Simulation and Prediction of Territorial Spatial Layout at the Lake-Type Basin Scale: A Case Study of the Dongting Lake Basin in China from 2000 to 2050

Abstract

The study of spatial layout in China is changing from land use/land cover to territorial spatial layout and is mostly carried out at the administrative division scale. However, the driving factors affecting the evolution of territorial spatial layout are not all influenced by administrative boundaries. As cities are mostly centered on lakes and water systems, the evolution of territorial spatial layout in lake-type basins must have its own special characteristics. Taking the Dongting Lake Basin (DLB), a representative lake-type basin in China, as an example, this study identifies its territorial spatial layout elements and couples the system dynamics model with the patch-generating land-use simulation model based on multi-layer perceptron artificial neural networks. This study identified the best combination of driving factors and research step size affecting the evolution of territorial spatial layout. An improved quantitative spatial coupling model was used to simulate the territorial spatial layout of the DLB from 2000 to 2050 and identify its evolutionary characteristics and trends at both the elemental level and for three types of space. The simulation and prediction of territorial spatial layout from the lake-type basin hydrology scale can provide a reference for the formulation of regional territorial spatial planning policies.

1. Introduction

In 2011, the State Council of China issued the National Plan for Main Functional Zones, which for the first time proposed the concept of “territorial space” that is similar to the nature of the environment. This space is an important spatial carrier for human production and life and socio-economic activities as well as ecological civilization construction including land, land waters, internal waters, territorial sea, and territorial air [1]. China’s approach to land development has undergone a transformation from urban expansion [2,3,4] to the coordinated development of production, living, and ecological space [5,6,7] and then was identified as the integrated development of ecological space, agricultural space, and urban space (three types of space) [8,9]. The three types of spatial identification research methods can be divided into two: the top-down decomposition of functional zoning and the bottom-up integration of land-use and -cover types, similar to the integrated approach within the framework of critical physical geography from Europe [10]. The former method is mostly used for qualitative analysis at macroscopic scales, such as the national and provincial levels with administrative districts as evaluation units, while the latter method is mainly employed for quantitative analysis at medium and micro scales, such as cities, counties, towns, and villages with land-use and -cover types as evaluation units [11,12]. Some scholars have also combined the two methods to conduct comprehensive research, but most took the administrative division as the evaluation unit without paying attention to the inconsistency between units and administrative districts with regard to the factors affecting the territorial spatial layout, with even less attention being paid to the empirical evidence regarding the territorial spatial layout of lake-type basins.

With the rapid socio-economic development and urbanization in China, the spatial layout of China’s territorial space is undergoing drastic changes, leading to the increasingly prominent contradiction between the utilization and protection of China’s territorial space and new challenges regarding the sustainable development of China’s territorial space [13,14]. The current territorial spatial planning measures need to clarify the evolution mechanism of the territorial spatial layout in the past on the one hand while grasping the development trend of its natural expansion on the other. The simulation method of territorial spatial layout evolution is the same as that of land-use and land-cover changes (LUCC), which can be divided into three categories according to its development history [15,16,17,18]: (1) quantity change prediction, explaining the problem of “How”; (2) spatial pattern change prediction, explaining the problem of “Where”; and (3) coupled models, which are generated because a single model cannot satisfy the complex characteristics of the land-use change process. Quantitative simulation models mainly include system dynamics (SD) [19,20], gray forecast (GF) [21], and Markov chain (MC) models [22], which can effectively simulate and predict the quantitative structure of land-use types but do not easily realize spatial structure simulation [19,20]. Spatial simulation models include cellular automaton (CA) [23], conversion of land use and its effects at a small regional extent (CLUE-S) [24,25], the future land-use simulation model (FLUS) [26], the agent-based model (ABM) [27], and the patch-generating land-use simulation model (PLUS) [28], all of which can all be used to simulate the spatial layout of land types but are deficient in quantitative simulation. The coupled model aims to combine the advantages of multiple models, improve the simulation accuracy of a single model, and provide a strong guarantee for land-use research [29,30,31]. In recent years, the quantitative spatial coupling model has become the main means of land-use and spatial pattern research because it can effectively simulate land use and cover, the number of land spatial types, and their rapid changes in space. At present, land-use and land-cover change simulation mainly use quantitative models to obtain the quantitative structure of each land type and then use spatial models to obtain the spatial layout under different future scenarios. The SD-CLUE-S models are the most frequently used. He et al. [32] used the SD model to numerically simulate each land type at the macro level with Chengdu City as the study area and used the CLUE-S model to conduct spatial distribution at the micro level for the values, successfully exploring the future urban growth model under different scenarios. Wu et al. [33] established a coupled SD-CLUE-S model and found that the comprehensive model could better simulate the dynamic changes in the landscape ecosystem value in the Baoshan area of Shanghai. Liang et al. [34] simulated and predicted the effects of land-use change on carbon storage at the pixel scale and regional scale in Zhangye Oasis from 2000 to 2018 based on SD-CLUE-S and integrated valuation of ecosystem services and trade-offs models (InVEST). Taking Hefei City, China, as an example: Yu et al. [35] used the multi-objective planning (MOP) model to forecast the land demand of the territorial space and generated the PLUS model to build a computer simulation of the territorial spatial layout, which successfully simulated the future development direction of construction land. For the selection of spatial simulation models, scholars compared the models according to the simulation results. Liang et al. [36] studied land use in Wuhan city and found that the PLUS model was better than the FLUS model in simulating the historical land-use change process. By comparing the simulation results of the SD model combined with the PLUS, FLUS, and CLUE-S models, Jiang et al. [37] found that the PLUS model was superior to the FLUS and CLUE-S models in terms of simulation accuracy and also had obvious advantages in landscape pattern simulation, with the overall simulation effect being significantly better than that of the FLUS and CLUE-S models. The choice of a coupled model is only one aspect of guaranteeing the accuracy of the simulation predictions. In the process of coupled model construction, the choice of driving factors is also crucial. The evolution of the territorial spatial layout is a complex system, and it is not possible to consider all the driving factors completely. There is a limit to the number of driving factors that can be obtained given the limitations of data availability. However, a higher number of driving factors does not mean a higher accuracy between their combined impact effect and the potential for change at the time of actual development [38]. At the same time, all the driving factors are not static but evolve with the natural and socio-economic development and territorial spatial layout, and the driving factors vary somewhat at different study steps. Therefore, identifying the best combination of driving factors among the limited driving factors and the best research step can improve the scientificity and accuracy of the coupled model. How to combine the best combination of driving factors and the best research step with the coupled model is one of the research issues in this study.

