Introduction

The ecological environment is fundamental for human well-being and development. Ecological quality (EQ) reflects the balance between human activities and natural processes, directly influencing the quality of life and public health1. Scientific assessment and monitoring of EQ constitute a fundamental component for deciphering ecosystem stability, resilience, and adaptability2. Such evaluations enable the identification of critical thresholds and the early warning of potential ecological shifts3. Research on EQ is evolving to encompass diverse themes encompassing:, quantitative assessment frameworks4drivers of spatiotemporal variation such as policy interventions5land use6natural factors7,8. ecological restoration efficacy9 and tourism activities10.These studies reflect methodological maturation in EQ research.

Quantifying EQ presents inherent scientific challenges11. Traditional metrics like the Air Quality Index (AQI)12Water Quality Index (WQI)13and Soil Health indicators14 lack the resolution to characterize EQ holistically. In contrast, composite indices overcome the inherent limitations of single-metric approaches,, facilitating a more holistic evaluation of EQ. Notable frameworks include the Environmental Quality Index (EQI)15the Ecological Index (EI)16the Pressure-State-Response (PSR) framework17and ecosystem service assessment models18. Satellite remote sensing technology has revolutionized EQ monitoring methodologies, offering unprecedented data accessibility, spatial coverage, and multidecadal temporal continuity-enabling multiscale diagnostics of ecosystem factors such as vegetation19land use20air quality21and water resources22. The Remote Sensing Ecological Index (RSEI), extracted from satellite imagery, is a comprehensive indicator that integrates greenness, humidity, dryness, and heat indices to characterize the ecological attributes of a region23. The RSEI has been utilized in long time series24multi-scenario simulations25 and has exhibited its efficacy in EQ monitoring and assessment across multiple spatial scales, ranging from urban regions26 to river basins27marshland ecosystems28and industrial zones29. To enhance the pertinence of EQ assessments, researchers have further refined the RSEI by incorporating supplementary indicators tailored to specific study areas. Prominent examples of these indicators include the Comprehensive Salinity Index (CSI)30the Soil Erosion Indicator (SEI)31and Aerosol Optical Depth (AOD)32. These studies have confirmed the applicability and reliability of RSEI in assessing EQ.

To formulate effective management strategies, policymakers must comprehensively understand the multiscale spatial heterogeneity and temporal trajectories in Eq33.. Key research focuses on three areas: (1) Identifying long-term trends using moving averages and filtering methods34(2) Detecting local changes through Theil-Sen slope estimation and Mann-Kendall tests35. (3) Analyzing spatial patterns with spatial autocorrelation36 and standard deviation ellipse37. However, global-level analysis alone may miss regional differences because natural and socioeconomic factors vary across areas To address this, researchers combine GeoDetector and geographically weighted regression (GWR)38. GeoDetector detects spatial differentiation and assesses the impact of various factors on EQ, while GWR models offer a more refined understanding of regional impacts on Eq39.. These methods have been successfully used in studies related to soil40ecological environments41and urban infrastructure42. Importantly, human activity regulations significantly shape EQ. Growing evidence highlights policies’ roles in protecting ecosystems43air pollution control44flood management45and ecological security46.

Designated as a UNESCO World Natural Heritage site, the Li River Basin’s tower karst landscape possesses significant ecological conservation value. Prior studies on its ecological environment have addressed tourism impacts47land use pattern48and landscape ecological risks49. However, three critical research gaps persist: (1) the lack of long-term EQ monitoring; (2) unresolved interactions between natural conditions, human activities and EQ across spatial scales; and (3) unexplored linkages between environmental policies and ecosystem responses. To resolve these limitations, this study Built a 27-year RSEI using Google Earth Engine (GEE). Tracked EQ dynamic through HP filtering and Theil-Sen slope analysis. Quantified driving factors via integrated GeoDetector, GWR, and text mining. These outputs provide reliable data support for tailoring conservation strategies in the Li River Basin.

Study area and data sources

The Li River originates from Mao’ er Mountain, the highest peak in southern China (2,141.5 m). As the upper reaches of the Gui River, a tributary of the Xinjiang River within the Pearl River system, the Li River Basin is located in the northeastern part of Guilin, Guangxi, China, spanning a total area of 12,128.7 km² (Fig. 1). The region lies within the subtropical monsoon climate zone, characterized by long summers, mild winters, and abundant annual precipitation. Moreover, the Li River Basin is distinguished by its unique karst hydrogeological characteristic, which governs groundwater flow, influences vegetation water, and contributes to minor subsidence. These natural conditions support diverse vegetation types and a distinctive ecosystem. Renowned for its scenic landscapes and cultural heritage, the Li River Basin is among China’s earliest tourism-developed regions, with millions of visitors annually and a globally recognized tourist destination.

Fig. 1
figure 1

Geographical location and elevation map of the study region. The actual scene was obtained from the Visual China website (https://www.vcg.com/). The vector map data for China and the Guangxi Zhuang Autonomous Region was obtained from the National Platform for Common GeoSpatial Information Services (https://www.tianditu.gov.cn/), with map approval number GS (2024) 0650. The delineation of the study area boundary was conducted using ArcMap 10.2 (https://www.esri.com/). The Digital Elevation Model (DEM) was obtained from the Geospatial Data Cloud (http://www.gscloud.cn/), and the base map is a hillshade layer provided by the Map World map service (http://lbs.tianditu.gov.cn/server/MapService.html), imported into ArcMap 10.2 via the GIS server.

