Novel Method for Evaluating Wetland Ecological Environment Quality Based on Coupled Remote Sensing Ecological Index and Landscape Pattern Indices: Case Study of Dianchi Lake Wetlands, China

Abstract

Wetlands serve as crucial ecological buffers, significantly influencing temperature reduction, carbon storage, regional climate regulation, and urban wastewater treatment. To elucidate the relationship between wetland landscape patterns and ecological environment, and to accurately assess lake ecosystems, this study proposes a semi-supervised classification method based on RSEI and K-Means. By integrating landscape pattern indices, the Remote Sensing Ecological Index (RSEI), and disturbance proximity, a comprehensive evaluation of the ecological quality of the Dianchi wetlands was conducted. The results indicate that the RSEI-K-Means method, with K set to 50, achieved overall accuracies (OAs) and Kappa values of 0.91 and 0.88, surpassing the SVM’s 0.85 and 0.80. This method effectively combines ecological and landscape indices without relying on extensive training samples, enhancing accuracy and speed in wetland information extraction and addressing the challenges of spatial heterogeneity. This study reveals that from 2007 to 2009, and 2013 to 2015, landscape patterns were significantly influenced by the rapid expansion of Kunming city, exacerbating wetland fragmentation. Notably, significant ecological quality changes were observed in 2009 and 2013, with gradual recovery post-2013 due to strengthened environmental protection policies. The RSEI disturbance proximity analysis indicated that the affected areas were primarily concentrated in regions of high human activity, confirming the method’s high sensitivity and effectiveness. This study can help in wetland ecosystem research and management.

1. Introduction

Wetland ecosystems, situated in the transitional zone between terrestrial and aquatic ecosystems, are one of the Earth’s three major ecosystems [1]. They provide critical habitats for a vast array of vegetation, animals, and microorganisms [2], serving as one of the planet’s largest carbon sinks [3] and playing a pivotal role in global biodiversity and ecosystem services such as water purification [4]. Wetlands are crucial ecological buffers with significant impacts on reducing surrounding temperatures, increasing carbon storage, regulating regional climate change, and treating urban wastewater [5,6]. As water resources become increasingly scarce and the quality of the ecological environment continues to degrade, the water resources contained within wetlands are becoming ever more important [7]. The ecological functions of wetlands are influenced by their area, quality, and environmental changes, making the study of wetland landscape patterns and their ecological effects globally important for understanding wetland functions and trends [8]. Currently, both domestic and international research on wetland ecological effects primarily focuses on methods for ecological environment assessment, ecosystem service evaluation, and environmental disturbance analysis. Numerous methods for evaluating ecological environment quality have also emerged [9]. The most widely used wetland assessment indices, such as Hydrogeomorphic (HGM) indices and Indices of Biological Integrity (IBI), are all single-factor wetland assessment systems [10]. Wetland ecological environment quality assessment is mainly used to quantify the extent to which wetlands are disturbed by human activities and natural factors [9], widely applied in the study of extreme changes in wetlands at different scales and their driving factors [10]; the research results indicate that climate change poses a threat to global Ramsar wetlands, necessitating efforts to mitigate climate change and reduce human disturbances to protect these wetlands [11]. The tidal wetlands in the Yellow Sea region have been severely damaged due to coastal development, with land use and the invasion of non-native species being the main causes of coastal wetland degradation [12]. Coastal wetlands face significant pollution risks, primarily from heavy metal contamination [13]. However, most risk assessments applied to these wetlands focus on single factors.

The Dianchi lake wetland is located in the central Yunnan province, China. Dianchi lake wetlands can isolate and weaken the direct impact of human activities, reduce the wastewater levels of the Dianchi lake, provide habitat for many animals and plants, and has an important surrounding ecological barrier. However, due to the influence of human activities and natural factors, Dianchi lake wetlands face many challenges, like many other wetlands, including human interference, land use change, and environmental degradation. This study focuses on the Dianchi lake wetland, which not only provides ideas for protecting the regional ecosystem, but also provides valuable experience for the restoration and management of wetlands worldwide.

Traditional evaluation methods often rely on static data, which fail to adequately capture the dynamic changes in wetland ecosystems over time. As a result, subtle changes in wetlands due to factors like seasonal fluctuations, climate change, and human disturbances are difficult to assess accurately. Most of these methods struggle to address the spatial heterogeneity within wetlands, making it challenging to precisely describe the ecological differences between various regions within a wetland.