The catchment area of a water system is called the basin. Catchments that end in lakes are called lake-type basins, while catchments that end in estuaries or the sea are called river-type basins. Since ancient times, people have lived and built around water, so the territorial spatial layout is more complex in lake-type basins with a multi-level water network. The Dongting Lake Basin (DLB) is an indispensable part of the Yangtze River Basin and is known as the kidney of the Yangtze River in China. The DLB plays a significant role in natural flooding, flood storage, and discharge, protecting the flood control safety of the middle reaches of the Yangtze River and assuring the ecological security of the Yangtze River Basin. At the same time, the rapid economic development of the DLB has led to rapid urban expansion, and the contradiction between urban-agricultural-ecological space has become increasingly prominent. The territorial spatial layout of the DLB is extremely complex and special, relying on the radial water system with Dongting Lake as the center. Therefore, the DLB is of great research value. Fewer studies have been conducted on the future territorial spatial layout of the DLB, and most studies have been conducted on land-use simulation and prediction. Yang et al. paid attention to the coupling relationship between land use, habitat quality [39], and ecological risk [40] in the DLB for a long time. Relying on the suitability evaluation of land space development, Yang et al. [41] evaluated the land suitability of the DLB based on the multi-factor spatial superposition method and discussed the land suitability of the DLB under different functional directions such as ecological protection, agricultural production, development, and construction. Building a territorial spatial layout for high-quality development is important for supporting the promotion of regional sustainable development. As the representative of a lake-type basin, the DLB needs to keep up with the national pace of territorial space planning. To grasp the evolution mechanism of the territorial spatial layout and the future natural development trends in the DLB is the prerequisite for the optimization of the territorial spatial layout and the key issue that needs to be addressed urgently. In this study, the improved SD-PLUS model was constructed to simulate and predict the evolution and future development trend of territorial spatial layout in the DLB and to provide guidance for regional territorial spatial development.

2. Materials and Methods

2.1. Data Sources and Study Area

2.1.1. Data Sources

The data sources were derived from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (RESDC) (http://www.resdc.cn/, accessed on 2 January 2023), including China’s Multi-Period Land-Use Land-Cover Remote Sensing Monitoring Dataset (CNLUCC), in 2000, 2005, 2010, 2015, and 2020. Data precision is 30 m × 30 m. This is currently one of the open-access, high-resolution land-use remote sensing monitoring data products in China. This dataset divides land-use types into 6 primary types and 25 secondary types.

2.1.2. Study Area

Dongting Lake is one of five major freshwater lakes in China. It is a famous area for fish and rice. In the north, Songzikou, Taiping Kou, and Wuchikou divert the water of the Yangtze River into the lake. In the south, the Xiangjiang River, Zijiang River, Yuanjiang River, and Lishui River flow into Dongting Lake, forming a radial water system centered on Dongting Lake, that is, the DLB. The DLB is located in the south of the middle reaches of the Yangtze River and north of the Nanling Mountains in China (107°26′ E–114°20′ E and 24°36′ N–30°27′ N). It flows through Hunan, Guizhou, Hubei, Guangxi, Chongqing, Jiangxi, and Guangdong, with a total area of 2.62 × 107 hm² (Figure 1). It covers 14.6% of the Yangtze River Basin [42]. The DLB is surrounded by mountains on three sides in the southeast and west. The terrain gradually inclines from the south to the center and northeast, showing a horseshoe shape that opens to the north.

2.2. Construction of Territorial Space Classification System

There are various ways of classifying territorial space, and to regulate the order of land development and reasonably manage different spaces, Chinese policies require the delineation of three types of space and the implementation of multiple regulations. Therefore, we take urban-agricultural-ecological spaces as the base and combine them with the current land-use situation in the DLB to construct a territorial spatial classification system. The land-use/land-cover remote sensing monitoring data released by the Chinese Academy of Sciences and the Guidelines on Land Use and “Guidelines for Land Use and Sea Use Classification for Land Space Survey, Planning and Use Control” were referred to in the process of constructing the system. Urban space includes urban areas, and agricultural space includes farmland and village. Ecological space includes forest land, garden, grassland, wetland, water, and other lands. The territorial spatial layout in the DLB in 2000, 2005, 2010, 2015, and 2020 is drawn in Figure 2.

