Ecosystem Health Assessment of the Liaohe Estuary Suaeda heteroptera Wetland Based on a Coupled PSR–Entropy Weight–PLSR Model

Abstract

The Suaeda heteroptera wetland in the Liaohe Estuary is a typical coastal wetland in northern China. This study presents a coupled PSR–entropy–PLSR model to assess ecosystem health and its driving factors, using long-term Landsat data from 1995 to 2024. The results show that the Ecosystem Health Index (EHI) dropped from 0.61 in 1995 to 0.20 in 2010, and then rebounded to 0.66 in 2024. The PLSR analysis identified four key drivers: Suaeda heteroptera carbon storage, mean patch area, aquaculture development intensity, and vegetation recovery rate. The simplified PLSR model constructed using these indicators achieved a cross-validation R² of 0.967. This coupled model provides a simple, efficient, and reliable method for the rapid assessment and long-term monitoring of coastal wetland ecosystem health.

1. Introduction

As transitional zones between terrestrial and marine ecosystems, coastal wetlands play a vital role in carbon sequestration, climate regulation, and the preservation of biodiversity. China has extensive and diverse wetlands, accounting for 10% of the global total and ranking first in Asia and fourth worldwide. However, over the past three decades, under the dual pressures of climate change and intensive human land development—including large-scale land reclamation, agricultural expansion, and urbanization in coastal areas—the natural characteristics of wetlands have been altered, resulting in reduced natural wetland area, disrupted ecological and hydrological processes, and elevated pollution levels. The Chinese government attaches great importance to wetland conservation and restoration. To this end, it has launched a series of ecological restoration projects, such as the Blue Bay Action, to promote the systematic restoration of typical coastal zones and coastal wetlands.

The Liaohe Estuary Wetland is one of the most representative coastal wetlands in northern China. Its halophytic vegetation, dominated by Suaeda heteroptera, forms a unique “red beach” landscape and plays a vital role in regional climate regulation, hydrological regulation, and the conservation of habitats for rare waterbirds. However, under the dual pressures of climate change and human activities, the Liaohe Estuary Wetland faces issues such as the degradation of Suaeda heteroptera communities, shrinking wetland areas, and increasing landscape fragmentation, with the health of the ecosystem under continuous threat. In response to habitat fragmentation and other issues in the Liaohe Estuary wetland ecosystem, the government has successively implemented ecological restoration projects such as converting farmland back to wetlands, replanting Suaeda heteroptera, and dredging tidal channels. As these restoration projects continue to advance, there is an increasingly urgent need to scientifically evaluate their effectiveness and systematically understand the wetland’s evolution process.

Ecosystem health assessment is an effective technical approach for identifying wetland degradation characteristics, revealing driving mechanisms, and formulating management strategies. Since the 1990s, frameworks such as VOR, PSR, and their derivatives have been widely applied in the field of wetland ecosystem health assessment. Although these methods differ in indicator selection and weighting approaches, they all reflect ecological pressures, system states, and restoration responses resulting from both human activities and natural disturbances, indicating that the indicator system approach can effectively reflect ecosystem health levels [1,2,3,4,5,6]. Compared with other frameworks, the PSR (Pressure–State–Response) model has a distinct advantage in that it systematically incorporates pressure indicators induced by human activities and the subsequent changes in ecological state. Moreover, its strong internal logic and clear causal relationships allow it to be flexibly adapted to the specific characteristics and research objectives of an ecosystem, thereby better revealing existing ecological problems. This makes the PSR model particularly suitable for human-dominated wetland ecosystems. However, as the number of indicators increases, existing methods commonly face issues such as highly subjective weight determination, pronounced indicator redundancy and multicollinearity, and increased complexity in result interpretation. To address these shortcomings, this study introduces Partial Least Squares Regression (PLSR). PLSR is a multivariate statistical method that extracts latent variables to maximize the covariance between independent and dependent variables, and it can effectively handle small samples, high-dimensional data, and collinearity. In the field of ecological assessment, PLSR has been used to identify key driving variables among multiple environmental factors [7]. At the same time, the Variable Importance in Projection (VIP) values derived from PLSR can objectively screen core indicators, enabling the construction of a simplified PLSR-based evaluation model and facilitating routine monitoring of coastal wetland ecosystem health.

Based on this, this study takes the Suaeda heteroptera wetland in the Liaohe Estuary as the study area. Integrating remote sensing interpretation, landscape pattern analysis, carbon storage estimation, and socio-economic statistical data, we constructed an ecosystem health assessment indicator system based on the PSR–entropy weight method and calculated the Ecosystem Health Index (EHI) from 1995 to 2024. On this basis, we further introduced PLSR coupled with the PSR–entropy weight method to identify key driving indicators affecting health changes and established a simplified PLSR evaluation model. This study aims to systematically study the change characteristics of the ecosystem health of the Liaohe Estuary coastal wetland from three aspects: health evolution process, core driving factors, and indicator simplification. The results provide a scientific basis for the precise conservation, ecological restoration, and sustainable management of the Suaeda heteroptera wetland and offer a methodological reference for the health assessment of similar coastal salt marsh ecosystems.

2. Materials and Methods

2.1. Study Area

The Liaohe Estuary Wetland is located at the mouth of the Liao River in Panjin City, Liaoning Province, with geographical coordinates of 121°28′–121°58′ E and 40°45′–41°05′ N (Figure 1). Situated in the northern part of the Liaodong Bay of the Bohai Sea and within the core area of the Liao River Delta, it represents a typical estuarine wetland type in northern China. It has been included in the United Nations’ List of Wetlands of International Importance [8]. It was included in the International List of Important Wetlands in 2005 and incorporated into the National Wetland Park development program in 2022 [9,10]. The wetland is dominated by Suaeda heteroptera, a typical halophytic vegetation, which, together with reed communities, forms a coastal salt marsh system, creating the distinctive “Red Beach” landscape. Additionally, the wetland provides a vital habitat and breeding ground for rare species such as the red-crowned crane, black-billed gulls, and spotted seals. It also fulfills ecological functions such as maintaining biodiversity, sequestering blue carbon, protecting the coastline, and purifying water quality [11,12].

The Liao River Estuary region entered a phase of large-scale oil development in the 1970s. Following the establishment of Panjin City in 1984, the intensity of human activities has continued to increase, and land use has gradually shifted from reed and grain cultivation to a diversified structure encompassing aquaculture, oil and gas extraction, and tourism [13]. Under multiple pressures from aquaculture, water conservancy projects, urban expansion, and tourism development, wetland fragmentation has intensified, Suaeda heteroptera communities have deteriorated, and ecosystem health faces sustained pressure [14]. Between 2010 and 2024, the government focused on suspending human activities that damage coastal wetlands. A series of coastal wetland restoration projects were launched on the eastern and western banks of the Liao River estuary. Large-scale aquaculture dikes were removed from both banks to address issues such as the overall degradation of the estuary’s coastal wetlands, obstructed hydrological connectivity, impaired wetland ecosystem functions, insufficient estuarine hydrodynamics, inadequate ecological water replenishment, damaged habitats for rare species, and reduced shoreline erosion protection capacity. These efforts aim to achieve the integration and connectivity of the Liao River wetlands.

2.2. Data Collection and Processing

2.2.1. Data Collection and Preprocessing

The remote sensing image data were obtained from the U.S. Geological Survey (USGS, https://earthexplorer.usgs.gov/). We selected 30 m resolution Landsat satellite images acquired between August and October in 1990, 1995, 2000, 2005, 2010, 2015, 2020 and 2024. All images had cloud cover below 10% across the study area (

Table 1. Information of Landsat Images.

Satellite	Sensor	Imaging Date	Spatial Resolution (m)	Orbit Number (Path/Row)
Landsat 5	TM	15 October 1990	30	120/032
Landsat 5	TM	26 August 1995	30	120/032
Landsat 5	TM	6 September 2000	30	120/032
Landsat 5	TM	6 September 2005	30	120/032
Landsat 5	TM	6 October 2010	30	120/032
Landsat 8	OLI	2 September 2015	30	120/032
Landsat 8	OLI	14 August 2020	30	120/032
Landsat 9	OLI	2 September 2024	30	120/032

). The 1990 data were used solely as the initial period for the transfer matrix and were not included in the analysis of other content in this paper. Preprocessing, including radiometric calibration, atmospheric correction, cropping, and image enhancement, was performed in ENVI 5.6. Meteorological data were obtained from the National Meteorological Science Data Center (Dawo Station, http://data.cma.cn/), while social statistics were sourced from the Panjin Municipal Bureau of Statistics and the Panjin Municipal Bureau of Culture, Tourism, and Radio and Television.

