Introduction

Wetlands are among the most significant ecosystems on Earth1,2 contributing to a range of fundamental ecological functions and services3 including water conservation and purification, provision of aquatic products and water resources, and flooding control4. However, human activities (e.g., land use changes and dam construction) and natural environmental changes (e.g., global warming) have greatly impacted the biological communities (e.g., invertebrates, aquatic plants, and birds), physical habitats (quantity and quality), and ecosystem functions and services in wetlands5,6. It is estimated that over 35.0% of the global wetland area has been disappeared during 1970 to 20157. In addition, wetlands receive large quantity of anthropogenic pollutants such as heavy metals, nutrients, and persistent organic pollutants through various ways leading to the rapid loss of biodiversity and degradation of 50% of natural wetlands8,9. The degradation of wetlands not only negatively affected natural ecosystems, but also threatened human health and caused great economic losses. Therefore, the protection and restoration of wetlands is becoming one of the major concerns for local and central government in many regions and countries. These actions including establishing the Ramsar Convention, returning farmland to lake, ecological restoration, and establishment of natural reserves to achieve the goals of improving water quality, increasing biodiversity, and recovery of hydrological connections10. Nevertheless, one challenge in protecting and restoring wetlands is that we lack information of the long-term and dynamic monitoring of environmental changes across multiple scales11.

Satellite remote sensing technology has emerged as a vital tool for ecological environment monitoring12 primarily due to its broad coverage, cost-effectiveness, extensive time series, and capability for repeated periodic observations when compared to traditional methods13. Currently, there are two main directions in the research of monitoring the change of eco-environmental quality (EEQ) based on remote sensing technology. One is the quantitative evaluation of ecological environment quality based on a single generalized quality index; the other is to build a comprehensive evaluation system based on different quality indicators. For example, Jiang et al. (2021) used the Normalized Difference Vegetation Index (NDVI) to assess the spatial-temporal changes of the ecological environment at different scales from 1998 to 2018 in China14. However, as the EEQ is usually influenced by many factors and the driving factors may differ among studies15. Monitoring and evaluating the EEQ based solely on one single index may not be adequate to encapsulate the intricate dynamics of ecological environment changes16. The pressure-state-response (PSR) model stands out as it is founded upon a comprehensive evaluation index and is employed for the assessment of ecosystem health17. Another option is the Ecological Environment Status Index (EI) which was introduced by the China Ministry of Environmental Protection in 200618. The revised EI index was a prevalent tool to holistically represent different information such as biological abundance, vegetation coverage, water network density, land stress, and overall ecological quality19. However, although these comprehensive ecological indices reflect the ecological conditions more broadly, they do have some limitations, including the difficulty in data collection, the inability to merge extensive remote sensing and socio-economic datasets, and the lack of persistence in the spatial visualization of the final evaluation results20. Moreover, the assignment of weights to different indicators within these models often involves subjectivity which may affect the objectivity of the final evaluation results21.

To better solve the above problems, Xu et al. (2013) proposed the remote sensing ecological index (RSEI), a new method of ecological environmental quality assessment based solely on remote sensing data22. This index offers a visual representation of regional ecological quality based on captured remote sensing images, making it easier for researchers to monitor and evaluate the ecological conditions of continuous land cover and long-term regional series compared to the PSR model and EI index23. The RSEI utilizes principal component analysis (PCA) to integrate four ecological indicators: greenness, humidity, dryness, and heat, which represent by NDVI, WET, NDBSI, and LST, respectively. These four indicators are easy to obtain and do not require manual weight allocation, thus providing an effective and fair means for obtaining regional ecological environment information. In addition, they have advantages of avoiding changes or errors in weight definitions caused by differences in individual characteristics24. Consequently, the RSEI has been extensively applied in the monitoring and assessment of the EEQ of cities25 mining zones26 wetlands11,27 lakes28 and watersheds6. These successful applications of RSEI demonstrate its strong flexibility and reliability for assessing ecological health of various ecosystems11,29.

Even though, the ecological environment can be affected by many anthropogenic and natural factors6 most previous studies did not try to find out the driving factors of regional EEQ neither how these factors individually or interactively to influence EEQ30. In order to analyze the changes and implement targeted restoration measures, it is necessary to identify potential driving factors of ecological environment changes31. In terms of the spatial heterogeneity of EEQ, one question remains unclear is whether the driving factors have stronger effects on the ecological environment independently or interactively32. The Geo-detector model (GDM) has been proposed as a promising way to quantitatively analyze the driving factors of EEQ33. As a statistical model, GDM can not only detect spatial changes and identify potential driving factors, but can also quantitatively analyze the impacts of individual factors on ecological environment and identify the types of interactions between driving factors34. This model has been applied in many fields such as natural sciences, social sciences, and environmental sciences35,36.