Full size image

This study utilized a range of data sources, including surface reflectance datasets from Landsat 5 TM, Landsat 7 ETM+, and Landsat 8 OLI, as well as China’s map data (map approval number GS (2024) 0650). Additional data incorporated included elevation, precipitation, and temperature data50along with socioeconomic indicators such as land use51GDP52population density, scenic spot lists, Nighttime-light Dataset53and policy text data. Table 1 contains a detailed enumeration of these data sources.

Table 1 Data Sources.
Full size table

Data preprocessing followed standardized scientific protocols: (1) Landsat surface reflectance (SR) product (30 m spatial/16-day temporal resolution) was processed via JavaScript programming on the GEE platform to calculate RSEI. The SR products on GEE were pre-corrected for radiometric, geometric, and atmospheric factors, eliminating the need for additional preprocessing. Annual composites integrated data from Landsat 5 TM (355 scenes: 1995–2011), Landsat 7 ETM+ (14 scenes: 2012), and Landsat 8 OLI (198 scenes: 2013–2021). Cloud removal was achieved using the QA band derived from the CFMASK algorithm, and only images with cloud cover of less than 50% were selected. All calculations employed the annual median synthesis method, effectively mitigating the impact of missing data in Landsat 7 ETM + SLC-off products54. (2) 1995 population density was estimated using the natural population growth rate and population density data from 2000. (3) The coordinates of tourist attractions were obtained from the list of sites using the Baidu Coordinate Picker System (https://api.map.baidu.com/lbsapi/getpoint/). These coordinates were then converted into the WGS-84 format using GIS technology to ensure consistency with international standards. A vector database of scenic spots was created in ArcMap 10.2.

To ensure consistency and comparability, all data were projected into the WGS_1984_UTM_Zone_49 N coordinate system in ArcMap 10.2, with raster image resolution standardized to 30 m and images clipped according to the vector boundary of the Li River Basin.

Research framework

The workflow of this study is outlined in Fig. 2. Initially, four remote sensing indices—WET, NDBSI, LST, and NDVI—were derived from the image dataset on the GEE platform, and PCA was applied to construct the RSEI from 1995 to 2021. Spatial and temporal trends in EQ within the Li River Basin were then analyzed using the Hodrick-Prescott filter, Theil-Sen slope estimation, and the transfer matrix. Finally, the factors driving changes in EQ were quantitatively assessed with GeoDetector and GWR, while policy impacts were analyzed using text mining.

Fig. 2
figure 2

Technical Flowchart. The map is designed using Adobe Photoshop software (https://www.adobe.com/products/photoshop.html).

Full size image

Methods

Ecological quality evaluation by RSEI

The utilization of RSEI to assess EQ is both effective and reliable55. This index integrates four ecological indicators that reflect conditions easily perceived by humans: greenness, wetness, dryness, and heat. To mitigate the impact of water bodies on these metrics, the Modified Normalized Difference Water Index (MNDWI) was applied in this study56. Detailed calculations for each indicator are presented in Table 2. The equations use symbols to represent reflectance values across various spectral bands: ρblue, ρgreen, ρred, ρnir, ρswir1, and ρswir2 for blue, green, red, near-infrared, short-wave infrared 1, and short-wave infrared 2 respectively. Other variables in the equations are CT, representing sensor temperature, ρ, with a value of 1.438 × 10−2 mK, ε, denoting surface emissivity, and λ, symbolizing the wavelength of thermal infrared. Specifically, λTM equals 11.435 μm, and λOLI is 10.9 μm.

Table 2 Methodology for calculating indicators.
Full size table

Normalization was applied to address unit and value range disparities to standardize indicator scales. Principal Component Analysis (PCA) was then performed, utilizing the first principal component (PC1) as the initial RSEI (RSEI₀). Component loading signs denote ecological contributions: positive loading reflects a beneficial effect, while negative loading suggests a detrimental impact. Notably, when loading polarities contradicted ecological expectations (NDVI/Wet negative or NDBSI/LST positive), sign correction was implemented via 1 – PC1 transformation57. Finally, the standardized RSEI value is obtained by normalizing the initial composite index (RSEI₀) using the following formula:

$$\:{RSEI}_{0}=\left\{\begin{array}{c}\:\:\:\:PC1\left[f\left(Greenness,Wetness,Heat,Dryness\right)\right],\:\:\:\:\:\:\:\:{V}_{Greeness},{V}_{Wetness}>0\\\:1-PC1\left[f\left(Greenness,Wetness,Heat,Dryness\right)\right],\:\:\:{V}_{Greeness},{V}_{Wetness}<0\end{array}\right.$$
(1)
$$\:RSEI=({RSEI}_{0}-{RSEI}_{0min})/({RSEI}_{0max}+{RSEI}_{0min})$$
(2)

The RSEI is measured on a scale from 0 to 1, with higher values indicating better EQ. It is divided into five distinct categories: poor (0 to 0.2), fair (0.2 to 0.4), moderate (0.4 to 0.6), good (0.6 to 0.8), and excellent (0.8 to 1)23.

Spatiotemporal evolution analysis of ecological quality

Hodrick-Prescott filter

The Hodrick-Prescott (HP) filter decomposes time-series data into its trend and cyclical components, thereby facilitating the analysis of trends in the RSEI data. Given that the RSEI series is represented as {Yt} for t = 1, 2, …, n, where n denotes the sample size, this series comprises a long-term trend c omponent (YtT) and a short-term fluctuation component (YtWT). The model is represented by the following equation:

$$\:{Y}_{t}={Y}_{t}^{T}+{Y}_{t}^{W}$$
(3)

The HP filter can separate YtT from Yt. The minimization of a specific function is required to achieve this:

$$\:\text{m}\text{i}\text{n}\left\{\sum\:_{t=1}^{T}{\left({Y}_{t}-{Y}_{t}^{T}\right)}^{2}+\lambda\:\sum\:_{t=1}^{T}{\left[\left({Y}_{t+1}^{T}-{Y}_{t}^{T}\right)-\left({Y}_{t}^{T}-{Y}_{t-1}^{T}\right)\right]}^{2}\right\}$$
(4)

It is evident that as the value of λ increases, the trend line becomes smoother. In this study, we refer to Ravn and Uhlig’s study58 and choose λ = 6.25.