2.3. Improved SD-PLUS Coupling Model

2.3.1. MLP-ANN Model

The evolution of territorial spatial layout is the product of the dual driving forces of nature and humanity, and there are many types and numbers of drivers. Meanwhile, the driving factors for different research steps show different influences on the evolution of territorial spatial layout. Therefore, judgment of the best combination of driving factors and research steps is a prerequisite for further development of the territorial spatial layout simulation model. To obtain the best research step size, 5-year and 10-year step sizes were set for analysis, respectively. Since the target year of simulation is 2020, the basic data periods of 5-year and 10-year step sizes are 2010–2015 and 2000–2010, respectively. Cramer’s V was used to represent the explanatory power of the driving factors of the change in territorial spatial layout [38]. A higher Cramer’s V indicates more explanatory power for the spatial layout changes. When the Cramer’s V of the driving factor is 0.15 or higher, it indicates that the factor has explanatory power for a spatial layout change. When the Cramer’s V value is 0.4 or higher, it indicates that the factor has good explanatory power for a spatial layout change. In this study, we need to exclude the driving factors with a Cramer’s V value of 0 and no explanatory power for all country space types (i.e., the Cramer’s V value of this factor is less than 0.15 for all country space types). Firstly, the explanatory power of each driving factor was ranked by calculating the Cramer’s V value, and multiple driving factor combinations were obtained by adding the driving factors one by one. After that, the multi-layer perceptron artificial neural networks model (MLP-ANN) was applied to test the accuracy of different driving factor combinations on the simulation of a potential change in territorial spatial layout. Half of the samples were randomly selected by the model for training, and the other half were used for validation. Finally, the best study step was determined by comparing the magnitude of the accuracy rate of territorial spatial transformation for the best combination of drivers among the two study steps, and the combination of drivers corresponding to the best study step with the highest accuracy rate was also determined as the best combination of drivers. The MLP-ANN model is more objective and holistic than the traditional driver research methods such as logistics regression models and multi-criteria evaluation (MCE) models. The MLP-ANN model is based on the Cramer’s V to determine the combination of drivers with the highest value of simulation accuracy determined through experimentation. Both the logistics model and MCE model are based on subjective judgment in selecting the drivers of spatial layout evolution, and the resulting combination of drivers does not necessarily represent a high simulation accuracy [38]. The MLP-ANN model explores the drivers to better reflect the complex natural and human factors that influence the evolution of territorial spatial layout, ensuring that the simulation results are more consistent with the actual situation.

2.3.2. Quantitative Structure Prediction Model

System dynamics (SD) is a quantitative prediction model based on mechanistic processes, which is a class of models based on cybernetics, system theory, and information theory to study the structure, function, and dynamic behavior of feedback systems and is often used to solve nonlinear system problems. The construction of an SD model influenced by specific driving factors and its application to the prediction of land-use demand on a time scale is also a hot topic in landscape architecture, geography, landscape ecology, and other disciplines. The architecture of an SD model enables it to better simulate the nonlinear, dynamic, and feedback effects of complex land-use systems and has advantages in reflecting driving factors and scenario simulation. However, the system dynamics model lacks the spatial concept in land-use simulation and does not have the ability to handle the spatial factors and spatial feedback of each element to predict the spatial changes in land classes [43]. Therefore, the system dynamics model is often combined with a geographical model with spatial analysis capabilities for spatial prediction simulations [44]. The specific steps include constructing the model subsystem system, determining the variable causality, and structural flow diagrams along with model calibration.

2.3.3. Spatial Pattern Prediction Model

The patch-generating land-use simulation model (PLUS) is a cellular automaton (CA) model based on raster data that can be used to simulate patch-scale LUCC. The PLUS model integrates two main parts: land-use expansion analysis strategy (LEAS) and CA based on multiple random seeds (CARS), which is widely used in LUCC simulation, policy formulation, land-use change mining, urban planning, ecological security early warning, etc. In the LEAS module, the best combination of driving factors for the best research step size is input into the model to obtain the transfer probability diagram of each territorial space type. In the parameter setting of the model CARS module, with reference to the area transfer matrix of the Markov chain module in the previous step-long segment, an area transfer ratio over 0.5% is considered to have the ability to transfer, i.e., set to 1, and vice versa, and the transition matrix parameter table is set considering the special characteristics of the construction land. The proportion of the expansion area of the territorial space type of the previous step-long section to the total expansion area of the study area is taken as the neighborhood weight of this type. In addition, for other model-operating parameters, we used the FoM (figure of merit) proposed by Pontius [45] as the evaluation index and carried out pixel-by-pixel accuracy evaluation of the simulation results of different parameter combinations, finally determining the model parameter setting.

2.3.4. Model Construction Process

The evolution of land spatial layout has both quantitative and spatial characteristics, and the use of the above single simulation model of land spatial layout evolution cannot effectively simulate the future land spatial layout. The comprehensive application of multiple models is the inevitable trend in the development of land-use models. On the basis of determining the best research step size and driving factor combination, based on the above combination of the characteristic advantages and operating principles of the quantitative structure prediction model (SD) and the spatial pattern prediction model (PLUS), the two types of models were coupled to construct an improved SD-PLUS coupled model to simulate and predict the dynamic changes in the spatial layout of the DLB. The model construction process is shown in Figure 3.

3. Results

3.1. The Best Driving Factor Combination and Study Step Size

Based on the principles of typicality and scientificity, we comprehensively considered the availability and quantification of data in the study area and initially selected a total of 18 driving factors affecting the spatial layout of the county area in the DLB from two categories of natural driving factors and social driving factors according to local conditions. The specific factors are shown in

Table 1. Preliminary selection of driving factors for the evolution of land spatial layout.