Dongting Lake is an important regulating and storing lake in the middle Yangtze River basin, and it ranks as the second largest freshwater lake in China37,38. The water level of Dongting Lake fluctuates between 17.97 (February) and 28.00 m (July) with a mean level of 22.77 m. The lake begins to shrink on October and starts to enlarge on March each year with a change of water surface area up to 50–80%, and the water surface area may increase or decrease for > 1400 km2 within one month due to extreme climate39. Dongting Lake Wetland (DLW) is one of the most crucial wetlands in the world and an important Ramsar site40. The DLW has endured intensified and sustained anthropogenic and natural stressors (e.g., dam construction, sedimentation, reclaim land from lake, pollution, and climate warming) during the past decades, resulting in serious environmental and ecological problems including biodiversity and habitat loss, water quality degradation, and a marked shrinkage of the lake’s surface area41. These factors collectively contribute to the substantial decline of the EEQ in the DLW. Although, Hunan Provincial Government initiated the “Three-Year Action Plan for the Special Improvement of the Ecological Environment of Dongting Lake (2018–2020)” by initiating a series of projects such as returning farmland to the lake and ecological dredging42 which significantly improved the EEQ of DLW. These actions do not guarantee that the EEQ would continue to improve in the future because it takes time for the recovery of EEQ, and climate changes (e.g., extreme drought and flooding) and pollution (e.g., high level of phosphorus) still remain great threats which may reverse the above positive effects6. Moreover, the majority of previous studies conducting EEQ monitoring were focused on Dongting Lake basin and the Dongting Lake area during growing season (June–September)6,43. Whereas, the monitoring of EEQ in Dongting Lake during dry season (October–March), when the DLW services as important wintering habitats for thousands of birds and many other organisms40 may have been largely ignored10. Because most areas of DLW are submerged by water during the growing season, the EEQ may differ between flooding and dry seasons. For instance, Miao et al. (2024) utilized Landsat data and employed the Time Series Harmonic Analysis (HANTS) method to reconstruct an index and evaluate the ecological quality of Yuxi City. They found that there were significant differences in the RSEI between the plant growth period and the non-plant growth period, with the greenness of the plant growth period being higher than that of the non-growth period44. Zhao et al. (2023) evaluated the ecological environment quality of Otindag Sandy Land based on the newly revised remote sensing ecological index and found that the ecological environment quality varied at different time scales (year-round, growing season and non-growing season), and the implementation of ecological restoration measures should not ignore the seasonal characteristics of local ecological quality45. Furthermore, there is knowledge gap regarding to the long-term comprehensive monitoring and evaluation of the EEQ of DLW, and the spatial differentiation of EEQ. Accordingly, we used four remote sensing ecological factors to calculate the RSEI of the DLW in dry seasons from 1990 to 2020, and analyzed the spatial–temporal changes and spatial differentiation characteristics of the EEQ. We also analyzed and identified the main factors influencing the EEQ of the DLW by applying the Geo-detector model (GDM). We aimed to provide a scientific basis for the comprehensive environmental management and sustainable development of the DLW.

Study area

Dongting Lake (28°30′–29°38′ N, 112°18′–113°15′ E), situated in the northern region of Hunan Province and on the southern bank of the Jingzhou section of the Yangtze River, is the second-largest freshwater lake in China40. This lake is influenced by subtropical monsoon climate46 which experiences significant seasonal fluctuations in water level and surface area40. The rainy season is from April to September and the dry season is from October to March. The large seasonal water level variations foster diverse habitats and biodiversity including many endemic species. Consequently, the DLW has been recognized as an international important wetland under the Ramsar Convention47,48. Phragmites and grass are the main vegetation type in DLW, which play fundamental roles in supporting ecosystem functions including providing food resources for birds and removing pollutants49,50,51,52. Dongting Lake was once the largest freshwater lake in China. However, climate changes and human activities greatly reduced water area by nearly half since the twentith century49. Therefore, it is urgent to conduct systematic evaluation and research on the ecological environment of DLW. This study concentrated on the DLW, with special focus on the natural wetland area encompassed by the large dike of Dongting Lake (Fig. 1).

Fig. 1
figure 1

Location of the study area and scope of East Dongting Lake (E-DL), Hengling Lake (HLL), South Dongting Lake (S-DL) and West Dongting Lake (W-DL). The figure is created using ArcMap 10.2, https://www.arcgis.com.