2.2.2. Land Cover Information Extraction Based on an Object-Oriented Classification Method

Using an object-oriented classification method to extract land cover information from the Suaeda heteroptera wetland, this study employed multi-scale image segmentation combined with multi-dimensional features and the Random Forest (RF) algorithm to classify land cover from time-series imagery of the Liaohe Estuary wetland from 1990 to 2024. The segmentation parameters in this study were optimized through human–computer interaction and repeated trial optimization, to ensure spectral homogeneity of patches and well-fitted boundaries, and were set as follows: scale parameter = 70, shape factor = 0.3, compactness factor = 0.5. This parameter combination ensures spectral homogeneity within Suaeda heteroptera, tidal flats, and aquaculture embankments while obtaining objects with relatively smooth boundaries.

Based on common wetland classification systems [15,16,17,18,19], the actual conditions of the study area, and the specific objectives of this study, the Liaohe Estuary wetlands were classified into three major categories: constructed wetlands, natural wetlands, and other land uses. These were further subdivided into six subcategories: aquaculture embankments, riverine ponds, reed wetlands, Suaeda heteroptera wetlands, tidal flats, and built-up areas. The overall classification accuracy and Kappa coefficient ranged from 80.56 to 87.43% and from 77.32 to 83.16%, respectively (

Table 2. Validation of classification accuracy.

Year	Overall Accuracy (%)	Kappa Coefficient (%)
1990	80.56	78.52
1995	83.53	80.16
2000	85.32	82.36
2005	82.13	78.25
2010	81.06	77.32
2015	85.23	80.25
2020	86.68	82.35
2024	87.43	83.16

), indicating that the classification results are reliable and sufficient to meet the research requirements for subsequent landscape pattern analysis and ecosystem health assessment. The selected feature types are shown in

Table 3. Feature Selection Based on an Object-Oriented Classification Method.

Feature Type	Feature Name	Number of Features
Spectral Characteristics	Mean B, Mean G, Mean R, Mean SWIR 1, Mean SWIR 2 Standard deviation B, Standard deviation G, Standard deviation R, Standard deviation SWIR 1, Standard deviation SWIR 2	10
Geometric Characteristics	Length/Width, Roundness, Shapeindex	3
Vegetation Index	NDVI, SAVI	2
Building Index	NDBI	1
Water Index	NDWI	1

, and the analysis results are presented in Figure 2.

2.2.3. Extraction of Landscape Pattern Indices and Carbon Storage Estimation of the Suaeda heteroptera Wetland

Landscape pattern indices can reflect the composition and spatial configuration of a landscape in terms of patch area, number, distribution, aggregation, and spatial relationships. In this study, using Fragstats 4.3 software, we analyzed the characteristics of landscape pattern changes in the Liaohe Estuary wetlands by selecting the following indicators at both the patch and landscape levels: total patch area (CA), proportion of landscape area occupied by patches (PLAND), number of patches (NP), mean patch area (AREA_MN), as well as the aggregation index (AI), Shannon diversity index (SHDI), and edge density (ED).

We used the Transformed Soil Adjusted Vegetation Index (TSAVI) to build a remote sensing inversion model for the aboveground biomass of Suaeda heteroptera (Equations (1) and (2)). An allometric growth model established in previous studies based on field measurement data from Suaeda heteroptera plots in the Shuangtaizi Estuary (Liaohe Estuary) was adopted to estimate belowground biomass from aboveground biomass [20] (Equation (3)). After obtaining aboveground and belowground biomass, the biomass was converted to carbon storage using measured carbon content parameters. Based on field measurements of Suaeda heteroptera conducted in the Shuangtaizi Estuary from 2015 to 2016, we adopted the two-year average values as carbon content coefficients. The aboveground and belowground carbon content coefficients were 25.96% and 32.15%, respectively [21] (Equation (4)). The total carbon storage of Suaeda heteroptera was further calculated by multiplying the carbon storage per unit area by the area of a single pixel and the total number of pixels (Equation (5)).

$$\mathit{T}\mathit{S}\mathit{A}\mathit{V}\mathit{I}={\displaystyle \frac{\mathit{a}\left({\mathit{R}}_{\mathit{N}\mathit{I}\mathit{R}}-\mathit{a}{\mathit{R}}_{\mathit{R}\mathit{e}\mathit{d}}-\mathit{b}\right)}{\mathit{a}{\mathit{R}}_{\mathit{N}\mathit{I}\mathit{R}}+{\mathit{R}}_{\mathit{R}\mathit{e}\mathit{d}}-\mathit{a}\mathit{b}}}$$

(1)

$${\mathit{Y}}_{\mathit{A}\mathit{G}}=\mathbf{1.141}\times \mathit{T}\mathit{S}\mathit{A}\mathit{V}\mathit{I}-\mathbf{0.107}$$

(2)

$${\mathit{Y}}_{\mathit{B}\mathit{G}}=\mathbf{0.205}{\mathit{Y}}_{\mathit{A}\mathit{G}}^{\mathbf{0.979}}$$

(3)

$${\mathit{C}}_{\mathit{v}\mathit{e}\mathit{g}}={\mathit{Y}}_{\mathit{A}\mathit{G}}\times \mathbf{0.2596}+{\mathit{Y}}_{\mathit{B}\mathit{G}}\times \mathbf{0.3215}$$

(4)

$${\mathit{C}}_{\mathit{t}\mathit{o}\mathit{t}\mathit{a}\mathit{l}}={\mathit{C}}_{\mathit{v}\mathit{e}\mathit{g}}\times \mathbf{900}\times \mathit{N}$$

(5)

In the equations, $R_{Red}$ and $R_{NIR}$ represent the reflectance in the near-infrared and red bands, respectively; a and b are the soil line coefficients. Referring to the study by Li Wei et al. on biomass inversion of Suaeda heteroptera in the Liaohe Estuary, a is taken as 0.791 and b as 0.043 [22]. $Y_{AG}$ represents aboveground biomass, and the constants 1.141 and 0.107 were obtained from a comparative analysis between field-measured data and multiple models based on the study by Mou Meng [23]. $Y_{BG}$ represents belowground biomass. $C_{veg}$ is carbon storage per unit area. $C_{total}$ is the total carbon storage of Suaeda heteroptera. N represents the number of pixels. The spatial resolution of the Landsat imagery is 30 m, and the area of a single pixel is 900 m².

2.3. Development of an Ecosystem Health Assessment Indicator System

2.3.1. Principles and Construction of the Evaluation Indicator System

As a typical composite habitat where land and sea meet, wetland ecosystems are characterized by multidimensionality and the interplay of multiple factors. The purpose of conducting wetland ecosystem health assessments is to quantitatively identify external pressures on the ecosystem, its current state, and its response to changes, thereby providing a basis for decision-making regarding wetland conservation, restoration, and management. Taking into account the regional ecological characteristics, study period, and data conditions of the Suaeda heteroptera wetland at the Liao River estuary, this study primarily adhered to the principles of scientific rigor, systematicness, representativeness and regional adaptability when constructing the ecosystem health assessment indicator system [24,25,26,27]. Additionally, drawing on existing research findings [28], a total of 11 assessment indicators were ultimately selected, comprising 5 pressure-level indicators, 4 state-level indicators, and 2 response-level indicators. Based on the direction of each indicator’s impact on ecosystem health, they were classified into positive and negative indicators. During the indicator selection process, this study examined the correlations among indicators in light of their ecological significance and conducted comparative analyses of indicators conveying similar information to avoid significant redundancy in the evaluation system. By comprehensively considering the dominant ecological processes and evaluation objectives in the study area, indicators with strong representativeness and good explanatory power were ultimately retained to form the evaluation indicator system for this study.

