Multi-Scenario Simulation of Production-Living-Ecological Space in the Poyang Lake Area Based on Remote Sensing and RF-Markov-FLUS Model

Abstract

With industrialization and urbanization, the competition among land production, living, and ecological (PLE) spaces has intensified. Particularly in ecological reserves, competition among various types of land use restricts the coordinated development of PLE space. To explore spatial sustainable development, this study starts from a PLE spatial perspective, based on Landsat long time series images. Object-based image analysis (OBIA) and landscape index analysis were selected to monitor the spatial and temporal land use and landscape pattern changes in the Poyang Lake region (PYL region) from 1989 to 2020. The RF-Markov-FLUS coupled model was used to simulate spatial changes in 2030 under four scenarios: production space priority (PSP), living space priority (LSP), ecological space priority (ESP), and an integrated development (ID). Finally, the goal-problem-principle was used to enhance PLE space. The results showed that: (1) production space and ecological spaces decreased in general from 1989 to 2020 by 3% and 7%, respectively; living space increased by 11%. (2) From 1989 to 2020, the overall landscape spread in the Poyang Lake (PYL) area decreased, connectivity decreased, fragmentation increased, landscape heterogeneity increased, and landscape geometry became more irregular. (3) Compared with the other three scenarios, the ID scenario maintained steady production space growth in 2030, the expansion rate of living space slowed, and the area of ecological space decreased the least. (4) Spatial pattern optimization should start with three aspects: the transformation of the agricultural industry, improving the efficiency of urban land use, and establishing communities of “mountains, water, forests, fields, lakes and grasses”. The results provide scientific planning and suggestions for the future ecological protection of Poyang Lake area with multiple scenarios and perspectives.

1. Introduction

Territorial space is the location and environment of human existence, the material basis on which a country and its inhabitants live [1]. It can be divided into three types of production, living, and ecological spaces (PLE space) [2]. Since the 1990s, with industrialization and urbanization, the competition among production, living, and ecological spaces has intensified [3]; urban space has expanded [4]; ecological space has been threatened [5]; agricultural space development has been restricted [6,7]; space use has been inefficient; and the allocation of various spatial resources has been unreasonable [8], all of which restricts the sustainable development of national land space [9]. The sustainable use of multifunctional land has attracted great attention from policymakers and scholars in various countries. In March 2019, the United Nations Environment Programme (UNEP), in its Global Environment Outlook report (GEO-6), called on countries to completely abandon unsustainable production and consumption patterns, actively pursue sustainable development goals, and develop proactive environmental policies and improve implementation standards [10]. Some countries in Europe and Asia are seeking a reasonable path for regional development by drawing on coordinated territorial spatial development in other countries [11]. To achieve orderly spatial development and promote balanced and sustainable national space, China proposed national spatial function planning at the 18th National Congress of the Communist Party of China in 2012. This emphasizes a spatial pattern of “promoting intensive and efficient production space, livable and moderate living space, and beautiful ecological space with picturesque scenery” as a plan for optimizing PLE space [12]. In 2021, China released revised the Implementing Regulations of the Land Administration Law, which for the first time included three control lines in the regulations. It integrated the layout of ecological, agricultural, and urban functional spaces, and implemented the ecological protection red line, a permanent basic cropland protection red line, and an urban development boundary (Figure 1) [13], further clarifying the implementation requirements and spatial control boundary construction of the “three control lines” in spatial planning.

Economic production-oriented land use development patterns have long resulted in competition and conflict in PLE space. This has intensified with accelerated urbanization. How to restrain disorderly expansion with bottom-line thinking, safeguard ecological security with spatial optimization, and coordinate the relationship between ecology, agriculture, and urban areas have become a major proposition in the development and protection of territorial land space [14].

The existing global PLE space research has focused on establishing an assessment framework [11,14,15,16,17,18,19], spatial and temporal evolution and drivers [20], functional identification, and optimal regulation [11,19,21], etc. Among them, Zong et al. established a comprehensive index system with a PLE spatial perspective to find the best land development pattern in Yunnan Province [15]. Wang et al. assessed ecological risk sources in the Poyang Lake eco-economic region based on production, living, and ecological function indicators [22]. Holmes et al. examined the resource use intensity and functional variability of seven categories of functional areas for the identification of rural spaces in Australia [23]. Citing models in PLE space studies to quantify multifunctional spaces, Yu et al. used coupled models to quantify land-use functions and measure their spatio-temporal coordination and conflicts [20]. Liu et al. introduced the latent Dirichlet allocation (LDA) topic model to identify multifunctional mixed land-use patterns and spatial distributions [24]. The literature has shown that PLE space research has focused on the analysis of the current situation and functional identification of national land space, and study areas are mostly in developed administrative regions and economic circles. Fewer studies have been conducted on the dynamic simulation of PLE space in ecological reserves, and there is a lack of such research on reserves in the less-developed central regions of China. Therefore, multi-scenario simulation modeling, projecting future development trends in PLE space in ecological reserves, and understanding the mechanism of long-term evolution in PLE space will deepen the related research.