Full size image

Data collection and pre-processing

Long-term remote sensing imagery were sourced from the United States Geological Survey (USGS) (http://www.usgs.gov/). As the DLW is a typical river and lake recharge wetland, its hydrological characteristics show obvious seasonal variations. During the rainy season, the lake area mainly presents as an open water landscape. At this time, the water bodies of rivers and lakes have largely submerged wetland vegetation, resulting in the blurring of ground features in this area on remote sensing images and making the identification of wetland boundaries difficult. As mentioned above, we exclusively utilized images captured during dry seasons of eight representative years (all were normal years). A normal year refers to a year in which there are no extreme climate events or excessive water level fluctuations. Meanwhile, when collecting images, this study referred to the water level data collected by the Chenglingji Hydrological Station of Dongting Lake for selection, eliminating to the greatest extent the impact of different water levels in the dry season within the year on the ecological environment quality of Dongting Lake. For details, please refer to Table S5 and Fig. S6. We only used images with cloud coverage < 20% to minimize uncertainties arising from excessive cloud interference and data integrity issues. Images that did not satisfy this criterion were replaced by adjacent years, therefore, the intervals between the eight representative years are not equal. All images were L1TP level data, alleviating the need for geometric correction. The requisite remote sensing image data for computation were derived after undergoing radiometric calibration and atmospheric correction pre-processes (Table 1). The elevation data used in this study was the ASTER GDEMV2 30 m dataset which were sourced from the Geospatial Data Cloud (https://www.gscloud.cn/). Hydrometeorological data including mean annual precipitation (PRE), mean annual temperature (TEMP), and mean annual potential evapotranspiration (PET) were obtained from the National Earth System Science Data Center (https://www.geodata.cn) and were presented as 1 km × 1 km raster datasets. The data is based on the global 0.5° climate data released by CRU and the global high-resolution climate data released by WorldClim, and was generated in China through the Delta spatial scaling scheme. The verification was carried out using the data of 496 independent meteorological observation points, and the verification results were reliable. The units of precipitation and potential evapotranspiration are 0.1 mm, and the unit of temperature is 0.1 °C. Socioeconomic indicators such as Gross Economic Product (GDP) and population density (POP) were downloaded from the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences (https://www.resdc.cn/), originating from China’s GDP Spatial Distribution Kilometer Grid dataset and China’s Population Spatial Distribution Kilometer Grid dataset, respectively. The vegetation factor (FVC) was obtained by calculating the NDVI of the study area using remote sensing images through ENVI5.6 software, and then inverting the vegetation coverage of the study area using the pixel bipartite model based on the NDVI data. The data type was raster data. The landscape factors (LSP) were obtained by classifying the landscape pattern of Dongting Lake Wetland from 1990 to 2020 using the maximum likelihood method through ENVI5.6 software. In the selection of training samples, the classic visual interpretation method was adopted to independently supervise and classify remote sensing images. After classification, the error matrix was constructed through the confusion matrix method, and the accuracy was verified by calculating the overall accuracy of the classification.

Table 1 Landsat image series data of Dongting lake wetland from 1990 to 2020.
Full size table

Results

Overall characteristics and correlation of indicators

Principal component analysis results

The PCA results showed that the eigenvalue contribution rate of the first principal component (PC1) exceeded 50%, with an average value of 61.61% (Table S1). In addition, the cumulative contribution rate was above 80%, indicating that PC1 had concentrated the majority of the characteristics observed in the four component indicators. These results suggested the high reliability of the four indicators to construct RSEI model. Furthermore, this model can be employed to analyze the temporal and spatial changes of EEQ in the DLW. In PC1, the characteristic values of NDVI and WET(both had positive impacts on the EEQ) were positive, while NDBSI and LST (both showed negative impacts on EEQ) were negative. In PC2–PC4, the indicators were uncertain, which was difficult to explain ecological phenomena, so it was reasonable to use PC1 to construct RSEI.

Correlation analysis between RSEI and evaluation index

In order to provide a more intuitive representation of RSEI, we examined the correlation between RSEI and the four evaluation indicators (Fig. 2). According to the load values of each index on PC1, RSEI positively correlated with NDVI and WET, while negatively correlated with NDBSI and LST. However, the strength of correlations may differ among years and among the four indicators.

Fig. 2
figure 2

The relationships between remote sensing ecological index (RSEI) and the four indicators (i.e., NDVI indicates greenness, WET indicates humidity, NDBSI indicates dryness, and LST indicates heat) in the Dongting Lake Wetland in the eight representative years.

Full size image

Spatial-temporal changes of EEQ in DLW

Temporal changes of EEQ in DLW

The RSEI values generally showed an initial decreasing and then fluctuating increasing trend (Fig. 3). According to the changes of RSEI values, we grouped DLW into three stages: degradation (1990–2001), fluctuation (2001–2008), and recovery (2008–2020). The lower quartile values in 1990, 1994, 2004, 2008 and 2013 were all higher than 0.4, indicating that the EEQ in most areas were at relatively high level. In addition, the RSEI values of 1994, 2013, 2016 and 2020, were at the excellent level (0.8–1), but the upper quartile values of the representative years did not exceed 0.6, indicating that most RSEI values were concentrated in the range of 0.4–0.6 and the changes were relatively stable.

According to the RSEI class distribution map of eight representative years, the area and proportion of each ecological class were calculated (Fig. S1). We calculated the sum of proportions at the poor, fair, and moderate level (MFP%), and the sum proportions of good and excellent level (EG%). There was a notable increase in the proportion of MFP%, from 72.87 to 94.11% during 1990–2001 (Table 2). Conversely, the proportion of EG% significantly decreased from 27.13 to 5.89%. This result further suggested the decline of EEQ during this period. From 2004 to 2008, the EEQ exhibited a trend of degradation. However, from 2008 to 2013, there was a trend of recovery. Finally, from 2013 to 2020, the EEQ fluctuated and increased.

Fig. 3
figure 3

Violin plot showing the changes of remote sensing ecological index (RSEI) in the Dongting Lake Wetland in the eight representative years during 1990–2020. The circle and the horizon line represent mean and median values, respectively.

