A Prediction of Estuary Wetland Vegetation with Satellite Images

Abstract

Estuarine wetlands are the transition zone between marine, freshwater, and terrestrial ecosystems and are more ecologically fragile. In recent years, the spread of exotic vegetation, specifically Spartina alterniflora, in the Yellow River estuary wetlands has significantly encroached upon the habitats of native species such as Phragmites australis, Suaeda glauca Bunge, and Tamarix chinensis Lour. With advances in land prediction modeling, predicting wetland vegetation distribution can aid management and decision-making for ecological restoration. We selected the core area as the study object and coupled the hydrological model MIKE 21 with the PLUS model to predict the potential future distribution of invasive and dominant species in the region. (1) Based on the fine classification results from satellite images of GF1/G2/G5, we gained an understanding of the changes in wetland vegetation types in the core area of the reserve in 2018 and 2020. (2) Using public data such as ERA5 and GEO as input for basic environmental data, using MIKE 21 to provide high-spatial-resolution hydrodynamic parameters for the PLUS model as an environmental driver, we modeled the spatial distribution of various wetland vegetation in the Yellow River estuary wetland in Dongying under different artificial restoration measures. (3) We predicted the 2022 distribution of typical vegetation in the region, used the classification results of GF6 as the actual distribution, compared the spatial distribution with the actual distribution, and obtained a kappa coefficient of 0.78; the predicted values of the model are highly consistent with the true values. This study combines the fine classification results of vegetation based on hyperspectral remote sensing, the construction of a coupled model, and the prediction effect of typical species, providing a reference for constructing and optimizing the vegetation prediction model of estuarine wetlands. It also allows scientific and effective decision-making for the management of ecological restoration of delta wetlands.

1. Introduction

Coastal wetlands, as critical zones of land–sea interaction, are highly sensitive ecosystems. Climate change and intensified human activities pose significant challenges to their sustainable use [1]. Globally, 52.4% of coastal wetlands have been degraded since 1986–2000 [2], primarily due to climate change, human interference, and biological invasions [3].

Hydrological conditions measure the level, volume, timing, and frequency of water inflows and outflows in wetlands, which can determine their structure and function and affect biodiversity and primary production. Rapidly rising sea levels will directly inundate coastal wetlands and alter their original properties by intensifying erosion, flooding, and storm surges and increasing the salinity of estuaries, aquifers, and wetlands, resulting in the degradation and loss of coastal ecosystem functions and structures [4]. The Millennium Ecosystem Assessment (MEA 2005) [5] analyzed the effects of direct drivers on wetlands. In order to protect and maintain the biodiversity of coastal wetlands, numerous coastal wetland ecological restoration projects and practices have been carried out in China, achieving specific results [6]. These efforts have led to the development of technologies for vegetation restoration [7], hydrological restoration [8,9], micro-topography modification [10,11], and other techniques. The former is a type of bioremediation, while the latter are forms of environmental restoration. Environmental remediation provides the basis for bioremediation. Bioremediation can also improve the physicochemical quality of environmental factors such as water and soil in regional wetlands. This poses a challenge for assessing restoration effectiveness and predicting the future of estuarine wetlands [12,13]. Whether the aforementioned measures minimize interference and impact on the surrounding environment, allocate biological resources reasonably, and consider the integrity, diversity, coordination, and sustainability of the ecosystem. By simulating and predicting the distribution of vegetation, providing ecological restoration decision support, and selecting reasonable management measures, it can monitor and warn of changes in vegetation distribution, take proactive measures for ecological restoration, and reduce economic losses. Ecological managers aim to simulate vegetation distribution under artificial restoration measures to improve the feasibility and effectiveness of future restoration measures, reduce repair costs, and improve repair efficiency. Predicting future wetland species distributions is thus an important tool for evaluating wetland ecological restoration.

The prediction accuracy of a model depends on the classification accuracy and spatial resolution of the input remote sensing data [14] along with the model [15], parameters, or methods used for data assimilation. High-resolution environmental input data enhance the simulation accuracy of vegetation distribution models. Medium-resolution Landsat images provide continuous, long-term data for wetland classification, modeling, and management. They play a vital role in monitoring wetland vegetation over large areas and extended time periods [16,17,18]. High-resolution spectral data also offer greater spatial and temporal detail [19,20], crucial for understanding land use patterns, regional planning, and future land use predictions.

Several ecological models, including GARP, CART, Domain, BIOCLIM, ENFA, and Maxent, are currently available [21]. The Maximum Entropy (Maxent) model is widely used to study the geographic distribution of key species in ecological reserves [22,23,24]. However, Maxent does not account for interspecies relationships or provide distribution patterns for multiple species in the same area. Scholars have combined Maxent with GIS spatial analysis to simulate the potential distribution of Suaeda glauca in the Yellow River estuary nature reserve, identifying highly suitable areas for this species [25].

The selection and construction of models are crucial for improving the accuracy of vegetation distribution prediction. Single-species models can no longer meet the needs of research, and multi-species distribution models are usually based on cellular automata to simulate various land patches. A meta-cellular automaton is a local grid dynamic model discrete in time, space, and state, with the ability to simulate the spatiotemporal dynamics of complex systems [26].

Land use/land cover (LULC) models derived from meta-cellular automata, such as CA Markov, FLUS, PLUS, and CLUE-S, offer innovative ways to simulate multi-vegetation distributions [27]. LULC modeling helps analyze ecological and environmental impacts, land use changes, and regional planning [28]. For instance, Qiu and Lu used a CA Markov model to simulate the invasion extent of M. intermedia in Xiangshan Bay [29]. Similarly, Guo applied the FLUS model to study mangrove forest distribution in the Greater Bay Area [30].

However, several of the above models have limitations. FLUS will not run with more than 14 types; CA Markov requires data completeness, and adding other drivers is inconvenient; and the complex background of CLUE-S is too computationally intensive and has insufficient predictive power [15]. The PLUS model is an improved version of the FLUS and, compared with CA Markov, FLUS, and other LULC models, can obtain the contribution of drivers to various types of land use expansion and their development probability. The model can analyze the mechanism of land use change over a certain period with better explanatory ability, higher prediction accuracy, and simpler arithmetic than others. It contains a new mechanism of multi-class seed growth, which can better simulate the patch-level change in multi-class land use and can be coupled with other models to improve the prediction accuracy [31,32,33,34].

This model can be used to explore the driving factors of the evolution of various land use types and their impact on evolution. It can simulate the generation and evolution of natural types of patches such as grasslands and forests at a finer resolution and be used to analyze the mechanisms of changes in multiple types of land use patches in the spatiotemporal domain. Despite its advantages, the PLUS model has drawbacks. It does not account for artificial restoration or natural dynamics. Coastal wetlands add complexity due to factors like tides and salinity, which greatly influence vegetation growth [32]. To accurately simulate vegetation distribution in coastal wetlands, it is crucial to incorporate their unique hydrodynamic processes.

Hydrological models have been extensively applied in ecological restoration projects, and three-dimensional hydrological models represented by MIKE 21/3, SWAT, DELFT 3D, etc., have also been widely used in ecological restoration projects for watersheds, wetlands, and nearshore environments [35,36]. The hydrodynamic module of the MIKE 21 model mainly relies on gridding the terrain and inputting factors such as bed roughness, boundary conditions, meteorology, and Coriolis force to analyze elements such as water level and flow velocity. The gridded terrain input also makes it possible to implement measures into the model such as manual excavation of tidal trenches in ecological restoration projects. When MIKE 21 is used to establish a two-dimensional hydrodynamic model of the Yellow River estuary wetland, it can output parameter distributions such as flow velocity, water level, and water temperature and salinity with high spatiotemporal resolution, which helps to accurately simulate and predict the growth pattern of vegetation in the Yellow River estuary.