Multi-scenario land-use change simulation can be predicted and analyzed by various models, and many models have been applied for future land-use change. These include the Markov chain model for calculating land demand [25,26], the Conversion of Land Use and its Effects at Small regional extent (CLUE-S) model for allocating land-use change demand space [27,28], the cellular automata(CA) model for simulating complex spatial patterns of land use [29,30], the artificial neural network (ANN) model for dealing with non-deterministic identification and classification [31], and coupled models that combine multiple advantages of different models [32,33,34,35]. The above models can achieve good simulation accuracy in applications but lack the comprehensive ability to simulate spatial and temporal changes in land use. Compared with traditional simulation models, the emerging hybrid model FLUS (Future Land Use Simulation) developed by Professor Liu Xiaoping’s team at Sun Yat-sen University can simulate land-use change with high accuracy by coupling multiple models [36], combining the strengths of Markov models for quantitative prediction, ANN models for nonlinear estimation of interactions and competition between land-use types, and CA models for spatial evolution [37,38]. This model can effectively deal with the uncertainty and complexity of land-use change under the combined influence of natural and human factors [36]. It can also be widely applied to different research directions and scales according to the research purposes, such as land use dynamic simulation [39,40], simulating the development boundaries of administrative towns [37,39], estimating watershed ecosystem service values [41,42], and assessing urban ecological risk [43,44]. Random forest (RF), an integrated learning method, has better accuracy and generalization than linear regression, and is especially suitable for the prediction and simulation of nonlinear and small sample data. It can better meet the impact factor screening needs [45]. Therefore, this paper coupled the RF-Markov-FLUS model to set up multi-scenarios and simulate future PLE space change under the influence of different spatial policy priorities. The introduction of the random forest approach to improve model accuracy and predict future land pattern evolution more accurately is of theoretical significance for understanding the complex dynamic evolution mechanism of land-use patterns. It is of theoretical significance to understand the complex dynamic evolution mechanism of land-use patterns and is also of practical significance to explore the spatial management and optimization of ecological reserves and to implement ecological control strategies.

The Poyang Lake (PYL) area is located in northern Jiangxi Province, with flat terrain [46], and is the direct economic hinterland of the Yangtze River Delta, the Pearl River Delta, and other important economic plates. It is the most important growth node being accelerated in the less-developed central region of China [47]. It is also a relatively fragile ecological area in Jiangxi Province [48], with good conditions for studying coordinated ecological and economic development. Therefore, this study chose the PYL area, a typical area in the central region that is significantly affected by human activities.

Based on the above background, this study attempted to answer several questions: (1) How has the structure and quantity of territorial space in the PYL area changed in the past three decades? (2) Under the set constraints, which scenario best promotes the sustainable development of national land? (3) How can territorial space development be enhanced?

We used a Landsat long time series of remote sensing images and land-use classification data interpreted by OBIA. The RF-Markov-FLUS coupled model was used to simulate future spatial changes of the PYL area under multi-scenarios to provide multiple perspectives on the layout of the implementation of territorial spatial planning in the PYL region, optimize the “three control lines” results, and provide a reference for a spatial development pattern with balanced population, resources, and environment, and unified economic, social, and ecological benefits. The aims of our research were:

(1)

To investigate regional land-use patterns and determine the development changes in regional space.

(2)

To simulate the territorial spatial pattern of the PYL area in 2030 under multiple scenarios using the RF-Markov-FLUS model and predict the spatial change characteristics.

(3)

To construct an enhancement path for sustainable development of PLE space from the perspective of the goal-problem-principle.

2. Materials and Methods

2.1. Study Area

The PYL area is located on the south bank of the middle and lower reaches of the Yangtze River and north of Jiangxi Province (28°10′–29°51′N, 115°23′–117°45′E). The study area includes 12 counties and urban areas in Jiujiang, Nanchang, and Shangrao cities (Figure 2). The area has a subtropical humid monsoon climate, with an annual average temperature of 16–20 °C, abundant rainfall, a long frost-free period, four distinct seasons, and small temperature fluctuation with few high-temperature days [49]. The western and northern parts of the PYL area are dominated by mountains and hills, and the central area around the lake is low and flat and dominated by plains. The water system is centered on Poyang Lake. Five major rivers of Xinjiang, Fuhe, Ganjiang, Raohe, and Xiuhe flow into Poyang Lake, which is the largest freshwater lake in China and an ecologically important area in China, and an important base for food, oil, cotton, and fish in the middle reaches of the Yangtze River [50].

The total area of the whole region is 20,321.53 km², with a resident population of 11,037,400 people at the end of 2020 and a GDP of 698.638 billion yuan, accounting for 27.21% of the GDP of Jiangxi Province. Due to the rapid population growth and urbanization, there is a huge demand for land and natural resources, and Jiangxi Province proposed the Poyang Lake Ecological City Cluster Plan in 2015 to respond to the “The Belt and Road Initiative” national strategy and the Yangtze River Economic Belt [51]. Therefore, the PYL area is an ideal area to study PLE spatial coordination.

2.2. Data

In this study, remote sensing image data over Poyang Lake were selected from a total of 21 sets of Landsat-5 TM and Landsat-8 OLI images with less than 5% clouds in 1989, 1995, 2000, 2005, 2010, 2015, and 2020 during the abundant water period (May-September). Images were obtained from the United States Geological Survey (USGS) website [52], and ENVI5.3 software was used to perform image pre-processing steps such as radiometric calibration, atmospheric correction, mosaicking, and cropping to derive the images of the study area for seven periods. The land-use data were interpreted using the OBIA method for remote sensing images, and Google Earth high-resolution images were referred to for comparative interpretation of land classes. The accuracy of the interpreted remote sensing images was checked, and the Kappa coefficients were all greater than 85%, so the data accuracy met the research requirements. The images were classified into six land cover types: cropland, forest, grassland, water, construction land, and unused land.

The data used for land use simulation included four categories of natural factors, socio-economic factors, accessibility factors, and limiting factors, with a total of 16 drivers of land-use change. The slope and slope direction data were derived from elevation DEM data. Precipitation, average annual temperature, population density, and average land GDP data were interpolated using the inverse distance weight method. POI data and accessibility factor data were calculated using the Euclidean distance tool. All data were projected to WGS_1984_UTM_Zone_50N with an accuracy of 30 m and output in TIFF format. The list of data is shown in

Table 1. List of data and sources.