Full size image

In this study, the area transfer matrix of each RSEI grade was calculated and the Sankey diagram was drawn to show the transformation of each RSEI grade in the DLW in each time period (Fig. S2). All RSEI of each level had been transformed into others, and the transformations were mainly concentrated in the levels of fair, moderate, and good.

Table 2 Changes in area proportion of each RSEI level from 1990 to 2020.
Full size table

Spatial changes of EEQ in DLW

The overall spatial distribution of EEQ in DLW showed higher values in the east while lower values in the west region (Fig. 4). The EEQ of excellent and good area were mainly distributed in the middle of the E-DL with a circular shape. The rest excellent and good area were distributed in the middle part of W-DL and in the middle part of HHL. The areas with moderate and fair EEQ were mainly found in the south, west and southeast of E-DL, and they basically showed a fixed patchy distribution. The rest moderate and fair regions were distributed in the south of HHL, the middle of S-DL and the north and south of W-DL. The regions with poor EEQ were mainly distributed in the south and west of E-DL, in the north of S-DL, and in the north and south of W-DL.

Fig. 4
figure 4

Spatial distribution of the eco-environmental quality of Dongting Lake Wetland in the eight representative years during 1990–2020. The figure is created using ArcMap 10.2, https://www.arcgis.com.

Full size image

Spatial autocorrelation analysis of RSEI in DLW

To ensure information integrity and the accuracy of quantitative evaluation, we combined internal characteristics of DLW, and RSEI images of the eight representative years were resampled using 1 km × 1 km grids. We got 1857–2106 sample points per year as water area in DLW fluctuated annually. Subsequently, the RSEI value, NDVI, WET, NDBSI and LST of the sampling points were projected in three-dimensional space, and the spatial distribution relationship between each index and RSEI were analyzed. There were positive correlations between RSEI values with both NDVI and WET values whereas, RSEI negatively correlated with NDBSI and LST (Fig. 5).

Fig. 5
figure 5

Three-dimensional scatter plots of the RSEI value and each index in the Dongting Lake Wetland sample points: (a) correlation of NDVI, WET, and RSEI; (b) correlation of NBDSI, LST, and RSEI.

Full size image

We found that the global Moran’s I statistic exceeded the 1% significance level in all years. Our scatter points were mainly distributed in the first and third quadrants of the image indicating positive spatial correlation and the spatial aggregation was significant (Fig. 6). The temporal changes of Moran’s I index were generally similar to that of RSEI and EEQ.

Fig. 6
figure 6

Moran scatter plots of RSEI in Dongting Lake Wetland in the eight representative years during 1990–2020: (a) 1990, 1994, 2001, and 2004; (b) 2008, 2013, 2016, and 2020.

Full size image

The spatial distribution of EEQ in the DLW exhibited a clustered pattern (Fig. 7). The distribution of the H-H cluster regions was analogous to that of the excellent and good EEQ levels, whereas the distribution of the L-L cluster regions was analogous to that of medium, fair, and poor EEQ levels. The L-H cluster regions were typically situated in the periphery of the H-H cluster regions, while the H-L cluster were generally distributed in the periphery of the L-L cluster regions.

Fig. 7
figure 7

LISA clustering map of RSEI in Dongting Lake Wetland in the eight representative years during 1990–2020. The figure is created using ArcMap 10.2, https://www.arcgis.com.

Full size image

Driving factors of EEQ in DLW

Single factor detection analysis

In order to gain further insight into the driving factors of EEQ change in the DLW, we grouped the influencing factors into two distinct categories: Remote sensing ecological assessment indicators (NDVI, WET, NDBSI, LST) were recognized as internal factors, and the remaining factors were considered as external drivers (Table 3). A 1 km × 1 km grid was established in the study area, and the natural breakpoint method was employed for the classification of the independent variables. We conducted classification and grading based on the actual situation of the driving factors and the characteristics of the DLW. Based on the previous classification criteria, the natural breakpoint method was used to classify the driving factors such as NDVI, WET, NDBSI, LST, elevation, slope, average annual precipitation, average annual potential evapotranspiration, average annual temperature, GDP, population density, landscape pattern, and vegetation coverage into five levels. Then, the aforementioned values were input into the GDM for factor detection, with the aim of detecting the impact of each factor on the EEQ. In addition, this process allowed us to ascertain the explanatory power of each factor with regard to the RSEI. Our results showed that the explanatory power of each index for RSEI was sufficient, and the P-values of each factor were significantly correlated at the 0.01 level. With regard to interannual variation, all 13 influencing factors exhibited alterations in their respective q values. With the exception of PRE and PET, the q value ranking of the remaining influencing factors exhibited only minor fluctuations (Tables S3 and S4).

Table 3 Results of single factor detection during 1990–2020.
Full size table

The average q values corresponded to the overall average driving forces and different impact factors from 1990 to 2020 (Table 3). As for the four internal driving factors, all q values were > 0.5, ranking in the first four places. Specifically, NDBSI and WET played a dominate role in regional EEQ. Regarding to external driving factors, the EEQ were mainly affected by vegetation coverage (FVC) (q = 0.417) and landscape pattern (LSP) (q = 0.347) which showed the strongest explanatory power for the spatial differentiation characteristics of RSEI. PET also explained > 10% of the variations in EEQ, and GDP (q = 0.074) and POP (q = 0.098) had significant impacts on EEQ. Other factors including PRE, TEMP, elevation (DEM), and slope (SL) had weak influences on RSEI (q < 0.060).

