Simulation and Prediction of the East Dongting Lake Wetland Landscape Based on the PLUS Model

Abstract

The East Dongting Lake Wetland, an internationally vital reserve, faces growing ecological threats, necessitating enhanced predictive research on its landscape dynamics. Using the PLUS model and Markov chain method, this study analyzes landscape changes (2010–2022) and simulates 2030 patterns under two scenarios. The key findings reveal the following: (1) poplar plantations plummeted from 28.65% to 2.79% due to restoration policies (e.g., tree removal), while grasslands surged from 21.43% to 59.64%; mudflats and water bodies fluctuated naturally. (2) Natural drivers dominated changes—precipitation and elevation influenced water bodies and grasslands the most, whereas road proximity primarily affected poplar plantations. (3) The PLUS model proved effective for small-scale wetland predictions. (4) Simulations showed divergent 2030 outcomes: under natural development, poplar plantations would rebound to 57.86 km², whereas ecological regulation—restricting plantations and expanding grasslands to 882.70 km²—better supported biodiversity. This study underscores policy-driven restoration success and the PLUS model’s utility for local-scale simulations, offering actionable insights for Dongting Lake’s management and a methodological framework for wetland conservation.

1. Introduction

Wetlands, among the most productive ecosystems on Earth, play a crucial role in regulating regional hydrology, preserving biodiversity, sequestering carbon, mitigating emissions, and improving water quality [1]. However, in recent decades, wetlands have suffered significant loss and degradation, resulting in ecological and environmental challenges at both regional and global scales. These changes have significantly affected the sustainable development of regional ecosystems, economies, and societies, positioning the simulation of wetland landscape change trends as a key research focus in resource and environmental science [2,3].

The degradation of wetlands is often driven by a combination of natural factors (e.g., climate change, hydrological alterations) and anthropogenic activities (e.g., urbanization, agricultural expansion, and policy interventions). Understanding these dynamics requires robust modeling approaches that can capture both gradual changes and abrupt shifts caused by policy implementations or extreme events.

Research on wetland landscape change currently spans multiple spatial scales, ranging from national and provincial to municipal and local levels. Key research areas include the spatiotemporal characteristics of wetland changes, driving factors, landscape pattern evolution, and future trend simulations. However, the complexity of factors influencing wetland systems, coupled with the stochastic nature of human activities and policy interventions, makes predicting wetland change trends a frontier research challenge. Traditional models often assume linear evolution of spatial patterns based on historical trajectories, which may fail to account for nonlinear or abrupt changes, such as those driven by sudden policy shifts (e.g., ecological redline enforcement or large-scale afforestation programs). Commonly used models for predicting wetland landscape change include the Markov model, system dynamics (SD) model, cellular automata (CA) and its enhanced versions, the conversion of land use and its effects (CLUE) model, and the future land use simulation (FLUS) model [4,5]. However, previous studies have shown that these models are limited in analyzing the driving factors of change and face challenges in balancing numerical prediction accuracy with spatial pattern forecasting capabilities [6]. For instance, while Markov models excel in quantifying transition probabilities, they often lack spatial explicitness, whereas CA models may struggle with capturing the complex interactions between multiple driving factors.

The Patch-Generating Land Use Simulation (PLUS) model, developed in [7], is a CA-based model utilizing raster data. It is widely employed to investigate the driving mechanisms of multi-class land expansion and to predict land use landscape evolution at the patch level. With its high predictive accuracy and capacity to effectively identify the driving factors of land use changes [6], the PLUS model has gained increasing attention in wetland landscape prediction in recent years. The model’s integration of a land expansion analysis strategy (LEAS) and a multi-class random patch seed mechanism (CARS) allows for a more nuanced understanding of both gradual and abrupt changes, making it particularly suitable for regions subject to strong policy interventions or natural disturbances. However, its applications have primarily focused on medium- and large-scale predictions, as demonstrated in large watershed simulations [6] with relatively few studies addressing its use in local-scale wetland landscape prediction such as that of specific lake ecosystems [7].

East Dongting Lake is a national nature reserve for wetland ecosystems and rare bird species, and it is one of the 21 internationally important wetland nature reserves designated by the Chinese government and listed under the Ramsar Convention on Wetlands. However, studies on predicting and simulating wetland change trends in this region remain limited [8,9]. The region has experienced dramatic landscape transformations, such as the plantation land due to the “Three-Year Action Plan for Ecological Environment Special Rectification of Dongting Lake (2018–2020),” highlighting the need for models that can incorporate such policy-driven changes. This study examines the East Dongting Lake wetland using the PLUS model, integrated with the Markov model’s quantitative predictive capabilities, to analyze wetland change processes and identify key driving factors and their contributions. The study subsequently simulates wetland change trends in 2030, offering scientific support for the protection, planning, and management of the East Dongting Lake wetland while providing a methodological reference for future wetland prediction studies. By incorporating dynamic transition rules and scenario-based approaches, this research addresses the limitations of linear assumptions and provides a framework for predicting both gradual and abrupt changes in wetland landscapes.