The pressure layer indicators are used to characterize the external pressures exerted on the Liaohe Estuary Suaeda heteroptera wetland ecosystem by human activities and environmental fluctuations. A total of five indicators were selected for this layer, reflecting the degree of external disturbances on the wetland in terms of tourism, aquaculture development, resource extraction, and climatic disturbances. The condition layer indicators are used to characterize the current state of the Liaohe Estuary Suaeda heteroptera wetland ecosystem in terms of landscape structure and ecological functions. This layer comprises four indicators that reflect the state of the wetland ecosystem in the study area in terms of patch scale, spatial aggregation characteristics, landscape heterogeneity, and carbon storage. The response layer indicators are used to characterize the dynamic feedback processes of the Liao River Estuary Suaeda heteroptera wetland under conditions of natural recovery and artificial restoration. A total of two indicators were selected for this layer, reflecting the ecosystem’s response to external disturbances and management measures through the restoration of Suaeda heteroptera habitats and the process of converting farmland back to wetlands. The calculation formulas for some of these indicators are as follows:

(1) Average temperature: The temperature anomaly is used to characterize the degree of deviation of the annual temperature from the long-term average for the study period. It is calculated using the following formula:

$$\Delta \mathit{T}={\mathit{T}}_{\mathit{i}}-\overline{\mathit{T}}$$

(6)

In the equation, $T_i$ represents the average temperature in year i; $\overline{T}$ represents the long-term average temperature over the study period.

(2) Precipitation anomaly: Precipitation anomalies are used to characterize the degree of deviation in precipitation from the long-term average for the study period. The formula is as follows:

$$\Delta \mathit{P}={\mathit{P}}_{\mathit{i}}-\overline{\mathit{P}}$$

(7)

In the equation, $P_i$ represents the annual precipitation in year i; $\overline{P}$ represents the long-term average precipitation.

(3) Suaeda heteroptera recovery rate: The Suaeda heteroptera recovery rate is used to characterize the proportion of bare sand flats that have converted to Suaeda heteroptera during the study period, reflecting the expansion and recovery processes of Suaeda heteroptera communities. The calculation formula is as follows:

$${\mathit{C}}_{\mathit{t}\to \mathit{c}}={\displaystyle \frac{\Delta {\mathit{S}}_{\mathit{t}\to \mathit{c}}}{{\mathit{s}}_{\mathit{t}}\times \Delta \mathit{t}}}$$

(8)

In the equation, $\Delta S_{t\to c}$ represents the area of tidal flats converted to Suaeda heteroptera; $S_t$ represents the initial total area of tidal flats; and $\Delta t$ represents the number of years elapsed.

(4) Rate of decommissioning: The rate of decommissioning is used to characterize the rate of change in the area of aquaculture enclosures per unit of time. A positive value indicates an acceleration in the process of decommissioning enclosures and restoring wetlands, while a negative value indicates the expansion of aquaculture enclosures. The calculation formula is as follows:

$$\mathit{V}={\displaystyle \frac{{\mathit{S}}_{\mathbf{1}}-{\mathit{S}}_{\mathbf{2}}}{\Delta \mathit{t}}}$$

(9)

In the equation, $S_1$ and $S_2$ represent the areas of the aquaculture enclosures at the beginning and end of the study, respectively; $\Delta t$ represents the number of years between the two time points.

The indicator system for assessing ecosystem health in the study area is shown in

Table 4. Evaluation System for Suaeda heteroptera Wetland Indicators at the Liao River Estuary.

PSR Category	Code	Indicator Name	Unit	Indicator Direction
Pressure (P)	P1	Number of tourists	thousands of visitors	Negative (−)
	P2	Aquaculture development intensity	km2	Negative (−)
	P3	Electricity consumption in the oil and natural gas extraction industry	10,000 kWh	Negative (−)
	P4	Temperature deviation from normal	°C	Negative (−)
	P5	Precipitation deviation from normal	mm	Negative (−)
Status (S)	S1	Mean patch area of Suaeda heteroptera	hm2	Positive (+)
	S2	Aggregation index of Suaeda heteroptera	%	Positive (+)
	S3	Shannon diversity index	--	Positive (+)
	S4	Suaeda heteroptera carbon storage	×106 kg	Positive (+)
Response (R)	R1	Suaeda heteroptera recovery rate	%	Positive (+)
	R2	Aquaculture withdrawal rate	km2/a	Positive (+)

.

2.3.2. Classification of Ecosystem Health Levels

After establishing the evaluation indicator system, it is necessary to classify the ecosystem health indices obtained from subsequent calculations into grades to compare the characteristics of changes in the health levels of the Suaeda heteroptera wetland ecosystem at the Liao River estuary across different years. Referring to domestic and international wetland evaluation standards [29,30], this study employs an equidistant grading method within the real number interval [0, 1] to classify the ecosystem health of the Liaohe Estuary Suaeda heteroptera wetland into five levels: Unhealthy, Generally Unhealthy, Fairly Healthy, Healthy, and Very Healthy. When the ecosystem health index is 1, it corresponds to the optimal state of wetland ecological health; when the ecosystem health index is 0, it corresponds to the worst state of ecosystem health. The specific classification is shown in

Table 5. Classification of Wetland Health Levels.

Grade	Health Assessment Index	Wetland Ecological Health Status
Unhealthy	EHI ≤ 0.2	The natural state of the wetland ecosystem has been severely disrupted; the system exhibits low vitality and significant structural imbalance, making it difficult for its ecological functions to operate normally. The Suaeda heteroptera wetland habitat has deteriorated markedly, with severe landscape fragmentation, poor system stability, high sensitivity to external disturbances, and insufficient resilience. Under the continuous influence of human disturbance and other pressures, ecological abnormalities are particularly pronounced, and the ecosystem is in a state of marked or severe degradation.
Generally Unhealthy	0.2 < EHI ≤ 0.4	The natural state of the wetland ecosystem has been significantly disrupted; its vitality is low, the integrity of its structural organization has declined, and its ability to perform ecological functions is limited. The quality of the Suaeda heteroptera wetland habitat has deteriorated, the landscape pattern has become increasingly fragmented, and the ecosystem’s resilience to external pressures and capacity for recovery are weak. The impacts of human disturbance and environmental stressors are pronounced, ecological anomalies are on the rise, and the ecosystem is already showing signs of degradation.
Fairly Healthy	0.4 < EHI ≤ 0.6	The natural state of the wetland ecosystem has been somewhat affected, resulting in periodic fluctuations in its structure and function, and a decline in stability. The landscape pattern of the Suaeda heteroptera wetland has undergone certain changes; in some areas, habitat contraction, increased fragmentation, or uneven recovery processes may be observed. The impacts of human disturbance and environmental fluctuations are relatively pronounced; while the system can still maintain its basic ecological functions, its sensitivity has increased and its resilience has relatively weakened.
Healthy	0.6 < EHI ≤ 0.8	The wetland ecosystem is generally in good natural condition, with high levels of system vitality and organizational structure, and its primary ecological functions are performed relatively stably. The wetland landscape pattern is generally well-structured, the Suaeda heteroptera habitat is well-maintained, and the system exhibits a certain degree of resilience and recovery capacity. External pressures are minimal or manageable; while localized changes exist, overall stability is good, and the ecosystem is generally in a stable state.
Very Healthy	0.8 < EHI ≤ 1.0	The wetland ecosystem as a whole remains in a relatively good natural state, exhibiting strong vitality, a relatively intact structural organization, and stable ecological functions. The Suaeda heteroptera wetland habitat is well-preserved, with a reasonably balanced landscape pattern, and the system demonstrates a strong capacity to adapt to and recover from external disturbances. Human disturbance is relatively minimal, ecological anomalies are not evident, and the ecosystem is generally in a stable and sustainable condition.

.