Interaction detection analysis

The interaction among internal and external factors were manifested as either enhancement or nonlinear enhancement (Fig. 8). Among the internal remote sensing ecological factors, WET and NDBSI had the strongest interaction explanatory power (0.8), and the interaction explanatory power of WET with other factors were > 0.6. The explanatory power of NDVI and LST interactions with other factors were > 0.5. The results showed that the remote sensing ecological evaluation index had significant influence on the EEQ of DLW. For external driving factors, the interaction between FVC and LSP had the greatest explanatory power on EEQ (q value = 0.63). In addition, any interactions including FVC and LSP had q values greater than 0.4 and 0.3, respectively.

Fig. 8
figure 8

Interactive detection results of factors affecting eco-environmental quality of Dongting Lake Wetland from 1990 to 2020. The larger the q value is, the more the independent variable explains the spatial differentiation of the dependent variable. Note DEM (X1): elevation; SLP (X2): slope; PRE (X3): precipitation; PET (X4): potential evapotranspiration; TEMP (X5): atmospheric temperature; LSP (X6): landscape pattern; GDP (X7): gross economic product; POP (X8): population density; NDVI (X9): greenness index; WET (X10): humidity index; NDBSI (X11): dryness index; LST (X12): heat index; FVC (X13): vegetation coverage.

Full size image

Driving force analysis of landscape pattern change in DLW

Our results showed that all types of landscape were being transformed into each other, and that the transformation was mainly focused on phragmites, grass, mudflat and water (Fig. S3). In general, the phragmites area first increased and then tended to stabilize, while the grass area first decreased and then fluctuated increased. Mudflat area first increased, then decreased and stabilized, and water area first decreased and then stabilized. The area of residential area remained relatively stable, and the proportion were generally lower than 1% (Table S7).

Phragmites were usually distributed in patches, mainly distributed in the northwest and south of E-DL, the north of HLL, S-DL, and W-DL. The grass was distributed around the central waterbodies of E-DL resulting in a circular pattern. In addition, the southern part of HLL, the central and southern part of W-DL showed lamellar distribution, and the central part of S-DL showed scattered distribution. The mudflat was mainly distributed around water areas, and the resident area showed a fixed distribution, which was basically unchanged (Fig. S4).

We calculated the RSEI mean values of main landscape types of DLW to investigate the impacts of landscape pattern changes on EEQ. As a large area of water in DLW was masked and the proportion of residential area was small, the water area and residential area were excluded in further calculation. The mean RSEI of phragmites and mudflat land were lower while that of grass was higher (Table 4). Therefore, the proportion of phragmites and mudflat (PM) land was grouped and compared with the proportion of grass (GR) (Fig. 9). The changes of RSEI indicated that when the proportion of PM was higher than GR, the mean value of RSEI in the year was lower indicating downward trend in EEQ (Fig. 3). Conversely, when the proportion of GR was higher than PM, the mean value of RSEI in the year was higher indicating an upward trend in EEQ. Generally, the changes of PM and GR ratio were consistent with the changing trends of EEQ.

Table 4 RSEI values of main landscape types in Dongting lake wetland from 1990 to 2020.
Full size table
Fig. 9
figure 9

Changes of main landscape proportion of Dongting Lake Wetland in the eight representative years during 1990–2020. PM: phragmites and mudflats, GR: Grass.

Full size image

Discussion

Temporal and spatial characteristics of EEQ in DLW

Our results demonstrated that the mean RSEI of the DLW during 1990 to 2020 ranged from 0.45 to 0.54, which was comparable to or slightly lower than the values observed in wetlands11,27 large floodplains or river basins6,29 while lower than forested or densely vegetated areas which usually exceeding 0.6324. This result indicated that the EEQ of DLW can be classified as medium level based on the classification criterion. Furthermore, the overall EEQ of DLW fluctuated annually, yet the intensity of changes between different RSEI levels were relatively low, with the majority of changes occurring between the “fair”, “moderate”, and “good” levels. The differential EEQ among ecosystems were mainly attributed to the comprehensive effects of climate conditions, soil characteristics, vegetation structures, and human activities6. Compared with the lower EEQ in desert areas, the EEQ in forest areas are usually at higher level because they have denser vegetation cover53.