Theil-sen slope Estimation

The Theil-Sen slope estimator is a widely employed technique for evaluating trends in substantial time-series raster data59. This method quantifies RSEI trends, providing insight into long-term changes in regional EQ. The estimator calculates the median of the slopes, offering a robust measure that minimizes the influence of outliers60. The following equation is employed to calculate the Theil-Sen slope:

$$\:\beta\:=Median\left(\frac{{X}_{j}-{X}_{i}}{j-i}\right),\forall\:j>i$$
(5)

Xj represents the data values for year j, while Xi refers to those for year i. β denotes the slope, with a value of β ≥ 0.0005 signifying improvement, while β< −0.0005 indicates degradation38. A β value ranging from − 0.0005 to 0.0005 indicates no significant change.

Hurst

The Hurst index, introduced by Hurst61is a widely employed metric for evaluating the degree of correlation in long-time-series data. It establishes a correlation between future fluctuations and historical patterns, facilitating predictions based on past data. The Hurst index is calculated using the rescaled range (R/S) method. For detailed calculation formulas and principles, please refer to references62,63.

Driving factors analysis of ecological quality

Multi-dimensional interacting factors with complex mechanistic process drive EQ dynamics3. Based on existing research and regional natural and economic conditions, this study identifies eight key drivers, including slope (Sp), elevation (Dem), temperature (Tem), precipitation (Pre), population density (Pop), land use intensity (LUI), GDP, and tourism activity intensity (TRI).

Tourism activities were classified into daytime and nighttime types. In the Li River Basin, daytime tourism predominantly involves visiting tourist attractions. So, its intensity is measured by the kernel density index of A-level scenic spots and rural tourism areas. Given that tourism is the primary industry in the Li River Basin and plays a pivotal role in the nighttime economy, previous studies have demonstrated a strong spatial correlation between nighttime light intensity and the distribution of tourism activities64. Therefore, nighttime light data were used to proxy for nighttime tourism intensity. The intensity of tourism activity is obtained by equally weighting and superimposing the daytime and nighttime tourism indices.

The comprehensive index of land use degree proposed by Zhuang Dafang et al.65 measures land use intensity, where different land use types are assigned graded index values. The index for other land types is 1, for forests and grasslands is 2, for cultivated land is 3, and for artificial surfaces is 4. The calculation formula is as follows:

$$\:La=100\times\:\sum\:_{i=1}^{n}{A}_{i}\times\:{C}_{i}$$
(6)

where La denotes land use intensity. Ai represents the unique metric used to categorize the intensity level of land use for the i-th tier. Ci denotes the percentage of the total regional area allocated to the i-th level of land use intensity. Additionally, the variable n is the parameter that ranks the various degrees of land use intensity.

GeoDetector

GeoDetector is a powerful tool for analyzing spatial variability for individual factors and their interactions66. In this study, the factor detector quantified the contribution of each driver to RSEI spatial variation, while the interaction detector evaluated the combined effects of two factors. The q-values from both methods range from 0 to 1. For further details on the principles and mathematical formulation, see Wang and Xu’s study in 201767. The grid-scale was set to 1 km, consistent with data precision.

Spatial autocorrelation analysis

Understanding spatial dependence is essential for identifying heterogeneity patterns and describe the distribution of geographic features68. Spatial autocorrelation analysis- an indispensable method for assessing dependency in spatial data – was performed using Global Moran’s I to evaluate RSEI’s spatial autocorralation. Moran’s I range from − 1 to 1, with values greater than 0 indicating positive spatial correlation and values less than 0 indicating negative spatial correlation. A larger absolute value indicates stronger spatial dependence, while values near 0 suggest a random spatial distribution. Statistical significance was set at P < 0.05, with Z-score used for testing. For detailed formulas and principles, refer to69,70.

Geographically weighted regression (GWR)

GWR incorporates spatial locations and local characteristics71offering a more accurate understanding of how these factors influence RSEI in different regions. The formula used for the calculation is provided below:

$$\:{y}_{1}={\beta\:}_{0}({u}_{i},{v}_{i})+{\beta\:}_{1}({u}_{i},{v}_{i}){x}_{i1}+{\beta\:}_{2}({u}_{i},{v}_{i}){x}_{i2}+\cdots\:+{\beta\:}_{k}({u}_{i},{v}_{i}){x}_{ik}+{\epsilon\:}_{i}$$
(7)

In this context, yi denotes the dependent variable’s measurement at the i-th location. The coordinates (ui, vi) correspond to point i. xik denotes the value of the k-th independent variable at point i. In the regression model for the i-th point, β₀(ui, vi) serves as the location-specific intercept, while βk(ui, vi) denotes the coefficient corresponding to the k-th predictor variable, anchored at the geographical center coordinates (ui, vi) of the i-th region. Finally, εi denotes the stochastic error component inherent to the observation at point i. GWR analysis will be carried out in ArcGIS software, with FIXED selected as the kernel type, the corrected Akaike Information Criterion (AICc) selected for the bandwidth method, and the number of points in each grid cell used as spatial weights.