In recent years, the exotic vegetation Spartina alterniflora in the wetlands of the Yellow River estuary has been expanding, which has significantly limited the survival space of native species such as Phragmites australis, Suaeda glauca Bunge, and Tamarix chinensis Lour. Therefore, the invasion of Spartina alterniflora will impact the biodiversity and material cycle of the invaded site [37]. In response, several ecological restoration projects have been implemented. The existing ecological restoration simulation predictions only target a single species or have not yet integrated the important environmental driving factor of coastal wetlands, hydrodynamics, which cannot be combined with artificial restoration and policies to predict the effectiveness of ecological restoration projects. Our study focuses on the core area of the Yellow River estuary wetlands, where we coupled the hydrological model MIKE 21 with the PLUS model to predict the potential future distribution of invasive and dominant species in the region. (1) The aim was to understand the changes in wetland vegetation types in the core area of the protected area in 2018 and 2020 based on the fine classification results from remote sensing satellites. (2) The spatial distribution of multiple wetland vegetation in the Dongying Yellow River Mouth Wetland under different artificial restoration measures was modeled using MIKE 21 to provide high-spatial-resolution hydrodynamic parameters for the PLUS model as environmental drivers. (3) The distribution of typical vegetation in the region in 2022 was predicted, the spatial distribution with the actual distribution was compared, and the accuracy was assessed based on several indicators. This research integrates hyperspectral remote sensing classification, a coupled model approach, and predictions for key species. It offers a valuable reference for improving estuarine wetland vegetation prediction models. Additionally, it supports scientific decision-making for the restoration and management of delta wetlands.

2. Materials and Methods

2.1. Research Area

The Yellow River Mouth Wetland (YREW) (Figure 1) is located at the intersection of the atmosphere, rivers, oceans, and land in the modern Yellow River Delta. Due to its rich integration of various material and dynamic systems, YREW supports abundant resources and a unique ecosystem structure forming the most complete, largest, and youngest wetland ecosystem in China [38]. The YREW is covered with various wetland vegetation, such as Spartina alterniflora, Phragmites australis, Suaeda heteroptera, and Tamarix chinensis. It also provides an important habitat for rare and endangered bird species. However, Spartina alterniflora is an invasive species that was introduced to the YREW in 1985 to prevent coastal erosion. Due to its strong adaptability, Spartina alterniflora has squeezed the growth range of other native vegetation in the YREW. The dense distribution of Spartina alterniflora in the tidal flat weakens the dynamics of waterlogging and destroys biodiversity. The coverage of Spartina alterniflora in YREW increased from 6.15 km² to 43.43 km² between 2009 and 2018, whereas that of Suaeda heteroptera decreased by 78.17% in 2014 compared with 1984.

Since 2010, the governments of Shandong Province and Dongying City have launched various ecological restoration projects to protect native vegetation and combat invasive species in the YREW. The measures include returning aquaculture ponds to wetlands, planting the native vegetation Suaeda heteroptera, modifying the micro-bathymetry, and increasing the land for bird habitats. These actions have produced some benefits, but continuous restoration efforts are needed to maintain the fragile ecosystem. This study models the restoration measures taken in 2020, utilizing the MIKE 21-PLUS model to forecast vegetation distribution in 2022 at Dongying City to evaluate the effectiveness of the restoration measures.

2.2. Data Source

2.2.1. Data Source of MIKE 21 Model

The topographic data for the MIKE 21 mesh were downloaded from the British Oceanographic Data Centre (https://www.gebco.net/ accessed on 30 April 2021), with a spatial resolution of 400 m. The daily water temperature and salinity data used for building the boundary conditions in MIKE 21 were obtained from the Copernicus Marine Environment Monitoring Service (CMEMS, https://resources.marine.copernicus.eu/ accessed on 24 December 2022).

2.2.2. Data Source of PLUS Model

The data input to the PLUS model are divided into two parts: land use data and driving forcing data. The land use data of the YREW from 2018 and 2020 were derived from the satellite images fused with the panchromatic image of GF-1 with a spatial resolution of 2 m, the multispectral image of GF-2 with a spatial resolution of 8 m, the panchromatic image of GF-2 with a spatial resolution of 1 m, and the hyperspectral image of GF-5 with a spatial resolution of 30 m, the technical parameters of GF5 are shown in Table 1. In order to improve the spatial–spectral resolution of the images, the GS method in Envi is used to fuse these images. The land use data from 2022 were obtained from the panchromatic image of GF-6 with a spatial resolution of 2 m. Through the collaborative classification of multispectral and hyperspectral data, the pixels in these satellite images were classified into 11 types: Spartina alterniflora (SA), Phragmites australis (PA), Tamarix chinensis (TC), Suaeda heteroptera (SH), Phragmites australis in tidal flat (PATF), bare tidal flat (TF), bare salt marsh (SM), willow, water, road, and plowland.

As described before, the MIKE 21 model provides hydrological driving forcing data, including water depth, current speed, water salinity, and temperature. The climate variables, such as precipitation, air temperature, and wind speed, are derived from the ERA5 dataset (https://cds.climate.copernicus.eu/cdsapp#!/home, accessed on 9 August 2021). The soil temperature and soil water content are also from this website The total nitrogen and phosphorus of the soil are from the National Earth System Science Data Center (https://gre.geodata.cn/ accessed on 7 August 2021). The pH and soil salinity were sampled during the field survey in July 2020. Besides the soil, seawater, and climate factors, the growth of coastal wetland vegetation is also influenced by the regional topography and location. The elevation, slope, and aspect data can be read using ArcMap 10.6 from the DEM data (https://www.gscloud.cn/search, accessed on 20 February 2020)). The distance to the river or coastline strongly influences coastal vegetation growth in terms of freshwater flushing and human activities. By depicting the outline of the river and coastline on the map, we calculated the Euclidean distance of every pixel in the satellite images using ArcMap 10.6.

All 18 kinds of driving forcing data and 3-year land use data were processed into the same spatial range and resolution (30 m) and saved as “.tif” images with the same format.

The data sources mentioned above are shown in Table 2.

2.3. Ideas for the Construction of the Coupled MIKE 21-PLUS Model

Many researchers are studying the development of wetland vegetation by applying cellular automaton. Ge established a cellular automaton model to simulate the distribution of Spartina alterniflora and Phragmites australis in Yangtze River estuary wetland [39,40]. Yi developed a vegetation population growth–competition dynamic model by coupling cellular automaton with an improved logistic function in the Yellow River estuary wetland and simulated the distribution of suitable habitats for Phragmites australis, Tamarix chinensis, and Suaeda Salsa under different hydrological years [41]. The two studies are based on the cellular conversion rules formulated by the researchers according to their own previous experience and work. For example, the optimal intervals of soil salinity and water content for the growth of the salt marsh vegetation in the model are delimited according to the observers’ experience, which means that subjective factors may affect the model results. This approach limits the possibility of expanding the model’s application to other restoration areas.

To address the limitations of subjective cellular automaton models, the MIKE 21-PLUS model provides an objective approach to simulate vegetation distribution under various restoration scenarios (Figure 2). In coastal wetland restoration projects, topographic and geomorphic modification can enhance hydrological connectivity and improve soil physicochemical properties. Cutting and deep plowing the invasive vegetation and reseeding the native vegetation are key measures to restore native ecosystems. In this study, the MIKE 21 model provides the distribution of water temperature, water salinity, water depth, and current speed to the PLUS model and is also used to simulate the artificial restoration of landforms, which is reflected in the differences in current speed. The PLUS model is utilized to output the contribution of the driving factor and the future vegetation pattern. The modeling of cutting and reseeding vegetation is implemented in the PLUS input file. By comparing the vegetation simulation results under restoration with the natural state, we aim to assess the long-term viability of restoration measures in preserving native vegetation.