Type	Properties	Time	Data Type	Data Source
Land Use Data	Remote Sensing Images	1989, 1995, 2000, 2005, 2010, 2015, 2020	Raster (15 m/30 m)	USGS (https://earthexplorer.usgs.gov/) (accessed on 10 September 2021)
Natural factors	Elevation	2010, 2015, 2020	Raster (30 m)	Resource and Environment Sciences and Data Center (https://www.resdc.cn/) (accessed on 30 September 2021)
	Slope
	Slope orientation
	Precipitation			WorldClim-Global Climate Data (http://www.worldclim.org/) (accessed on 30 September 2021) Local weather bureau and weather bulletin
	Average annual temperature
Socio-economic factors	Population density	2010, 2015, 2020	Raster (1 km)	Nanchang, Jiujiang, Shangrao City Statistical Yearbook
	GDP per capita
	POI kernel density (airports, bus stations, buildings, schools, hotels, supermarkets, banks, etc.)	2010, 2015–2021	Vector	Baidu Map Open Platform (https://lbsyun.baidu.com/) (accessed on 20 October 2021)
Accessibility factors	Distance from city roads	2010–2020	Vector	OpenstreetMap Website (https://openstreetmap.org/) (accessed on 15 October 2021) National Catalogue Service For Geographic Information (http://www.webmap.cn) (accessed on 15 October 2021)
	Distance to railroad
	Distance to the water system
	Distance to urban center
	Distance to rural settlements
Limiting factors	Permanent basic cropland protection red line	2017	Vector	Jiangxi Provincial Natural Resources Bureau
	Urban development boundary	2019
	Ecological protection red line	2018

.

3. Methodology

In this study, the multi-time Landsat series remote sensing images were preprocessed in the ENVI5.3 platform and interpreted by the OBIA method with the help of the eCognition Developer 64 platform, and the land-use types of the study area were classified into cropland, forest, grassland, construction land, water, and unused land according to the classification method of third-round national land inquisition and the purpose of the study.

Secondly, nine landscape pattern indices were selected for calculation using the Fragstats 4.2 software platform to analyze the trend and current situation of landscape pattern change in the PYL area from 1989 to 2020. Then, 13 driving factors and 3 limiting factors were selected to simulate land-use change in 2030 using the RF-Markov-FLUS model, and PLE spatial pattern changes in the PYL area in 2030 were predicted under 4 scenarios of production space priority (PSP), living space priority (LSP), ecological space priority (ESP), and integrated development (ID). Finally, the PLE spatial enhancement path was constructed from the perspective of the goal-issue-principle.

Finally, according to the evolution characteristics and function mechanism of the territory spatial pattern, existing spatial pattern problems and utilization function issues were diagnosed, the path framework of territory spatial function improvement was established, and a path for territory spatial function improvement was put forward. The workflow of this study is shown in Figure 3.

3.1. The Object-Based Image Analysis (OBIA) Method

When extracting feature information from remote sensing images, the main purpose is to compare the spatial structure information differences in the image features, which differ in shape, size, texture, and spectral information. The current research mainly focused on pixel-based feature information classification and object-based feature discrimination techniques. Traditional pixel-based classification can result in misclassification and omission due to the “same object and different spectrum, different object and same spectrum” problem, and serious “salt-and-pepper noise”. The OBIA method is a new classification method proposed as remote sensing image resolution has increased. In it, individual objects are the basic unit for classification instead of image elements, and images can be segmented according to the structure and texture information of the objects to obtain more accurate classification information [53]. This study combined national land-use classification standards and current land use in the study area to classify the spatial use types of PYL area into three major categories from the perspective of PLE space: production space (mainly cropland), living space (mainly construction land), and ecological space (mainly water, forest, grassland, and unused land) (

Table 2. The spatial classification system for the PYL area.

Category I	Category II	Description
Production space	Cropland	Land where crops are grown, including paddy fields, irrigated land, and dry land, which are important for agricultural production and function.
Living space	Construction land	Land mainly used for housing and ancillary facilities, including urban and rural residential land, and land for supporting commercial services and other facilities.
Ecological space	Water	Areas such as inland water bodies, mudflats, ditches, and marshes. Wetlands have water conservation and purification functions and are important ecological lands.
	Forest	Land where trees, bamboos, and shrubs grow. Forests play important roles in ecosystem regulation and biodiversity, and are important ecological lands.
	Grassland	Land where herbaceous plants mainly grow, including natural pasture, marsh grassland, artificial pasture, and other grasslands. Grasslands have functions such as ecological landscape and water connotation and are important ecological lands.
	Unused land	Land types other than those mentioned above, including saline land, sandy land, bare land, and other unused land, all of which are natural land cover types and are important ecological lands.

).

Using the eCognition Developer 64 software platform, appropriate image segmentation scales, shape factor weights, compactness weights, and other parameters were set. DEM data were supplemented to extract image spectral features, and the normalized difference vegetation index (NDVI) (Equation (1)) [54], normalized difference built-up index (NDBI) (Equation (2)) [55], and modified normalized differenced water index (MNDWI) (Equation (3)) [56] were constructed. The nearest neighbor method was used for training and rule extraction of selected samples to obtain the land-use classification results for the Poyang Lake area over 7 periods in the last 30 years:

$$\mathrm{NDVI}=\frac{{\rho }_{nir}-{\rho }_r}{{\rho }_{nir}+{\rho }_r}$$

(1)

$$\mathrm{NDBI}=\frac{{\rho }_{mir}-{\rho }_{nir}}{{\rho }_{mir}+{\rho }_{nir}}$$

(2)

$$\mathrm{MNDWI}=\frac{{\rho }_{green}-{\rho }_{mir}}{{\rho }_{green}+{\rho }_{mir} }$$

(3)

where ${\rho }_{nir}$, ${\rho }_r$, ${\rho }_{green}$, and ${\rho }_{mir}$ represent the reflectance in the near-infrared, red, green, and mid-infrared wavelengths, respectively.