Compared to traditional SVM methods, this study’s semi-supervised classification approach based on the RSEI demonstrates higher accuracy and flexibility in extracting wetland landscape patterns and conducting ecological evaluations. This method not only provides more precise wetland information, but also reveals ecological quality trends through multi-temporal analysis, offering a scientific basis for developing wetland conservation and ecological restoration policies. By addressing the limitations of existing evaluation methods in dynamic change detection, spatial heterogeneity, and multi-scale analysis, this study allows for new perspectives and methodologies for wetland ecosystem research and management.

2. Materials and Methods

2.1. Study Area

The Dianchi lake (24°40′~25°02′ N, 102°37′~102°48′ E) is a tectonic fault subsidence lake located in the southern part of Kunming, Yunnan Province. It has a perimeter of approximately 140 km and covers an area of about 298 km², making it the sixth largest freshwater lake in China. The lake exhibits typical plateau lake characteristics, including weak water flow, low self-purification capacity, and a fragile ecological environment. Positioned at the center of the Kunming Basin, it extends from Chenggong District in the east to the foothills of the Western Hills in the west, bordered by Daguan Park to the north and Jinning County to the south, all within Kunming city. The surrounding wetland resources of the Dianchi lake are abundant, but due to human activities and natural factors, the wetland area has decreased by over 40% compared to the Eastern Han Dynasty over the past 2000 years. Although this study’s time series is relatively short, it effectively captures significant ecological environmental changes due to increased human activities, policy shifts, and climate change in the research area during this period. This makes the findings representative within the studied timeframe. Current human activities, such as urban sprawl, infrastructure development, and agricultural expansion, continue to degrade the wetlands. For example, the conversion of natural landscapes into built-up areas has led to reduced habitat connectivity and increased surface runoff, while agricultural runoff has introduced excessive nutrients into the wetlands, fostering eutrophication. These activities are closely linked to the decline in ecological quality as revealed by the RSEI analysis.

We selected a 3 km buffer zone; the main reason for this was that a 3 km buffer zone included not only the Dianchi lake wetland, but also the land use types with human activities as the main influencing factor, effectively revealing the gradient change in landscape fragmentation and the intensity of the wetland around the Dianchi lake and human activities. The study area covers approximately 366.30 km². The location map of the research area uses a Landsat 8 OLI image from 26 January 2023 displayed through RGB synthesis, with bands B4, B3, and B2, as shown in Figure 1.

2.2. Data Source

Landsat 5 and Landsat 8 satellites were both launched by the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS). They were launched in 1984 and 2013, respectively. Landsat 5 carried the Multispectral Scanner (MSS) and Thematic Mapper (TM), with TM having a total of 7 bands. Landsat 8 is equipped with the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS). Due to the special geographical and climatic conditions of Kunming city, the seasonal climatic change is not large, the remote sensing images selected in this study are generated from March to May of each year, and the selected time has less cloud content, which can reduce the impact of clouds on the research area. For this study, a total of 11 images from Landsat 5 TM and Landsat 8 OLI sensors were selected. All images were downloaded from the official USGS website (https://earthexplorer.usgs.gov accessed on 1 April 2024). To promote the accuracy of the model, the cloud content of the optical remote sensing images was 5%.

We used ENVI to preprocess the selected remote sensing images to ensure data accuracy and consistency. Radiometric calibration transformed the sensor DN values into radiance values, correcting sensor-related distortions to enhance the interpretability of the images. Atmospheric correction reduced the impacts of particulates, gasses, and scattering in the atmosphere, ensuring that reflectance values more accurately represent the true spectral characteristics of surface objects, thereby improving the reliability of landscape classification and ecological analysis. Finally, the images were cropped to the study area, optimizing processing efficiency and focusing the analysis on the research area. This process enhanced the quality and consistency of the remote sensing images, laying a solid foundation for subsequent landscape pattern and ecological environment analysis.

2.3. Analytical Approaches

2.3.1. Landscape Pattern

Landscape pattern indices can quantify complex landscape patterns. In this study, representative landscape pattern indices were selected at the landscape scale to evaluate the ecological environment of the Dianchi lake. In this study, due to the limited applicability of patch-level landscape indices in landscape pattern analysis, we selected eight indices from the class and landscape levels. These indices can effectively depict the overall spatial structure of the landscape and land use, and they are sensitive to changes in the spatial structure of land use in areas with high levels of human activity. The selected landscape pattern indices and their ecological significance are presented in

Table 1. Calculation formula and explanation of landscape pattern index.