2.4. MIKE 21-PLUS Model Development

2.4.1. MIKE 21 Model Configuration

MIKE 21, part of the MIKE (2020) series (https://www.mikepoweredbydhi.com/) developed by the Danish Hydraulic Institute (DHI), is a two-dimensional ocean hydrological model used to simulate water dynamics. We constructed a two-dimensional model to obtain the hydrological parameters of the YREW based on MIKE 21. The model domain covers the YREW and extends to the western part of Laizhou Bay (Figure 1) to obtain a more accurate tidal boundary condition. In this study, the model adopts a dense transition triangular mesh, that aligns closely with the coastline to simulate the estuarine flow field (Figure 3). Factors such as tides and major inflows into the sea are considered when defining flow and tidal boundaries during model construction. Tidal forcing is added at the open boundary. Based on global tidal model data, the prediction time is set in the system, the open boundary is determined, and a mesh file is generated. The tidal boundary condition file contains the elevation changes within the boundary mesh. The area is illustrated in the figure below. Once boundary information is provided, grid and node relationships are automatically generated to maintain a uniform transition of grid density. The spatial resolution of the model gradually decreases from the Yellow River estuary to its two sides.

The total number of model mesh nodes and cells is 17,367 and 33,996, respectively. The spatial resolution of the model mesh varies from 300 m at the YREW to 1 km at Laizhou Bay. This setting balances simulation time and spatial resolution. Because the YREW is located at the boundary of land and sea, we set the “Flood and dry” grid to enhance the stability of the model. The drying, flooding, and wetting depths are 0.005 m, 0.05 m, and 0.1 m, respectively, which are default values. Wind forcing was not added to the model since tidal forces are the primary driver in Laizhou Bay. The tidal boundary condition file, which contains the elevation variation in boundary grids, was produced using the “MIKE 21 Toolbox”. The initial tidal elevation and velocity were both set to zero. The time step varied from 0.01 to 30 s to make the CFL (Courant–Friedrich–Levy) number less than 1, which is a standard for the stability of numerical models. The simulation period was 2020, and the output interval was 1 h. We input the average water temperature, salinity, depth, and current speed values from May to July into the PLUS model as vegetation grows and extends rapidly during this period.

2.4.2. MIKE 21 Model Validation

The water salinity results in Laizhou Bay simulated using MIKE 21 were verified using the data observed in August 2020 by the North China Sea Administration, and the validation result is shown in Figure 9. The coefficient of determination $R^2$ and the NSE (Nash–Sutcliffe efficiency) (Nash and Sutcliffe, 1970) were calculated to evaluate the simulated performance. Values of $R^2$ and NSE close to 1 indicate a better simulation. The $R^2$ and the NSE are calculated as Formulas (1) and (2)

(1)$R^2={\displaystyle \frac{\sum _{i=1}^n{(y_i-\overline{y_i})}^2-\sum _{i=1}^n{(y_i-\widehat{y_i})}^2}{\sum _{i=1}^n{(y_i-\overline{y_i})}^2}}$

(2)$NSE=1-{\displaystyle \frac{\sum _{i=1}^n{(\widehat{y_i}-y_i)}^2}{\sum _{i=1}^n{(y_i-\overline{y_i})}^2}}$

where $y_i$ is the actual value, $\overline{y_i}$ is the average of the actual values, and $\widehat{y_i}$ is the predicted value.

2.4.3. PLUS Model Configuration

The PLUS model integrates the LEAS (land expansion analysis strategy) module, which assesses land development potential, and the CARS (cellular automata model based on multi-type random seeds) module [42], which simulates future land use patterns. In the LEAS module, the development potential of each type of land and the contributions of every driving force to land use are calculated using land expansion conditions (2018–2020) and environmental driving forces. The Markov chain predicts the probability of occurrence of events based on Markov process theory, without aftereffect features, where the state of the next step is determined only by the current state and is independent of past states [43]. The PLUS model incorporates the Markov chain to predict the pixel count of each type of land use. In the CARS module, the future (2022) land use pattern is simulated using the initial (2020) land use pattern, development potential files, and pixel prediction from the Markov chain.

Researchers can also set different land use scenarios by modifying the neighborhood weights and transition matrix in the PLUS model. The neighborhood weights of each land type are obtained by calculating the percentage of the expansive pixels of each type in the total expansive pixels. The range of neighborhood weights is 0–1, where a larger value indicates a greater influence on the neighborhood. In this study, we adjusted the model parameters based on the neighborhood range, domain weight, patch generation threshold, diffusion coefficient, random seed ratio, and transformation matrix rules of the region. The neighborhood weights were derived from the land expansion patterns of 2018 and 2020, as shown in Table 3. The transition matrix was defined by considering the competitive regulation of native and invasive vegetation, local government policies, and researchers’ experience. A value of 0 indicates that conversion is not allowed, while a value of 1 indicates that conversion is allowed. For example, Spartina alterniflora can transition to other vegetation types but not vice versa. The road and plowlands remain unchanged. The transition matrix is shown in Appendix A.

2.4.4. MIKE 21-PLUS Model Validation

The classified map of the YREW in 2022 was used to verify the model results. The kappa coefficient was calculated to evaluate the consistency between the simulation results and the actual land use conditions.

(3)$k={\displaystyle \frac{p_0-p_{\epsilon }}{1-p_{\epsilon }}}$

where $k$ represents the kappa coefficient, and $p_0$ is the overall classification accuracy, which is the ratio of the sum of the number of correctly classified samples of each class to the total number of samples, $p_{\epsilon }$ is the ratio of the sum of the “product of actual and predicted quantities” of each class to the “square of the total number of samples”.

At the same time, the Precision, Recall, and F₁-score were calculated to evaluate the model’s performance. They are calculated as Formulas (4)–(6)

(4)$Precision={\displaystyle \frac{TP}{TP+FP}}$

(5)$Recall={\displaystyle \frac{TP}{TP+FN}}$

(6)$F_1\text{-}score=2·{\displaystyle \frac{Precision·Recall}{Precision+Recall}}$

where TP represents True Positives, which is the number of samples correctly predicted as positive by the model results, where FP represents False Positives, which is the number of negative samples that were incorrectly predicted as positive by the model results, where FN represents False Negatives, which is the number of positive samples that were incorrectly predicted as negative by the model results.

2.5. Simulation of Scenario Settings (Artificial Restoration Program)

Spartina alterniflora has extensive salinity tolerance, high adaptability, strong reproductive dispersal ability, invasive competitiveness, and a wide ecological range. In comparison, indigenous plants exhibit relatively low competitiveness. Spartina alterniflora has occupied the ecological niches of large areas of saline Suaeda glauca Bunge, light beaches, and seagrass beds and continues to expand rapidly.

Ecological restoration work was planned for 2022 in the protected area to remove the exotic species of Spartina alterniflora from the nature reserve; restore the degraded salt marsh wetlands of the native species saline Suaeda glauca Bunge; and use integrated control techniques involving mowing, flooding, and plowing to carry out artificial control in the concentrated distribution area of Spartina alterniflora. Additionally, the joint implementation of alternative restoration efforts for the native species saline Suaeda glauca Bunge was planned. Therefore, this restoration area was selected to discuss the effectiveness of the ecological restoration work (Figure 4).

The artificial restoration measures realized in the coupled model were implemented in accordance with Sub-Project 1 of the Marine Ecological Protection Project of Dongying City in 2020 (Figure 4). In Figure 4, the region circled by the blue contour line represents the Spartina alterniflora cutting area (640 hm²), and the area within the red is the Suaeda heteroptera replanting area (2100 hm²). Meanwhile, several tidal creeks leading to the sea were excavated to enhance the water dynamics in this area.

In the research work, the impact of digging tidal trenches was simulated through changes in water depth in MIKE 21, and the cutting and replanting of plants were simulated through pixel filling of relevant categories in ArcGIS. After inputting into the PLUS model, the vegetation prediction distribution after artificial ecological restoration measures was obtained and compared with the vegetation prediction under natural conditions, providing a basis for evaluating the effectiveness of artificial restoration measures (Figure 5).

2.5.1. Trenching

Excavating the tidal creeks involves deepening the area’s topography, increasing seawater fluidity, reducing salinity on the surface of the salt marsh vegetation, and promoting its growth. Based on the distribution of the tidal creeks to be excavated in the restoration plan, the water depth of the MIKE 21 mesh in the same geographical location was adjusted accordingly. The artificial restoration area is part of the tidal flat, where the initial water depth of the grid in this area is +1 m and the spatial resolution of the grid is 300 m. The water depth of the newly excavated tidal creeks is uniformly set at −2 m. The modification is illustrated in Figure 6a.