To enhance the effectiveness and reliability of the GWR model, the study eliminates highly correlated and non-correlated factors based on Pearson correlation analysis. After that, appropriate variables are selected through OLS model multicollinearity tests and significance evaluation. The traditional regression models are based on Ordinary Least Squares (OLS) to estimate parameters. It is typically composed of a random error term εi and a set of parameters β0 and βi. The formula is as follows:

$$\:{y}_{i}={\beta\:}_{0}+\sum\:_{i=1}^{k}{\beta\:}_{i}{x}_{i}+{\epsilon\:}_{i}$$
(8)

Policy analysis of ecological quality

Text mining effectively extracts key terms related to ecological conservation and restoration from large textual datasets46. This study employs Python 3.8 to analyze policies concerning socioeconomic development and the natural environment in the Li River Basin. The analysis produces word clouds, providing data to evaluate the effectiveness of macro-level policies in ecological protection. The detailed procedures are as follows:

  1. a.

    Data acquisition and screening. The study collected 63 policy documents. To ensure a high level of relevance, the Python program was used to perform segmented screening on each policy text, extracting paragraphs where keyword combinations appeared, and saving them into an Excel file as a dataset.

  2. b.

    Text preprocessing. The Jieba library was utilized to execute tasks such as word segmentation, the elimination of stop words, special characters, and punctuation marks on the dataset.

  3. c.

    Text mining and quantitative analysis. The Gensim library in Python was utilized to call the gensim.models.TfidfModel is used for TF-IDF calculation to determine feature words and their weights.

  4. d.

    Generation of word clouds. Feature words with higher weights at each stage were extracted, and the WordCloud function in Python was used to generate the word cloud.

Results

Results of ecological quality evaluation based on RSEI

Table 3 demonstrates that PC1 accounts for more than 65% of the total variance. Moreover, the loadings for each indicator within PC1 exhibited a consistent distribution from 1995 to 2021. These results indicate that PC1 effectively captures the key information across the four metrics. Consequently, it is justified to construct the RSEI based on data from PC1, further supporting that the RSEI accurately reflects the EQ in the study area.

Table 3 Indicator loadings and first principal component contributions from 1995 to 2021.
Full size table

In the Li River Basin, RSEI values are higher at the edges and lower in the central areas, as depicted in Fig. 3. Regions classified as excellent or good are predominantly located in surrounding forests, which benefit from better vegetation cover, higher humidity, and less human disturbance. However, since 2015, there has been a gradual decline in areas categorized as excellent in the Mao’ er Mountain water source forest in the northern part of the basin. Areas classified as fair or poor are mainly found in the central and downstream regions, where urbanization, agricultural activities, and increased human practice lead to sparse vegetation, lower humidity, higher surface temperatures, and more frequent droughts. From 1995 to 2020, RSEI classifications in these areas generally improved, with a decrease in fair and poor classifications and an increase in good and excellent classifications. This positive trend is probably due to ongoing ecological restoration and conservation efforts.

Fig. 3
figure 3

Spatial and temporal distribution of different levels of RSEI in Li River Basin. The map is designed using ArcMap 10.2 software (https://www.esri.com/).

Full size image

Spatiotemporal evolution of ecological quality

Figure 4(a) shows that the average RSEI in the Li River Basin ranged from 0.54 to 0.66 between 1995 and 2021. The mean values of the individual indicators displayed varying trends: the NDVI increased by 2.12%, the WET increased by 4.04%, the NDBSI decreased by 4.40%, and the LST decreased by 1.78%. These changes suggest that the positive effects of increased moisture and vegetation cover primarily drove the increase in RSEI values. At the same time, the reduction in drought and heat also contributed to the overall improvement in RSEI.

In Fig. 4(b), the RSEI-HP trend highlights a general upward trend in the average RSEI, with peaks in 2003 and 2009. This trend can be divided into three phases: Stage I (1995–2003): The average RSEI was 0.59, indicating a low overall level with steady annual increases. Stage II (2004–2009): The average RSEI rose to 0.60, reflecting consistent improvement, albeit with greater fluctuations. Stage III (2010–2021): The average RSEI reached 0.62, indicating continued improvement with less fluctuation and a more stable upward trend.

Fig. 4
figure 4

Time-Series Changes in the Li River Basin from 1995 to 2021. (a) mean values of indicators and RSEI, (b) RSEI trend.

Full size image

Figure 5 shows that from 1995 to 2000, 2005 to 2010, and 2015 to 2020, the main transitions in RSEI levels were from “moderate to good” and “good to excellent,” leading to increased coverage of areas classified as good and excellent. In contrast, from 2000 to 2005, and from 2010 to 2015, the predominant transitions were from “moderate to fair,” “good to moderate,” and “excellent to good,” leading to a rise in areas rated as moderate and fair, and a decrease in areas rated as excellent.

Fig. 5
figure 5

RSEI transfer matrix for Li River Basin from 1995 to 2020.

Full size image

Figure 6 shows that from 1995 and 2020, the RSEI in the Li River Basin generally improved, with 61.06% of areas showing improvement, primarily in construction and agricultural areas, likely driven by urban green space development and ecological restoration efforts. However, some urban areas deteriorated, possibly due to the urban heat island effect, increasing LST and reducing RSEI. Areas with deterioration, accounting for 31.52%, were mainly in upstream regions like Mao’ er Mountain and Qingshitan Reservoir, where steep terrain, heavy rainfall, and karst topography exacerbate soil erosion. Human activities such as logging, land expansion, reforestation, and tourism have further stressed the local environment. In karst areas, the combination of rainfall and human disturbance worsens soil erosion72negatively affecting EQ.