In the context of scenario setting, the study acknowledges the limitations of assuming linear evolution of spatial patterns. For example, the sharp decline in poplar plantations land from 2019 to 2022 was driven by policy interventions (e.g., poplar tree felling), which cannot be captured by linear projections alone. To address this, the study defines two scenarios—natural development and ecological regulation—explicitly accounting for policy impacts and nonlinear dynamics. This approach enhances the model’s ability to predict future changes under varying governance and environmental conditions.

The findings underscore the importance of integrating policy-driven changes into predictive models, as purely linear or historical trajectory-based assumptions may lead to significant inaccuracies. Future research could further refine the PLUS model by coupling it with dynamic evaluation frameworks or machine learning algorithms to better capture the complex, nonlinear interactions between natural and anthropogenic drivers.

2. Materials and Methods

2.1. Study Area

The East Dongting Lake Wetland is situated in the eastern region of Dongting Lake, the second-largest freshwater lake in China, and is administratively governed by Yueyang City. Geographically, the wetland lies between 28°59′–29°38′ N latitude and 112°43′–113°15′ E longitude (Figure 1). The region is distinguished by broad river channels characteristic of major lake systems, surrounded by extensive marshlands. This area is abundant in water, biological, and land resources, making it one of the most biodiverse regions in China. In recent decades, the combined effects of natural factors and human activities have led to substantial transformations in the wetland, posing a potential threat to regional biodiversity.

2.2. Methods, Data Sources and Preprocessing

This study investigated wetland landscape changes in East Dongting Lake using integrated geospatial data analysis. The research combines remote sensing imagery with environmental and socioeconomic datasets to enable comprehensive modeling of landscape dynamics.

The evolution of wetland landscapes in the study area is shaped by a combination of diverse driving factors. Considering the unique natural and social environment of the study area, along with relevant research findings, nine driving factors were selected for the analysis: natural factors including DEM, slope, temperature, and precipitation; socio-economic factors such as population density and GDP; and spatial accessibility factors including main roads, railways, and rivers.

The remote sensing and environmental data used in this study are summarized in

Table 1. Types and sources of research data.

Type	Data	Source
Vector Data	Vector data of the study area	Chinese Academy of Sciences Resource and Environmental Science Data Center
Remote Sensing Data	2010 Landsat 7 ETM+	AI Earth (Earth Science Cloud Platform)
	2013 Landsat 7 ETM+
	2016 Landsat 7 ETM+
	2019 Sentinel-2 MSI
	2022 Sentinel-2 MSI
Natural Environmental Data	Elevation	Resource and Environmental Science Data Center
	Slope	Extracted from DEM data
	Temperature	Chinese Academy of Sciences Resource and Environmental Science Data Center
	Precipitation	Chinese Academy of Sciences Resource and Environmental Science Data Center
	Rivers	Acquired from Open Street Map (Open Street Map Foundation, London, UK)
Socioeconomic Data	Population Density	Resource and Environmental Science Data Center
	Per Capita GDP
Transportation Network Data	Roads	Acquired from Open Street Map
	Railways

. The analysis was based on Landsat 7 imagery (acquired on 17 September 2010, 8 August 2013, and 16 August 2016) and Sentinel-2 imagery (acquired on 16 August 2019, and 20 August 2022); the Random Forest classification method was applied to this imagery to analyze the distribution characteristics of wetland landscape types in East Dongting Lake from 2010 to 2022. Initially, East Dongting Lake was selected as the study area, and these images from 2010 to 2022 were gathered as source data for analysis. Following preprocessing of the remote sensing imagery, all nine driving factors (DEM, slope, temperature, precipitation, population density, GDP, and distances to roads, railways, and rivers) were standardized to a 0–1 range using min–max normalization to ensure comparability across variables in the PLUS model analysis.

Wetland landscape types in East Dongting Lake from 2010 to 2022 were classified into four categories: poplar plantations, grassland, mudflats, and water bodies. A Random Forest classifier was developed using sample data and classification features for interpretation, generating wetland landscape distribution maps for East Dongting Lake across various years and periods from 2010 to 2022.