The EEQ of DLW showed a trend of initially deterioration and then improvement during 1990–2020. The declining EEQ during 1990 and 2001 was mainly attributed to a series of severe floods including the extreme flooding event in 1998 in the Dongting Lake region54. Frequent flooding events had the effect of exacerbating the vulnerability of ecosystems55. The gradual increasing EEQ during 2001–2004, were probably due to the implementation of local government’s environmental protection policy, and the operation of the Three Gorges Dam Project (TGD) in 2003 which slowed down the rate of sediment deposition and reduced the quantity of water from the Yangtze River into Dongting Lake56. Consequently, water areas significantly decreased while the mudflats were transformed into vegetation and grassland54. This has led to an increase in vegetation coverage. The slightly deterioration of EEQ between 2004 and 2008 was also caused by abnormal climate, including an extreme drought in Dongting Lake in 200657. The adverse effects of drought on decreasing EEQ may be attributed to the dropped water level and shrinking water surface area which resulted in increasing bare land and deteriorative habitats for animals and plants58. In addition, the operation of TGD may exacerbate the severity of drought by changing the hydrological connection and water balance between the Dongting Lake and the Yangtze River59. This may lead to an intensification of anthropogenic droughts in the DLW, such as the earlier occurrence of dry season46. There was a discernible improvement in the EEQ of DLW from 2008 to 2020, which attained a state of stability. This was because the efficacious implementation of the special ecological remediation plan for Dongting Lake and the government’s endeavors to reinstate the wetland ecological environment6.

The EEQ of the DLW exhibited a distinct spatial pattern with higher EEQ in the eastern while lower EEQ in the western region. This pattern largely corresponded with the spatial distribution of nature reserves in the Dongting Lake. East Dongting Lake National Nature Reserve was established in 1982 and it is the largest nature reserve in the Dongting Lake. The EEQ of “excellent” and “good” level generally distributed in the core regions of each nature reserve, while other levels were mostly distributed in the buffer and experimental zones. The core area of nature reserve usually prohibit or severely restrict human activities to maintain the health and integrity of protected ecosystems60. It played an important role in reducing pollution, increasing biological biodiversity, maintaining natural hydrological dynamics and connections, and enhancing the resistance and resilience to natural and anthropogenic stressors, all of which may contribute to the increasing high EEQ61. Buffer zones and experimental zones were designed to protect the core area while allowing low-moderate human activities. Therefore, human activities were relatively frequent, which may impose some negative impacts on the ecological balance and biodiversity of wetlands11.

Spatial differentiation characteristics of EEQ in DLW

Our results showed that the EEQ of the DLW exhibited a discernible spatial aggregation and a positive correlation. Furthermore, the changes of Moran’s I index were similar to that of RSEI, which is similar with the results of previous studies in Erhai Lake Basin and Jianghan Plain29,62. Each cluster region had certain distribution characteristics. The H-H cluster areas were mainly distributed around the water bodies in the middle of E-DL (core protected areas of nature reserve), where most areas were grassland, with low level of human activities63. The high density of grass with well-developed roots making the soil had a high water-holding capacity, with further improvement in the EEQ64. The L-L cluster areas were mainly located in the south of E-DL, in the north of HLL and S-DL. Most of these areas were covered with phragmites and mudflats. However, most reed in Dongting Lake were planted and harvested for economic benefits which resulted in large number of reed factories being operated during the past decades65. In addition, the improper ways to harvest, transport, and process of phragmites may induce problems of pollution and habitat loss to cause further ecological problems66. Therefore, these activities may impose great negative effects on the ecological and environmental quality, even though the existence of reed benefited many freshwater ecosystems67. Likewise, although the area of mudflats increased by 45.46 km2 during 1995–2020 and mudflats were crucial habitats for birds and other animals68,69. The mudflats had been significantly impacted by human activities, including illegal sand mining and unpermitted vegetation planting, all of which may overridden the positive effects on EEQ and resulted in the overall low EEQ.

Driving mechanism of ecological environment of DLW

Our results suggested that the EEQ of DLW was synergistically (i.e., enhanced) influenced by internal and external factors, and internal factors (NDBSI and WET) had the greatest explanatory power because RSEI was coupled by them53. The positive ecological benefits by WET while negative ecological effects by NDBSI agrees with the finding in the Ebinur Lake Wetland in Xinjiang11. WET and NDBSI can reflect climate changes to some extent. In normal years, WET was positively correlated with EEQ because the long-term water retention function of wetlands and the growth of dense vegetation on the surface require a lot of water11. Increased dryness was usually induced by the decreasing of rainfall and lower water levels, which can expose plant roots, affecting their growth and reproduction. This can lead to irreversible changes in wetland hydrological conditions, such as a shift from seasonal flooding to prolonged drought70.

In addition to internal factors, external factors, such as vegetation coverage also showed strong positive effects on the EEQ. This result is in accordance with other studies30. High level of vegetation cover can not only mitigate the impact of human activities on wetlands, but also provide ideal habitats for numerous plants and animals71. Plants such as reed, sedge, and many other gramineous plants are the foundation for the formation of DLW ecosystem services and functions, especially in preventing soil erosion and can effectively improve the EEQ53. One good example of this is the Weihe River Basin which was the first region area on the Loess Plateau to implement ecological improvement projects, including returning farmland to forest and grassland. These projects significantly increased the vegetation coverage leading to the expansion of surface water area and the improved EEQ72. Landscape change also played a role in affecting the EEQ of DLW. Different landscape cover types provided suitable environments for a variety of organisms, and this habitat diversity was key to support high biodiversity6,73. This was because the increase of bare land will reduce water infiltration, while more vegetated area can increase water evaporation and transpiration, and promoted the water cycle74 which also reflects the synergistic effect between landscape pattern and vegetation cover. The changed landscape pattern were mainly attributed to human activities73 such as rapid development of urbanization6. As indirect influential factors, GDP and POP showed significant negative impacts on the EEQ of DLW. Because higher population density and higher GDP mean the existence of high intensity human activities, especially the replacement of natural landscapes (e.g., wetlands and forests) by human-dominated landscapes73.