Fig. 6
figure 6

Spatial Distribution of EQ Changes in the Li River Basin from 1995 to 2020. The map is designed using ArcMap 10.2 software (https://www.esri.com/).

Full size image

Driving factors analysis of ecological quality

Driving factors identify based on geodetector

Figure 7 presents the results of the factor detection analysis. LUI is the primary driver of spatial variation in RSEI, with all q-values above 0.45. The DEM follows as the second most significant factor. Among other natural factors, Sp and Tem rank third and fourth on average, while Pre holds the lowest rank. Among the anthropogenic factors, Pop and GDP rank fifth and seventh, respectively, indicating their relatively minor influence on the spatial differentiation of the RSEI. Notably, from 1995 to 2020, the q-value for TRI increased from 0.12 to 0.22, highlighting the growing influence of tourism on RSEI spatial differentiation.

Fig. 7
figure 7

Factor detector results of impact factors in Li River Basin in 1995, 2000, 2005, 2010, 2015, and 2020. Sp represents Slope, and Dem represents Elevation, Tem represents Temperature, Pre represents precipitation, Pop represents population density, LUI represents land use intensity, and TRI represents tourism activity intensity.

Full size image

Figure 8 illustrates the results from the interaction detector. The findings indicate that the combined effects of two factors on RSEI variations are greater than the individual impact of each factor. Interactions involving LUI show q-values above 0.50, underscoring the key role of LUI in the evolution of RSEI. While interactions among human factors are relatively weak, their impact increases when paired with natural factors. For instance, the q-values for the interactions between Pop, GDP, DEM, and Sp are all above 0.37.

Fig. 8
figure 8

Interaction detector results of impact factors in Li River Basin in 1995, 2000, 2005, 2010, 2015, and 2020. Sp represents Slope, and Dem represents Elevation, Tem represents Temperature, Pre represents precipitation, Pop represents population density, LUI represents land use intensity, and TRI represents tourism activity intensity.

Full size image

Results of spatial autocorrelation analysis

Before conducting the GWR modeling analysis, we assessed the spatial autocorrelation in the RSEI and its driving factors within the Li River Basin. At a 1 km grid scale, the Moran’s I indices for RSEI in 1995, 2000, 2005, 2010, 2015, and 2020 were 0.55, 0.57, 0.61, 0.60, 0.53, and 0.56, respectively. All |Z| values were greater than 1.96 and passed the significance test (P < 0.05), indicating a significant positive spatial autocorrelation in the distribution of RSEI across the basin.

Driving factors analysis based on GWR

Despite spatial autocorrelation in the drivers and RSEI, concerns regarding correlation and multicollinearity remain. To address these concerns, the Pearson correlation coefficients were utilized to evaluate the relationships between the driving factors, with variables exhibiting high or no correlation being removed. This process included the removal of variables such as Tem and Pre. Tem shares a similar spatial distribution with RSEI, indicating a high risk of collinearity. Meanwhile, Pre shows a weak correlation with RSEI at the regional scale and fails to reflect EQ changes effectively. As a result, both Tem and Pre were excluded from the analysis. Next, the OLS model was employed in ArcGIS to test combinations of the remaining six factors, as illustrated in Table 4. In this study, all explanatory variables had VIFs below 7.5, confirming their suitability for further analysis. The significance of the impact of Pop and GDP on the regression equation was unstable, with some years failing the p-value test. Therefore, we excluded Pop and GDP. In 1995, TRI was not significant (p > 0.1), but it became important in all other years (p < 0.05), with a VIF below 7.5. This suggests a generally stable relationship between TRI and RSEI, with the 1995 result reflecting the early stage of tourism development rather than a broader trend. In 1995, tourism development was limited, characterized by low intensity and a confined spatial distribution, resulting in a weak impact on RSEI. As tourism progressed, activities became more evenly distributed, and intensity increased, strengthening the relationship between TRI and RSEI. Ultimately, we selected DEM, Sp, LUI, and TRI as the driving factors for the GWR model.

Table 4 OLS test result. Sp represents slope, and dem represents elevation, pop represents population density, LUI represents land use intensity, TRI represents tourism activity intensity.
Full size table

Table 5 shows that the R² and AICc values for the GWR model exceed those from the OLS analysis. This finding suggests that the GWR model provides a superior fit to the data and facilitates a more comprehensive understanding of the influence of each factor on RSEI. For further analysis, the GWR regression coefficients for the drivers are visualized for each year.

Table 5 Estimate parameters of the GWR model.
Full size table

There are considerable spatial variations in the effects of the four driving factors on the RSEI. Specifically,

  1. (1)

    The correlation between slope and RSEI is predominantly positive (Fig. 9a). However, the regression coefficients show a declining trend over the years within the basin, while some adjacent mountainous regions have shifted to negative coefficients. This indicates a diminution in the positive influence of slope on RSEI and the emergence of adverse effects in neighboring areas.

  1. (2)

    (2) Elevation demonstrates a primary positive correlation with RSEI (Fig. 9b), suggesting that areas at higher elevations tend to exhibit higher RSEI values. The regression coefficients show that the area with positive values in the surrounding mountains has gradually expanded over time. In contrast, negative values are concentrated in the central part of the study area, with the area spreading over time. These findings indicate that the positive impact of elevation on RSEI has increased in the mountainous regions. In contrast, its negative impact has become more pronounced in the central region, where human activity is more intense.