The selection of driving factors was based on the unique natural-social system characteristics of the East Dongting Lake wetland [7]. Natural factors (e.g., DEM, precipitation) primarily govern hydrological processes, while human activity factors (e.g., road distance) reflect spatial heterogeneity in policy interventions (e.g., the 2018–2020 poplar tree removal campaign). All factors were normalized to a 0–1 range using min–max standardization to eliminate unit differences. The chosen factors account for the following:

• Natural processes: DEM and slope influence flooding frequency [10], while temperature and precipitation directly regulate wetland vegetation growth.
• Human impacts: Road and railway proximity captures accessibility for plantation management, and GDP/population density reflects regional development pressure [3].
• Data reliability: All datasets were sourced from open-access repositories (e.g., Resource and Environmental Science Data Center) to ensure reproducibility.

Slope data were extracted from DEM data using the Slope tool in ArcGIS 10.8 software (developed by Environmental Systems Research Institute, Inc. (ESRI), Redlands, CA, USA). The Euclidean Distance tool was applied to vector data of main roads, railways, and rivers to extract the accessibility factor raster data. All data were clipped, resampled, and projected to ensure consistency in boundaries, projection, and spatial resolution across the datasets.

2.3. Methods

2.3.1. PLUS Model

The PLUS model integrates a novel land expansion analysis strategy (LEAS) with a multi-class, random patch seed-based cellular automata (CARS) model [7]. In this model, LEAS employs the Random Poplar Plantations classification algorithm to examine the relationship between land class expansion and various driving factors, determining the development probability of each land class and the contribution of these factors to their expansion over a specified period. The CARS model, in contrast, applies a threshold-decreasing, multi-type random patch seed mechanism to simulate the evolution of patches for different land types. The simulation parameters include a transition matrix, neighborhood weights, and land use demand.

The transition matrix is primarily derived from historical land use data and expert knowledge. The neighborhood weight parameter represents the expansion intensity of different land types, reflecting the expansion capacity of each land class under spatial driving factors. Land use demand is predicted using either the Markov chain or linear regression model within the PLUS framework.

Policy interventions (e.g., poplar tree removal mandates under the ‘Three-Year Action Plan’) were incorporated through the following: (1) transition constraints: prohibiting conversion to poplar plantations in ecological regulation scenarios; (2) neighborhood weights: setting low expansion capacity for poplar plantations (0.053) to reflect suppressed regrowth; (3) binary masks: using 2022 actual poplar distribution as a spatial constraint to limit simulated expansion.

To ensure the replicability and computational efficiency of the model implementation, this study deployed the PLUS model using Python (v3.9) (Python Software Foundation, Wilmington, DE, USA), leveraging its robust ecosystem of geospatial analysis and machine learning libraries such as GDAL (v3.6) for raster data processing, NumPy (v1.24) for numerical computations, and scikit-learn (v1.2) for algorithmic support. The source code of the PLUS model is publicly accessible on GitHub (https://github.com/), enabling researchers to reproduce the analytical workflow, verify results, and adapt the model to specific research contexts. All simulations were conducted in a high-performance computing environment to handle large-volume geospatial datasets and complex multi-class land use simulations efficiently. Visual Studio Code (v1.78) was utilized as the primary integrated development environment (IDE), facilitating seamless debugging, code versioning, and integration of diverse spatial datasets throughout the modeling process.

This study applies the model to predict the trends of patch-level changes across different wetland types, providing both theoretical and practical insights.

2.3.2. Accuracy Validation Methods

The accuracy of the PLUS simulation results was evaluated using the Kappa coefficient and the FoM coefficient. The FoM coefficient is used to quantitatively evaluate simulation accuracy at the cellular scale. A higher value indicates greater simulation accuracy, with typical values ranging from 0.01 to 0.25 [6]. The FoM coefficient is presented in Formula (1):

$$\mathrm{FoM}={\displaystyle \frac{\mathrm{B}}{\mathrm{A}+\mathrm{B}+\mathrm{C}+\mathrm{D}}}$$

(1)

In the formula, A denotes the area of error where land use has changed, but was predicted to remain unchanged; B represents the area where the prediction is accurate; C indicates the area of error where the prediction was incorrect; and D refers to the area of error where land use has remained unchanged, but the prediction shows a change. Limiting large-scale human activities, including planned regulatory measures (e.g., the felling of poplar trees), allows wetland landscape types to primarily evolve under the influence of natural factors.

2.3.3. Scenario Setting

To investigate future trends in wetland landscape changes under different development objectives, two scenarios—natural development and ecological regulation—are defined to predict the spatial patterns of wetland landscape types in the study area by 2030. These scenarios are based on the characteristics of the East Dongting Lake wetland as a national nature reserve and integrate relevant research findings. In the natural development scenario, the current status and driving factors of 2022 serve as the baseline. It is assumed that the spatial pattern of the wetland landscape in the study area will evolve linearly, following its historical trajectory. This scenario envisions the evolution of wetland landscape types primarily driven by natural influences, with changes occurring due to natural succession [3]. In the ecological regulation scenario, the impact of poplar plantations on biodiversity is considered, including the continued deforestation of poplar trees and restrictions on converting existing grasslands into poplar plantations. This scenario predicts a multi-factor development of wetland landscape types, driven by both natural factors and human activities.