Level 1	Level 2	Number	Level 3
Natural driving factors	Climate	X1	Average annual precipitation
		X2	Average annual temperature
	Terrain	X3	Elevation
		X4	Slope
		X5	Aspect
	Soil	X6	Soil types
	Biology	X7	Biodiversity value
Social driving factors	Location of transportation	X8	Distance from railway
		X9	Distance from highway
		X10	Distance from administrative center
		X11	Distance from river
	Economic development	X12	GDP per capita
		X13	Multi-year average gross output value of the primary industry
		X14	Multi-year average gross output value of the secondary industry
	Social life	X15	Multi-year average population distribution
		X16	Multi-year average sales of social consumer goods
	Policy	X17	Multi-year average government investment in fixed assets
		X18	Multi-year average public expenditure

.

The explanatory power of each driving factor on the change in territorial spatial layout within the 5-year and 10-year steps is shown in

Table 2. The explanatory power of each driving factor on the change in land spatial layout for the 5-year research step.

Factor	Total Cramer’s V	Cramer’s V of Different Types of Land Space
		Forest Land	Garden	Grassland	Wetland	Water	Other Land	Urban	Farmland	Village
X1	0.0938	0.2092	0.0451	0.0936	0.0752	0.0957	0.0052	0.0374	0.1903	0.0228
X2	0	0	0	0	0	0	0	0	0	0
X3	0.1523	0.3023	0.0330	0.1953	0.0937	0.1629	0.0026	0.0990	0.2910	0.0818
X4	0	0	0	0	0	0	0	0	0	0
X5	0	0	0	0	0	0	0	0	0	0
X6	0.1452	0.0000	0.3044	0.0412	0.1369	0.1435	0.1963	0.0052	0.0800	0.2951
X7	0.1908	0.0000	0.3967	0.0239	0.0642	0.2967	0.2446	0.0134	0.1075	0.3646
X8	0.0681	0.1131	0.0104	0.0378	0.0520	0.0330	0.0065	0.1335	0.1066	0.0296
X9	0.0672	0.1271	0.0121	0.0378	0.0449	0.0290	0.0044	0.1136	0.1270	0.0307
X10	0.0992	0.1888	0.0127	0.0606	0.0549	0.0184	0.0060	0.1611	0.1954	0.0658
X11	0.0689	0.1391	0.0122	0.0405	0.0472	0.1106	0.0032	0.0627	0.1101	0.0429
X12	0.0820	0.0915	0.0210	0.0710	0.0691	0.0828	0.0044	0.1701	0.0710	0.0321
X13	0.1265	0.0000	0.2218	0.0411	0.1375	0.2002	0.1614	0.0083	0.0538	0.2143
X14	0.1079	0.0000	0.1704	0.0232	0.1202	0.1072	0.1313	0.0047	0.1760	0.1578
X15	0.1889	0.0000	0.1590	0.0124	0.0321	0.1637	0.5277	0.0008	0.1072	0.0216
X16	0.1169	0.0000	0.1943	0.0353	0.1406	0.0987	0.1317	0.0055	0.1780	0.1913
X17	0.1060	0.0000	0.1734	0.0355	0.1174	0.1215	0.1175	0.0067	0.1639	0.1557
X18	0.0850	0.1425	0.0438	0.1078	0.0786	0.0847	0.0058	0.0908	0.1425	0.0385

and

Table 3. The explanatory power of each driving factor on the change in land spatial layout for the 10-year research step.

Factor	Total Cramer’s V	Cramer’s V of Different Types of Land Space
		Forest Land	Garden	Grassland	Wetland	Water	Other Land	Urban	Farmland	Village
X1	0.0931	0.2101	0.0347	0.0935	0.0695	0.0991	0.0055	0.0373	0.1908	0.0227
X2	0	0	0	0	0	0	0	0	0	0
X3	0.1516	0.2990	0.0319	0.1953	0.0878	0.1663	0.0029	0.0924	0.2930	0.0822
X4	0	0	0	0	0	0	0	0	0	0
X5	0	0	0	0	0	0	0	0	0	0
X6	0.1404	0.0000	0.2986	0.0326	0.1521	0.1140	0.1631	0.0063	0.0755	0.3080
X7	0.1923	0.0000	0.3975	0.0204	0.0638	0.3090	0.2431	0.0138	0.0983	0.3670
X8	0.0665	0.1097	0.0109	0.0376	0.0571	0.0330	0.0065	0.1248	0.1092	0.0291
X9	0.0617	0.1236	0.0100	0.0375	0.0407	0.0287	0.0046	0.0860	0.1304	0.0306
X10	0.0991	0.1869	0.0086	0.0603	0.0405	0.0149	0.0069	0.1648	0.1978	0.0651
X11	0.0685	0.1400	0.0099	0.0402	0.0507	0.1075	0.0032	0.0601	0.1116	0.0429
X12	0.0814	0.0899	0.0187	0.0755	0.0732	0.0817	0.0049	0.1668	0.0724	0.0316
X13	0.1229	0.0000	0.2207	0.0392	0.1377	0.1672	0.1750	0.0086	0.0536	0.2142
X14	0.1069	0.0000	0.1670	0.0283	0.1203	0.0961	0.1399	0.0050	0.1703	0.1591
X15	0.1926	0.0000	0.1606	0.0102	0.0319	0.1248	0.5471	0.0008	0.1221	0.0218
X16	0.1176	0.0000	0.1920	0.0394	0.1405	0.0910	0.1376	0.0056	0.1818	0.1926
X17	0.1055	0.0000	0.1708	0.0388	0.1176	0.0967	0.1341	0.0075	0.1646	0.1568
X18	0.0847	0.1352	0.0488	0.1079	0.0788	0.0876	0.0060	0.0836	0.1431	0.0388

. Based on the principle of factor elimination, seven factors, i.e., X2, X4, X5, X8, X9, X11, and X18, were eliminated, and the overall explanatory power of the remaining driving factors was ranked as follows: 5-year step: X7 > X15 > X3 > X6 > X13 > X16 > X14 > X17 > X10 > X1 > X12; 10-year step: X15 > X7 > X3 > X6 > X13 > X16 > X14 > X17 > X10 > X1 > X12.