The comparison of the current velocity in the simulated area with and without tidal creek shows that the current velocity in the tidal creek is significantly higher than in the surrounding area (Figure 6b), while the current velocity distribution in the restoration area without tidal creeks is more uniform (Figure 6c).

2.5.2. Mowing and Replanting

The removal of invasive species and the replanting of restoration species are important artificial means of wetland ecological restoration. This model adopts the method of filling pixels to reflect the policy impact on the corresponding areas in the ecological restoration plan.

2.5.3. Natural State

For the blank control, we modeled the vegetation distribution in 2022 as a natural case without any artificial restoration measures. The result obtained for the natural case can be used to validate the model’s performance and reveal the effectiveness of artificial ecological restoration projects.

3. Results

3.1. Temporal and Spatial Distribution of Vegetation in the Yellow River Estuary Wetland from 2018 to 2022

This article focuses on the current situation of large wetland areas and limited labels in the Yellow River estuary. Hyperspectral (HS) images were used to help classify multispectral (MS) images of large scenes. A few-shot MS-HS image collaborative classification method combining feature distribution enhancement (FDE) and subdomain alignment was used, and coupled dictionary learning and distance-preserving probability distribution were integrated [44,45]. The classification results were verified for accuracy through wetland field surveys, with a classification accuracy of 0.85, which meets the input requirements of the model. Figure 6 shows the variation in LULC in the YREW from 2018 to 2022 as classified using satellite images; these data were corrected with a tidal model (Figure 7). The change in vegetation area is shown in Table 4.

3.2. Changes in Vegetation Distribution After Ecological Restoration

The aim of the restoration project is to remove the invasive vegetation Spartina alterniflora (green outline in Figure 8) and replant the native vegetation Suaeda heteroptera in the restoration area (yellow outline in Figure 8). This study adopted the method of transferring the corresponding pixel type, meaning that Spartina alterniflora pixels were converted into tidal flat pixels to represent artificial cutting, and bare salt marsh pixels were converted into Suaeda heteroptera pixels to represent artificial reseeding. ArcMap 10.6 was used to transform the Spartina alterniflora pixels (91,234.1 km²) in the restoration area based on the 2020 map into bare tidal flat pixels and to increase the Suaeda heteroptera pixels (46,732.1 km²) on the tidal flat along the coastline (Figure 8). The revised vegetation distribution in 2020 was input into the PLUS model as an unnatural case compared with the natural case.

3.3. Validation of MIKE 21 Simulation Results

The coefficient of determination $R^2$ between the observed and modeled salinity is 0.474, and the NSE (Nash–Sutcliffe efficiency) [46] is -2.44. Although the simulated salinity values are slightly lower than the observed values (Figure 9), they approximately represent the spatial pattern of seawater salinity in Laizhou Bay in August 2020 (Figure 9). Since the environmental driving factors input to the PLUS model are normalized, the reasonable simulation of the salinity spatial distribution pattern by the MIKE 21 meets the research needs, allowing for further research to be carried out based on these results.

According to the literature review, the tidal current in this sea area is basically a regular semi-diurnal current, which is a reciprocating flow along the coast. The rising tide flows from northwest to south and the falling tide flows from south to northwest, with a flow velocity of about 20–80 cm/s. There is a time difference of about 6 h between the rising and falling tides in the Laizhou Bay area and Bohai Bay area, which verifies that the MIKE21 simulation results meet the input requirements of PLUS. According to survey data from the North Sea Environmental Monitoring Center of the State Oceanic Administration in the Yellow River estuary and adjacent waters, in August 2020, the surface temperature of the Yellow River estuary varied between 24.87~27.57 °C. The bottom water temperature varies between 23.13~26.95 °C. In August 2020, the surface salinity of the Yellow River estuary varied between 14.333~31.232‰, and the bottom salinity varied between 27.398~31.324‰, verifying the results of the MIKE21 simulation.

3.4. Validation of MIKE 21-PLUS Simulation Results

The kappa coefficient is commonly used to verify data consistency. Theoretically, the calculated values of the kappa coefficient range between −1 and 1. A kappa coefficient close to 1 indicates a high consistency between the modeled and actual land use patterns. However, the kappa coefficient has certain limitations related to consistency verification.

The following figure shows the predicted results under natural conditions, the predicted results under artificial comprehensive restoration, and the actual vegetation distribution in 2022 (Figure 10). The accuracy of the model constructed in this study was evaluated using Formula (3) based on the predicted results after comprehensive restoration and the actual distribution. Table 5 shows the pixel values of the predicted and actual results image, calculate the Recall and Precision using Formulas (4) and (5), F₁-score according to Formula (6).

The overall accuracy obtained from the confusion matrix is 0.84. The k coefficient calculated according to Section 2.5.2 is 0.78, indicating that the predicted values of the model are highly consistent with the true values. The difference between the two categories, water body and salt marsh, in Table 5 is likely due to misjudgment caused by tidal water cover at the time of data acquisition. The requirements of simulation for wetland ecological simulation and prediction can be satisfied.

The F₁-score are shown in Table 4. The performance evaluation of the model is included in Section 4.4.

4. Discussion

4.1. Changes in the Spatial and Temporal Distribution of Wetland Vegetation in the Yellow River Estuary, 2018–2022 (Purely Observational Data Analysis)

Spartina alterniflora occupied a large area in the northeast of the YREW compared to other wetland vegetation and extended toward the coastline. It is clear that this invasive species has significantly limited the living space of other vegetation. Phragmites australis was mainly distributed along the old and new channels of the Yellow River. The distribution pattern of Tamarix chinensis is similar to that of Phragmites australis, but the former was more concentrated downstream of the river channel. Suaeda heteroptera was widely distributed in the YREW, but its spatial pattern was dispersed in the north, east, and south of the map. Phragmites australis in the tidal flat primarily gathered at the bifurcation of the new and old river channels. Willows appeared sporadically along both sides of the new channel of the Yellow River.

In 2018, the total area of the nature reserve only reached 41.71 km². At present, the majority of Spartina alterniflora is distributed in the area from the south of the Ludong Sea dike at the current mouth of the Yellow River to the area north of the 121 platform of the Dawenliu management station, and south of the 121 platform. Additionally, a small amount of newborn Spartina alterniflora is found in the area of the 1002 management station.

During 2018–2020, the area of Spartina alterniflora increased by 2.82 km², and its spatial distribution pattern remained relatively stable. However, it severely squeezed the habitat of Suaeda salsa. The area of Phragmites australis increased by 10.45 km², with its growing region extending closer to the downstream portion of the river channel. In contrast, the area of Tamarix chinensis decreased by 9.58 km². Suaeda heteroptera’s area increased by 13.31 km², mainly in the southern part of the YREW.

Between 2020 and 2022, the government carried out artificial restoration by mowing Spartina alterniflora, dredging the tidal channel, and implementing a flooding cofferdam. This reduced the area of Spartina alterniflora by 13.84 km², while the tidal flat area of Phragmites australis increased by 8.18 km², mainly downstream of the estuary. Over these two years, Phragmites australis and tidal flat Phragmites australis gradually formed a competitive situation with Spartina alterniflora; however, the area of alkali flats decreased by 14.19 km² and was mainly distributed on the landward side, still under the stress of M. alterniflora. The area of Suaeda glauca Bunge decreased by 14.19 km²; it was primarily distributed on the landward side and remained under stress from Spartina alterniflora. Zhang [43] pointed out that stable sediment transport to Laizhou Bay resulted in a continuous increase in land area at the Yellow River estuary gate, which caused the area of Spartina alterniflora in this region to continue increasing. The results showed that the distribution of these wetland vegetation types in the study area was concentrated on both sides of the Yellow River and scattered in areas farther away from the river channel, which is consistent with Yi’s study [47].

In summary, the remote sensing vegetation distribution input into the MIK21-PLUS model is similar to that of other scholars. In practical work, satellite historical images may not be complete, and multiple images need to be fused to meet the requirements of the model input. This result also verifies the reliability of the method used in this paper for image fusion and wetland vegetation collaborative classification.