Landscape Pattern Index	Formulas and Explanations	Ecological Significance
Patch Density (PD)	$\mathrm{PD}={\displaystyle \sum _{i=1}^M{N}_i}/C$
	where M represents the total area of all landscape types in the study area. C represents the quantity of individual landscapes in the study area. PD reflects the degree of landscape fragmentation and the level of human activity interference with the landscape.	The larger the PD, the higher the degree of landscape fragmentation.
Contagion Index (CONTAG)	$\mathrm{CONTAG}={\displaystyle \sum _{i=1}^m{\displaystyle \sum _{j=1}^n{p}_i{g}_{ik}/{\displaystyle \sum _{k=1}^m{g}_{ik}}\times \ln (p_i)g_{ik}/{\displaystyle \sum _{k=1}^m{g}_{ik}}\times 100/2\ln (m)}}$
	where M represents the total number of patch types, gik represents the probability that two randomly selected adjacent patches belong to types i and k, respectively. CONTAG reflects the degree of aggregation and the contagion trend of different patch types within the landscape.	A high CONTAG value indicates that landscape patches are highly aggregated, while a low value suggests that the lands cape is more fragmented and dispersed.
Landscape Division Index  (DIVISION)	$\mathrm{DIVSION}=1-{\displaystyle \sum _{j=1}^n{a}_{ij}/A}$
	where a represents the area of the j-th patch of the i-th landscape type, and A represents the total landscape area. DIVISION reflects the degree of segmentation and fragmentation in the landscape.	A higher DIVISION value indicates a greater degree of landscape separation.
Shannon’s  Diversity  Index (SHDI)	$\mathrm{SHDI}=-{\displaystyle \sum _{i=1}^m\left[P_i\ln (P_i)\right]}$
	where P represents the proportion of the total landscape area occupied by patch type, and i represents the number of patches. SHDI reflects the degree of landscape fragmentation and fragmentation.	A high SHDI value indicates a higher diversity within the landscape.
Shannon’s  Evenness  Index  (SHEI)	$\mathrm{SHEI}=-{\displaystyle \sum _{i=1}^m\left(P_i\ln P_i\right)/\ln m}$
	where P represents the proportion of the total landscape area occupied by patch type, and i represents the number of patches. SHEI reflects the uniformity of various patch categories on the landscape.	The SHEI value ranges between 0 and 1, with values closer to 1 indicating a more even distribution of patch types.

.

2.3.2. Remote Sensing Ecological Index

The preprocessed remote sensing imagery undergoes band arithmetic to calculate four ecological indices: Normalized Difference Vegetation Index (NDVI), Wetness Index (WET), normalized differential build-up and bare soil index (NDBSI), and Land Surface Temperature (LST). The RSEI (Remote Sensing Ecological Index) is particularly effective in evaluating regional ecological environments. The RSEI and RSEI₀ range from 0 to 1, and the closer the value is to 1, the better the ecological environment quality is. For semi-supervised classification, the four remote sensing ecological evaluation index bands of RSEI are fused through band merging. This approach not only reduces computational complexity but also yields robust results for the classification of the Dianchi lake wetland landscapes.

The calculation formula for RSEI is

$$\mathrm{RSEI}=f(\mathrm{NDVI},\mathrm{WET},\mathrm{NDBSI},\mathrm{LST})$$

(1)

$${\mathrm{RSEI}}_0=1-\left\{\mathrm{PC}1\left[f(\mathrm{NDVI},\mathrm{WET},\mathrm{NDBSI},\mathrm{LST})\right]\right\}$$

(2)

2.3.3. RSEI Interference Proximity

In the past, scholars considered extreme changes in NDVI at the same or similar times in different years at the same location as interference occurrences; NDVI cannot represent the situation of buildings or other non-vegetation factors well in the study area. RSEI, coupling four ecological environment evaluation indicators, can better reflect the ecological environment changes in a region. In this study, RSEI is introduced into NDVI interference identification:

$$D(x,y)=\left\{\left[f_{\tau 1}(x,y)-f_{\tau 2}(x,y)\right]-m\right\}/\sqrt{S_{\tau 1}^2+S_{\tau 2}^2-2co{v}_{\tau 1,\tau 2}}$$

(3)

where D (x, y) represents the standard change intensity index value of the raster image at row x and column y; f_τi is the RSEI value during time; and m is the average deviation based on pixel level. S_τi is the standard deviation of f_τi; cov_τ1,τ2 in the equation refers to the covariance between f_τ1 and f_τ2. Based on research experience and relevant data (such as urban planning and development, population growth, etc.), a fixed ratio (30%) of the empirical distribution of D (x, y) is set as the threshold for interference.

