and multiple decision making for modeling historical changes in Kirkuk City, Iraq. J Indian Soc Remote Sens, 42 (1) (2014), pp. 165-178.

[69]	M. Beroho, H. Briak, E.K. Cherif, I. Boulahfa, A. Ouallali, R. Mrabet, et al. Future scenarios of land use/land cover (LULC) based on a CA-Markov simulation model: case of a Mediterranean watershed in Morocco. Remote Sens, 15 (4) (2023), p. 1162.

[70]	Y. Tsai, A. Zia, C. Koliba, G. Bucini, J. Guilbert, B. Beckage. An interactive land use transition agent-based model (ILUTABM): endogenizing human-environment interactions in the western Missisquoi watershed. Land Use Policy, 49 (2015), pp. 161-176.

[71]	Arbab NN. Application of a spatially explicit, agent-based land use conversion model to assess water quality outcomes under buffer policies [dissertation]. Morgantown: West Virginia University; 2014.

[72]	N.N. Arbab, A.R. Collins, J.F. Conley. Projections of watershed pollutant loads using a spatially explicit, agent-based land use conversion model: a case study of Berkeley County, West Virginia. Appl Spat Anal Policy, 11 (1) (2018), pp. 147-181.

[73]	B. Egoh, M. Rouget, B. Reyers, A.T. Knight, R.M. Cowling, A.S. van Jaarsveld, et al. Integrating ecosystem services into conservation assessments: a review. Ecol Econ, 63 (4) (2007), pp. 714-721.

[74]	I. Dullinger, F. Essl, D. Moser, K. Erb, H. Haberl, S. Dullinger. Biodiversity models need to represent land-use intensity more comprehensively. Global Ecol Biogeogr, 30 (5) (2021), pp. 924-932.

[75]	S.M. Manson. Agent-based modeling and genetic programming for modeling land change in the Southern Yucatán Peninsular Region of Mexico. Agric Ecosyst Environ, 111 (1-4) (2005), pp. 47-62.

[76]	T. Wang, X. Tu, V.P. Singh, X. Chen, K. Lin, R. Lai, et al. Socioeconomic drought analysis by standardized water supply and demand index under changing environment. J Cleaner Prod, 347 (2022), p. 131248.

[77]	W. Mo, Y. Zhao, N. Yang, Z. Xu, W. Zhao, F. Li. Effects of climate and land use/land cover changes on water yield services in the Dongjiang Lake Basin. ISPRS Int J Geo-inf, 10 (7) (2021), p. 466.

[78]	K. Zhu, X. Qiu, Y. Luo, M. Dai, X. Lu, C. Zang, et al. Spatial and temporal dynamics of water resources in typical ecosystems of the Dongjiang River Basin, China. J Hydrol, 614 (Pt B) (2022), p. 128617.

[79]	Y. He, X. Chen, Z. Sheng, K. Lin, F. Gui. Water allocation under the constraint of total water-use quota: a case from Dongjiang River Basin, South China. Hydrol Sci J, 63 (1) (2018), pp. 154-167.

[80]	YP Cai, GH Huang, ZF Yang Q. Tan. Identification of optimal strategies for Energy management systems planning under multiple uncertainties. Appl Energy, 86 (4) (2009), pp. 480-495.

[81]	Cai YP, Huang GH, Lu HW, Yang ZF, Tan Q. I-VFRP: an interval-valued fuzzy robust programming approach for municipal waste-management planning under uncertainty. Eng Optim 2009; 41(5):399-418.

[82]	W Zhou,W Zhang, Y. Cai. Laccase immobilization for water purification: a comprehensive review. Chem Eng J, 403 (2021), p. 126272.

[83]	C Dong, Q Tan, GH Huang, YP Cai. A dual-inexact fuzzy stochastic model for water resources management and non-point source pollution mitigation under multiple uncertainties. Hydrol Earth Syst Sci, 18 (5) (2014), pp. 1793-1803.

[84]	Y. Xu, Y Cai, T Sun, XA Yin, Q Tan, J Sun, et al. Ecological preservation based multi-objective optimization of coastal seawall engineering structures. J Cleaner Prod, 296 (2021), Article 126515.

RIGHTS & PERMISSIONS

THE AUTHOR