2.4. Calculation of the PSR–Entropy Health Index

2.4.1. Standardization of Evaluation Metrics

To eliminate dimensional differences and magnitude disparities among all indicators, and to standardize the direction of the indicators, this study employed the range-normalization method to convert the raw indicator data into dimensionless form. For indicators positively correlated with wetland ecosystem health, a positive normalization method is applied; for those negatively correlated, an inverse normalization method is applied, thereby ensuring consistency in the direction of the indicators and enabling comparability. Let the matrix of raw data be $Z_{ij}$. For positive indicators:

$${\mathit{Z}}_{\mathit{i}\mathit{j}}={\displaystyle \frac{{\mathit{Z}}_{\mathit{i}\mathit{j}}-\mathit{m}\mathit{i}\mathit{n}\left({\mathit{Z}}_{\mathit{j}}\right)}{\mathit{m}\mathit{a}\mathit{x}\left({\mathit{Z}}_{\mathit{j}}\right)-\mathit{m}\mathit{i}\mathit{n}\left({\mathit{Z}}_{\mathit{j}}\right)}}$$

(10)

Regarding negative indicators:

$${\mathit{Z}}_{\mathit{i}\mathit{j}}={\displaystyle \frac{\mathit{m}\mathit{a}\mathit{x}\left({\mathit{Z}}_{\mathit{j}}\right)-{\mathit{Z}}_{\mathit{i}\mathit{j}}}{\mathit{m}\mathit{a}\mathit{x}\left({\mathit{Z}}_{\mathit{j}}\right)-\mathit{m}\mathit{i}\mathit{n}\left({\mathit{Z}}_{\mathit{j}}\right)}}$$

(11)

In the equation, $Z_{ij}$ represents the dimensionless standardized value of the j-th evaluation indicator in the i-th year; $Z_{ij}$ represents the raw observed physical value of that indicator; $max\left(Z_j\right)$ and $min\left(Z_j\right)$ represent the maximum and minimum values of this indicator, respectively, within the study period from 1995 to 2024. After standardization, the values of each indicator are converted to the interval [0, 1]. To ensure accuracy, a minimum value of 0.001 is added to any value of 0.

2.4.2. Calculation of Evaluation Indicator Weights

The entropy method is an objective weighting approach based on the concept of information entropy. In multi-criteria evaluation, the entropy method calculates the information entropy of each indicator across the evaluation objects to reflect the degree of dispersion of the indicator data and its effective information content, and determines the weights accordingly [31,32]. Generally, a larger variation of an indicator corresponds to lower information entropy. This indicates that the indicator provides a greater amount of useful information, plays a stronger role in distinguishing the comprehensive evaluation results, and thus corresponds to a higher weight; conversely, if the differences in the indicator are small, the information entropy is larger, and the weight is relatively lower [33]. The entropy method can reduce human interference to a certain extent and is suitable for the data characteristics of multi-indicator, long-term time-series ecosystem health assessments [34,35]. The steps for weight calculation using the entropy method are: constructing a proportion matrix, calculating information entropy, calculating the coefficient of variation, and calculating weights. The results of the weight calculation for the evaluation system are shown in

Table 6. Results of Weight Calculations for the Evaluation System.

PSR Category	Code	Metric Name	Layer Weight	Total Weight	Rank
Pressure (P)	P1	Number of tourists	0.04551	0.30642	11
	P2	Aquaculture development intensity	0.07730		5
	P3	Electricity consumption in the oil and natural gas extraction industry	0.05603		10
	P4	Temperature deviation from normal	0.06577		7
	P5	Precipitation deviation from normal	0.06181		9
Status (S)	S1	Mean patch area of Suaeda heteroptera	0.15882	0.43046	2
	S2	Aggregation index of Suaeda heteroptera	0.10121		3
	S3	Shannon diversity index	0.09411		4
	S4	Suaeda heteroptera carbon storage	0.07632		6
Response (R)	R1	Suaeda heteroptera recovery rate	0.20026	0.26312	1
	R2	Aquaculture withdrawal rate	0.06286		8

.

Table 6 provides the objective weights for each evaluation indicator. A higher weight value indicates that the indicator contributes more information to the time-series evaluation covering the period from 1995 to 2024.

2.4.3. Calculation of the Ecosystem Health Index

Based on the weighted calculation results, the Ecosystem Health Index (EHI) for the Suaeda heteroptera wetlands at the Liao River estuary from 1995 to 2024 was calculated using the following formula.

$${\mathit{E}\mathit{H}\mathit{I}}_{\mathit{i}}={\displaystyle \sum _{\mathit{j}=1}^{\mathit{n}}}{\mathit{W}}_{\mathit{j}}{\mathit{Z}\prime }_{\mathit{i}\mathit{j}}$$

(12)

In the equation, ${EHI}_i$ represents the ecosystem health index for the i-th year; $W_j$ represents the entropy weight for the j-th indicator; and $Z_{ij}^{\prime }$ prime represents the standardized value of the j-th indicator for the i-th year.

2.5. Calculation of the PLSR-Based Health Index

Partial Least Squares Regression (PLSR) is a regression analysis method suitable for small sample sizes, multiple variables, and conditions where correlations exist among independent variables. This method compresses the information from the independent variables into a small number of new synthetic components by extracting a limited set of latent variables, and then establishes a regression relationship between the independent and dependent variables. Compared with ordinary multiple linear regression, PLSR has better applicability under conditions of relatively few samples, a large number of indicators, and potential multicollinearity among variables [36,37,38].

To analyze the relationship between each evaluation indicator and changes in the Ecosystem Health Index (EHI), identify key factors affecting wetland health changes, and construct a simplified representation model suitable for routine monitoring, this section introduces Partial Least Squares Regression (PLSR) for analysis. First, an initial full-indicator PLSR model was constructed based on all 11 indicators of the PSR framework. Key indicators were comprehensively screened by calculating Variable Importance in Projection (VIP) values and considering the ecological significance of the indicators as well as the data availability for subsequent monitoring. This screening aimed to eliminate redundant variables and lay the foundation for the subsequent construction of a simplified PLSR model. Finally, the PLSR identification results were compared with the entropy weight method results. The steps are shown in Figure 3.

2.5.1. Construction of the Full-Indicator PLSR Model

This study constructed and calculated the PLSR model using MATLAB R2022b, primarily employing the ‘plsregress’ function for model training. The initial PLSR model was constructed with the Ecosystem Health Index (EHI), calculated using the PSR–entropy method, as the dependent variable and 11 evaluation indicators as independent variables. The independent variables include tourist numbers, aquaculture development intensity, electricity consumption in the oil and gas extraction industry, temperature anomalies, precipitation anomalies, Mean patch area of Suaeda heteroptera, Aggregation index of Suaeda heteroptera index, Shannon diversity index, Suaeda heteroptera carbon storage, Suaeda heteroptera recovery rate, and aquaculture conversion rate. The dataset consists of seven time periods: 1995, 2000, 2005, 2010, 2015, 2020, and 2024. Due to differences in the units and orders of magnitude of the evaluation indicators, they were uniformly standardized in MATLAB. The basic form of the PLSR model can be expressed as:

$$\mathit{Y}=\mathit{X}\mathit{B}+\mathit{E}$$

(13)

In this equation, X represents the standardized matrix of independent variables, Y represents the vector of dependent variables, B represents the matrix of regression coefficients, and E represents the residuals. PLSR constructs a model by extracting latent variables from X. These latent variables must retain as much information from the original independent variables as possible while maintaining a strong correlation with the dependent variables.

The number of latent variables is a key parameter in PLSR modeling. Too few latent variables can result in insufficient information extraction, while too many may reduce the model’s stability. This study employs Leave-One-Out Cross Validation (LOOCV) to determine the number of latent variables, calculates the prediction error under different conditions of latent variable counts, and selects the number of latent variables yielding the smallest cross-validation error as the model parameter. Model evaluation metrics include Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). The calculation formulas are as follows:

$$\mathit{R}\mathit{M}\mathit{S}\mathit{E}=\sqrt{{\displaystyle \frac{\mathbf{1}}{\mathit{N}}}{\displaystyle \sum _{\mathit{i}=\mathbf{1}}^{\mathit{N}}}{\left({\mathit{y}}_{\mathit{i}}-{\widehat{\mathit{y}}}_{\mathit{i}}\right)}^{\mathbf{2}}}$$

(14)

$$\mathit{M}\mathit{A}\mathit{E}={\displaystyle \frac{\mathbf{1}}{\mathit{N}}}{\displaystyle \sum _{\mathit{i}=\mathbf{1}}^{\mathit{N}}}\left|{\mathit{y}}_{\mathit{i}}-{\widehat{\mathit{y}}}_{\mathit{i}}\right|$$

(15)

$${\mathit{R}}^{\mathbf{2}}=\mathbf{1}-{\displaystyle \frac{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{N}}{\left({\mathit{y}}_{\mathit{i}}-{\widehat{\mathit{y}}}_{\mathit{i}}\right)}^{\mathbf{2}}}{{\sum }_{\mathit{i}=\mathbf{1}}^{\mathit{N}}{\left({\mathit{y}}_{\mathit{i}}-\overline{\mathit{y}}\right)}^{\mathbf{2}}}}$$

(16)

In these equations, RMSE denotes the root mean square error, MAE denotes the mean absolute error, R² denotes the coefficient of determination, N denotes the number of samples, $y_i$ denotes the target value for the i-th sample, ${\widehat{y}}_i$ denotes the model-predicted value for the i-th sample, and $\overline{y}$ denotes the mean of the target value sequence.