3. Results and Analysis

3.1. Analysis of RSEI-K-Means Wetland Landscape Classification

Based on the results of wetland landscape classification using RSEI-K-Means with varying values of K, the landscape distribution map of the Dianchi lake for the years between 2003 and 2023 was created (Figure 2). Classification accuracy was evaluated using field survey data and high-resolution remote sensing imagery. As shown in Figure 3, the average overall accuracy (OA) and Kappa coefficient for RSEI-K-Means are 0.8384 and 0.8079, respectively, indicating generally high-quality landscape classification results. Notably, when K = 50, the OA and Kappa values for RSEI-K-Means reach 0.8755 and 0.9104, respectively, demonstrating the superiority of RSEI-K-Means in wetland landscape classification. In this study, the classification accuracy was highest when the number of clusters (K) was set to 50.

From Figure 3, it can be observed that the OA and Kappa values of the classification results exhibit certain fluctuations across different time points and classification methods. Class 30 shows the least stability, particularly before 2015, with OA and Kappa values of 0.76 and 0.73, indicating poor classification accuracy. In contrast, Class 50, Class 100, and SVM demonstrate better classification precision, with Class 50 achieving OA and Kappa values of 0.91 and 0.88 in 2021. However, using OA and Kappa to validate the classification accuracy of semi-supervised classification is somewhat unfair, especially given that it does not require the time-consuming preparation of training samples and still achieves higher classification accuracy than SVM. Therefore, RSEI-K-Means has advantages over SVM in wetland landscape classification.

Based on the landscape classification results with K = 50, the study area was divided into five regions based on permanent or semi-permanent landscapes (Figure 2l). This division allows for a detailed examination of the wetland ecological environment evolution and the correlation between landscape patterns and the ecological environment in wetland parks and surrounding areas. Figure 4 illustrates the transition areas of wetland landscape types from 2003 to 2023. It is evident that the area of lake wetland within the study area remains relatively stable, while the predominant landscape pattern evolution within the area is characterized by changes in “Vegetation-Others.”

In this study, a semi-supervised classification method based on RSEI was proposed in previous research. To further validate the advantages and transferability of the RSEI-based semi-supervised classification method, this section applies the method to a different study area from that used in the earlier research. By extracting lake wetland landscape types using the RSEI semi-supervised classification method, we aim to verify its transferability and general applicability.

The Fuxian lake, located in Chengjiang city, Yuxi, Yunnan Province, lies to the southeast of the Dianchi lake and shares similar geological and climatic conditions. Both are plateau lakes that have increasingly been affected by human activities in recent years. The results indicate that between 2003 and 2023, the vegetation landscape around the Fuxian lake has increased (Figure 5). This development has been promoted by the combined effects of human and climatic factors, suggesting a trend toward an improving ecological environment around the Fuxian lake.

3.2. Landscape Pattern Index

To further investigate the evolution of wetland landscape patterns in the Dianchi lake, eight wetland parks surrounding the lake were extracted (Figure 6), and landscape pattern indices for these parks and their respective areas were calculated. The results of the landscape pattern index calculations indicate (Figure 7) that the PD values for the Daguan and Haigeng wetlands remain high throughout the study period. In contrast, the PD values for other wetland parks decrease over time and shift southward with the development of Kunming city. The variation in PD values suggests that the fragmentation level of wetland park patches initially decreases and then increases across the circular study area.

The analysis of landscape pattern indices for wetland parks and their surrounding areas (Figure 8) reveals that indices such as PD, DIVISION, SHDI, and SHEI exhibit a linear increase, indicating a rise in the fragmentation and shape diversity of wetland park landscapes. Conversely, the CONTAG index shows a linear decrease, signifying a reduction in the aggregation of landscape patches and an increase in edge dispersion. This trend suggests that the landscapes within the study area are becoming increasingly fragmented.

Spearman’s rank correlation coefficients were used to analyze the relationships between landscape pattern indices (

Table 2. Correlation analysis of wetland park and its segmented landscape pattern index.