3. Results

3.1. Landscape Structure Analysis of Wetland Types in East Dongting Lake

Understanding the direction of landscape pattern evolution and its driving factors is essential for optimizing regional landscapes and ensuring the coordinated development of ecology and socioeconomics. To examine the spatial distribution of wetland landscape types in East Dongting Lake, a supervised classification method using Random Poplar Plantations was applied to classify data from five time periods between 2010 and 2022, with data collected every three years (Figure 2).

Based on this, the distribution characteristics and evolutionary patterns of wetland landscape types in East Dongting Lake over this period were analyzed.

The wetland landscape of East Dongting Lake exhibited remarkable spatiotemporal variations (Figure 3,

Table 2. Area and proportion of wetland landscape types in East Dongting Lake from 2010 to 2022.

Year		Poplar Plantations	Grassland	Mudflat	Water
2010	Area (km2)	405.62	303.42	416.59	289.07
	Proportion (%)	28.65	21.43	29.42	20.50
2013	Area (km2)	406.25	411.71	273.75	322.99
	Proportion (%)	28.66	29.04	19.31	22.84
2016	Area (km2)	425.13	420.84	244.60	324.13
	Proportion (%)	30.06	29.77	17.30	22.87
2019	Area (km2)	387.92	451.16	389.99	173.14
	Proportion (%)	26.27	33.77	27.58	12.38
2022	Area (km2)	39.49	843.33	216.37	313.01
	Proportion (%)	2.79	59.64	15.30	22.27

). The spatial patterns presented here are derived from a robust classification process. The classification accuracy for all generated maps was assessed using a confusion matrix, yielding an overall accuracy of over 85%, which is sufficient for the study’s objectives. Poplar plantations showed a continuous decline from 405.62 km² (28.7%) in 2010 to 39.49 km² (2.8%) in 2022, with an abrupt 89.8% reduction during 2019–2022 due to systematic removal under the “Three-Year Action Plan”. This policy-driven change triggered grassland expansion from 303.42 km² (21.4%) to 843.33 km² (59.6%), effectively restoring habitats for wetland biodiversity. While mudflats decreased by 48.1% (416.59 to 216.37 km²) through sediment-vegetation feedbacks, water bodies exhibited hydrologic resilience—maintaining 20–23% coverage in normal years but dropping to 12.4% during the 2019 drought, demonstrating climate-mediated dynamics rather than absolute stability [3].

3.2. Wetland Change Trend Prediction in the Study Area Based on the PLUS Model

The LEAS module of the PLUS model software was utilized for wetland type expansion analysis, while the CARS module was applied to predict wetland change trends in the study area.

3.2.1. Wetland Landscape Type Expansion Analysis Strategy Module (LEAS)

The wetland landscape classification data for East Dongting Lake from 2010 and 2022 were overlaid with land use type data for these two periods to identify areas where land types have changed. These changes were subsequently input into the LEAS module to generate land development probability maps for each wetland type from 2010 to 2022, and to assess the contribution of driving factors to the expansion of each wetland type (Figure 4 and Figure 5). The LEAS module’s random forest algorithm effectively captured non-linear relationships between driving factors and wetland transitions, while minimizing overfitting through ensemble learning—a key advantage over traditional linear regression approaches.

The results indicate significant spatial heterogeneity in the expansion probabilities of different wetland types (Figure 4). Grassland expansion probability (0.149) dominates the transition matrix, primarily reflecting policy mandates (Hunan Forestry Department, 2018). This high transition probability aligns with the “Three-Year Action Plan” implementation period (2018–2020), demonstrating the model’s capacity to integrate abrupt policy-driven changes into landscape projections. The PLUS model’s innovation lies in revealing how elevation (β = 0.114) and river proximity (β = 0.119) spatially constrained the implementation efficiency of poplar removal (Figure 5b), with 72% of new grasslands occurring within 1 km of waterways. This spatial pattern suggests that hydrological connectivity significantly facilitated grassland restoration efforts, as areas near waterways were more accessible for mechanical poplar removal and natural revegetation.

Table 3. RMSE of wetland type expansion probability from 2000 to 2022.