Fig. 9
figure 9

The spatial distribution of regression coefficients from the GWR model for the RSEI in the Li River Basin spans from 1995 to 2020. Sp represents Slope, Dem represents Elevation, LUI represents land use intensity, and TRI represents tourism activity intensity. The map is designed using ArcMap 10.2 software (https://www.esri.com/).

Full size image
  1. (3)

    The correlation between land use intensity and RSEI is negative (Fig. 9c), suggesting that increased land use intensity impedes RSEI improvement. Over time, the regression coefficients have consistently increased absolute values, with the highest values concentrated in the surrounding water source forests. This trend indicates that the negative impact of land use intensity on RSEI has strengthened, particularly in the water source forest areas, where the suppression of RSEI improvement is most pronounced.

  1. (4)

    The influence of tourism activity intensity on RSEI has varied significantly over the years (Fig. 9d). The regression coefficients reveal that the area with negative values has expanded over time, especially in the middle and lower reaches, while the positive values in the upstream Mao’ er Mountain area have steadily increased. These shifts suggest that the impact of tourism on RSEI in the middle and lower reaches has become more negative. At the same time, its positive effect in the upstream regions, particularly in Mao’ er Mountain, has continued to grow.

Results of policy analysis

Based on the RSEI-HP trend analysis, we divided the text mining process into three periods (1995–2003, 2004–2009, and 2010–2021) and generated word clouds to visualize Guilin’s macro-policy focus across phases (Fig. 10). The recurrent prominence of terms like “Urban greening,” “Ecology,” and “Tourist destination” underscores the city’s consistent emphasis on ecological protection and tourism development.

In the first phase, policies centered on urban green spaces and forest resources, with national programs (e.g., the National Ecological and Environmental Protection Outline, and the National Natural Forest Protection Program), enhancing vegetation cover and humidity, thereby improving EQ. The second phase shifted toward broader ecological goals, supported by initiatives like the Li River Tributary Environmental Comprehensive Management Plan and the Ecological Guangxi Construction Plan Outline. Sanitary city construction further enhanced urban ecological conditions. In the third phase, the policies focus on sustainable and tourism destination development. Key plans—such as the Guilin International Tourism Destination Construction and Development Plan and the Policies Supporting Guilin’s National Sustainable Development Agenda Innovation Demonstration Zone - promoted eco-tourism. Complementary policies (e.g., the Guilin Li River Ecological Protection and Restoration Plan (2019–2025) and the Guilin Water Pollution Prevention and Control Action Plan (2018–2020)) - supported ecosystem management and ecological restoration.

Overall, the evolving policies in the Li River Basin reflect a continued commitment to the sustainable use of integrated mountain-water resources and the comprehensive management of ecosystems.

Fig. 10
figure 10

Word clouds of macro policy keywords for each stage in Guilin City.

Full size image

Discussion

Selection and driving mechanisms of factors

The impact of natural factors on RSEI

Topography emerges as a principal driver of RSEI spatial heterogeneity; this study reveals a remarkable congruence between RSEI distribution and elevation and slope patterns. This correlation can be attributed to topography’s influence on EQ. It shapes the biological growth environment73,74 and restricts the extent and intensity of human activities75. The Li River Basin’s surrounding mountains (741–2061 m) provide favorable climate conditions for plant growth, while the rugged terrain limits human disturbance and preserves natural integrity. In contrast, the low-elevation flat areas in the central region are more susceptible to development, damaging vegetation cover, and degrading EQ. Additionally, areas with more significant relief and steeper slopes suffer a more pronounced negative impact, likely due to the karst landscape’ poor soil physical and chemical properties. Heavy rainfall and runoff accelerate soil erosion and water loss, further deteriorating Eq76.. In light of these findings, ecological protection strategies must be tailored to different topographical regions. A key aspect of this strategy is to regulate human activities, and vegetation restoration efforts should focus on plants suited to karst landscapes. Additionally, implementing soil and water conservation projects is essential to mitigate soil erosion and water loss.

Climate factors influence vegetation growth and ecosystem stability by regulating energy flow and nutrient cycling within ecosystems77. Previous research has shown that both the amount and spatial-temporal pattern of precipitation affect NDVI78. Since NDVI is a key component of RSEI, precipitation was expected to influence RSEI strongly. However, this study found precipitation’s relatively weak driving effect on RSEI. Wang et al.79 reported the mean annual precipitation (MAP) threshold of 1337–1438 mm for karst regions of southern China. Beyond this range, vegetation’s response to precipitation diminishes or even becomes negative. MAP has steadily increased in the Li River Basin in recent years, reaching 1314, 1349, 1376, 1424, 1484, and 1493 mm, exceeding the upper threshold. This persistent rise likely reduces vegetation sensitivity to rainfall, explaining precipitation’s limited effect on RSEI. These findings suggest a non-linear relationship between climate factors and ecosystem responses. Once critical thresholds are exceeded, internal ecosystem feedback may shift. Therefore, vegetation management strategies must adapt to changing precipitation regimes, emphasizing maintaining existing vegetation communities under altered climatic conditions.

The impact of human activities on RSEI

Results indicated that human activities significantly influence EQ in the Li River Basin. Given the world-renowned status as a tourist destination, the impact of tourism on the region’s ecology cannot be overlooked. Therefore, this study incorporates tourism activity intensity alone with GDP, population density, and land use intensity as key factors. The findings reveal that GDP and population density have minimal impact on RSEI, while land use intensity is the primary driver of RSEI spatial variation. Furthermore, tourism activity has steadily increased and now significantly negatively affects RSEI. These results underscore human activities’ mounting ecological challenges in the Li River Basin.