Paragraph	Wetland Park	Spearman’s Correlation	Paragraph	Wetland Park	Spearman’s Correlation
Guandu–Chenggong	Haidong	0.674	Caohai	Daguan	0.806
	Wangguan	0.683		Haigeng	0.784
	Laoyuhe	0.749		Haihong	0.788
Jinning-down	Dongda	0.722	Xishan	Xihua	0.716

). The results indicate a significant positive correlation between the landscape pattern indices of wetland parks and their respective segments at a 99.9% confidence level.

Combining ecological models, causal analysis, and spatiotemporal dynamics can provide a deeper understanding of the interactions among these variables. Urbanization, land use changes, and development projects surrounding the wetland are likely key factors influencing landscape pattern indices. For instance, areas like Daguan and Haihong may exhibit higher fragmentation and diversity in landscape patterns due to significant human activities in their vicinity, which could increase these indices. Based on the values of the landscape pattern indices, there may be regional differences in the degree of landscape fragmentation and the connectivity between different landscape patches. Such fragmentation is often closely linked to human disturbance, which in turn can impact the wetland’s ecological functions and biodiversity.

3.3. RSEI

Classification and time series analysis of ecological environment quality using RSEI (Figure 9 and Figure 10) reveals that significant changes occurred in the study area in 2009 and 2013, with an overall cyclical pattern of “increase-stabilize-decrease.” This indicates that the ecological environment quality began to recover after 2013.

To further explore the relationship between the landscape pattern and ecological environment of the Dianchi lake wetlands, an RSEI model was established to quantitatively study the functional relationship between the RSEI of the Dianchi wetlands and NDVI, WET, LST, and NDBSI (

Table 3. RSEI regression model coefficient table.

Year	NDVI	WET	NDBSI	LST	Gap
2003	0.578	0.145	−0.395	−0.421	0.093	12.92%
2009	0.488	0.126	−0.351	−0.372	0.109	17.79%
2015	0.562	0.098	−0.327	−0.270	−0.063	−9.56%
2023	0.680	0.105	−0.418	−0.163	−0.203	−25.92%
Mean value	0.577	0.118	−0.373	−0.306	−0.016	−2.30%

). The 3D spatial distribution and projections of RSEI and these four indicators were plotted (Figure 11). A random forest analysis was conducted to assess the importance of the relationship between RSEI and the selected landscape pattern indices (Figure 12).

It was found that the ecological indicator contributing the most to the ecological index of the Dianchi lake wetlands is NDVI, while the indicator with the least contribution is WET. Both NDVI and WET have a positive correlation with RSEI, while NDBSI and LST have a negative correlation. The year 2009 marked a turning point for the Dianchi lake wetlands, as the influence of NDVI and WET began to decrease compared to that of NDBSI and LST, with the gap between them continuing to widen.

The figure shows that the tilt of the scatter plot projection for NDVI and WET, which have a positive correlation with RSEI, is much less pronounced compared to that of NDBSI and LST, which have a negative correlation with RSEI. This indicates that the combined impact of NDVI and WET on the ecological environment of the Dianchi lake wetlands is relatively smaller than that of NDBSI and LST. Consequently, the positive influence of NDVI and WET on the Dianchi lake wetlands is less significant, and the ecological conditions of the wetlands are deteriorating rapidly.

The importance of landscape pattern indices within different landscape types, as indicated by RSEI, reveals the varying impacts of these indices on ecological environment quality. In wetland and vegetation types, connectivity (CONTAG) and diversity (SHDI) have higher importance, highlighting their roles in maintaining ecosystem stability. In other landscape types, patch density (PD) and fragmentation (DIVISION) have a more significant impact on RSEI, suggesting that these areas are subject to stronger human activity interference.

The statistical results of the disturbance situation in the five regions of the study area are as follows (

Table 4. Statistics of disturbance area in Dianchi wetland.

	Guandu–Chenggong	Jinning-Up	Jinning-Down	Xishan	Caohai
Undisturbed (%)	33	36	70	50	63
Disturbed (%)	67	64	30	50	37

):

Finally, combining RSEI with the disturbance proximity algorithm, it was found that the regions with high levels of disturbance are the Guandu–Chenggong segment and the Jinning-up segment.