Wetland Type	2010–2013	2013–2016	2016–2019	2019–2022	2010–2022
Poplar Plantations	0.133	0.126	0.137	0.138	0.053
Grassland	0.145	0.135	0.149	0.149	0.128
Mudflat	0.100	0.102	0.141	0.137	0.092
Water	0.105	0.096	0.034	0.067	0.098

shows that the analysis of driving forces behind the expansion of each wetland type across different time periods results in a root mean square error (RMSE) of less than 0.15 for all cases. The consistently low RMSE values (range: 0.053–0.149) validate the model’s stability across different wetland types and time intervals, with particularly high accuracy for grassland transitions (RMSE = 0.128). This suggests that the analysis is highly accurate and reliable, thereby making the results credible and providing strong data support for further wetland change analysis. The robustness of these findings is further reinforced by the model’s Kappa coefficient of 0.73, demonstrating substantial agreement in land change modeling according to standard validation criteria [7].

The contributions of driving factors to the expansion of grassland for each period from 2010 to 2022 are summarized in Figure 6. Based on the analysis of Figure 5 and

Table 4. Contribution of factors to water expansion for each time period from 2010 to 2022.

Factor	2010–2013	2013–2016	2016–2019	2019–2022	2010–2022
DEM	0.115	0.111	0.116	0.114	0.114
GDP	0.047	0.046	0.054	0.051	0.045
Road Distance	0.103	0.093	0.101	0.105	0.102
Population	0.056	0.054	0.056	0.048	0.055
Precipitation	0.132	0.131	0.128	0.135	0.132
Railway Distance	0.037	0.042	0.036	0.038	0.039
River Distance	0.117	0.119	0.119	0.122	0.119
Slope	0.107	0.106	0.102	0.106	0.105
Temperature	0.124	0.107	0.112	0.115	0.111

, it can be concluded that between 2010 and 2022, precipitation contributed the most to the expansion (or contraction) of water, followed by DEM. While road proximity was identified as the dominant spatial determinant of poplar plantation distribution patterns, the study period actually saw a dramatic net loss of poplar plantation cover. This apparent paradox can be explained by two key mechanisms: (1) the model’s driving factor analysis captures intrinsic landscape suitability—residual poplar plantation patches showed strong spatial association with road networks, reflecting historical plantation management practices; (2) these natural spatial patterns were subsequently overridden by systematic ecological restoration policies implemented during the study period. The top three contributing factors to grassland expansion were DEM, river distance, and precipitation (Figure 7). The two most influential factors in mudflat expansion were precipitation and temperature. Overall, the primary factors influencing the expansion or contraction of each wetland type were precipitation, DEM, river distance, temperature, and road distance. This suggests that natural factors primarily drive wetland changes in the study area, with human activities being mainly influenced by road distance.

The overlay analysis of multiple raster layers from the multi-period wetland landscape data obtained between 2010 and 2022 accurately revealed the expansion areas of each wetland type during this period.

Taking grassland expansion as an example, the contribution analysis of the influencing factors, as discussed above, shows that DEM is the most significant driving factor (Figure 7). DEM significantly influences the distribution and expansion of grasslands, as they tend to thrive in low-lying areas near water bodies and mudflats. These areas typically feature fertile soils and favorable moisture conditions, which are ideal for grassland growth and expansion.

Further overlay analysis of the grassland expansion area with the elevation data of the study area (Figure 8) reveals that grasslands are primarily concentrated in lower elevation, flat areas. During this period, grasslands expanded not only in low-elevation mudflat areas but also into forested areas. This expansion is primarily driven by large-scale deforestation, demonstrating that human regulatory activities can influence and even control wetland changes in the region.

3.2.2. CA Model Based on Multiple Random Patch Seeds (CARS)

The parameters of the CA model include wetland landscape type demand prediction, transition matrix, and neighborhood weight. The wetland landscape type demand parameters, based on the 2022 land use data, are divided into two scenarios: natural inertial development and ecological regulation. The demand quantities for each wetland type in 2030 are predicted using the Markov Chain method. Specific data are provided in

Table 5. Markov model-based land use demand prediction for East Dongting Lake in 2030 (unit: pixel count).

Scenario	Poplar Plantations	Grassland	Mudflat	Water
2022 Wetland Status	43,876	937,038	240,415	347,789
2030 Natural Development	57,856	929,508	239,528	342,226
2030 Ecological Regulation	8900	984,102	241,105	335,011

.

The transition matrix represents the change rules between different land use types, indicating whether transformations between types are allowed. Based on the constraints of the two prediction scenarios outlined in the previous methods, the corresponding transition cost matrices are established. The specific parameters are provided in

Table 6. Transition matrix parameter settings.

Wetland Landscape Type	Natural Development Scenario				Ecological Regulation Scenario
	Poplar Plantations	Grassland	Mudflat	Water	Poplar Plantations	Grassland	Mudflat	Water
Poplar Plantations	1	0	0	0	1	1	1	1
Grassland	0	1	1	1	0	1	1	1
Mudflat	0	1	1	1	0	1	1	1
Water	0	1	1	1	0	1	1	1

, where 1 indicates that the transition is permitted, and 0 indicates that it is not.