Land use intensity is a key metric for assessing the extent of human exploitation and utilization of land resources. Extensive land use development frequently disrupts the structure and functionality of natural ecosystems, adversely affecting biodiversity habitats, vegetation growth, and soil conditions80. The negative effects of land use intensity vary with land use type. Notably, forested areas experience the most pronounced adverse impacts. example ecosystem are inherently fragile81and highly sensitive to anthropogenic activities82. Escalating land use intensity triggers intensive human interventions that devastate vegetation83compromise soil integrity, and exacerbate soil erosion and water depletion84ultimately EQ degradation. For example, the encroachment of construction land in the Qingshi Tan Reservoir region has reduced vegetation cover, while tourism exploitation in Haiyang Mountain has accelerated vegetation degradation. Furthermore, illegal quarrying and excessive felling in Mao’ er Mountain have compromised the vegetation and soil’s regenerative capacities, worsening soil erosion and related issues. Nevertheless, initiatives such as establishing ecological red lines and converting croplands back to forests have restored vegetation and soil stability in certain regions. Liu et al.85 verified that implementing the returning farmland to forests initiative in the Li River Basin has effectively restored vegetation cover and soil stability. Conversely, the ecological integrity of central agricultural lands and construction zones is less affected by land use intensity. This is because the simper ecosystem configurations of these land uses during agricultural and urban development enhance their ecological resilience and stability86. Additionally, enforcing regional ecological conservation measures has substantially alleviated pressures from escalating land use intensity. In conclusion, devise strategies for forestland conservation, sustainable agricultural land management, and the ecological enhancement of construction zones to maintain and improve the ecological health of the Li River Basin.

GWR results indicate that the influence of TRI on RSEI in the Li River Basin exhibits marked spatial heterogeneity closely linked to regional differences in tourism development modes. Since 2013, only 10% of tourism land in Guilin has been converted for construction, while in the upper reaches—where ecological resources are scarce and valuable—the majority is retained as ecological landscape areas with tourism development primarily focused on eco-tourism. This low-intensity model sustains ecosystem integrity, and research shows that upper-reach tourism intensity positively impacts RSEI, confirming that eco-tourism can foster conservation-oriented governance. In contrast, Yangshuo County in the mid-lower basin—characterized by a high density of scenic spots—exhibits the highest rate of tourism land conversion, elevated visitor numbers, and intensive tourism activities. These factors have led to land hardening, ecological fragmentation, and habitat loss, increasing ecosystem stress and the likelihood of ecological degradation. Consequently, the negative response of EQ to tourism disturbance is particularly evident in the mid-reaches. Implementing targeted spatial planning for tourism land is necessary to mitigate these pressures and ensure ecological security. High-intensity tourism zones require clearly defined capacity limits, the widespread adoption of eco-tourism practices, support for eco-friendly hotels and resorts, and better integration of tourism with ecological management to achieve sustainable development. Future research should focus on identifying the threshold effects of tourism on ecosystems and exploring synergistic strategies to align tourism development with ecological conservation.

The study found that GDP and population density have weak and inconsistent direct effects on the spatial differentiation of RSEI. Comparative experiments (Table 6) reveal that when LUI is included, the significance of GDP and population density becomes unstable or even disappears. However, once LUI is removed, their relevance is restored. This likely results from the shared variance between LUI and these two variables. LUI is a comprehensive indicator reflecting land-use changes and development intensity, including urbanization and population distribution. These factors are closely tied to economic activities and population patterns, partially explaining the influence of GDP and population density on RSEI variations. Consequently, including LUI in the model diminishes the independent contributions of GDP and population density. This finding suggests that GDP and population density impact the ecological environment through land-use patterns and resource allocation. This is consistent with the views of Jiang et al.87and the synergistic interplay among the three factors warrants further exploration. In conclusion, efforts to conserve and improve the ecological environment must address the growth of GDP and population density and consider how these growths influence the environment through land use. Furthermore, promoting green development, upgrading industrial structures, and planning land use rationally are crucial to balancing economic growth with ecological protection.

Table 6 Comparison of OLS test results for pop and GDP in each year after controlling for LUI. Sp represents slope, and dem represents elevation, pop represents population density, LUI represents land use intensity, TRI represents tourism activity intensity.
Full size table

Policy impact on ecological quality

The ecological impacts of policies often unfold over varying time scales and exhibit delayed effects. Relying on fixed-period analysis risks obscuring these dynamics and may misrepresent the relationship between policy and ecological outcomes. The study applies the HP filter to address this to segment policy effects into distinct phases, aligning key policy interventions with inflection points in EQ trends. By linking major policies and events with changes in the mean RSEI value (Fig. 11), the analysis shows that sustained efforts in ecological protection, sustainable development, and tourism transformation have driven steady improvements in RSEI across the Li River Basin. This progress reflects a growing synergy between ecological conservation and economic growth. The coordinated implementation of policies across different periods highlights the critical role of policy integration in supporting long-term sustainability.

Fig. 11
figure 11

Trends in RSEI mean values and key events in Li River Basin from 1995 to 2021.

Full size image

In the first phase, the Li River Basin prioritized conserving forest resources and developing urban green spaces. The RSEI increased significantly, growing at a rate of 15.98%, and peaked in 2003, underscoring the critical role of basic ecological infrastructure in enhancing EQ. First, forest protection policies proved highly effective, increasing the area of high-quality forests from 5,015.20 km² in 1995 to 6,438.73 km² in 2003. Second, with an emphasis on “Green Space” and “Urban Green,” Guilin City advanced the development of road green spaces and urban parks. Consequently, the high-quality ecological zones in the central urban area expanded by 15%, indicating a clear improvement in the urban EQ. These foundational measures resulted in sustained regional EQ improvements, highlighting the positive impact of ecological construction on regional environmental development.