3.2. The RF-Markov-FLUS Model

3.2.1. Selecting the Drivers of Land-Use Change

Accurate screening of driving factors is the basis for revealing land-use change patterns and predicting the development suitability of various land uses. The random forest algorithm can discriminate the importance of input variables by a multi-decision classification tree, and its effect is better than that of a single classifier [57]. In this study, the random forest (RF) algorithm in R was used to filter 16 drivers (Table 1) from 4 aspects: natural, accessibility, socio-economic, and limiting factors, using 2015–2020 land use data as the dependent variable and factors as the independent variables. After iterative training, the accuracy was 93.14%, which reflected high reliability. The selected factors were used as parameter inputs for the FLUS model simulation.

3.2.2. Quantity Forecast

Markov chain (Equation (4)) is a forecasting method created by Russian mathematicians in the 20th century, based on probability theory to analyze land-use change patterns and trends. It uses a transfer probability matrix to forecast the future demand for land-use types [26,33]:

$$X_{t+1}=X_t\times P_{ij}$$

(4)

where $X_t$ and $X_{t+1}$ are the states of the land at times t and t + 1, respectively, and t is the year; $P_{ij}$ is the state transfer probability matrix, which is the probability of transferring land type i to land type j.

In this study, the Markov model in IDRISI software was used to forecast the land-use quantity in 2030 based on the current land-use quantity in 2010, 2015, and 2020. The calculated values were used as parameter inputs for the second module of the FLUS model.

3.2.3. Spatial Distribution Simulation

The FLUS model is a future land use model that combines human–nature interactions to simulate land use under multi-scenarios [36]. FLUS is available for free download at http://www.geosimulation.cn/FLUS.html (accessed on 7 August 2021).

In the first module of the model, the factors are imported into the module and the probability diagram of land-use suitability is obtained by ANN, which can effectively correlate nonlinear relationships. The suitability probability was calculated by the following equation (Equation (5)) [36]. The obtained probability map of land-use suitability was used for the second module:

$$P_{p,k}^t={{\displaystyle \sum }}_j{w}_{jk}\times S\left(N_j\left(p,q\right)\right)$$

(5)

where $P_{p,k}^t$ is the suitability probability of the tuple $p$ to convert to a $k$ class of sites at time$t$, and$t$ is a year; $w_{jk}$ is the weight of the hidden layer and the output layer; and $S\left(N_j\left(p,q\right)\right)$ denotes the function correspondence between the output layer and the hidden layer.

The second module, a CA model with roulette and adaptive inertial competition mechanisms, simulates the spatial changes in future land use. In it, the adaptive inertia coefficient is adapted to the difference between the current amount of each land type and the scale of demand and iteratively adjusted to approach the target scale [36]. The equation for the adaptive inertia coefficient is as follows (Equation (6)):

$$\mathrm{Inertia} { }_k^t=\{\begin{array}{ll} \mathrm{Inertia} { }_k^{t-1}\hfill & \mathrm{if} \left|D_k^{t-1}\right|⩽\left|D_k^{t-2}\right|\hfill \\ \mathrm{Inertia} { }_k^{t-1}\times \frac{D_k^{t-2}}{D_k^{t-1}}\hfill & \mathrm{if} D_k^{t-1}<D_k^{t-2}<0\hfill \\ \mathrm{Inertia} { }_k^{t-1}\times \frac{D_k^{t-1}}{D_k^{t-2}}\hfill & \mathrm{if} 0<D_k^{t-2}<D_k^{t-1}\hfill \end{array}$$

(6)

where Inertia ${ }_k^t$ denotes the adaptive inertia coefficient of land class k at iteration time t, and $D_k^{t-1}$ is the difference between the current land class area and the target size at time t − 1.

In the CA model, parameters such as the conversion probability matrix, domain factor index, restricted area setting, and future land-use demand are combined to deal with the complexity and uncertainty of land-use change under the interactions between human activities and the surrounding environment, and to improve the simulation accuracy.

3.3. Landscape Pattern Index

The analysis of landscape pattern dynamics is a focal area of landscape ecology research. The shape, size, number, and spatial combination of landscape patches reflect the intensity of human activities, the degree of economic development, land-use level, and other characteristics [58]. Analysis of the landscape index can reflect changes in landscape patterns, quantitative indicators of the landscape structural composition, and spatial configuration characteristics [59]. Based on the relevant literature and the current situation of the study area, the following landscape pattern indices in the class/landscape metrics were selected for this study: the Mean Fractal Index Distribution (FRAC_MN), Interspersion and Juxtaposition Index (IJI), the Landscape Shape Index (LSI), the Landscape Division Index (DIVISION), the Cohesion of Patches Index (COHESION), the Contagion Index (CONTAG), the Shannon Diversity Index (SHDI), the Shannon Evenness Index (SHEI), and the SPLIT Index.

Table 3. Landscape indices and their ecological significance.

Metrics	Index	Ecological Significance
Class/Landscape	FRAC_MN	Reflects the complexity of the shape in the range of spatial scales. The larger the value, the more convoluted the shape is from the regular geometry.
	IJI	IJI converges to zero when the distribution of nodes of a particular patch type in the landscape becomes uneven.
	LSI	As the landscape shape becomes irregular and the edges lengthen, LSI increases and has no maximum limit.
	DIVISION	The value approaches 1 when the area weight and patch size of that patch type in the landscape decreases.
	COHESION	When the connectivity of a patch type in the landscape decreases, the value approaches 0. As the proportion of that type of patch composition in the landscape increases, the value increases.
Landscape	CONTAG	Reflects the degree of fragmentation of landscape patches, and the value approaches 0 when all patch types are maximally fragmented and randomly distributed.
	SHDI	Reflects landscape heterogeneity, and higher values reflect more diverse land use and higher fragmentation.
	SHEI	Reflects the degree of uniformity in the distribution of different ecosystems in the landscape, and values close to 1 indicate more uniformity.
	SPLIT	Reflects the degree of dispersion of landscape patches, and a maximum value occurs when the whole landscape is maximally refined.

contains each landscape index and its ecological significance.