The neighborhood weight parameter represents the expansion intensity of a wetland type, indicating the ability of each wetland landscape type to expand under the influence of external driving factors. The threshold ranges from 0 to 1, with values closer to 1 indicating a stronger expansion capability for that type [11]. A higher value indicates that the type is less likely to convert to other land types and has a greater expansion ability; conversely, a lower value indicates that it is more likely to be converted into other land types and is more susceptible to being occupied by them. In this project, continuous adjustments and tests in the PLUS software led to the determination of domain weight values with higher simulation accuracy (

Table 7. Neighborhood weight parameter settings.

Wetland Landscape Type	Poplar Plantations	Grassland	Mudflat	Water
Neighborhood weight	0.053	0.662	0.077	0.208

).

Certain policies impose restrictions on changes to designated wetland types, such as open water areas and specific nature reserves. Based on the ecological protection policies of East Dongting Lake, water body areas were extracted from the 2022 land use data and converted into a binary constraint map with values of 0 or 1. In this map, a value of 0 indicates that conversion is prohibited, while a value of 1 indicates that conversion is permitted. Under this assumption, water bodies cannot convert into other wetland types, while poplar plantation land, grassland, and mudflats are eligible for conversion. Therefore, before performing the CARS operation, water bodies were separated from other areas within the study region and loaded into the PLUS model to generate a map of convertible areas (Figure 3).

3.2.3. Model Performance and Validation Results

A visual comparison between the simulated wetland distributions for 2022 and the actual remote sensing interpretation is presented in Figure 9, while Figure 10 shows the predicted changes for 2030. The accuracy validation results indicate a Kappa coefficient of 0.73 and an FoM coefficient of 0.19.The overall simulation accuracy is high, making it suitable for simulating future wetland type changes in the East Dongting Lake region.

Figure 9. Spatial distribution of wetlands. (a) PLUS model simulation results; (b) remote sensing interpretation results.

Figure 9. Spatial distribution of wetlands. (a) PLUS model simulation results; (b) remote sensing interpretation results.

Figure 10. Predicted wetland changes in 2030 under two scenarios: (a) natural development and (b) ecological regulation. Note: the dark green patches in (a) represent low-coverage secondary sprouting forests, the total area of which is accurately quantified in Table 8.

Figure 10. Predicted wetland changes in 2030 under two scenarios: (a) natural development and (b) ecological regulation. Note: the dark green patches in (a) represent low-coverage secondary sprouting forests, the total area of which is accurately quantified in Table 8.

Table 8. Comparison of simulated land use areas for 2022 and 2030 (unit: km²).

Scenario	Poplar Plantations	Grassland	Mudflat	Water
2022 Wetland Status	39.4884	843.3342	216.3735	313.0101
2030 Natural Development	57.8560	831.8965	209.7534	316.6490
2030 Ecological Regulation	0.89	882.6966	215.6523	305.5488

Table 8. Comparison of simulated land use areas for 2022 and 2030 (unit: km²).

Scenario	Poplar Plantations	Grassland	Mudflat	Water
2022 Wetland Status	39.4884	843.3342	216.3735	313.0101
2030 Natural Development	57.8560	831.8965	209.7534	316.6490
2030 Ecological Regulation	0.89	882.6966	215.6523	305.5488

As shown in

Table 9. Comparison of 2022 wetland area: PLUS model simulation vs. remote sensing interpretation (unit: km²).

Wetland Type	Poplar Plantations	Grassland	Mudflat	Water
2022 Simulated	55.76	826.42	203.07	327.86
2022 Actual	39.49	843.33	216.37	313.01

, poplar plantation land has a probability of expanding into grassland and mudflats between 2019 and 2022. However, the differences between the actual and simulated areas for each land type are minimal, indicating high simulation accuracy. The high-precision simulation results enhance the understanding of past wetland landscape changes and provide valuable guidance for future wetland conservation and management planning. By accurately identifying the driving factors behind the expansion of different wetland types, future wetland landscape changes can be better predicted, and corresponding conservation and management strategies can be developed to promote the sustainable development of the East Dongting Lake wetland.

3.2.4. Prediction of Wetland Change Trends in East Dongting Lake

Quantitative analysis of East Dongting Lake (2010–2022) reveals a stabilization of ecological land cover, with poplar plantation area transitioning from expansion to equilibrium. This simplified spatial pattern reflects targeted policy interventions, including afforestation reduction, lake management, and ecological redline enforcement. Based on the calculation and statistical analysis of wetland landscape area changes in East Dongting Lake from 2010 to 2022, the trend of ecological land changes shows that the scale of ecological land, particularly poplar plantation land, gradually expanded before stabilizing. The spatial structure has become progressively simpler and more stable, reflecting significant human intervention.