As shown in

Table 7. Cross-validation results for the full-feature PLSR model with different numbers of latent variables.

Number of Latent Variables	RMSE	MAE	R2
1	0.1432	0.1246	0.1685
2	0.1103	0.0875	0.5071
3	0.102	0.0766	0.5779
4	0.1019	0.076	0.5787
5	0.1043	0.0772	0.5588
6	0.1043	0.0772	0.5588

, as the number of latent variables increases, the model’s cross-validation error generally follows a trend of first decreasing and then stabilizing. When the number of latent variables is 4, both RMSE and MAE reach their minimum values, and R² is at its highest; therefore, the number of latent variables for the PLSR model of all PSR indicators is set to 4.

2.5.2. Identification of Key Indicators

After constructing the full-variable PLSR model, this study employed Variable Importance in Projection (VIP) values to identify indicators that contribute significantly to changes in ecosystem health [39]. VIP values comprehensively reflect the contribution of each variable to latent variable extraction and the explanation of the response variable. Referring to common empirical criteria in existing studies [40,41,42], variables with VIP > 1 are typically considered important variables, as they are believed to make a strong contribution to model interpretation; variables with 0.8 ≤ VIP < 1 are regarded as candidate variables with some explanatory power, and their retention can be further determined based on ecological significance, indicator representativeness, and data availability; variables with VIP < 0.8 are generally considered to have a relatively weak contribution. The calculation formula is as follows:

$${\mathit{V}\mathit{I}\mathit{P}}_{\mathit{j}}=\sqrt{\mathit{p}·{\displaystyle \frac{{\sum }_{\mathit{a}=\mathbf{1}}^{\mathit{A}}{\mathit{S}\mathit{S}\mathit{Y}}_{\mathit{a}}\left(\frac{{\mathit{w}}_{\mathit{j}\mathit{a}}^{\mathbf{2}}}{{\displaystyle {\sum }_{\mathit{j}=\mathbf{1}}^{\mathit{p}}}{\mathit{w}}_{\mathit{j}\mathit{a}}^{\mathbf{2}}}\right)}{{\sum }_{\mathit{a}=\mathbf{1}}^{\mathit{A}}{\mathit{S}\mathit{S}\mathit{Y}}_{\mathit{a}}}}}$$

(17)

In the equation, ${VIP}_j$ represents the variable importance projection value for the j-th indicator, $p$ is the number of independent variables, A is the number of latent variables, ${SSY}_a$ is the sum of squares explained by the a-th latent variable for the dependent variable, and $w_{ja}$ is the weight of the jth indicator on the a-th latent variable.

As shown in

Table 8. VIP Values for Each Feature in the Full-Feature PLSR Model.

Rank	Metric Name	VIP Value
1	Suaeda heteroptera carbon storage	1.6669
2	Mean patch area of Suaeda heteroptera	1.5359
3	Aquaculture development intensity	1.4868
4	Suaeda heteroptera recovery rate	1.1677
5	Shannon diversity index	0.8991
6	Number of tourists	0.745
7	Electricity consumption in the oil and gas extraction industry	0.5563
8	Aquaculture withdrawal rate	0.4888
9	Aggregation index of Suaeda heteroptera	0.4648
10	Temperature anomaly	0.2903
11	Precipitation anomaly	0.2761

and Figure 4, the VIP values for the four indicators—Suaeda heteroptera carbon storage, mean patch area, aquaculture development intensity, and recovery rate—are all greater than 1, indicating that they are key factors driving changes in the EHI. The VIP value for the Shannon diversity index is close to 1, suggesting it also has a significant influence. The remaining indicators have relatively low VIP values and thus contribute limited explanatory power to the model. Therefore, Suaeda heteroptera carbon storage, mean patch area, aquaculture development intensity, Suaeda heteroptera recovery rate, and the Shannon diversity index were selected as the indicators for the subsequent simplified PLSR model.

2.5.3. Construction of the Simplified-Indicator PLSR Model

Although the full-variable PLSR model can reflect the overall relationship between each indicator and the Ecosystem Health Index (EHI), an excessive number of input variables in a limited sample size increases model complexity, thereby reducing the interpretability and practical value of the results. Therefore, this study constructs a simplified PLSR model based on the identified key indicators, aiming to reduce dimensionality while retaining core information, thereby enhancing the model’s stability and practicality.

The selection of indicators for the simplified model was based on the VIP values from the full-indicator PLSR analysis, retaining variables with strong explanatory power for the EHI [30]: Suaeda heteroptera carbon storage, mean patch area, aquaculture development intensity, Suaeda heteroptera recovery rate, and the Shannon diversity index. According to the PSR framework, the above indicators correspond to pressure (aquaculture development), state (carbon storage, patch area, and diversity), and response (recovery rate), respectively, covering the three dimensions of pressure, state, and response. They collectively characterize the key features of the ecological health changes of the Suaeda heteroptera wetland in the Liaohe Estuary.

For modeling, the EHI was used as the dependent variable and the five screened indicators as independent variables, and the analysis was performed using the ‘plsregress’ function in MATLAB R2022b. The independent variables were standardized and normalized using the same methods as in the full-indicator model. The number of latent variables was determined using leave-one-out cross-validation: by comparing the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²) for different numbers of latent variables, we selected the parameter combination with the best cross-validation performance. The results are shown in

Table 9. Cross-validation results for the simplified PLSR model with different numbers of latent variables.

Number of Latent Variables	RMSE	MAE	R2
1	0.0635	0.0532	0.8365
2	0.0303	0.0273	0.9627
3	0.0285	0.0224	0.967
4	0.0587	0.0507	0.8603
5	0.1706	0.1402	−0.1794

.

As shown in Table 9, when the number of latent variables is 3, the simplified PLSR model exhibits the smallest cross-validation error and the highest coefficient of determination; therefore, the number of latent variables for this model is set to 3.

Based on this, a simplified PLSR model is constructed, whose general form can be expressed as:

$$\mathit{Y}={\mathit{b}}_{\mathbf{0}}+{\mathit{b}}_{\mathbf{1}}{\mathit{X}}_{\mathbf{1}}+{\mathit{b}}_{\mathbf{2}}{\mathit{X}}_{\mathbf{2}}+{\mathit{b}}_{\mathbf{3}}{\mathit{X}}_{\mathbf{3}}+{\mathit{b}}_{\mathbf{4}}{\mathit{X}}_{\mathbf{4}}+{\mathit{b}}_{\mathbf{5}}{\mathit{X}}_{\mathbf{5}}$$

(18)

In the equation, Y represents the ecosystem health index, X₁–X₅ represent aquaculture development intensity, Mean patch area of Suaeda heteroptera, carbon storage of Suaeda heteroptera, Suaeda heteroptera recovery rate, and Shannon diversity index, respectively, and b₀ is the constant term. b₁ to b₅ are the regression coefficients corresponding to each indicator. This simplified model is used to fit changes in the ecosystem health index; the regression coefficients of the simplified PLSR model are shown in

Table 10. Regression coefficients of the simplified PLSR model.

Variable	Regression Coefficient
Constant terms	0.4386
Suaeda heteroptera carbon storage	0.05213
Average Suaeda heteroptera patch area	0.002584
Aquaculture development intensity	−0.00421
Suaeda heteroptera recovery rate	0.003107
Shannon diversity index	0.178093

.

3. Results

3.1. Results of the Health Assessment of the Suaeda heteroptera Wetland Ecosystem at the Liao River Estuary

To evaluate the fitting performance of the simplified PLSR model for changes in the Ecosystem Health Index (EHI), this study employed two evaluation methods: full-sample fitting and cross-validation using the leave-one-out method. The former measures the model’s explanatory power for the existing data, while the latter tests its stability under small-sample conditions. Evaluation metrics included the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The fitting and cross-validation results for the simplified PLSR model are presented in

Table 11. Fitting and Cross-Validation Results for the Simplified PLSR Model.