3.4. Multi-Scenario Spatial Constraints and Parameter Settings

In this study, the parameters for spatial constraints, the transfer probability matrix, neighborhood factors, and the conversion cost matrix were set by the FLUS model to construct the land use pattern of each spatial type under four scenarios: PSP, LSP, ESP, and ID. The domain factor indicates the difficulty of conversion between land-use types, with a value ranging from 0–1, where 1 represents the high expansion capacity of that type and a lower likelihood of conversion to another type [60]. The conversion cost matrix indicates the difficulty of conversion from the current land-use type to future land-use types, where one means conversion between two land-use types is allowed and zero means conversion is not allowed.

Different spatial constraints and transfer probability matrices were set for each of the four scenarios. The spatial constraints were the three restricted conversion area vector data for the ecological protection red line, the permanent basic cropland protection red line, and urban development boundary. These were converted into binarized raster TIFF files, where a value of 1 indicated conversion and 0 indicated no conversion transfer probability. Multi-scenario planning was implemented by modifying the transfer probabilities in specific experiments.

Under the PSP scenario, the permanent basic farmland protection red line was used as a restriction and the probability of transferring productive land to construction land was reduced. The probability of transferring construction and ecological land to productive land increased, and the demand for agricultural land increased to meet the priority development of productive space.

Under the LSP scenario, the urban development boundary was used as a restriction and reduced the probability of transferring construction land to cropland, water, forest, and grassland. It increased the probability of transferring production and ecological land to construction land.

Under the ESP scenario, the area within the ecological protection red line was treated as a restricted area, and the conversion of ecological protection land to other land within the red line was strictly controlled. At the same time, the probability of transferring water, forests, and grasslands to construction land was reduced, and the probability of transferring productive land to ecological land increased to achieve effective environmental protection.

Under the ID scenario, the transfer probability of the above three scenarios was integrated to make numerical adjustments. The three red lines of national land space (ecological protection red line, permanent basic cropland protection red line, and urban development boundary) were combined to form the restriction conditions for coordinated and symbiotic development of living space, production space, and ecological space.

3.5. The Space Type Dynamic Degree

The space type dynamic degree indicates the change in a certain space type in the study area within a time frame [16]. The expression is as follows (Equation (7)):

$$K=\frac{U_b-U_a}{U_a}\times \frac{1}{T}\times 100\%$$

(7)

where K is the dynamic degree of a space type during the study period; $U_a$ and $U_b$ are the number of a space type at the beginning and end of the study period, respectively; T is the length of the study period, and when the time period T is set to a year, the value of K is the annual rate of change in the space type in the study area.

4. Results

4.1. Accuracy Verification

The land use classification data were verified using a sample-based confusion matrix. The accuracy results of the extracted land classification for the seven periods were 90.03%, 91.45%, 92.64%, 90.37%, 93.98%, 89.81%, and 90.29%, all of which were greater than 85%, providing basic data for subsequent high-precision simulations.

The RF-Markov-FLUS model simulated 2020 land-use data based on the 2000 and 2010 land-use data, and the results were verified against the real land surface for accuracy. With a random sampling of 10%, the Kappa coefficient was 0.912 and the simulation accuracy was as high as 90.3%, which met the needs of land-use change simulation. Parameters such as the influence factors and domain weights also met the simulation needs and the RF-Markov-FLUS model had good applicability in this study.

4.2. Analysis of the Evolution of PLE Space from 1989 to 2020

4.2.1. Analysis of Spatial and Temporal Changes

This paper used the dynamic degree to measure the net rate of land-use change [61]. Spatial-temporal changes in PLE in the PYL area were compared and analyzed (

Table 4. Change in the PLE area in the PYL area from 1989 to 2020 (km²).

	Production Space	Living Space	Ecological Space
Year	Cropland Area	Construction Land	Water	Forest	Grassland	Unused Land
1989	4695.60	1080.98	4905.83	4884.20	3839.61	902.39
1995	4856.34	1181.21	4183.11	5005.16	4197.74	898.07
2000	4679.42	1303.56	4311.90	4823.64	4047.56	1155.96
2005	4655.59	1366.57	4639.13	4780.30	3915.82	966.79
2010	4591.02	1893.56	4712.12	4499.16	3368.24	1260.10
2015	4547.87	2420.23	3750.40	4471.04	3760.14	1379.22
2020	3977.70	3225.70	4978.33	4407.06	2876.11	856.33
Area of change for each space	−717.9	2144.72	−1414.19
dynamic degree (%)	−0.58	2.14	−0.35

, Figure 4), and the results showed that the largest percentage of ecological space in the national land area was 69% on average per year during 1989–2020, followed by production space and living space, with 23% and 9% on average per year, respectively. During the study period, living space had the highest dynamic degree of 2.14%, which increased by 2144.72 km² with the largest growth rate.

Figure 5 shows that during the study period, production space was mainly distributed in the central lakeside plain of the PYL area; living space was mainly distributed in urban development areas, such as Nanchang City in the southwest and Jiujiang City in the north; and ecological land was mainly distributed around the ecological reserves, such as Lushan area in the north, Poyang Lake in the middle, and Meiling Forest Park in the west.