From Figure 13, it is evident that there are significant differences between NDVI disturbance identification and RSEI disturbance identification in detecting wetland disturbances. Particularly in areas with frequent human activity and low vegetation coverage, RSEI disturbance identification proves to be more effective in identifying disturbed areas within the study region. It reduces the “identification error” and provides higher clarity and boundary precision compared to NDVI, resulting in a more accurate representation of the disturbed regions.

RSEI combines four key ecological indicators: greenness, wetness, dryness, and heat. Compared to NDVI, which focuses solely on vegetation health, RSEI demonstrates higher sensitivity to human activity disturbances. The heat and dryness indicators play a crucial role in RSEI, especially in areas significantly affected by human activities such as urban expansion, agricultural land use, and infrastructure development. In these areas, surface temperatures and bare soil areas often increase, directly reflecting disturbances. The multidimensional nature of RSEI allows it to capture more complex disturbance patterns by considering interactions between different ecological factors. For example, a region may show good NDVI values while still experiencing significant human impact due to declining humidity, rising surface temperatures, or expanding bare land, all of which can be effectively detected through RSEI.

4. Discussion

In this study, we employed a semi-supervised classification method based on the RSEI and K-Means clustering to evaluate the ecological environment and landscape patterns of wetlands surrounding the Dianchi lake. Our findings reveal significant changes in the wetland landscape structure, which are closely associated with variations in ecological quality. By identifying key landscape pattern indices, we provided insights into the factors influencing wetland health and proposed strategies for ecological restoration. This discussion will explore the implications of these findings, address potential limitations, and suggest avenues for future research.

4.1. RSEI-K-Means Provides Effective Means for Wetland Landscape Classification

This study reveals that analyzing the classification and pattern changes in the Dianchi lake wetlands demonstrates the effectiveness of the RSEI-based semi-supervised classification method for extracting wetland information. This approach often outperforms traditional SVM classification methods, particularly when the number of classifications is set to 50, likely due to RSEI’s integration of multiple ecological indicators, which enhances its ability to capture and represent landscape features in complex wetland environments [14]. This highlights how semi-supervised classification, with effective classifier selection and appropriate training samples [15,16], can reduce classification errors and improve accuracy, thereby reducing the time and expertise required for achieving classification goals.

Comparative analysis with previous studies shows that the wetland extraction results align well with those of Zhao F [17] and Yang K [18] within the same study time series. Guo F [19] et al. have demonstrated the superiority of semi-supervised classification for extracting coastal wetland information using hyperspectral and LiDAR data. Traditional SVM classifiers differentiate feature types by maximizing class margins [20], which becomes challenging in areas with indistinct boundaries, such as water–land transitions, where SVM struggles to define strict classification boundaries [21]. In contrast, the RSEI-K-Means classification used in this study, which couples various ecological quality indices, provides better comprehensiveness and sensitivity to ecological changes, addressing the issue of unclear boundaries between transitional land types. However, RSEI’s calculation, which requires masking large water bodies to reduce PCA load distribution [22], may impact classification accuracy. The superior performance of the 50 class model compared to the 100 class model may be related to the sensitivity of RSEI to wetland landscape heterogeneity [23]. We believe this result is primarily influenced by factors such as “data algorithm complexity,” “classification granularity and landscape heterogeneity,” “wetland landscape complexity,” and “the impact of the RSEI composite index.” This indicates that this level of classification granularity can effectively capture the landscape characteristics of the Dianchi lake wetland. As a complex ecosystem, the Dianchi wetland has diverse landscape types and uneven spatial distribution. When K < 50, the classification results were relatively coarse, and the classification model failed to adequately represent the wetland’s landscape diversity, with some critical ecological features being overlooked, leading to reduced accuracy. Conversely, when K > 50, the model tended to overfit minor changes and less significant features in the landscape, which also decreased classification accuracy. Our research found that the RSEI, which integrates multiple ecological indices, showed strong performance when applied to wetland classification. However, for different levels of classification granularity, the advantages of the RSEI might not be fully realized, leading to blurred boundaries between categories and making it challenging to accurately reflect the wetland’s actual ecological state.

Additionally, the remote sensing images used in this study, generated predominantly in the first half of each year, are suitable for Kunming city, where climate and vegetation types show minimal seasonal and temporal variation [24]. For other study areas, acquiring remote sensing images from multiple seasons and time periods may be necessary to ensure higher overall classification accuracy.