The effects of potential evapotranspiration and climatic factors on the EEQ were dependent. Through evapotranspiration, wetlands increase atmospheric humidity and reduce atmospheric temperature, thus facilitating the formation of rainfall. It contributes to the mitigation of extreme arid climates and plays an important role in the hydrological regulation of wetlands75. Because DLW was located in the subtropical humid monsoon climate zone, the temperature was suitable and the precipitation was sufficient64. We found that precipitation and temperature had less impact on the ecological environment in normal years than in years of extreme weather (e.g. extreme droughts and floods). Although precipitation and temperature were not the main factors in influencing the ecological environment, it did not mean that we could ignore their impacts76. Topographic factors (elevation, slope) were the least explanatory factors for EEQ in our study. This is because the DLW was located in the Dongting Lake plain. The topography exhibited minimal variation in elevation and slope. The effects of elevation and slope on the distribution of aquatic ecosystems in this terrain were very limited69.

Future directions and limitations

Although our study advanced the understanding of EEQ in DLW, there is still several limitations such as data sources. The MNDWI was used to mask the water information in this study. In other words, we focused exclusively on monitoring the EEQ of non-water bodies within the DLW. However, the water body was an important part of the ecosystem, which was not fully represented in our current research. In fact, the Xiangjiang River, Zijiang River, Yuanjiang River, Lishui River and other water bodies played an important role in maintaining the ecological security and ecological balance of the DLW to influence the EEQ6. Future studies should enhance the precision of ecological environmental assessments by incorporating a broader range of data sources and utilizing multi-source remote sensing data fusion techniques. This would enable a more thorough analysis of the evolution patterns of EEQ. Concurrently, the incorporation of data pertaining to land resources, road traffic, and water quality into the model resulted in a notable enhancement in its evaluation accuracy, thereby facilitating a more precise assessment of the EEQ changes observed in the DLW.

Conclusions

In conclusion, we firstly found that the mean values of RSEI of DLW from 1990 to 2020 varied between 0.45 and 0.55, which suggests that the EEQ was at a medium level. The temporal changes of EEQ can be grouped into three stages: degradation (1990–2001), fluctuation (2001–2008), and recovery (2008–2020). The EEQ also showed significant spatial differences with a general higher level in the eastern while lower level in the western DLW, which closely related to the distribution of nature reserves. Secondly, the spatial distribution of EEQ in the DLW demonstrated a positive correlation and exhibited a cluster pattern, with the High-High cluster areas were associate with grassland while the Low-Low cluster areas were characterized by phragmites and mudflats. This result highlights that the protection or recovery of grassland would impose positive effects on the EEQ of surrounding environments, whereas, the decrease of EEQ may also cascade to the surrounding environments. Finally, the EEQ of DLW is influenced by both natural and human factors and these factors synergistically interacted with each other. Among them, vegetation coverage and landscape pattern were the dominant external driving factors, followed by potential evapotranspiration and socio-economic. The proportion of reeds and mudflats is positively correlated with the degradation of the EEQ of DLW. This is caused by human harvesting of reeds and grassland degradation. The proportion of grassland is positively correlated with the improvement of the EEQ of DLW, which indicates that vegetation coverage can enhance the EEQ. In addition, for key landscape types and risk areas with low RSEI levels, it is recommended to implement targeted local management strategies, including ecological restoration projects (such as vegetation restoration and the establishment of a long-term ecological water replenishment mechanism).

Methods

Calculation of RSEI

RSEI consists of four coupled components: NDVI, WET, NDBSI, and LST77. Before conducting the principal component analysis, it is necessary to normalize the four indicators to ensure that the data range of the four indicators are within [0,1]. The closer the RSEI value is to 1, the better the EEQ23. We classified the RSEI values into five groups: poor (0–0.2), fair (0.2–0.4), moderate (0.4–0.6), good (0.6–0.8) and excellent (0.8–1.0)24. The greenness is represented by the Normalized difference vegetation index (NDVI), which reflects the vegetation physical characteristics such as vegetation coverage and plant growth in an area78. We employed the wet component derived from the Tasseled cap transformation as the humidity index79. The dryness is represented by the Normalized difference impervious surface index (NDBSI), which combines the Index-based built-up index (IBI) and Soil Index (SI) for its computation80. Land surface temperature (LST) was used to represent the heat index, and the atmospheric correction method in the single-channel algorithm was used to retrieve the heat index62. We adopted the Modified normalized difference water index (MNDWI) to circumvent the confounding effects of such large water bodies on normalization and the subsequent influence on the eigenvalue of WET and NDBSI in the RSEI calculation49,81. This index serves to mask and eliminate the vast water regions from our study area prior to computing the four evaluation indices. The calculation formulas of each index are shown in Table S8.