4.2.2. Transfer Matrix Analysis

A transfer matrix represents the transformation of each type of site from the beginning to the end of the study [61]. This study used the ENVI5.3 platform to calculate the transfer matrix results from 1989 to 2020. The results are shown in

Table 5. Spatial transfer matrix in the PYL area from 1989 to 2020 (km²).

1989/2020	Production Space	Living Space	Ecological Space	Total Transfers Out
Production space	2413.36	996.05	1446.29	2442.34
Living Space	72.63	600.84	507.11	579.74
Ecological Space	1474.95	2348.33	10,457.85	3823.28
Total transfers in	1547.58	3344.38	1953.4

and Figure 6. During the study period, PLE space in the PYL area was dominated by a transfer out of ecological space and a transfer in of living space.

4.2.3. Landscape Pattern Analysis

Figure 7 presents the variability of each indicator for the seven periods in different landscapes at the class level.

(a)

The FRAC_MN index indicates the geometric complexity of each landscape type. During the study period, the production space and ecological space areas increased while the living space area decreased. This indicated that ecological and production spaces became more complex in shape, whereas living space expanded regularly from the center to the surroundings, with spatial continuity and unity. The landscape complexity decreased.

(b)

The IJI index reflects the overall dispersion and juxtaposition between specific landscape patch types. The production and living spaces exhibited more or less the same increase while ecological space increased and then decreased. Compared with the other two spaces, the IJI index for ecological space decreased from a maximum value in 1989 to a minimum value in 2020. Ecological space connectivity increased with the landscape patches of the other two spaces, indicating that the conservation of ecological space is easily constrained by the surrounding human activities.

(c)

The LSI index reflects the irregularity of landscape patches. During the study period, the production, living, and ecological space areas all increased, and for all of them, 2005 and 2010 were turning points for growth rate increases. The most significant change occurred in ecological space, which indicated that under the influence of human activities, the landscape regularity change was hindered, and fragmentation occurred.

(d)

The DIVISION index reflects the degree of separation of dominant landscape patches and patch integration. As Figure 7 shows, although the production and living space index values were higher, the changes were more stable compared to ecological space. Ecological space decreased and then increased, indicating that the dominant patches of ecological space decreased and were more dispersed from each other.

(e)

The COHESION index reflects the degree of aggregation and connectivity of each patch. The cohesion of patches decreased in production space and increased in living space, but ecological space remained stable, which indicated that the continuous expansion and protection of the landscape can improve connectivity. In general, the complexity, fragmentation, and dispersion of production and ecological space gradually increased, whereas the living space change was more steady, and the degree of connectivity and overall degree improved.

Figure 8 reflects the changes in each index PYL area over the landscape during the study period. The FRAC_MN and LSI indices showed the edge curvature of the landscape shape. The small increase in FRAC_MN and the large increase in the LSI values during the study period indicated that the overall landscape shape became detached from the regular geometry and more complex. The SHDI and SHEI indices increased, and the distribution of each patch type was non-uniform, which reflected increasing landscape fragmentation and landscape heterogeneity in the region. The increase in the DIVISION and SPLIT index and the decrease in the CONTAG index indicated that landscape aggregation decreased, the degree of integration of the landscape was weakened, and there were no dominant patches. From 1989 to 2020, the COHESION index was stable, indicating that the overall landscape connectivity in the PYL area changed smoothly.

4.3. Multi-Scenario Simulation Results

Figure 9 and

Table 6. Area of each space and its percentage in 2030 under the PSP, LSP, ESP, and ID scenarios.

Scenario	PSP	LSP	ESP	ID
Production space (km2)	4162.01	3889.45	3819.99	3881.52
Percentage of the whole area (%)	20.49	16.3	18.8	18.94
2020–2030 rate of change (%)	0.91	-3.28	−0.77	−0.64
Living space (km2)	4343.28	5183.69	4318.21	4415.94
Percentage of the whole area (%)	21.38	25.51	21.25	21.54
2020–2030 rate of change (%)	5.5	9.63	5.38	5.67
Ecological Space (km2)	11,811.95	11,823.39	12,179.04	12,200.01
Percentage of the whole area (%)	58.14	58.19	59.94	59.52
2020–2030 rate of change (%)	−6.41	−6.36	−4.61	−5.03

show the simulation results for the four scenarios. Overall, most of the production space was regularly distributed in Xinjian District and flat areas such as northeastern Nanchang County and northwestern Yugan County while other areas were more scattered. Living space was concentrated in the urban area of Nanchang City, the urban area of Jiujiang City, and other town centers. Ecological space was concentrated around Poyang Lake and northeast Poyang County, west Yongxiu County, south Jiujiang City, and other areas where the nature reserves were widely spread. In addition, three representative areas were selected to study the spatial details under multi-scenarios in this study, including the obvious urban sprawl area in Nanchang (Figure 10a area), the concentrated cropland area in the northern part of Yugan County (Figure 10b area), and the national ecological reserve in Lushan (Figure 10c area).

In the PSP scenario, production, living, and ecological spaces accounted for 20.49%, 21.38%, and 58.14% of the total area, respectively. When the conversion of cropland to other land was restricted, production space increased by 0.91%, living space increased by 5.50%, and ecological space decreased by 6.41% compared with 2020. This showed that decreasing production space was effectively controlled in the future, but ecological space became the main source of living space expansion to maintain the human demand for the construction and development of living space. From Figure 10, we know that the ecological space in zone c was threatened mainly by the development of production space. In this scenario, ecological space not only served the production space but also made concessions for the living space, and production space and living space were mutually constrained. This was not conducive to construction development or environmental protection.