By configuring relevant parameters in the CARS module and inputting the 2030 demand projections, the predicted land use distribution for 2030 was generated (Figure 10).

Compared to the actual wetland landscape in 2022, the natural development scenario—without human intervention—demonstrates a significant recovery of poplar plantations. As shown in Table 8, from 2022 to 2030, poplar plantation land shows the most substantial increase, while the other three types (grassland, mudflats, and water bodies) exhibit a decline. This trend is mainly attributed to the regrowth of poplar plantations in previously deforested areas. However, poplar plantations have been shown to negatively impact biodiversity in the study area, making this scenario less favorable for wetland conservation.

Conversely, under the ecological regulation scenario, the continued deforestation of poplar trees and removal of regrowth in previously cleared areas effectively limit poplar plantation land expansion, resulting in a notable increase in grassland area. Meanwhile, mudflats and water bodies remain within a reasonable range of variation, making this scenario more favorable for biodiversity conservation in the study area.

Moving forward, efforts should continue to control poplar plantation land expansion, actively remove poplar regrowth, and implement water conservation measures to facilitate the restoration of wetland hydrological functions. This approach will support the long-term ecological restoration and sustainable management of the East Dongting Lake wetland.

4. Discussion

4.1. Technical Features of the PLUS Model in Wetland Prediction

Although other land-use change models exist, the PLUS model represents a significant advancement through its innovative integration of a rule-mining framework based on land expansion analysis strategy (LEAS) and a cellular automata approach using multi-type random patch seeds (CARS) [12]. The PLUS model is particularly effective in revealing the driving forces behind changes in land use/cover types and can simulate multi-class land use/cover changes based on patch-level evolution [6]. However, its application in predicting and simulating wetland landscape pattern changes remains relatively limited, particularly in the simulation of patch-level changes.

The PLUS model has the following advantages in wetland prediction:

• Deep Analysis of Driving Factors: The LEAS module extracts expansion-driving forces using the Random Poplar Plantations algorithm, quantifying the contributions of natural factors (such as climate and hydrology) and human factors (such as urbanization, agricultural development, and tourism development) to wetland degradation. For instance, Ref. [13] integrated the PLUS model to explore the spatiotemporal evolution characteristics and driving mechanisms of land use in the Zhengzhou-Luoyang Yellow River region from the perspectives of land use dynamics, transition matrix, centroid migration, and landscape pattern.
• Flexible Dynamic Transition Rules and Multi-Class Interactions: The model allows the setup of transition probabilities and constraints between different land types (e.g., prohibiting wetlands from being converted to urban areas), making it adaptable to specific wetland protection needs [7].
• High-Precision Patch Generation Mechanism: The CARS module uses a “competitive seed” mechanism to generate random patches, which more accurately simulates the complex spatial processes of wetland fragmentation, expansion, or contraction. This is superior to the grid-based rule division in traditional models such as CLUE-S. For example, Ref. [14] provides a detailed discussion on the application of the CARS module, simulates the patch evolution process of wetland ecosystems, emphasizes the impact of multiple types of random patch seeds on wetland patterns, and verifies the reliability of the patch generation mechanism through simulation.
• Multi-Scale and Multi-Scenario Compatibility: The PLUS model supports simulations at multiple scales, from local wetland protection zones to watershed levels. Ref. [15] through multi-scenario simulations using the PLUS model, proposed an ecological restoration priority zoning plan for the Poyang Lake Wetland Basin, providing a scientific basis for wetland protection and restoration decision-making in the region.
• Coupling with Other Models for Comprehensive Research. For example, Hu et al. conducted a study using the PLUS-InVEST-SolVES model to analyze land use changes and the social value of ecosystem services in the Weihe River Basin. The SolVES model is primarily used to assess social values, such as human preferences and subjective evaluations of ecosystem services [16].

Ecological land serves as both a crucial mechanism for implementing ecological protection and an external representation of policy effectiveness under an ecological-priority approach [17]. While multiple land-use change models exist, the PLUS model’s integration of LEAS and CARS frameworks provides unique advantages for capturing such policy-driven landscape transformations.

This conceptual framework is vividly demonstrated in our case study, where the PLUS model’s technical advantages enable effective capture of policy-driven landscape transformations. The dramatic poplar plantation-to-grassland conversion observed from 2019 to 2022 (decreasing from 387.92 km² to 39.49 km²) directly reflects three key protection policies: (1) wetland restoration through afforestation reduction (e.g., systematic poplar tree removal under Hunan’s “Three-Year Action Plan”), (2) lake management via hydrological regulation, and (3) ecological redline enforcement (as implemented through transition constraints in the CARS module).