4.2. Human Activities Have Exacerbated the Fragmentation of the Landscape Patterns in the Dianchi Wetland

Segmented analysis of the study area shows that landscape pattern evolution is influenced by both natural landscape types and human activities. Significant changes in landscape types occurred during 2007–2009 and 2013–2015, periods characterized by rapid urban development in Kunming city. The impact of human activities is evident in population growth and economic development. Since the relocation of Kunming’s municipal government in 2003 and the initiation of related urban planning, there have been large-scale changes in the wetland landscape patterns of the Dianchi lake due to the expansion of regional urban boundaries [25]. Following the government relocation, changes in urban planning and rising land prices led to the southward shift in agricultural land in Chenggong District [26]. During this period, Kunming continued its rapid expansion and implemented more infrastructure projects, such as road and airport expansions, which often involved wetland areas, further fragmenting and degrading the wetland landscape.

Changes in wetland landscape patterns can also impact the regional ecological environment. Rapid urbanization generally causes significant harm to wetland vegetation [27]. Studies indicate that once land use and land cover (LULC) changes stabilize [28,29], the restoration of vegetation and wetlands becomes particularly challenging, as lake wetland ecosystems are highly sensitive to landscape fragmentation [30]. Analysis of wetland park landscape pattern indices further reveals spatial distribution characteristics of wetland fragmentation. The indices for wetland parks and their surrounding areas generally show high correlation, though the degree of dispersion varies among different wetland parks. This suggests that each park operates independently while being influenced by similar factors in its surrounding area.

The radial variation in PD values indicates that the fragmentation level of wetland parks is influenced by human activities. Although the overall PD values have increased, the trends differ across regions, possibly reflecting the combined effects of urban expansion and environmental protection measures. Additionally, the evolution of wetland landscape patterns due to human activities can impact lake water quality [31]. Gong Y et al. found that seasonal variations in BOD, DO, and TN are significantly affected by LPI, with higher LPI values for farmland and construction areas associated with poorer water quality [32].

4.3. The Combined Effects of Human Activities and Landscape Patterns Have Made the Evolution Mechanism of Ecological Environment Quality in the Dianchi Wetland More Complex

Time series analysis of the RSEI indicates that the ecological quality of the study area experienced significant changes in 2009 and 2013. Notably, after 2013, as environmental protection policies were strengthened, the ecological quality gradually improved. The RSEI disturbance proximity results also indicate that areas most affected by disturbances are primarily concentrated in regions of human activity and governmental planning, which aligns with earlier findings. Yang H et al.’s assessment of RSEI for the Dianchi lake basin from 1990 to 2020 reached similar conclusions [33].

This study’s results reveal that the ecological environment quality of the Dianchi lake wetland has shown a cyclical trend of “decline-stability-rise.” During the early stages of the study period, the rapid urbanization and economic development of Kunming led to significant land use changes in the Guandu–Chenggong and Jinning areas. This period coincided with the southward shift in agricultural land and government institutions observed in this study, which directly impacted the wetland’s ecosystem structure, increasing pollution in both the wetland and surrounding areas. According to water quality data from the 24 h surface water quality monitoring system of the Ministry of Ecology and Environment of the People’s Republic of China, the water quality in the Dianchi lake was generally better in the southwest compared to the northeast, resulting in a decline in ecological environment quality.

After 2009, as urban expansion reached a certain scale, direct human interference with the wetland decreased, and the wetland ecosystem entered a relatively stable phase. During this time, Kunming’s infrastructure development reached a more advanced state, the speed of urban development slowed down, and the ecological environment quality stabilized. Following 2015, the evolution of the Dianchi lake wetland’s landscape patterns also had some impact on ecological environment quality. During the period of rapid fragmentation, the connectivity between wetland landscape patches was poor, threatening biodiversity and ecological diversity. However, with the continuous strengthening of ecological protection policies and the government’s initiatives to manage the Dianchi lake and implement wetland ecological restoration measures, external disturbances to the wetland gradually lessened. As a result, the connectivity between landscape patches improved, and the ecological environment quality began to rise during this period.

A comprehensive analysis of RSEI and landscape type evolution reveals that RSEI trends are not static. For instance, this study highlights that the period from 2013 to 2015 may reflect the preliminary results of ecological projects such as reforestation. Under similar conditions, the ecological environment in the northwestern mountainous region of the Dianchi lake is superior to other areas, possibly due to Kunming’s rapid urbanization, which generates more heat compared to natural conditions, thereby promoting vegetation growth [34,35].