In the LSP scenario, production, living, and ecological spaces accounted for 16.30%, 25.51%, and 58.19% of the total area, respectively. Compared with 2020, the expansion of living space increased in 2030 to meet the demand for human production and living space. With an increase of 9.63%, it was the fastest growth rate compared to other scenarios while production and ecological space decreased by 3.28% and 6.36%, respectively. This indicated that the growth of living space was converted by a significant decrease in production space and a small decrease in ecological space to prioritize living space development. Area a in the LSP scenario in Figure 10 shows that the living area in the Nanchang urban area was saturated with a substantial extension to the north to occupy production space. In areas b and c, living space gradually encroached on ecological space from the periphery. Under this scenario setting, the material needs of people and urban construction were met to promote economic development while regional ecological security and cropland protection were neglected, threatening ecosystems and food security.

In the ESP scenario, the production, living, and ecological spaces accounted for 18.80%, 21.25%, and 59.94% of the total area, respectively. Compared with 2020, production space decreased by 0.77%, living space increased by 5.38%, and ecological space decreased by 4.61% as living space was constrained and ecological sustainable development in the region was promoted. Compared to the first two scenarios, the loss of ecological space was effectively reduced, the expansion of living space was controlled, and the decrease in production space was reduced. Under the ESP scenario in Figure 10 for areas a and b, it can be seen that the encroachment of living space into production space and ecological space was reduced, and living space was mainly transferred from urban areas. This scenario can substantially improve ecosystem quality, and strong policy protection would add to ecological protection, but it also suppresses the development of living space and production space, and it is difficult to obtain new results in socio-economic development.

In the ID scenario, the production, living, and ecological spaces in 2030 accounted for 18.94%, 21.54%, and 59.52% of the total area, respectively. Compared with 2020, the production and ecological spaces decreased less, by 0.64% and 5.03%, respectively, while living space increased moderately by 5.67% under the reasonable demand regulations of each space type. This scenario focused on the rational allocation of each space, which can balance the conflicts between the demand for food output, socio-economic development, and ecological protection. It promotes reasonable land use and optimizes the structure of each space.

5. Discussion

5.1. Discussion of the RF-Markov-FLUS Model

CA, Markov, CLUE-S, and other models are commonly used in current land-use simulation research. The traditional CA model has strong spatial simulation capability, but it does not obtain conversion rules for meta-automata in a linear way, and the influence of subjective factors is large. It also fails to objectively reflect stochastic evolution and future development trends in land-use change [62]. The Markov model conducts superior simulation over long time periods but is unable to simulate spatial pattern change and needs to be combined with a spatial prediction model [63]. The CLUE-S model has a strong spatial allocation function but is limited in land quantity prediction [36].

No single traditional model can adequately study the complex changes in a long time series in most regions and coupled models combining the advantages of quantitative and spatial prediction are the inevitable trend in land-use model development. ANN-CA, Logistic-CA-Markov models, and other coupled models are widely used, and the RF-Markov-FLUS coupled model was the main research method used in this paper. The RF-Markov-FLUS model greatly eliminated the influence of subjective factors, simulated complex land transformations, and improved the simulation accuracy. The RF-Markov-FLUS model can be used to more accurately predict future land-use patterns, and the high accuracy screening mechanism of the RF algorithm can be used to improve the ability of Markov-FLUS to predict future land-use pattern changes, which is important for predicting land-use patterns in small- and medium-scale regions.

Although the model accuracy met the requirements of the study, the RF-Markov-FLUS model had certain limitations. First, it predicted land-use demand as a constant value, and the spatial allocation of the raster was quantitatively limited. Meanwhile, the model used historical land-use data to simulate the future land-use quantity but did not consider the unknown changes in the future population and economic structure, which caused deviations between the simulation results and the actual situation. Second, the model failed to set the weights of the driving factors when inputting the normalized driving factor data, and their degree of influence on land-use changes was not reflected. Therefore, in the future, we can consider adding population and economic structure changes, and other factors in the prediction, and improve model parameters and increase the factor weight settings to make the simulation results more realistic.

5.2. Comparison of Multi-Scenario Simulation Results

In this study, due to PLE space research demands, the spatial patterns of the PYL area in 2030 under the PSP, LSP, ESP, and ID scenarios were simulated. The results showed that in the LSP scenario, production space was compressed, and the area decreased significantly due to urban expansion, whereas in the PSP scenario, the production space area had a stable increasing trend due to the rigid control of the permanent basic cropland protection red line. In order to meet the spatial needs of urban residents for production and living, living space increased in all scenarios but at varying rates, especially in the LSP scenario. Ecological space decreased in all four scenarios, but it was not strictly protected by the ecological protection red line in the PSP and LSP scenarios and was easily transformed into production and living space. The area was significantly reduced.

In recent years, the Yangtze River Economic Belt has been oriented to “well-Coordinated Environmental Conservation and Avoiding Excessive Development” and focused on ecological priorities and sustainable development [64]. The PYL area is an important hinterland for the development of the Yangtze River Economic Belt, which can promote coordinated development between ecology and economy [65]. The ID scenario proposed in this study maintained the smooth development of production and living spaces under the premise of protecting ecological space. Under this scenario, production space was stable, the growth rate of living space declined, and the decline in ecological space slowed down. Compared with the simulation results of the other three scenarios, the ID scenario can satisfy the needs of all three to a certain extent, whether it promotes intensive development of production space, or the livable and moderate living space and the beautiful mountains and clear water of ecological space. It provides an ideal development path for the coordinated development of multiple spaces in the PYL area [66].

5.3. The Optimization of Territorial Space Pattern Enhancement

Integrating the results of historical spatial and temporal land-use changes, landscape patterns, and future land multi-scenario simulation, it can be seen that the urbanization rate of the PYL area is accelerating. The PYL area city cluster is bordered by the Yangtze River in the north, Meiling in the west, and contains Poyang Lake and Mount Lu in the middle. There is a large amount of concentrated continuous cropland around, and urban development and construction threaten the surrounding ecological and production space. In addition, the reduction of ecological space area and landscape fragmentation are partly attributed to the anthropogenic enclosure of lakes, uncontrolled sand mining, and farming [67], in which some lake resources are converted into cropland resources to promote agricultural development while excessive enclosure often damages the lake natural resources and destroys lake environments and storage functions [68].