The average simulation accuracy of the territorial space layout change potential increased with the increase in the number of combination driving factors. When the number of combination driving factors reached a certain number, the simulation accuracy showed a downward trend. With a greater number of combination driving factors, the simulation accuracy was not necessarily higher [38]. Based on the combination of driving factors whose overall explanatory power is higher than 0.15, driving factors were added one by one according to the order of the explanatory power of the driving factors for the combination. From

Table 4. The 5-year step size and 10-year step size corresponding to driving factor combination explanatory power.

Number	Factor Combination	Accuracy Rate/%
		5-Year Step	10-Year Step
1	X7, X15, X3	51.74	56.84
2	X7, X15, X3, X6	52.39	58.50
3	X7, X15, X3, X6, X13	52.98	59.90
4	X7, X15, X3, X6, X13, X16	53.99	60.44
5	X7, X15, X3, X6, X13, X16, X14	54.18	61.31
6	X7, X15, X3, X6, X13, X16, X14, X17	54.51	61.62
7	X7, X15, X3, X6, X13, X16, X14, X17, X10	55.93	63.26
8	X7, X15, X3, X6, X13, X16, X14, X17, X10, X1	55.02	62.60
9	X7, X15, X3, X6, X13, X16, X14, X17, X10, X1, X12	54.43	62.46

, we can see the accuracy rate of the changes in territorial space types in the 5-year and 10-year steps. The results showed that the overall explanatory power of the best combination of factors for the 5-year study step was 55.93% less than that of the best combination of factors for the 10-year study step, which was 63.26%. The difference between the two was large, so the simulated study step for this study was determined to be 10 years. The best driving factor combination corresponding to the two research steps is “elevation, soil types, biodiversity value, distance from the administrative center, multi-year average gross output value of the primary industry and the secondary industry, multi-year average population distribution, multi-year average sales of social consumer goods, multi-year average government investment in fixed assets”. Therefore, it is determined that this driving factor combination is the best driving factor combination for the simulation and prediction model of territorial spatial layout in the study area.

3.2. Demand and Spatial Layout Simulation Prediction

We divided the spatial distribution system dynamics model of the DLB into five subsystems: population subsystem, economic subsystem, productivity subsystem, environment subsystem, and resource subsystem. Since the minimum administrative unit of socio-economic data statistics is at the county level, the validity and scientificity of such data cannot be guaranteed if the data are spatialized and cut into the DLB. Therefore, the variable values of the county area involved in the DLB were collected. Repeated debugging was performed to determine the sequential relationship between the interactions of the two variables of different subsystems and to accurately establish the causal feedback relationships of the main variables of the subsystems (Figure 4). The stability of the model was tested by comparing the simulation results of the main level variables with the statistical true values at different time steps, with the results showing that the simulation error of the main level variables was within 10%, the model accuracy was good, the simulated values at different simulation steps showed the same trend, and the model construction was stable. The demand for each land space type from 2021 to 2050 was projected year by year, and the projected demand for three key time points in China’s socialist modernization construction, i.e., 2025, 2035, and 2050, was extracted (

Table 5. Forecast demand for each type of territorial space from 2021 to 2050 (unit: 100 hm²).

Year	Forest Land	Garden	Grassland	Wetland	Water	Other Land	Urban	Farmland	Village
2025	212,931.34	5765.65	20,314.87	2703.35	8430.38	40.46	6803.85	93,983.82	2941.38
2035	211,119.30	5811.50	20,183.29	3181.69	8354.89	36.55	9011.66	93,050.66	3165.54
2050	210,623.61	5816.72	19,958.69	3650.97	8308.25	30.57	9971.10	91,894.22	3660.97

).

Consistent with the data range of the SD model, the best combination of driving factors was brought into the LEAS module of the model to obtain the development probability of each territorial space type. The spatial layout map of the DLB in 2020 was obtained by cropping after running the CARS module (Figure 5). The accuracy was verified by Kappa test series group, and the results showed that the standard Kappa coefficient (K_standard) = 0.9181, the random Kappa coefficient (K_no) = 0.9382, the location Kappa coefficient (K_location) = 0.9190, and the stratified location Kappa coefficient (K_{locationStrata}) = 0.9190, all of which are higher than 0.9, indicating that the simulated raster map of the study area in 2020 is in high agreement with the actual raster map in that year, the PLUS model simulation is accurate, and the simulation is better and more credible.

3.3. Future Simulation Results of Territorial Spatial Distribution in DLB

The Five-Year Plans are an important part of China’s national economic plan, and 2025 will mark the completion of the 14th Five-Year Plan for China’s national economic and social development. It is remarkable that 2035 will be the year in which China basically achieves socialist modernization, and the general tasks and development steps of China’s socialist modernization in the first half of the 21st century refer to China fully realizing socialist modernization by 2050. Therefore, this 3-year period is a critical time for China’s socialist modernization. Based on the SD-PLUS model, the spatial layout of the DLB for the three nodes was predicted (Figure 6).