4.2. Comparison of Simulation Results in Restored Areas (Evaluation of the Effectiveness of Artificial Restoration)

The area was artificially ecologically restored in 2020, so this section mainly discusses the distribution status of invasive species and wetland dominant species simulated under natural and artificial restoration conditions in 2022, providing a basis for evaluating the effectiveness of artificial restoration measures. The natural state refers to the natural distribution of vegetation without artificial restoration, while the artificial restoration state refers to the vegetation distribution after measures such as excavation of tidal trenches and vegetation cutting and replanting.

The comparison of model results in the restoration region is shown in Figure 11. The differences in vegetation distribution between the natural and restored scenarios are clearly visible (Table 6). In 2020, the area of Spartina alterniflora in the restoration region was 5.05 km². Under the natural scenario, the simulated area of this invasive vegetation in 2022 is 5.06 km². The area of Suaeda heteroptera in 2020 and 2022 is 3.89 km² and 3.90 km², respectively. Similar changes occurred in other vegetation types, indicating that without human intervention, the area of vegetation changes minimally over a relatively short period.

As described in Section 3.4, the area of Spartina alterniflora decreased by 4.1 km², and that of the native Suaeda heteroptera increased by 2.1 km² in 2020, consistent with the actual restoration project. After implementing the restoration measures, the areas of Spartina alterniflora and Suaeda heteroptera in 2020 in the restoration region were 0.34 km² and 6.02 km²_, respectively. The modeled vegetation pattern in 2022 was similar to that in 2020 under the restored scenario, with the areas of Spartina alterniflora and Suaeda heteroptera increasing to 0.69 km² and 6.09 km²_, respectively. However, the areas of Phragmites australis and Tamarix chinensis remained unchanged.

Some researchers have conducted similar experiments. Zhang [43] simulated the distribution of Spartina alterniflora in the YREW with and without other native vegetation and found little difference in the results under the two scenarios. This suggests that the native plants had minimal effect on the population dispersal of Spartina alterniflora and that this invasive species has a strong ability to invade and disperse, along with a great competitive advantage.

4.3. Analysis of the Contribution of Environmental Elements

The LEAS module in the PLUS model can output the contributions of environmental driving factors to each type of land. In Figure 12, the analysis shows that the distance to the coastline, water salinity, current speed, and elevation are the main contributors to Spartina alterniflora’s spatial pattern formation. The contribution rate (0.177) of the distance to the coastline is much higher than the other factors. This is mainly a result of the initial location of Spartina alterniflora, which is close to the coastline. This invasive vegetation shows a seaward growth trend during 2018–2020, making the distance to the coastline the most influential environmental driving factor. Additionally, seawater salinity, current speed, and elevation play important roles in the spatial pattern formation of Spartina alterniflora.

For Suaeda heteroptera, the main driving factors are soil total nitrogen, distance to the river, air temperature, and soil salinity. However, soil total nitrogen is the most influential factor (0.174), consistent with Min Zong’s study [23]. The result indicates that the physicochemical properties of the wetland soil significantly impact the growth of this fragile native vegetation. Furthermore, the distance to the river also plays a key role in the spatial evolution of Suaeda heteroptera, which suggests that excavating some tidal creeks to enhance the hydrodynamics of the tidal flat and increase the flush frequency could help remove salinity in the surface of the salt marsh vegetation, benefitting its growth.

The main contributing factors for Phragmites australis are soil salinity, soil pH, soil phosphate, and distance to the river, similar to Suaeda heteroptera. Among these, soil salinity is the most influential factor (0.197).

Some researchers have demonstrated that the growth of salt marsh vegetation is closely related to the nitrogen and phosphate content of the wetland soil [8]. Micro-topography modification has also been regarded as an effective way to restore salt marsh vegetation [48]. Thus, natural forces based on ecological processes can help humans protect coasts, facilitate coastal zone restoration, and significantly reduce management and restoration costs [49]. Additionally, improvements to wetland models will significantly enhance the effectiveness of restoration engineering.

4.4. Comparison of Model Performance

In this study, we simulated the distribution of wetland vegetation in 2022 using the PLUS model and compared the F₁-score results with the proposed method. It was found that the prediction accuracy of vegetation types for ecological restoration that we are concerned about, such as Spartina alterniflora, Phragmites australis, and Suaeda salsa, is better for the proposed model than using PLUS alone. In Figure 13, all vegetation types studied are affected by seawater salinity. The excavation of tidal trenches leads to changes in the micro-topography, water flow velocity, and salinity of tidal flats. Therefore, the vegetation distribution prediction advantage of coupled hydrodynamic models is obvious.

The performance of a model may be affected by various factors, including data distribution and domain characteristics. There are too many water samples in the simulation area, and road samples are greatly affected by artificial and policy factors, so the differences in simulation using different models are relatively small.

4.5. The Significance of the MIKE 21-PLUS Model for Ecological Restoration

After artificial restoration and natural remediation, the area of Spartina alterniflora decreased by 10.94 km² from 2018 to 2022, while that of Suaeda glauca Bunge grew by only 0.88 km². Additionally, the area of bare and saline beaches increased by 19.81 km². These changes are attributed to the natural mortality of Spartina alterniflora following mowing and the replanting of Suaeda glauca Bunge.

The modeling results obtained after the construction of the MIKE 21-PLUS model to predict vegetation distribution in the Yellow River Mouth Delta Protected Area in 2022 showed that this model can accurately simulate the spatiotemporal pattern of wetland vegetation in the YREW under both natural growth and artificially restored scenarios. The area of each type of vegetation changes slightly under the natural growth scenario, which is consistent with the conditions shown by the satellite images from 2022.

By modifying topography and vegetation pixels, we successfully modeled the vegetation distribution under the restored scenarios in 2022. In the restored region, the area of Spartina alterniflora decreased by 4.37 km², while that of Suaeda heteroptera increased by 2.19 km²_. These variations are also reflected in the satellite images. Our findings also indicate that location and soil properties are the main environmental driving factors in forming spatial patterns for invasive and native vegetation. In summary, remote sensing results from the fine classification of coastal zone vegetation enable predictions of artificial restoration outcomes for coastal zones. This study demonstrated the potential of the MIKE 21-PLUS model in coastal wetland artificial restoration projects, helping decision-makers to develop more comprehensive and effective strategies.

The combination of MIKE 21 and PLUS overcomes the drawbacks of a single model, and it fully exploits the advantages of both models in terms of quantity predictions and high spatial resolution for simulating coastal vegetation distribution. The coupled model can serve as a valuable tool for managers and researchers in planning and evaluating wetland restoration projects. However, this study has some limitations. First, some environmental driving forces have a relatively low spatial resolution, such as soil properties and climate variables, which may reduce the simulation accuracy of the PLUS model. Second, the MIKE 21 model can only provide seawater parameters without soil properties. Third, due to data availability limitations, the LULC in 2022 was not derived from hyperspectral images, which may impact the accuracy of image classification. Additionally, the artificial restoration scenarios set in this study are somewhat limited. We need to incorporate more restoration strategies into the model, such as plowing the land to improve soil aeration, reducing pollution sources, and conducting benthic bio-breeding projects.

5. Conclusions

The distribution of wetland vegetation to some extent reflects the health status of wetlands. Simulating changes in key vegetation helps monitor landscape evolution trends and supports wetland management planning. This study analyzed vegetation distribution patterns in the Yellow River estuary wetland from 2018 to 2022. It introduced a new approach using the coupled MIKE 21-PLUS model to evaluate and predict vegetation distributions under coastal wetland restoration scenarios. By incorporating key environmental factors, such as hydrodynamics, and applying machine learning, the model predicts vegetation distribution for future years. At present, the experiments in the article are limited by vegetation ecological remote sensing data and classification methods, as well as the accuracy of hydrodynamic simulation, resulting in limited accuracy of coupled model vegetation prediction. If soil and other environmental assimilation data can be coupled in the prediction model, it will provide insights into the effectiveness and adaptability of artificial restoration methods, offering scientific guidance for restoring habitats and ensuring the ecological security of coastal wetland species.