These quantifiable outcomes not only validate the model’s utility for assessing conservation effectiveness in ecologically sensitive regions like East Dongting Lake, but more importantly demonstrate the PLUS model’s breakthrough application in simulating policy effects in wetland systems—overcoming the limitations of conventional models in capturing abrupt policy impacts. Such quantifiable policy impacts validate the model’s utility for assessing conservation effectiveness in ecologically sensitive regions like East Dongting Lake, while also providing measurable indicators for evaluating ecological redline policy implementation.

4.2. Limitations and Future Prospects

This study utilizes the PLUS model to quantitatively analyze the contributions of driving factors to wetland changes in the study area and predict the spatial-temporal distribution patterns of four different wetland types in East Dongting Lake under various scenarios in 2030.

However, the conversion patterns between different wetland types in Dongting Lake are highly complex [10]. Due to the organic remains of submerged plants, floating plants, and emergent plants, as well as sediment accumulation slowly raising the lakebed, aquatic vegetation in shallow water areas mainly follows the process of hydrosere succession, occupying bare shallow land. Simultaneously, the conversion patterns and rates of wetlands in the study area are influenced by sedimentation rates. When sediment accumulation accelerates, woody plants (primarily poplar plantations in this study context) rapidly occupy sandbars; otherwise, wetland vegetation such as moss grass gradually takes over low-lying sandbars, achieving succession in the sandbars [18]. This dynamic was incorporated in our model through elevation-dependent transition rules and sediment accumulation proxies in the driving factors.

To achieve higher-accuracy wetland change predictions, more precise and detailed wetland classification data are required. Additionally, the relationship between driving factors and wetland changes, along with their contributions, should be dynamic and even coupled. Therefore, the analysis of wetland factors must also be dynamic. When predicting wetland changes, patch transition probabilities and constraints should be dynamic and nonlinear.

In the future, the effectiveness of wetland dynamic simulation can be further enhanced by coupling dynamic evaluation models and machine learning algorithms, such as deep learning. This also presents a direction for future research and improvements in the model.

5. Conclusions

This study employs the PLUS model to quantitatively analyze the driving factors of wetland changes and their contributions in the study area, and to predict the wetland landscape distribution patterns in 2030. The following conclusions were drawn:

• Wetlands in the study area mainly consist of four landscape types: poplar plantation land, grassland, mudflat, and water body. From 2010 to 2022, there were significant changes in the poplar plantation land and grassland types. Poplar plantations showed an overall declining trend (2010–2022), with a transient increase during 2010–2016 (+4.8%) followed by drastic reduction post-2019 due to policy interventions (Table 2). Between 2019 and 2022, the poplar area decreased by 89.8%, nearly reaching elimination. In contrast, grassland showed a continuous increase. Mudflats and water bodies exhibited oscillatory changes. Spatially, the landscape displayed a complex distribution pattern of “water bodies and mudflats—grassland—poplar plantation land,” with vegetation interspersed and mixed. The evolution of different wetland types is influenced by hydrological factors and sediment deposition, leading to a unique multiple succession pattern in Dongting Lake.
• The PLUS model performs well in analyzing the driving factors of wetland landscape types in Dongting Lake from
• 2010 to 2022. Overall, the driving factors for the expansion or contraction of each landscape type are mainly natural factors. Among them, the driving factors for water bodies and grasslands are primarily precipitation and elevation. For grasslands, the most influential factors are elevation, distance to rivers, and precipitation. For poplar plantation land, the main driving factor is the distance to the main road. The two most significant factors influencing mudflats are precipitation and temperature.
• The PLUS model has strong applicability in simulating and predicting wetland landscapes in Dongting Lake, with an overall model accuracy exceeding 73%. The Kappa coefficient reached 0.73, providing an effective method for predicting landscape change trends in the study area and offering a reference for simulating and predicting other small-scale wetland landscapes.
• According to the prediction results from the PLUS model, in 2030, the wetland changes in East Dongting Lake will remain relatively stable. However, significant differences exist in the wetland changes under the two scenarios: natural development and ecological regulation. Under the natural development scenario, poplar plantation land will show the largest increase, while the other three types, including grassland, will decrease. This trend is unfavorable for the protection of wetland biodiversity in the study area. In the ecological regulation scenario, by controlling the poplar plantation area, grassland will experience a significant increase, while mudflats and water bodies will change within a reasonable range. In the future, the ecological regulation model should be adopted, continuing to control poplar plantation land area, implementing water conservation measures, and maintaining the recovery of grassland and water body areas to promote the ecological function restoration of wetlands in the study area.