Improvement in the function of territorial space requires a combination of problem and goal orientation. According to the territorial spatial positioning and the problems faced by development, the three aspects of agricultural industry transformation, improving the efficiency of urban land use, and creating communities of “mountains, water, forests, fields, lakes and grass” were addressed to alleviate inefficient use of arable land, sloppy use of urban space, fragile ecology, and human damage, and to establish a framework for upgrading the path (Figure 11).

Considering the resource endowment and development status of PLE space in the PYL area, the agricultural production function spatial distribution was optimally adjusted to realize the scale and industrialization of agricultural production, improve the efficiency of agricultural production space, and promote the intensive and efficient use of production space. Production space should be in terms of quantity, quality, and ecology, with lake and wetland agriculture, urban agriculture, and traditional agricultural panels as the base, inlaid with modern agricultural professional industrial parks and modern field complexes, to build a new pattern of agricultural development. The lake agriculture relies on the wetland resources of Poyang Lake, and is focused on the development of ducks, fish, crabs, and rice as the core fishing composite agricultural system with the advantages of the lake’s special agriculture. The urban agriculture segment, which is focused on the development of high-value-added agriculture, such as agricultural processing and livestock breeding, improves the overall environment of the city. In the traditional agricultural sector, we should develop leading industries such as high-quality crop planting and agricultural processing, form agricultural industrialization clusters, build professional agricultural production and processing bases, promote the construction of high-standard modern agricultural industrial parks and modern pastoral complexes, and build a new pattern of production space development with three sectors, multiple parks, and multiple nodes.

With the Central Rising Strategy and Yangtze River Economic Belt policy, Jiangxi Province proposes the promotion of a national urbanization strategy pattern in the Greater Nanchang Metropolitan Area [69]. The future expansion of the city cluster with Nanchang City as the core is inevitable. Combined with the urban development plan for the Poyang Lake area, a reconstructed living space pattern should build an urban functional area with the core municipal central business and leisure area, accelerate the integration process of Nanchang-Jiujiang, and form a core; upgrade the urbanization level of three urban clusters in Jinxian County, Yugan County, and Yongxiu County to form three clusters; and build a living space structure of “one core, three groups and many branches”. This living space pattern improves the connection between different regions to promote integrated living space development, drives coordinated development within the living space, reduces the gap in the carrying capacity and living standards between regions, cultivates urban core areas and strengthens their agglomeration and radiation capacity, strengthens the connection between different administrative units, builds cross-regional town clusters, and improves the efficiency of living space utilization in cross-regional spatial development.

The protection, development, and utilization of ecological space is an important guarantee for coordinated territorial space development, and the creation of “Mountain, Water, Forest, Field, Lake and Grass” communities is an important way to improve ecological space. Starting from the overall situation of the ecosystem, the ecological corridor of Mount Lushan-Meiling Mountains, the “one lake and five rivers” (Poyang Lake, Ganjiang River, Fu River, Xinjiang River, Rao River, and Xiu River) ecological corridor, will coordinate all natural ecology elements for overall protection and systematic restoration and management, and strengthen the capacity of ecosystem circulation. For biodiversity conservation, nature reserves with the main function of water conservation and biodiversity protection should be established, and the management of typical ecological reserves (including nature reserves, scenic spots, forest parks, wetland parks, and Yangtze River dolphin reserves) should be strengthened to save endangered species.

6. Conclusions

Future land-use changes have been simulated using coupled models for a long time, but there have been fewer studies of multi-scenario simulations that consider spatial policies. This study took the perspective of PLE space and used a coupled Markov and FLUS model to consider 16 driving factors. Land-use data from 1989 to 2020 were used, model accuracy was validated, and the spatial patterns of national land in Poyang Lake region in 2030 were simulated under four scenarios with spatial policy constraints. Historical and future land-use changes were analyzed and an optimum scenario for sustainable development in the region was determined. The main conclusions are as follows:

(1)

From 1989 to 2020, the production and ecological space areas in the study area both declined, by 717.9 and 1414.19 km², respectively. Living space expanded from the center to the periphery with the fastest growth rate, with an area increase of 2144.72 km², mainly dominated by the urban areas of Nanchang City and Jiujiang City, indicating the continuous expansion of urban living space.

(2)

Secondly, the PYL area experienced enhanced landscape fragmentation, landscape heterogeneity, landscape connectivity, and landscape dominance during the study period.

(3)

The overall accuracy of the RF-Markov-FLUS coupled model simulation was as high as 90.3%, and the Kappa coefficient of the model was 0.912. This showed it has strong applicability in the region and can predict the future spatial pattern of land use in all districts and counties in the PYL area.

(4)

According to the four scenario simulation results in 2030, production, living, and ecological spaces can be developed and protected to the greatest extent in the PSP, LSP, and ESP scenarios, respectively, but the ID scenario can more scientifically and reasonably lay out the spatial pattern of production, living, and ecology, realizing multiple functions of production, living, and ecology.

(5)

Based on the ID scenario and PYL area profile, we established a path framework to improve the territorial spatial function in three aspects: agricultural industry transformation, improved urban land use efficiency, and the creation of “mountain, water, forest, field, lake and grass” communities, which will help the government make more sustainable decisions in the PYL area.

In this study, a coupled model was used to study the spatio-temporal evolution and multi-scenario simulation of PLE space in the PYL area, but an accuracy comparison with other models was lacking. Therefore, comparison of the results between multiple models should be enhanced in future studies for a more in-depth study.