With the interaction between human activities and the natural environment in the research area, understanding the future development trends, environmental pressures, and threats in the ecosystem will require researchers to make quantitative predictions. The development of high-precision classification technology for satellite hyperspectral images and high-resolution environmental data assimilation products are prerequisites for improving simulation accuracy. Additionally, appropriate hydrological models and robust ecological restoration models are crucial for predicting wetland vegetation dynamics effectively.

Author Contributions

Conceptualization, M.Y. and Y.G.; methodology, M.Y., N.G. and Y.Y.; software, Y.Y.; formal analysis, N.G.; resources, X.S.; data curation, B.G.; writing—original draft, M.Y. and Y.Y.; writing—review and editing, M.Y. and Y.G. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data supporting the findings of this study can be obtained from the corresponding author upon reasonable request.

Acknowledgments

This study was supported by data from the Dongying Yellow River Delta Nature Reserve and the Hai River Basin Research Center of the Ministry of Ecology and Environment.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1. Research area-the Yellow River estuary wetlands.

Figure 2. Schematic diagram of the coupling models.

Figure 3. Grid range of MIKE 21.

Figure 4. Schematic diagram of the ecological restoration area of the Yellow River estuary delta.

Figure 5. The processing flow of MIKE 21-PLUS coupling model in artificial ecological restoration measures.

Figure 6. Hydrodynamic simulation of MIKE21 model. (a) Hydrodynamic simulation within the restoration area; (b) current velocity without tidal creek; and (c) flow velocity under tidal ditch conditions.

Figure 7. Changes in the distribution of features in the Yellow River estuary, 2018–2022.

Figure 8. Model realization process for mowing and replanting.

Figure 9. Comparison of observed and modeled salinity in Laizhou Bay. (Salinity data from the environmental survey of Laizhou Bay in August 2020 by Beihai Bureau of the Ministry of Natural Resources).

Figure 10. Comparison of simulation results based on MIKE 21-PLUS with actual results.

Figure 11. Comparison of simulation results between natural and artificial restoration scenarios in the restoration area.

Figure 12. Environmental drivers of the evolution of the distribution of Spartina alterniflora, Suaeda glauca Bunge, and Reed (Phragmites australis).

Figure 13. Comparison of F1-score for simulating vegetation distribution in the Yellow River estuary wetland in 2022 using MIKE21-PLUS and PLUS.

Table 1

Parameters of GF-5 hyperspectral images.

Parameter	Orbit (km)	Swath Width (km)	Spatial Resolution (m)	Number of Spectral Channels	Spectral Range (nm)	Spectral Resolution (nm)
Technical index	705	60	30	330	400–2500	VNIR: 5 nmSWIR: 10 nm

Table 2

Summary of data sources.

Model	Data	Spatial Resolution	Data Source
MIKE 21	Topography	15″	https://www.gebco.net/accessed on 30 April 2021
	Boundary condition: temperature, salinity	0.083°	https://resources.marine.copernicus.eu/accessed on 24 December 2022
PLUS	Land use condition (2018, 2020)	1 m	GF-5 HSI GF-1/2 PMS The fusion of panchromatic and hyperspectral images from GF-1, GF-2, and GF-5
	Land use condition (2022)	1 m	The multispectral images from GF-6
	DEM (elevation, aspect, slope)	30 m	https://www.gscloud.cn/searchaccessed on 20 February 2020
	Precipitation, air temperature, and wind speed	8 km	ERA5 https://www.gscloud.cn/searchaccessed on 9 August 2021
	Soil temperature, soil water content	8 km
	Total nitrogen and total phosphorus of soil	1 km	The National Earth System Science Data Center https://gre.geodata.cn/accessed on 7 August 2021
	pH and salinity of soil	/	Field survey sites
	Distance to river and coastline	30 m	Calculated from ArcGIS 10.6

Table 3

Field weights.

Spartina Alterniflora	Phragmites Australis	Tamarix Chinensis Lour	Suaeda Glauca Bunge	Phragmites Australis on the Tidal Flats	Naked Tide Beach	Salt Marsh	Water Body	Willow Forest	Road	Plow Land
0.056	0.130	0.032	0.152	0.039	0.145	0.292	0.101	0.031	0.004	0.017

Table 4

Changes in the area of each feature type in the Yellow River estuary wetlands, 2018–2022.

Particular Year	Vegetation Area (km2)
	Spartina Alterniflora	Phragmites Australis	Tamarix Chinensis Lour	Suaeda Glauca Bunge	Phragmites Australis on the Tidal Flats	Naked Tide Beach	Salt Marsh	Water Body	Willow Forest	Road	Plow Land
2018	41.71	37.71	24.52	23.77	19.37	101.34	17.79	247.94	2.94	2.88	1.25
2020	44.53	48.16	14.94	37.08	10.63	63.89	64.78	227.85	5.67	0.62	3.07
2022	30.77	34.32	37.98	22.89	18.81	109.45	29.49	223.94	2.79	0.89	0

Table 5

Comparison between the predicted and actual distribution of the MIKE21-PLUS model in 2022 by vegetation area (pixel).

Vegetation Type	Others	Spartina alterniflora	Phragmites australis	Tamarix chinensis Lour	Suaeda glauca Bunge	Phragmites australis on the Tidal Flats	Naked Tide Beach	Salt Marsh	Water Body	Willow Forest	Plow Land	Road	Recall
Others	48,285	530	1315	279	1180	166	1930	4876	1738	101	46	0	0.80
Spartina alterniflora	521	9831	0	0	0	0	0	0	0	0	0	0	0.95
Phragmites australis	1211	0	4577	0	0	0	0	0	0	0	0	0	0.79
Tamarix chinensis Lour	481	0	0	996	0	0	0	218	0	0	0	0	0.59
Suaeda glauca Bunge	1060	0	0	0	2702	0	0	0	0	0	0	0	0.72
Phragmites australis on the tidal flats	180	0	0	0	0	242	0	0	0	0	0	0	0.57
Naked tide beach	1630	0	0	22	0	0	7134	1278	0	0	0	0	0.71
Salt marsh	1088	0	0	0	0	0	0	7478	0	0	0	0	0.87
Water body	4541	0	7	0	0	0	524	94	50,173	0	0	0	0.91
Willow forest	111	0	0	0	0	0	0	0	0	305	0	0	0.73
Plow land	35	0	0	0	0	0	0	0	0	0	0	0	0
Road	0	0	0	0	0	0	0	0	0	0	0	618	0.89
Precision	0.82	0.95	0.78	0.77	0.70	0.59	0.74	0.54	0.97	0.75	0	0.93	-
F1-score	0.81	0.95	0.78	0.67	0.71	0.58	0.73	0.66	0.94	0.74	0	0.91

Table 6

Statistics on the area of each type of feature in the restoration area in its natural state and under artificial restoration scenarios.