3.3.1. Spatial Pattern Prediction Model

From the statistical data (

Table 6. Simulation and statistical results of territorial spatial distribution in the DLB.

Type	2025		2035		2050
	Area/100 hm²	Proportion/%	Area/100 hm²	Proportion/%	Area/100 hm²	Proportion/%
Forest land	153,836.40	58.74	152,676.13	58.30	152,453.65	58.22
Garden	4758.25	1.82	4800.83	1.83	4805.24	1.83
Grassland	13,380.28	5.11	13,252.16	5.06	13,899.23	5.31
Wetland	2300.30	0.88	2412.72	0.92	2307.48	0.88
Water	6588.29	2.52	6521.17	2.49	6480.49	2.47
Other land	28.92	0.01	26.48	0.01	22.45	0.01
Urban	5671.40	2.17	7484.93	2.86	8215.30	3.14
Farmland	73,230.11	27.96	72,460.51	27.67	71,572.03	27.33
Village	2085.74	0.80	2244.73	0.86	2123.80	0.81

), it can be seen that the spatial layout structure of the DLB from 2025 to 2050 is basically maintained in a stable state, with the largest area of forest land still accounting for more than 58%, followed by farmland, which slowly and gradually decreases in area but still makes up more than 27% in total. The garden area expands from 2025 to 2035 but basically remains stable after 2035. The area of grassland decreases by 128.12 km² during the 2025–2035 period but increases by 647.06 km² from 2035–2050. The wetlands region increases and then decreases, and the water area decreases, but the rate of change slows down. Other lands gradually decrease, and the village area increases significantly during the 2025–2035 period but decreases during the 2035–2050 period.

3.3.2. The Evolution of Type Structure

By combining the simulation data with the real area data of various territorial spatial types in the DLB during the 2000–2020 period, the spatial layout structure change map of the DLB during the 2000–2050 period was drawn (

Table A1. The statistical data of the territorial space-type area of the DLB from 2000 to 2020 and the forecast data from 2020 to 2050 (unit: 100 hm²).

Type	2000	2005	2010	2015	2020	2025	2035	2050
Forest land	158,317.95	158,423.49	157,820.07	156,312.75	155,485.66	153,836.40	152,676.13	152,453.65
Garden	1782.82	1908.96	2845.90	4090.51	4399.79	4758.25	4800.83	4805.24
Grassland	14,694.15	14,255.99	13,590.29	13,521.01	13,397.64	13,380.28	13,252.16	13,899.23
Wetland	1841.30	1781.18	1890.30	2168.45	2177.68	2300.30	2412.72	2307.48
Water	6381.81	6591.63	6669.53	6422.07	6584.18	6588.29	6521.17	6480.49
Other land	23.92	22.20	35.71	35.32	30.76	28.92	26.48	22.45
Urban	1296.56	1606.13	2768.67	3551.85	4509.17	5671.40	7484.93	8215.30
Farmland	75,560.27	75,211.00	74,274.32	73,798.85	73,250.96	73,230.11	72,460.51	71,572.03
Village	1981.71	2079.91	1985.69	1979.63	2043.82	2085.74	2244.73	2123.80

, Figure 7) to master the evolution dynamics and trends of the spatial layout of the DLB in the first half of the 21st century. According to the development of natural expansion, the development trend of each type of territorial space is as follows: The area of forest land changes greatly, decreasing rapidly from 2005 to 2025, with the rate of change slowing after 2025 and gradually stabilizing after 2035. The garden area first increases slowly from 2000 to 2005, then increases rapidly from 2005 to 2015, slowing down from 2015 to 2025. After 2025, it is basically stable and only fluctuates in a small area. The grassland region decreases from 2000 to 2010 and basically remained stable after 2015. The area of wetland decreases from 2000 to 2005 and then increases when it fluctuates. The region of water gradually expands from 2000 to 2010, rapidly decreases in 2015, recovers in 2020, remains stable in 2025, and slowly decreases in 2050. The total area of other lands accounted for a very small proportion and changed greatly from 2000 to 2010. After 2010, with the improvement in land surface modification technology, the other lands area gradually decreases. The growth rate in the urban area was relatively slow from 2000 to 2005 and relatively high from 2005 to 2035. After 2035, the expansion rate reduces. The farmland region decreases gradually over the past 50 years, but the rate of change slows down. The village area expands from 2000 to 2005, falls back in 2010, gradually increases from 2010 to 2035, and gradually decreases from 2035 to 2050. In the numerical simulation results of the SD model for the future, the grassland area gradually decreases, and the wetland and village regions gradually increase, while in the spatial simulation results, the reverse changes will occur within a certain range in 2050 due to the spatial conditions that cannot reach the preset target and constitute a local feedback to the total. Therefore, the simulation results are scientifically credible.

3.3.3. The Evolution of Three Types of Space

The spatial layout change map of the DLB in the three space area types was drawn from 2000 to 2050 according to the land area statistics of the three types of space (

Table A2. The statistical data of the three types of space area of the DLB from 2000 to 2020 and the forecast data from 2020 to 2050 (unit: 100 hm²).