Evaluation Methods	RMSE	MAE	R2
Full-sample fitting	0.0131	0.0104	0.993
Leave-one-out cross-validation	0.0285	0.0224	0.967

.

As shown in Table 10, the simplified PLSR model maintains high fitting accuracy and good cross-validation results even with fewer variables, indicating that the selected key indicators can effectively characterize the variation patterns of the Ecosystem Health Index (EHI). To further examine the degree of agreement between the model-predicted values and the observed values, the EHI values obtained using the PSR–entropy method for each year were compared with the model-predicted values. The results are shown in

Table 12. Comparison of EHI values calculated using the PSR metric system and fit values from the simplified PLSR model.

Year	PSR EHI	PLSR EHI
1995	0.61	0.6142
2000	0.52	0.5044
2005	0.27	0.3098
2010	0.20	0.1923
2015	0.40	0.4187
2020	0.41	0.3960
2024	0.66	0.6528

, and the fitting trends are illustrated in Figure 5.

As shown in Table 11 and Figure 5, The fitted trends of the simplified PLSR model are highly consistent with the observed EHI values. Between 1995 and 2010, the Ecosystem Health Index showed an overall downward trend, reaching a low point in 2010; after 2010, the index gradually rebounded, reaching a relatively high level by 2024. The simplified PLSR model effectively captures this dynamic process, indicating that the selected key indicators possess strong representational capacity for changes in ecosystem health within the study area.

An analysis of the directions of the model coefficients revealed that the Mean patch area of Suaeda heteroptera, Suaeda heteroptera carbon storage, Suaeda heteroptera recovery rate, and Shannon diversity index were all positively correlated with the ecosystem health index, while aquaculture development intensity was negatively correlated with the health index, consistent with the actual ecological changes observed in the study area. Specifically, the increase in the Mean patch area of Suaeda heteroptera reflects a trend toward the recovery of dominant vegetation patches and an improvement in wetland habitat integrity; the increase in Suaeda heteroptera carbon storage demonstrates enhanced carbon sequestration capacity of the ecosystem; the improvement in Suaeda heteroptera recovery rate indicates that ecological restoration measures have a positive effect on system health; and the rise in the Shannon diversity index suggests that the landscape composition is becoming more balanced and optimized. Conversely, increased aquaculture development intensity exacerbates human disturbance of wetland space and ecological processes, thereby constraining improvements in ecosystem health.

It should be noted that the simplified PLSR model constructed here is not intended to replace the ecosystem health assessment results based on the PSR–entropy method presented earlier. Its primary purpose is to identify key indicators influencing changes in health and to verify the effectiveness of these indicators in fitting the ecosystem health index. The model results indicate that, based on the comprehensive indicator system, changes in ecosystem health trends within the study area can be adequately reflected using only a small number of representative indicators. This not only validates the rationality of the indicator system established earlier but also demonstrates that the use of key indicators for simplified analysis is feasible in routine monitoring and rapid health assessment practices.

3.2. Analysis of Evaluation Results

3.2.1. Analysis of Ecosystem Health Trend

From 1995 to 2024, the Ecosystem Health Index (EHI) of the Suaeda heteroptera wetland ecosystem at the Liao River estuary showed a trend of first declining and then increasing (Figure 5). Using 2010 as a turning point, the ecosystem health of the study area can be divided into two phases: a decline phase and a recovery phase.

As shown in

Table 13. Ecosystem Health Index and Classification of the Suaeda heteroptera Wetland at the Liao River Estuary.

Year	1995	2000	2005	2010	2015	2020	2024
EHI Score	0.61	0.52	0.27	0.20	0.40	0.41	0.66
Health Rating	Healthy	Fairly Healthy	Generally Unhealthy	Unhealthy	Generally Unhealthy	Fairly Healthy	Healthy

, the ecosystem health index for the study area was 0.61 in 1995, with a health rating of “Healthy”; it declined to 0.52 in 2000, with the health rating changing to “Fairly Healthy”; it further declined to 0.27 in 2005, entering the “Generally Unhealthy” category; and it fell to 0.20 in 2010, the lowest value during the study period, with a health rating of “Unhealthy.” After 2010, the ecosystem health index gradually rebounded, reaching 0.40 and 0.41 in 2015 and 2020, respectively, with both years classified as “Generally Unhealthy”; by 2024, the index had risen to 0.66, and the health rating had recovered to “healthy.” Overall, the ecosystem health status of the study area exhibited a dynamic change process of “healthy → gradual decline → unhealthy → gradual recovery → healthy.”

(1)

1995–2010: Period of Declining Ecosystem Health

During this phase, the ecosystem health index declined steadily from 0.61 to 0.20, showing a clear downward trend. In 1995, the study area was in a generally healthy state; however, health levels began to decline in 2000, reaching a “Generally Unhealthy” status by 2005 and further deteriorating to an “Unhealthy” level by 2010. This indicates that during this period, the structure and function of the study area’s ecosystem were subject to continuous disturbance, and its health gradually deteriorated. Combined with the analysis of landscape patterns and changes in carbon storage, this phase was characterized by a reduction in Suaeda heteroptera wetland area, increased landscape fragmentation, and a significant decline in Suaeda heteroptera carbon storages, resulting in a corresponding weakening of ecosystem stability and self-sustaining capacity.

(2)

2010: The Lowest Point in Ecosystem Health

2010 was the year with the lowest Ecosystem Health Index (EHI) in the study area, with an EHI of only 0.20. According to the health classification criteria, the study area was classified as “Unhealthy” that year, indicating that the ecosystem was significantly affected by external disturbances, with both structural integrity and ecological functions at low levels. Considering the actual evolutionary context of the study area, the expansion of land reclamation for aquaculture during this phase continued to encroach upon natural intertidal zones and Suaeda heteroptera habitats, resulting in marked changes in the wetland landscape structure and causing ecosystem health to reach its lowest point during the study period. This year can be regarded as a turning point in the evolution of ecosystem health within the study area.

(3)

2010–2024: Period of Ecosystem Health Recovery

After 2010, the ecosystem health index in the study area generally showed a trend of recovery. In 2015 and 2020, the index rose to 0.40 and 0.41, respectively, with the health rating shifting from “Unhealthy” to “Generally Unhealthy.” By 2024, the index had further risen to 0.66, returning to the “Healthy” category. This indicates that ecosystem health in the study area gradually improved during this phase. Based on ecological restoration practices and the analysis presented earlier, the implementation of measures such as converting farmland back to wetlands, dredging tidal channels, and restoring vegetation led to a gradual recovery of the Suaeda heteroptera habitat area. Consequently, landscape patterns and carbon storage improved, driving an overall recovery in ecosystem health.

3.2.2. Analysis of Determinants of Health Changes

Changes in the health of the Suaeda heteroptera wetland ecosystem at the Liao River estuary result from the combined effects of various factors, including stressors, state, and responses. The weighting results calculated using the PSR–entropy method in the preceding section show that the total weight of the state layer is 0.43046, which is higher than that of the pressure layer (0.30642) and the response layer (0.26312). This indicates that changes in ecosystem health during the study period were primarily influenced by the evolution of the wetland’s internal ecological state, while also being jointly modulated by human activity pressures and ecological restoration responses. Combined with the PLSR analysis results in Section 2.5, it is evident that certain indicators possess high statistical significance, and their ranking generally aligns with that of the entropy method. This indicates that the indicator system constructed in the preceding section effectively reflects the primary driving processes of ecosystem health changes in the study area, as detailed in

Table 14. Comprehensive Analysis of Key Indicators Driving Changes in Ecosystem Health.