Parameters	Vegetation Area (km2)
	Spartina alterniflora	Phragmites australis	Tamarix chinensis Lour	Suaeda glauca Bunge	Phragmites australis on the Tidal Flats	Naked Tide Beach	Salt Marsh	Water Body	Willow Forest	Road	Plow Land
2020 Natural	5.05	0.09	0.01	3.89	0.23	4.85	7.08	1.94	0.08	0	0
2022 Natural	5.06	0.09	0.01	3.90	0.23	3.09	9.79	0.97	0.07	0	0
2020 Artificial	0.34	0.09	0.01	6.02	0.23	8.03	7.02	1.38	0.08	0	0
2022 Artificial	0.69	0.09	0.01	6.09	0.16	5.62	10.02	0.47	0.06	0	0
The difference in area	−4.37	0	0	2.19	−0.07	2.53	0.23	−0.5	−0.01	0	0

Appendix A. Transfer Matrix

Parameters	Spartina alterniflora	Phragmites australis	Tamarix chinensis Lour	Suaeda glauca Bunge	Phragmites australis on the Tidal Flats	Naked Tide Beach	Salt Marsh	Water Body	Willow Forest	Road	Plow Land
Spartina alterniflora	1	0	0	0	0	1	1	0	0	0	0
Phragmites australis	1	1	0	0	0	1	1	0	0	0	0
Tamarix chinensis Lour	1	0	1	0	0	1	1	0	0	0	0
Suaeda glauca Bunge	1	0	0	1	0	1	1	0	0	0	0
Phragmites australis on the tidal flats	1	0	0	0	1	1	1	0	0	0	0
Naked tide beach	1	1	1	1	1	1	1	0	1	0	0
Salt marsh	1	1	1	1	1	1	1	0	1	0	0
Water body	1	1	1	1	1	1	1	1	1	0	0
Willow forest	1	0	0	0	0	1	1	0	1	0	0
Road	0	0	0	0	0	0	0	0	0	1	0
Plow land	0	0	0	0	0	0	0	0	0	0	1

References

1. Zhong, J.Q.; Gao, M.F.; Zhao, Y.Q.; Peng, S.; Wang, M.T. Spatial variation of the cultural ecosystem services pattern and influencing factors in the coastal areas: Evidence from Ramsar Sites in China. Acta Geogr. Sin.; 2024; 79, pp. 76-96.

2. Davidson, N.C.; Finlayson, C.M. Extent, regional distribution and changes in area of different classes of wetland. Mar. Freshw. Res.; 2018; 69, pp. 1525-1533. [DOI: https://dx.doi.org/10.1071/MF17377]

3. Han, G.X.; Niu, Z.G.; Luan, Z.Q.; Wang, G.; Zhang, L.; Guan, B. Construction of healthy wetland ecosphere in estuarine delta: Theory and method. Chin. J. Appl. Ecol.; 2018; 29, pp. 2787-2796. [DOI: https://dx.doi.org/10.13287/j.1001-9332.201808.036]

4. Reed, D.; Wang, Y.; Meselhe, E.; White, E. Modeling wetland transitions and loss in coastal Louisiana under scenarios of future relative sea-level rise. Geomorphology; 2019; 352, 106991. [DOI: https://dx.doi.org/10.1016/j.geomorph.2019.106991]

5. Assessment, M.E. Millennium Ecosystem Assessment; Millennium Ecosystem Assessment: Washington, DC, USA, 2001.

6. Pang, B.; Yang, W.; Cui, B.; Zhang, S.; Xie, T.; Ning, Z.; Gao, F.; Zhang, H. Evaluation of the effect of vegetation restoration in the yellow river delta wetland biodiversity conservation project. Environ. Eng.; 2023; 41, pp. 213-221.

7. Sui, H.; Ma, X.; Yan, J.; Sui, H.C.; Ma, X.; Yan, J.G.; Zhang, S.Y. Seed coating methods and availability for applying in the restoration of salt marsh vegetation in coastal wetlands. J. Nat. Resour.; 2019; 34, pp. 2601-2614.

8. Wang, Q.; Xie, T.; Luo, M.; Bai, J.; Chen, C.; Ning, Z.; Cui, B. How hydrological connectivity regulates the plant recovery process in salt marshes. J. Appl. Ecol.; 2021; 58, pp. 1314-1324. [DOI: https://dx.doi.org/10.1111/1365-2664.13879]

9. Sun, R.; Wang, D.; Cao, H.; Wang, Y.R.; Lu, Z.H.; Xia, J.B. Ecological pervious concrete in revetment and restoration of coastal Wetlands: A review. Constr. Build. Mater.; 2021; 303, 124590. [DOI: https://dx.doi.org/10.1016/j.conbuildmat.2021.124590]

10. Qiu, D.; Ma, X.; Yan, J.; Shao, D.; Bai, J.; Cui, B. Bio-geomorphological processes and structures facilitate seedling establishment and distribution of annual plants: Implications for coastal restoration. Sci. Total Environ.; 2021; 756, 143842. [DOI: https://dx.doi.org/10.1016/j.scitotenv.2020.143842] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33302077]

11. Kevan, J.M.; Alexia, M.K.; Miao, G.F. Micro-topography altershydrology, phenol oxidase activity and nutrient availability inorganic soils of a coastal freshwater forested wetland. Wetlands; 2019; 39, pp. 263-273.

12. Nima, S.; Bottacin-Busolin, A.; Tregnaghi, M.; Nepf, H.; Marion, A. Variation in contaminant removal efficiency in free-water surface wetlands with heterogeneous vegetation density. Ecol. Eng.; 2020; 143, 105662.

13. Ji, Z.G.; Jin, K.R. An integrated environmental model for a surface flow c.onstructed wetland: Water quality processes. Ecol. Eng.; 2016; 86, pp. 247-261. [DOI: https://dx.doi.org/10.1016/j.ecoleng.2015.09.018]

14. Chen, G.; Zhuang, H.; Liu, X. Cell-level coupling of a mechanistic model to cellular automata for improving land simulation. Giscience Remote Sens.; 2023; 60, 2166443. [DOI: https://dx.doi.org/10.1080/15481603.2023.2166443]

15. Li, Z.; Jiang, W.; Peng, K.; Wang, X.; Deng, Y.; Yin, X.; Ling, Z. Comparative analysis of land use change prediction models for land and fine wetland types: Taking the wetland cities Changshu and Haikou as examples. Landsc. Urban Plan.; 2024; 243, 104975. [DOI: https://dx.doi.org/10.1016/j.landurbplan.2023.104975]

16. Sun, X.; Liu, H. Identification of Factors Affecting Expansion of Spartina alterniflora and Their Intensities Using Markov-CA Model. J. Ecol. Rural. Environ.; 2014; 30, pp. 38-43.

17. Zhu, C.; Tian, B.; Zhou, Y.; Fan, Z.Q. Wetland Change Analysis and Forecasting in Pudong New Area Using Markov and CLUE-S Model. J. Fudan Univ. (Nat. Sci.); 2015; 54, pp. 431–438+448. [DOI: https://dx.doi.org/10.15943/j.cnki.fdxbjns.2015.04.006]

18. Zheng, W. Analysis on the Temporal and Spatial Distribution Characteristics and Driving Factors of Vegetation in Newly-Born Wetland of the Yellow River Delta. Ph.D. Thesis; Jiangsu Ocean University: Lianyungang, China, 2022; [DOI: https://dx.doi.org/10.44354/d.cnki.gjsuy.2022.000025]

19. Zhu, Z.; Qiu, S.; Ye, S. Remote sensing of land change: A multifaceted perspective. Remote Sens. Environ.; 2022; 282, 113266. [DOI: https://dx.doi.org/10.1016/j.rse.2022.113266]

20. Chai, B.; Li, P. An ensemble method for monitoring land cover changes in urban areas using dense Landsat time series data. ISPRS J. Photogramm. Remote Sens.; 2023; 195, pp. 29-42. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2022.11.002]

21. Gao, Y. The Impact of Artificial Water Replenishment on the Suitability of Breeding Habitat for Red Crowned Cranes in Zhalong Nature Reserve. Ph.D. Thesis; Harbin Normal University: Haerbin, China, 2021.

22. Chang, Y.; Chang, C.; Li, Y.; Liu, M.; Lv, J.J.; Hu, Y.M. Predicting Dynamics of the Potential Breeding Habitat of Larus saundersi by MaxEnt Model under Changing Land-Use Conditions in Wetland Nature Reserve of Liao he Estuary, China. Remote Sens.; 2022; 14, 552. [DOI: https://dx.doi.org/10.3390/rs14030552]

23. Zong, M.; Han, G.; LI, Y.; Wang, G.Z. Simulation of potential distribution of dominant plant communities in coastal wetlands of the Yellow River Delta based on MaxEnt mode. Chin. J. Appl. Ecol.; 2017; 28, pp. 1833-1842.