Year	Ecological Space	Urban Space	Agricultural Space
2000	183,041.94	1296.56	77,541.98
2005	182,983.45	1606.13	77,290.91
2010	182,851.80	2768.67	76,260.01
2015	182,550.11	3551.85	75,778.47
2020	182,075.72	4509.17	75,294.78
2025	180,892.43	5671.40	75,315.84
2035	179,689.49	7484.93	74,705.24
2050	179,968.54	8215.30	73,695.83

, Figure 8). As can be seen from the figure, the ecological space area decreases slowly from 2000 to 2020, accelerates from 2020 to 2025, slows down from 2025 to 2035, and basically maintains a stable state after 2035. As the urban space occupies less than 4% of the total area, its area change is clear in the figure and is consistent with the above urban change trend. The change in agricultural space area is small from 2000 to 2005, gradually decreases from 2005 to 2020, slightly increases in 2025, and gradually decreases again from 2025 to 2050.

4. Discussion

4.1. Contribution of the Best Combination of Driving Factors to the Spatial Change in Land Types in the DLB

In the LEAS module of the PLUS model, the contribution value of the best combination of driving factors to the spatial change in each territorial spatial type in the DLB can be obtained, and the action mode affecting the spatial change in a certain territorial spatial type can be further analyzed (Figure 9). For the forestland area, elevation is always the driving factor with the greatest influence on regional forestland expansion, which is consistent with the common knowledge that forestland is generally distributed in mountainous areas. The change in forestland is mainly concentrated in the lower-elevation area of the DLB and is related to human activities. The population density is highest in the urban space. The concentrated distribution area of forestland is negatively correlated with the population distribution. The change in forestland mainly occurs in the area with a very small population. Compared with other driving factors, soil types are less likely to change over a short period of time and belong to the state variables. The forestland region change occurs in all soil types, and there is no obvious tendency. The variations in the garden area are mainly distributed in regions with low values of multi-year average fixed-asset investment and total secondary industry output, medium-value regions of biodiversity, and low-value regions of population. The values of multi-year average fixed-asset investment and secondary industry gross output are mostly dependent on construction land, so their values are negatively correlated with the growth in forest land and positively correlated with the decrease in forest land. The biodiversity of the garden region is in the middle among all the land spatial types, so it is mainly affected by the regional range of the median biodiversity value. As a matter of fact, the garden area is mainly located in the lower-elevation areas around the towns and cities, where the population distribution is low, so it is greatly affected by the areas with low values of multi-year population distribution. The grassland region is mainly distributed in the areas with higher elevation, and the distribution of grassland changes is more dispersed with more distribution in the mountainous areas in the western part of the DLB. Similarly, the population distribution in the area of change is less, and it is greatly affected by the low-value area of population distribution. The wetland change area is mainly around a large lake water system, and urban space is usually constructed around water systems, so it is greatly affected by population distribution. Meanwhile, the water and wetland areas are formed by natural precipitation gathered and deposited in low-elevation areas through surface runoff, so they are more influenced by elevation. In addition, the complex and diverse habitats in water areas provide optimum living conditions and breeding spaces for all types of organisms, and the value of biodiversity is high. The spatial changes in other lands have no obvious pattern and are more dispersed in space. Population growth is the most fundamental driving factor for the continuous expansion of construction land area, and the continuous urbanization process eventually leads to the continuous expansion of residential land area. The development of urban areas tends to give little consideration to the soil type of the subsoil and more consideration to the soil environment of the construction land. Therefore, the contribution of soil type to the change in urban area is minimal. The biodiversity value of farmland is in the middle position among all the land spatial types and is greatly affected by the median area of biodiversity value. The change in farmland area mainly occurs in the lower-elevation area of the DLB and is mainly influenced by human activity. Population growth and mobility in rural areas are the most important factors affecting the change in the village region. Meanwhile, the permanent basic farmland control line at all levels also restricts the expansion of the village region, and the number and distance of government sites greatly affect the direction and rate of change in the urban and village regions.

4.2. Accuracy Check Based on FoM

We introduced the FoM metric in the parameter-set setting of the PLUS model. The FoM only depends on the number of change cells in the simulation process, which can better identify the simulation accuracy. The accuracy value of the FoM is generally small. The FoM is a ratio: the numerator is the intersection of the simulated change and the original change, and the denominator is the union of the simulated change and the reference change [46,47]. Pontius et al. [46] used the unified model to simulate many cities with different sizes in the world and used the FoM index to verify the accuracy. The results showed that the simulation accuracy of each city was different, with some higher than 50% and some lower than 5%. Therefore, as it could not directly show a small FoM value, the simulation accuracy must be poor. Since there is no effective judgment criterion, in other studies, FoM values are mostly used to compare the results of simulations using different models for the same study area, and the FoM values are compared horizontally to determine the most suitable simulation model. In this study, it was not directly used as the final accuracy verification index, but the optimal parameter combination was determined through the simulation results, which also had a good effect. Furthermore, we tried to interpret the optimal FoM value (0.0482) of the simulated results of the PLUS model for the DLB. We consulted Professor Liang, the main developer of the PLUS model, who pointed out that if the net change in land use in the study area is not large, the FoM must not be high and suggested an evaluation of the FoM value of each sub-basin in the large area. We further obtained the FoM of four scales: large basin, sub-basin, municipal, and county district (

Table 7. Comparison of FoM values at different scales.

Scale	Name	Area/ hm²	FOM
Large basin	DLB	2.6 × 107	0.0482
Sub-basin	Xiangjiang River Basin	9.4 × 106	0.0475
	Zishui River Basin	2.6 × 106	0.0538
	Yuanjiang River Basin	9.0 × 106	0.0504
	Lishui River Basin	1.8 × 106	0.0499
	Dongting Lake area	3.4 × 106	0.0478
City	Changsha City	1.1 × 106	0.0758
District	Tianxin District	1.4 × 104	0.2046

). The results showed that there was an increasing trend in FoM values as the study area was reduced, the difference in FoM values between the large basin and sub-basin scales was small, and the FoM values increased significantly when the scales were reduced to municipal and county districts. However, how to verify the accuracy based on the individual FoM values still needs further in-depth study.