Metric	PSR Level	Entropy-Based Weighting	Weighting Order	PLSR VIP Value	Function Direction	Description of Primary Function
Number of tourists	P	0.04551	11	0.745	Negative	Reflects the intensity of visitor activity and has a certain impact on local habitat disturbance and management pressures.
Aquaculture development intensity	P	0.0773	5	1.4868	Negative	Characterizing the impact of the expansion of coastal aquaculture on wetland habitats and ecological processes is a key stressor during the decline phase.
Electricity consumption in the oil and gas extraction industry	P	0.05603	10	0.5563	Negative	Reflects the intensity of regional energy development activities and exerts a certain degree of disturbance on ecosystems.
Temperature deviation from normal	P	0.06577	7	0.2903	Negative	Characterizes abnormal climate fluctuations and exerts a background influence on wetland ecological processes.
Precipitation deviation from normal	P	0.06181	9	0.2761	Negative	Reflecting abnormal changes in precipitation, which indirectly affect moisture conditions and vegetation growth in wetlands.
Mean patch area of Suaeda heteroptera (AREA_MN)	S	0.15882	2	1.5359	Forward	It reflects the extent and integrity of dominant vegetation habitats and serves as a key indicator of changes in wetland structure.
Suaeda heteroptera aggregation index (AI)	S	0.10121	3	0.4648	Forward	Reflects the connectivity and aggregation of vegetation patches, serving as a complementary indicator of health changes.
Shannon diversity index (SHDI)	S	0.09411	4	0.8991	Forward	Reflects the balance of landscape composition and spatial heterogeneity, and provides some insight into changes in health.
Suaeda heteroptera carbon storage	S	0.07632	6	1.6669	Forward	It reflects the ecological functions of wetlands and the level of material accumulation, serving as a key functional indicator of changes in wetland health.
Suaeda heteroptera recovery rate	R	0.20026	1	1.1677	Forward	Characterizing the ecological restoration response process is a key positive factor in promoting the recovery of ecosystem health.
Rate of aquaculture conversion	R	0.06286	8	0.4888	Forward	Reflecting the process of returning land to wetlands plays a role in promoting ecological recovery.

.

From the perspective of stress factors, the Aquaculture development intensity is a significant negative factor affecting ecosystem health. This indicator has a weight of 0.07730 in the entropy method, ranking among the top factors; PLSR results show that its VIP value is greater than 1 and its regression coefficient is negative, indicating that the expansion of land-reclaimed aquaculture has a significant inhibitory effect on wetland health. From 1995 to 2010, the continuous expansion of aquaculture dikes encroached upon natural habitats, leading to a reduction in the distribution range of Suaeda heteroptera and increased landscape fragmentation, resulting in a corresponding decline in the ecosystem health index. After 2010, with the advancement of measures to reclaim wetlands and restore ecosystems, the Aquaculture development intensity decreased and the pressure of human disturbance eased, creating conditions for the recovery of ecosystem health. In contrast, indicators such as tourist numbers, electricity consumption in the oil and gas extraction industry, temperature anomalies, and precipitation anomalies were relatively weak in terms of both weight ranking and PLSR importance. This suggests that while they have some impact on ecosystem health, their influence is less significant than that of aquaculture development.

At the state level, the Mean patch area of Suaeda heteroptera, Aggregation index of Suaeda heteroptera, the Shannon diversity index, and Suaeda heteroptera carbon storage collectively reflect the structural and functional status of the wetland ecosystem and form the core of the health change analysis. Among these, the Mean patch area of Suaeda heteroptera has a weight of 0.15882, ranking among the top indicators, and also exhibits a high VIP value in PLSR, indicating its strong explanatory power regarding changes in the ecosystem health index. This indicator directly reflects the scale and integrity of the dominant vegetation habitat, and its trend is highly synchronized with the evolution of the health index: the decline in patch area around 2010 corresponds to a low point in health, while the gradual recovery after 2010 reflects enhanced habitat continuity. The weight of Suaeda heteroptera carbon storage is 0.07632, and it has the highest VIP value in PLSR, indicating its outstanding statistical explanatory power and its ability to characterize wetland vegetation recovery and material accumulation levels from an ecological function perspective. The Shannon diversity index and the aggregation of Suaeda heteroptera also carry significant weight in the entropy method, reflecting that changes in landscape pattern equilibrium and spatial connectivity are also important indicators of health evolution. Overall, state-level indicators performed exceptionally well in both methods, confirming that the internal structure and functional state of wetlands are central to the evolution of ecosystem health.

From the response layer perspective, the recovery rate of Suaeda heteroptera and the rate of grazing cessation reflect the role of ecological restoration and management interventions in influencing changes in ecosystem health. The weight of the Suaeda heteroptera recovery rate is 0.20026, the highest among all indicators, and its VIP value in PLSR exceeds 1, indicating that it has strong explanatory power regarding changes in ecosystem health in terms of both comprehensive evaluation and statistical relationships. Changes in this indicator suggest that health recovery in the study area stems not only from the alleviation of external pressures but is also closely related to the vegetation’s own recovery processes. The weight of the aquaculture withdrawal rate is 0.06286, indicating that the conversion of aquaculture areas back to wetlands plays a certain role in promoting health recovery; however, the PLSR results show that its direct explanatory power is weaker than that of the Suaeda heteroptera recovery rate, and it is more appropriate to regard it as a background management factor supporting health recovery.

In summary, the decline in ecosystem health in the study area was primarily driven by stress factors such as aquaculture development, whereas the recovery phase was closely associated with the recovery of Suaeda heteroptera, improvements in habitat structure, and enhanced ecological functions. The PSR–entropy method results highlighted the importance of the state and response layers, while the PLSR analysis further identified key indicators such as Suaeda heteroptera carbon storage, Mean patch area, aquaculture development intensity, and Suaeda heteroptera recovery rate. The two methods were generally consistent in identifying the primary driving factors, indicating that the constructed PSR indicator system can effectively reflect the main driving mechanisms behind changes in the ecological health of the Suaeda heteroptera wetland ecosystem at the Liao River estuary.

3.2.3. The Relationship Between Landscape Pattern, Suaeda heteroptera Carbon Storages, and Ecosystem Health

The spatial configuration and evolution of landscape patterns have a significant impact on material cycling, energy flow, and habitat stability in wetland ecosystems, while the carbon storage of Suaeda heteroptera species reflects the dynamic changes in the carbon accumulation capacity and ecological functions of wetland vegetation. These two factors characterize the state of wetland ecosystems from the perspectives of spatial structure and ecological function, respectively.

To further explore the relationship between changes in landscape structure, changes in ecological function, and ecosystem health, building upon the results of previous landscape pattern analysis and carbon storage estimation, this study selected landscape-level indices (edge density ED, Shannon diversity index SHDI) and patch-level indices (total patch area CA, patch proportion PLAND, number of patches NP, mean patch area AREA_MN, and aggregation index AI), as well as Suaeda heteroptera carbon storages, to conduct a correlation analysis with the PSR Ecosystem Health Index (EHI). The results of these comparisons are presented in Figure 5, Figure 6 and Figure 7.

(1)

The Relationship Between Changes in Landscape Diversity and Ecosystem Health Indices.

According to Figure 6, landscape-level indices reflect changes in the complexity and heterogeneity of the overall spatial structure within the study area. From 1995 to 2010, edge density (ED) decreased from 33.56 to 29.07, and the Shannon diversity index (SHDI) fell from 1.37 to 1.34, indicating that the complexity of landscape boundaries and spatial heterogeneity generally weakened during this period, with the structure tending toward uniformity. During the same period, the Ecosystem Health Index (EHI) decreased from 0.61 to 0.20, and the health rating gradually declined from “Healthy” to “Unhealthy,” indicating a strong temporal correlation between the decline in landscape diversity and the deterioration of ecosystem health. After 2010, the ED rebounded to 45.95, and the SHDI increased to 1.53, reflecting a renewed enhancement in landscape boundary complexity and spatial heterogeneity; simultaneously, the EHI rebounded to 0.66, and the health rating returned to “Healthy.” Overall, the improvement in landscape structural complexity and diversity showed clear synchrony with the recovery of ecosystem health.

(2)

Correlation between structural changes in Suaeda heteroptera patches and ecosystem health indices.

According to Figure 7, Suaeda heteroptera wetlands are the core natural vegetation type in the study area; the size, proportion, and spatial aggregation of these patches directly reflect the quality of the natural habitat. From 1995 to 2010, the total area (CA) and proportion (PLAND) of Suaeda heteroptera patches declined significantly, with the Mean patch area (AREA_MN) decreasing from 133.12 ha to 69.11 ha, reflecting a reduction in the scale and a weakening of the continuity of Suaeda heteroptera habitats. During the same period, the EHI continued to decline, indicating a significant correlation between the degradation of the Suaeda heteroptera habitat and the deterioration of ecosystem health. After 2010, the CA and PLAND of Suaeda heteroptera rebounded significantly, reaching 5098.23 hm² and 32.12%, respectively, by 2024, with the Mean patch area recovering to 101.96 hm², indicating improvements in the scale of the Suaeda heteroptera community and patch structure. Although the overall change in aggregation index (AI) was minor, the recovery of patch area and the optimization of structure have provided an important spatial foundation for the restoration of ecosystem health.