This recovery trend is consistent with changes in landscape pattern indices, indicating that the spatial structure and ecological functions of wetland landscapes have been somewhat restored and optimized under policy intervention. Previous studies have considered extreme changes in NDVI as indicative of disturbances [36]. However, RSEI demonstrates a more pronounced advantage over traditional NDVI in reflecting disturbance proximity. RSEI integrates not only vegetation cover but also other ecological factors, offering a more comprehensive view of wetland ecological changes [37]. The consistency between RSEI disturbance proximity results and landscape pattern changes further validates the effectiveness of RSEI in disturbance proximity analysis.

4.4. Promotion of RSEI-K-Means for Wetland Ecological Environment Quality Assessment

The RSEI method, which integrates multiple ecological indicators (such as greenness, dryness, heat, and humidity), is highly flexible and can be applied to other wetlands with similar environmental conditions. Wetlands that experience human disturbances, such as urbanization, agricultural expansion, or pollution, can benefit from the method’s ability to capture ecological changes over time. While this study focuses on the Dianchi lake, it is important to recognize that other wetlands may have different hydrological regimes, biodiversity, or climatic conditions. To apply RSEI and landscape pattern indices effectively in these cases, adjustments may be needed. For example, wetlands in tropical regions might need additional ecological indicators (e.g., humidity or biomass) integrated into RSEI to reflect their unique characteristics. Similarly, the weights assigned to various ecological factors in RSEI could be recalibrated depending on the specific threats and stressors in the new wetland system. The method proved effective at identifying landscape fragmentation and recovery trends in Dianchi. It can be scaled to larger or smaller wetlands by adjusting the spatial resolution of the remote sensing data, and the size of the study area. Although this study shows RSEI’s advantages over NDVI in detecting ecological changes, it is essential to discuss potential limitations. Wetlands with complex or less disturbed ecosystems may not respond in the same way as the Dianchi wetland, meaning that RSEI might not fully capture their unique dynamics.

Despite demonstrating the effectiveness of RSEI combined with landscape pattern indices in assessing wetland ecological environment quality changes, discussing potential limitations is crucial. Ecosystems that are more complex or less disturbed may not respond in the same way as the Dianchi wetland, indicating that RSEI might not fully capture their unique dynamics. Future research should explore ways to improve or enhance this approach by incorporating additional indicators or developing hybrid models that integrate RSEI with other ecological or remote sensing tools.

5. Conclusions

This study emphasizes the effectiveness of the RSEI-based semi-supervised classification method in wetland ecological assessments, particularly in heterogeneous environments like the Dianchi lake. Compared to traditional SVM methods, the integration of various ecological indicators not only enhances classification accuracy but also shows better adaptability in capturing the complex landscape features of wetland systems and identifying areas of human disturbance.

(1)

This multi-indicator coupling approach provides a more detailed reflection of wetland ecological quality and serves as a robust method for monitoring environmental changes. Given that RSEI incorporates multiple ecological indicators, this method is broadly applicable to regions with complex landscape features and high ecological heterogeneity, especially in ecologically sensitive areas like the Dianchi wetlands. When evaluating multiple ecological indicators, the comprehensiveness of RSEI allows for strong adaptability to different ecological factors. The design of this method specifically considers the spatial heterogeneity of human activities, making it suitable for wetlands or other ecosystems significantly affected by human disturbances, such as areas impacted by urban expansion or agricultural development (e.g., the Guandu–Chenggong area studied here). Furthermore, the analysis based on RSEI disturbance proximity can further reveal the direct or indirect effects of human activities on ecosystems.

(2)

The incorporation of human disturbance metrics into the analysis further deepens our understanding of how the spatial heterogeneity of anthropogenic activities impacts wetland landscapes. This offers valuable insights into the dynamic interactions between landscape patterns and ecological changes across different spatial scales, revealing the mechanisms of landscape–ecology feedback. Consequently, this study presents a novel pathway for enhancing decision-making in wetland conservation and restoration efforts, emphasizing the spatial and temporal dimensions of human–environment interactions.

(3)

Future research directions should focus on refining the methodology by integrating additional ecological indicators and testing its application in diverse wetland types and geographic contexts. Moreover, the potential of RSEI for long-term monitoring under different climate scenarios and its use in predictive ecological modeling deserve further exploration. Such work will contribute to a deeper understanding of wetland ecosystem resilience and offer guidance for future conservation policies aimed at mitigating human impacts while promoting sustainable wetland management.