24. Convertino, M.; Muñoz-Carpena, R.; Chu-Agor, M.L.; Kiker, G.; Linkov, I. Untangling drivers of species distributions: Global sensitivity and uncertainty analyses of MaxEnt. Environ. Model. Softw.; 2014; 51, pp. 296-309. [DOI: https://dx.doi.org/10.1016/j.envsoft.2013.10.001]

25. Zhang, H.; Gao, H.; Ma, M.X.; Shao, F.Z.; Wang, Q.; Li, G.Y.; Qiu, J.; Zhou, K.X. The impact of roads on the breeding grounds of crested ibis based on Maxent model. J. Appl. Ecol.; 2017; 28, pp. 1352-1359.

26. Wolfram, S. Cellular Automata and Complexity: Collected Papers; CRC Press: Boca Raton, FL, USA, 2018.

27. Geng, J.; Shen, S.; Cheng, C.; Dai, K.X. A hybrid spatiotemporal convolution-based cellular automata model (ST-CA) for land-use/cover change simulation. Int. J. Appl. Earth Obs. Geoinf.; 2022; 110, 102789. [DOI: https://dx.doi.org/10.1016/j.jag.2022.102789]

28. Zhu, K.; Li, Y.; Zhou, M. Land use scenario simulation of the main city of Chongqing based on the CLUE-S model. Resour. Environ. Yangtze Basin; 2015; 24, pp. 789-797.

29. Qiu, Y.; Lu, J. Dynamic simulation of Spartina alterniflora based on CA-Markov model—A case study of Xiangshan bay of Ningbo City, China. Aquat. Invasions; 2018; 13, pp. 299-309. [DOI: https://dx.doi.org/10.3391/ai.2018.13.2.10]

30. Guo, H.; Cai, Y.; Yang, Z.; Zhu, Z.C.; Ouyang, Y.R. Dynamic simulation of coastal wetlands for Guangdong-Hong Kong-Macao Greater Bay area based on multi-temporal Landsat images and FLUS model. Ecol. Indic.; 2021; 125, 107559. [DOI: https://dx.doi.org/10.1016/j.ecolind.2021.107559]

31. Wu, D.; Zhu, K.; Zhang, S.; Huang, C.Q.; Li, J. Evolution analysis of carbon stock in Chengdu-Chongqing economic zone based on plus model and invest model. Ecol. Environ. Monit. Three Gorges; 2022; 7, pp. 85-96. [DOI: https://dx.doi.org/10.19478/j.cnki.2096-2347.2022.02.10]

32. Liang, X.; Guan, Q.; Clarke, K.C.; Liu, S.S.; Wang, B.Y.; Yao, Y. Understanding the drivers of sustainable land expansion using a patch-generating land use simulation (PLUS) model: A case study in Wuhan, China. Comput. Environ. Urban Syst.; 2021; 85, 101569. [DOI: https://dx.doi.org/10.1016/j.compenvurbsys.2020.101569]

33. Hu, F.; Zhang, Y.; Guo, Y.; Zhang, P.P.; Lv, S.; Zhang, C.C. Spatial and temporal changes in land use and habitat quality in the Weihe River Basin based on the PLUS and INVEST models and predictions. Arid. Land Geogr.; 2022; 45, pp. 1125-1136.

34. Li, C.; Gao, B.; Wu, Y.; Zheng, K.J.; Wu, Y. Dynamic simulation of landscape ecological risk in mountain towns based on PLUS model. J. Zhejiang AF Univ.; 2022; 39, pp. 84-94. [DOI: https://dx.doi.org/10.1631/jzus.B2100508]

35. Zhang, Y.; Chang, G.; Huang, B.; Xu, W.J.; Zhang, Z.Y.; Liu, Z.F.; Zhang, S.H.; Ye, Z.H. Research on the Configuration of Restoration Measures for Mountains, Rivers, Forests, Fields, Lakes, and Grasslands with a Focus on Reducing Nitrogen Load in the Upper Reaches of the North Canal Based on the SWAT Model. J. Environ. Sci.; 2022; 42, pp. 13-22.

36. Hou, X.; Zhou, Q.; Xing, B.; Zhao, T.Y.; Song, W.C. Hydrodynamic and Water Quality Model of Constructed Wetland Ecological Restoration in Guanhe River Basin. Yellow River; 2020; 42, pp. 68-72.

37. Zhu, Y. Remote-Sensed Monitoring and Analysis of Invasive Alien Species Spartina alterniflora in Shandong Province Based on Deep Learning Classification Method. Ph.D. Thesis; First Institute of Oceanography: Qingdao, China, 2020.

38. Zhang, S.; Xu, T.; Wang, H. Status quo of wetland conservation and new management ideas in the Yellow River Delta. Wetl. Sci. Manag.; 2022; 18, pp. 69-73.

39. Ge, Z.; Cao, H.; Zhang, L. A process-based grid model for the simulation of range expansion of Spartina alterniflora on the coastal saltmarshes in the Yangtze Estuary. Ecol. Eng.; 2013; 58, pp. 105-112. [DOI: https://dx.doi.org/10.1016/j.ecoleng.2013.06.024]

40. Zhang, S.; Liu, Y.; Liu, Y.; Shao, D.D.; Sun, L.M.; Zheng, S.Y. Cellular automata simulation of population expansion dynamics of Spartina alterniflora in the Yellow River Delta. J. Beijing Norm. Univ. (Nat. Sci. Ed.); 2021; 57, pp. 121-127.

41. Li, S.H.; Ge, Z.M.; Tan, L.S.; Zhou, K.; Hu, Z.J. Coupling Scirpus recruitment with Spartina control guarantees recolonization of native sedges in coastal wetlands. Ecol. Eng.; 2021; 166, 106246. [DOI: https://dx.doi.org/10.1016/j.ecoleng.2021.106246]

42. Liang, Y.D.; Zhang, S.Q.; Zhu, D.M.; Li, Y.P. Research on urban development boundary delineation using coupled Markov Chain and PLUS models from the perspective of main functions. J. Geo-Inf. Sci.; 2024; 26, pp. 1702-1716. [DOI: https://dx.doi.org/10.12082/dqxxkx.2024.240064]

43. Zhang, C.; Chen, S.; Li, P.; Liu, Q.L. Spatiotemporal dynamic remote sensing monitoring of typical wetland vegetation in the Current Huanghe River Estuary Reserve. J. Oceanogr.; 2022; 44, pp. 125-136.

44. Guo, B.; Liu, T.; Gu, Y. Integrating Coupled Dictionary Learning and Distance Preserved Probability Distribution Adaptation for Multispectral–Hyperspectral Image Collaborative Classification. IEEE Trans. Geosci. Remote Sens.; 2022; 60, 5528013. [DOI: https://dx.doi.org/10.1109/TGRS.2022.3169216]

45. Guo, B.; Liu, T.; Zhang, X.; Gu, Y. Few-Shot Multispectral-Hyperspectral Image Collaborative Classification with Feature Distribution Enhancement and Subdomain Alignment. IEEE Trans. Geosci. Remote Sens.; 2024; 62, 5505117. [DOI: https://dx.doi.org/10.1109/TGRS.2024.3502236]

46. Nash, J.E.; Sutcliffe, J.V. River flow forecasting through conceptual models-part I-A discussion of principles. J. Hydrol.; 1970; 10, pp. 282-290. [DOI: https://dx.doi.org/10.1016/0022-1694(70)90255-6]

47. Yi, Y.; Xie, H.; Song, J.; Yang, Z.F. Simulation of salt marsh vegetation community’s suitable habitat in Yellow River Estuary II: Application. J. Water Resour.; 2021; 52, pp. 401-408.

48. Su, L.; Ge, L.; Xia, J.B.; Sun, J.; Zhao, X.M. Evaluating Suaeda salsa restoration after micro-topography reconstruction in the coastal beach of the Yellow River Delta, China. J. Trans. Chin. Soc. Agric. Eng.; 2021; 37, pp. 82-90.

49. Brière, C.; Janssen, S.K.; Oost, A.P.; Taal, M.; Tonnon, P.K. Usability of the climate-resilient nature-based sand motor pilot, The Netherlands. J. Coast. Conserv.; 2018; 22, pp. 491-502. [DOI: https://dx.doi.org/10.1007/s11852-017-0527-3]