4.3. Research on the Spatial Layout of Basin

In China, territorial spatial planning is a planning system that integrates and unifies spatial planning, such as main functional area planning, land-use planning, and urban and rural planning, which is carried out with administrative units as the boundary. In the planning process, it is necessary to evaluate the regional resources and the environmental carrying capacity, which refers to sustainable development and support, and the guarantee capacity of resources and the environment to maintain the natural environment along with social and economic activities under certain social, economic, resource, ecological, and environmental conditions in a certain period and spatial area. The key to the evaluation of the carrying capacity of resources and the environment is to study the “weak board elements” of regional territorial development. The constraints of water resources, land resources, climate, ecology, environment, and disasters on urban development and cultivated land use provide the upper limit quantity constraint and the spatial boundary constraint for the optimization of land-use structure in the optimization of territorial spatial pattern. China carries out its territorial spatial planning within the scope of administrative boundaries in consideration of its convenience in terms of management and regulation. However, it is worth noting that water resources, climate, ecology, environment, and disasters are all integrated systems rather than independent factors with administrative boundaries. Therefore, from the perspective of hydrology, conducting research on territorial spatial layout within the scope of river basins is another approach that China can take. The driving factors, study steps, and parameters of the SD-PLUS model that influence the evolution of the territorial spatial layout of basins with regard to different scales, regions, and developmental contexts need to be further adjusted.

The territorial spatial structure of the lake-type basin is mainly characterized by a circular structure, which is the main difference in spatial structure from the river-type basin. Elements such as water system, water quantity, water quality, aquatic life, topography, precipitation, and temperature at the regional scale generally exhibit the circling characteristics of the lake-type basin. The circular structure of the lake-type basin from the outside to the inside is mountain area, hilly region, piedmont area, lacustrine area, lakeshore, and the lake [48]. As the urban space with the greatest evolutionary initiative of the three types of space, its expansion direction and extent influence the spatial layout of the entire territory. The urban groups within the lake-type basin are influenced by the interaction of other urban groups in the same and adjacent circles, which also influence the evolution of other territorial space types. Therefore, the complexity of the territorial spatial layout is higher in lake-type basins than in river-type basins in general.

5. Conclusions

This study focuses on the best combination of drivers and the best study step size that affect regional territorial spatial layout and uses them as the basis for the SD-PLUS quantitative coupling model. Taking the DLB as an example, its territorial spatial layout was predicted for three key socialist modernization years with Chinese characteristics—2025, 2035, and 2050. By determining the Cramer’s V value of each driving factor under different research steps in the evolution of territorial spatial layout, the best driver combination was constructed, and the accuracy of different driver combinations on the simulation was compared based on the MLP-ANN model. The optimal driving factor combination for the research area was determined to be “elevation, soil types, biodiversity value, distance from the administrative center, annual gross output value of the primary industry, annual gross output value of the secondary industry, annual population distribution, annual sales of consumer goods, annual fixed-asset investment”, and the best research step was 10 years. This operation not only improves the accuracy of simulation but also improves the operational efficiency of the model. Compared with the accuracy (0.9074) of land-use simulation of DLB by others through the Markov-PLUS model [39], the improved SD-PLUS model used in this study has better simulation accuracy (0.9181). Therefore, the improved SD-PLUS model is both scientific and efficient. From 2020 to 2050, the spatial distribution structure of the DLB will basically maintain a stable state. From the perspective of the three types of space, the area of ecological space decreases slowly in the first 20 years, accelerates in the 2020–2025 period, slows down in the 2025–2035 period, and basically maintains a stable state after 2035. Urban space expands rapidly until 2035 and then slows down. The change in agricultural space area is small from 2000 to 2005, gradually decreases from 2005 to 2020, slightly increases in 2025, and gradually decreases from 2025 to 2050. These changing trends are closely related to China’s new urbanization construction, and the spatial layout of the country will reach a relatively stable state around 2035. Constructing territorial spatial layout for high-quality development is an important support for promoting sustainable regional development. Different from the study of territorial spatial layout with administrative divisions as the boundary, this study makes a scientific prediction of the future territorial spatial layout at the lake-type basin scale based on the natural expansion scenario, which is the basis for carrying out the simulation study of the territorial spatial layout under different constraint scenarios, and suggests the next step of constraint scenarios by simultaneously adjusting the variable equations of the SD model and the transition matrix of the PLUS model.

It is worth noting that in the reality of China’s territorial spatial planning practice, the territorial spatial layout does not present a natural expansion. At the most basic level, three control lines are required to constrain the territorial spatial layout, such as the urban development boundary, the permanent basic agricultural land boundary, and the ecological protection red line. As of now, these three control lines are still being defined by the relevant departments at the provincial, municipal, and county levels, and data are difficult to obtain. If relevant data are obtained in the future, they can be loaded into the PLUS model as a constraint factor for various types of territorial space for calculation. With regard to the scale of the study, the DLB selected for this study is a large lake-centric basin type and can only represent the spatial layout evolution of the lake-type basin. In the future, we can further explore different scales of basins, such as small basins with mainly tertiary rivers, the world’s largest river basins, and multi-basin junction areas. By comparing the evolution mechanism and development trend of the territorial spatial layout between different scales, a more complete system of research on the territorial spatial layout in basins can be formed.