(3)

Correlation between Changes in Areas of Human Disturbance and Ecosystem Health Indices.

According to Figure 6, the expansion and contraction of areas of human disturbance, such as aquaculture enclosures, are important external factors influencing landscape patterns and ecosystem health. From 1995 to 2010, the total area (CA) and proportion of area (PLAND) occupied by aquaculture dikes continued to increase. PLAND rose from 13.11% to 31.58%, and the Mean patch area (AREA_MN) increased from 61.03 hm² to 294.04 hm², indicating that artificial dikes were gradually developing into large-scale, contiguous structures. This process directly encroached upon natural intertidal zones and Suaeda heteroptera habitats, corresponding to the continuous decline in the Ecosystem Health Index (EHI) during the same period. From 2010 to 2024, the CA and PLAND of aquaculture dikes decreased significantly, with the Mean patch area shrinking to 33.43 hm², reflecting a reduction in the intensity of human disturbance and the gradual emergence of the “returning wetlands after aquaculture cessation” effect. Concurrently, the EHI continued to rise, indicating that the reduction in human disturbance is a crucial prerequisite for the restoration of ecosystem health.

(4)

Correspondence between changes in Suaeda heteroptera carbon storages and the Ecosystem Health Index.

According to Figure 8, Suaeda heteroptera carbon storages reflect the vegetation’s carbon accumulation capacity and serve as a key indicator of wetland ecological function. From 1995 to 2010, the carbon storage of Suaeda heteroptera in the study area decreased from 3.37 × 10⁶ kg to 0.95 × 10⁶ kg, and the average carbon storage per pixel decreased from 78.56 kg to 53.64 kg, indicating a marked decline in the vegetation’s carbon accumulation function. During the same period, the Ecosystem Health Index (EHI) decreased from 0.61 to 0.20, reflecting a synchronous decline in ecosystem function and health. After 2010, the carbon storage of Suaeda heteroptera significantly rebounded, reaching 3.10 × 10⁶ kg in 2015 and further increasing to 4.39 × 10⁶ kg in 2024, the highest value recorded during the study period. Concurrently, the EHI recovered to 0.66, indicating that the recovery of the Suaeda heteroptera community and the enhancement of vegetation carbon accumulation jointly promoted the improvement of ecosystem functional status and facilitated the gradual restoration of health levels. Thus, changes in Suaeda heteroptera carbon storages effectively reflect the evolution of wetland ecological functions and serve as a functional response indicator for assessing ecosystem health.

Overall, changes in the health of the Suaeda heteroptera wetland ecosystem at the Liao River estuary show a strong correlation with shifts in landscape patterns and changes in Suaeda heteroptera carbon storages. In the early phase, a reduction in natural habitat area, decreased landscape heterogeneity, expansion of human disturbance, and a decline in carbon storage collectively led to a deterioration in ecosystem health. In the later phase, as Suaeda heteroptera habitats recovered, artificial embankments were reduced, landscape structure improved, and carbon storage increased, the ecosystem health index showed a clear trend of recovery. This indicates that changes in landscape structure and the evolution of ecological functions together form the fundamental basis for changes in ecosystem health in the study area.

4. Discussion

This study combines the PSR framework, objective weighting based on the entropy method, and PLSR for key factor identification, effectively addressing the issues of subjective weight determination, indicator redundancy, and multicollinearity among indicators in traditional wetland health assessments. The PSR model establishes a logical chain of human activity pressure, ecological state change, and management response feedback, while the entropy method assigns weights based on the discrete characteristics of the data itself, objectively reflecting the contribution of each indicator to variability in the time series. Meanwhile, the simplified PLSR model introduced in this study reduces the number of indicators from 11 to 5 while maintaining high fitting accuracy, providing a methodological basis for the rapid diagnosis of wetland health using a small number of highly sensitive indicators under conditions of limited data availability.

From the ecological perspective of key driving factors, the VIP ranking results show that the carbon storage of Suaeda heteroptera and the average patch area are slightly more important than the intensity of aquaculture development. This finding reinforces the understanding that ecological status determines wetland health. Comparative studies of natural and artificial wetlands in the northwestern arid region indicate that wetland degradation in this area is primarily driven by urban expansion and agricultural reclamation. In the Liaohe Estuary wetlands, the core conflict centers on spatial competition in the intertidal zone; the expansion of aquaculture dikes has disrupted the integrity of Salicornia habitats and the connectivity of tidal channel systems, leading to vegetation degradation and the loss of carbon sink functions [30]. It is worth noting that climatic factors ranked low in both the entropy-based weighting and VIP values, indicating that, over the past 30 years, the intensity of disturbances to ecosystem health caused by intense local human activities has significantly exceeded the natural impacts of interannual climate fluctuations. This is consistent with the pattern revealed by Chen, X.; Zhang, M. and Zhang, W., where development activities in the Liaohe Estuary wetlands dominate landscape evolution [14].

Nevertheless, this study still has certain limitations. Although this study covers a 30-year period, only seven time-series samples were obtained, leading to a small sample size. This, to some extent, restricts the testing effectiveness and extrapolation capability of the PLSR model. In addition, the estimation of Suaeda heteroptera carbon storage mainly relies on the TSAVI inversion model and allometric growth equations established in previous studies, and applying these across the entire study period may introduce some uncertainty. Furthermore, due to the long time series, we were unable to adequately obtain field-measured data on hydrology, soil, and vegetation physiology and ecology, resulting in insufficient reflection of micro-processes within the wetland ecosystem. Future research should strengthen the synergistic application of remote sensing inversion and ground monitoring data, combine them with more frequent time-series observations, further refine the evaluation indicator system, and improve the accuracy and reliability of health diagnosis.

5. Conclusions

This study focuses on the Suaeda heteroptera wetland at the Liao River estuary. Based on remote sensing imagery and multi-source data from 1995 to 2024, an evaluation index system was constructed using the PSR framework. By combining the entropy method with the PLSR method, the study conducted a quantitative assessment of wetland ecosystem health and identified key driving factors. The health assessment results based on the PSR–entropy method indicate that the ecosystem health index in the study area exhibits a phased evolution with 2010 as a turning point, declining continuously from 0.61 (healthy) in 1995 to 0.20 (Unhealthy) in 2010, and then gradually recovering to 0.66 (healthy) by 2024. Weight analysis using the entropy method revealed that the weight of the state layer (0.4305) was higher than that of the pressure layer (0.3064) and the response layer (0.2631). Among these, the recovery rate of Suaeda heteroptera, the Mean patch area of Suaeda heteroptera, and the aggregation degree of Suaeda heteroptera ranked as the top three, serving as core indicators for changes in the health index. The PLSR model further identified Suaeda heteroptera carbon storage (VIP = 1.67), Mean patch area of Suaeda heteroptera (VIP = 1.54), aquaculture development intensity (VIP = 1.49), and Suaeda heteroptera recovery rate (VIP = 1.17) as key drivers influencing changes in the EHI. Among these, aquaculture development intensity acts as a negative driver, while the other three act as positive drivers, which is generally consistent with the weight rankings obtained using the entropy method. The simplified PLSR model constructed based on the above key indicators, with Suaeda heteroptera carbon storage, Mean patch area, aquaculture development intensity, Suaeda heteroptera recovery rate, and Shannon diversity index as independent variables, achieved a coefficient of determination (R²) of 0.967 and a root mean square error (RMSE) of 0.0285 using leave-one-out cross-validation. The fitted values showed a high degree of consistency with the EHI calculated using the PSR–entropy method in terms of trends, indicating that the five selected indicators can effectively characterize the evolutionary features of ecosystem health in the study area. In summary, the coupled application of the PSR–entropy method and PLSR not only retains the causal logic and comprehensive evaluation capabilities of the PSR framework but also achieves objective weight assignment for indicators through the entropy method. Furthermore, by utilizing PLSR, it resolves the issue of indicator multicollinearity under small-sample conditions, thereby achieving an effective transition from comprehensive evaluation to the identification of key factors. The streamlined PLSR model provides a practical and simplified approach for the rapid diagnosis and routine monitoring of coastal wetland health.