Spatial autocorrelation analysis

The spatial correlation analysis of EEQ can be used to describe the spatial uniform distribution of EEQ in the study area62. In this study, the spatial correlation of RSEI was analyzed using Global and Local Moran’s I autocorrelation. The global Moran’s I index reflects the correlation between the attribute values of adjacent spatial units, and can directly reflect the spatial aggregation and spatial anomaly of RSEI in DLW. The calculation formula is as following32:

$${\text{Global}}\;{\text{~moran's}}{{~}}\;I = \frac{{m \times \mathop \sum \nolimits_{{i = 1}}^{m} \mathop \sum \nolimits_{{j = 1}}^{m} W_{{ij}} \left( {D_{i} - \bar{D}} \right)\left( {D_{j} - \bar{D}} \right)}}{{\mathop \sum \nolimits_{{i = 1}}^{m} \mathop \sum \nolimits_{{j = 1}}^{m} W_{{ij}} (D_{i} - \bar{D})^{2} }}$$
(1)

Where, m is the total number of factors, Di represents the EEQ value of the i position, D represents the average EEQ of all factors in the study area, and Wij is the spatial weight.

Local spatial autocorrelation can test whether there is a cluster of variables in a local region. Within the LISA (Local Indicators of Spatial Association) clustering map, there are five distinct types of local spatial aggregation: high-high (H-H), low-low (L-L), high-low (H-L), low-high (L-H), and non-significant clustering82. The LISA index can be used to analyze the Moran’s I value of each spatial unit. The following formula is used for calculation:

$${\text{Local}}\;{\text{moran'}s}\;{{~}}I = \frac{{\left( {Di - \bar{D}} \right) \times \mathop \sum \nolimits_{{j = 1}}^{m} Wij\left( {Dj - \bar{D}} \right)}}{{\mathop \sum \nolimits_{{i = 1}}^{m} (Di - D)^{2} }}$$
(2)

Where, Local moran’s I represents Local Moran’s I index, and the calculation parameters are the same as Moran’s I index.

Geo-detector model

GDM are powerful tools for exploring and analyzing spatial data, as well as statistical methods for analyzing the driving factors53 these analyses including factor detectors, interaction detectors, risk zone detectors, and ecological detectors78. In this study, factor detector and interaction detector were used to investigate the explanatory power of spatial heterogeneity of EEQ in DLW and the relationship among the factors. The explanatory power of the factors was used to quantitatively measure the extent of the influence factors on the spatial differentiation of RSEI in DLW. The higher the q value of the influence factor, the stronger the explanatory power to the EEQ. Conversely, a lower q indicates a weaker explanatory power. The calculation formula of factor detector is as following78:

$$\:q=1-\left({\sum\:}_{h=1}^{L}{N}_{h}{\sigma\:}_{h}^{2}\right)/N{\sigma\:}^{2}=1-\left(-\frac{SSW}{SST}\right)$$
(3)

Where, 0 ≤ q ≤ 1. The closer the value of q is to 1, the stronger the factor’s ability to explain Y. Nh and N are the number of samples in layer h and the whole region, respectively. σ h and σ are the variance of Y in the h layer and the whole region, respectively. SSW is the sum of in-layer variances. SST is the total variance for the entire region.

The interaction detector was used to assess the effect of combined interactions between multiple independent variables X on the explanatory power of the dependent variable Y. It examines whether these interactions leading to an increase, decrease, or no change in explanatory power when compared to single variables83. There are five types of interaction between two factors which are detailly described in Table 5.

Table 5 Types of the interaction.
Full size table

Landscape classification

We used the maximum likelihood method, a prevalent technique in landscape ecology, for landscape classification46. ENVI5.6 software was employed to interpret the remote sensing images from each representative year listed in Table 1. To improve the maximum likelihood classification accuracy, as training samples, in this study, no less than 50 regions of Interest (ROI) were selected for each type of ground object. To ensure the representativeness of the samples, the selection of the test samples also comprehensively referred to the previous research results, the interpretation results of satellite images, and field investigation data. In the selection of training samples in this study, the classic visual interpretation method was adopted to independently supervise and classify remote sensing images. The relevant basis for visual interpretation can be found in Table S6 in the supplementary materials. Furthermore, to avoid the influence of spatial autocorrelation, the corresponding ROI of each remote sensing image is independently selected as the training sample. After the selection of training samples is completed, on the premise of ensuring that there is no spatial overlap in the distribution among the samples, a comparable number of test samples are selected for subsequent accuracy verification. The specific selection method of the samples is shown in Fig. S5.

Subsequently, these interpreted images were converted into vector data using ArcGIS, which was then reclassified to derive the landscape pattern classification results for the study area. The DLW encompasses five distinct landscape types: phragmites, grass, mudflat, water, and habitation. The selection of training samples followed principles of uniform distribution and representativeness. The sample separability tool in ENVI was used to calculate the degree of separation, with all sample separation values exceeding 1.80. The results in Table S2 of the supplementary materials show: The accuracy of the classification results of remote sensing images was evaluated by using the Kappa coefficient. The Kappa coefficient always exceeded 0.85, and the overall accuracy exceeded 90%. This indicates a high level of agreement between the classification outcomes and the actual landscape features, suggesting that the results of the remote sensing image classification are highly reliable and accurately reflect real-world conditions.