In the second phase, the mean RSEI displayed fluctuations. During this period, rapid industrial growth in Guilin City generated significant environmental challenges—such as pollutant emissions, acid rain, industrial pollution, and air pollution88—that exacerbated the urban heat island effect. The Li River Basin experienced its highest heat levels in 2004 and 2005, negatively affecting the EQ and leading to a decline in RSEI. However, the introduction of sustainable development strategies and the designation of Guilin City as a national health city in 2005, along with policies such as tributary protection in the Li River, the “Return farmland to the forest” program, and the promotion of “Energy conservation and emission reduction”, helped mitigate dryness and heat and increased humidity. These efforts contributed substantially to the region’s EQ improvement, with mean RSEI gradually rising to a second peak by 2009. During this phase, Guilin’s GDP growth rate was 1.5 times that of the preceding four years89underscoring the positive interaction and coordinated development of ecology and the economy under policy guidance.

In the early stage of the third phase, Guilin City emerged as an international tourist destination. Urban construction and tourism development in the Li River Basin accelerated rapidly. However, Guilin’s coordination of ecological, economic, and social development was not yet optimal90. This imbalance resulted in excessive quarrying, sand extraction, deforestation, uncontrolled restaurant expansion, and the discharge of industrial and ecological wastewater, which degraded the water and mountain resources and put significant pressure on the ecosystem. From 2009 to 2012, the RSEI declined. Since 2012, with the promotion of ecological civilization and the “Beautiful China” initiative, Guilin has regulated quarrying operations and standardized tourism development. Integrated management, ecosystem restoration, and actively promoting eco-tourism have improved the basin’s overall EQ. According to Shao et al.91Guilin City’s sustainable development has trended upward since 2012, strongly correlated with the degree of tourism resource development and utilization. This positive shift demonstrates growing coordination between ecology, economy, and society. Furthermore, the steady rise in mean RSEI since 2012 reflects the effectiveness of the sustainable development model tailored to local characteristics and comprehensive ecological governance.

Hurst index analysis (Fig. 12) suggests that future EQ changes in the Li River Basin will vary regionally. In the fourth stage (post-2021), improvements are mainly concentrated in water-source forest areas, reflecting adequate forest protection and restoration measures. Additionally, some urban areas in the middle reaches are expected to improve ecologically. Policy keywords analysis attributes this positive trend to sustainable development measures - such as “Zero-waste cities,” “Sponge cities,” and “Pollution prevention” policies- which have enhanced pollution control, urban greening, and rainwater management. However, deterioration is predicted for cultivated lands and typical karst regions in the middle and lower reaches. We hypothesize that agricultural development degrades fragile karst soils, worsening soil erosion. It is thus crucial to strengthen the management of human activities in karst areas, particularly through regulating agricultural expansion and urbanization, while promoting sustainable urban and farming practices to prevent over-exploitation. The continued implementation of the “Ecological compensation” policy and the enforcement of the Regulations on the Sustainable Use of Karst Landscape Resources in Guilin are vital for maintaining ecological stability and supporting the sustainable development of these regions.

Fig. 12
figure 12

Spatial distribution of RSEI future trends and feature words from 2022 to 2024. The map is designed using ArcMap 10.2 software (https://www.esri.com/).

Full size image

Future research directions

The Li River Basin is characterized by its distinct karst geomorphology. Future research should incorporate regional indicators tailored to its unique geological and ecological context to optimize the RSEI. Additionally, integrating data from multiple sensors could provide more comprehensive, high-resolution monitoring of terrain changes and vegetation growth in karst areas. Introduce ground-truth data calibration to enhance the accuracy and applicability of the RSEI and better capture ecological changes in the region. Moreover, this study has focused primarily on non-aquatic EQ, yet the basin’s water resources are equally crucial. Therefore, future research must employ more holistic evaluation methods to assess aquatic EQ changes, providing targeted scientific evidence for the protection and management of the Li River Basin. Lastly, this study used the Nighttime-light Dataset to proxy for nighttime tourism activities. However, this data may be confounded by commercial, residential, and industrial activities, complicating precise differentiation between tourism and non-tourism activities. Future research should integrate GPS big data92 and crowdsourced big data93 to build a multi-dimensional tourism monitoring system.

Conclusions

This study investigates EQ’s spatiotemporal evolution and driving mechanisms in the Li River Basin from 1995 to 2021. The analysis reveals three key findings. First, the RSEI exhibited a characteristic “high-center, low-periphery” spatial distribution with an overall fluctuating upward trend, peaking in 2003 and 2009, as evidenced by three-phase mean values of 0.59, 0.60, and 0.62. While 61.06% of ecologically improved areas were concentrated in construction and cultivated lands of the middle-lower reaches, 31.52% of degraded zones occurred in peripheral water-conservation forests. Second, land use intensity and elevation emerged as primary drivers of ecological changes, while tourism activity intensity demonstrated progressively increasing annual impacts. Local-scale analysis further indicates weakened positive slope effects, intensified negative elevation impacts in human activity-concentrated areas, and pronounced land use intensity pressures on forested lands, coupled with tourism’s escalating negative impacts on RSEI in the middle-lower reaches. Third, Guilin’s policy transition from single-element management to integrated mountain-water resource systems significantly enhanced regional EQ. These findings provide critical scientific support for sustainable landscape utilization in karst regions and offer transferable insights for managing similar global natural heritage sites.