Spatio-Temporal Changes and Driving Mechanisms of the Ecological Quality in the Mountain–River–Sea Regional System: A Case Study of the Southwest Guangxi Karst–Beibu Gulf

Abstract

This study investigates the spatio-temporal characteristics and driving mechanisms of ecological quality in the mountain–river–sea regional system using the Remote Sensing Ecological Index (RSEI) model, moderate-resolution imaging spectroradiometer (MODIS) data, and the Google Earth Engine (GEE) platform. The analysis, conducted at both the grid and county scales using spatial autocorrelation and geodetector, showed a notable improvement in ecological quality, with the average RSEI value rising from 0.549 in 2000 to 0.627 in 2022. The distribution pattern reveals superior quality in the northwest and inferior quality in central urban cores and coastal zones. Ecological quality exhibited significant spatial clustering, with high–high clusters in karst mountains and low–low clusters in urban and industrial zones. Geodetector analysis identified GDP and population density as dominant factors at the grid scale, and GDP and elevation at the county scale. By quantifying spatio-temporal variations and driving mechanisms of ecological quality across scales, this study provides a solid scientific foundation for regional ecological conservation and sustainable development.

1. Introduction

China’s socio-economic development has entered a new phase of high-quality growth marked by green and low-carbon initiatives. Yet, the development of ecological civilization currently stands at a critical stage. While environmental conservation efforts have yielded measurable outcomes, the structural pressures, root causes, and persistent challenges have not been fundamentally alleviated. This reality necessitates advancing ecological protection with heightened priority, broader vision, and intensified efforts moving forward [1]. Ecological quality reflects the integrated characteristics of ecosystems, including both internal elements and external environmental interactions, and serves as a critical foundation for quantitatively assessing regional ecological status [2]. The accurate and objective assessment of ecological quality not only facilitates the promotion of sustainable socio-economic development but also provides a robust scientific foundation for the formulation of ecological protection policies and the coordinated development between ecology and the economy, thereby promoting the continuous improvement of the ecological environment and advancing sustainable development.

Scholars worldwide have conducted extensive research, leading to significant progress. Existing methodologies can be broadly categorized into two types. The first category involves analysis based on single indicators, such as the Normalized Difference Vegetation Index (NDVI) [3], Land Surface Temperature (LST) [4], Fraction Vegetation Coverage (FVC) [5,6], and Net Primary Production (NPP) [7]. While these indicators reflect specific aspects of ecosystems, they fail to provide a comprehensive characterization or interpretation of the overall ecological quality of a region. The second category involves comprehensive analysis by coupling multiple indicators. For example, China’s Ecological Index (EI), proposed in 2006, enables macroscopic regional ecological assessments but has limited capacity to reveal spatial heterogeneity or support spatio-temporal dynamic analysis [8]. The Pressure–State–Response (PSR) model, prepared by the United Nations Development Programme (UNDP) and the United Nations Environment Programme (UNEP), is widely adopted in environmental assessments. However, it faces challenges, including complex indicator systems and a high degree of subjectivity in weight assignment [9]. The RSEI, proposed by Xu Hanqiu in 2013, addresses the limitations of conventional methodologies. By employing Principal Component Analysis (PCA) to integrate multiple indicators, the RSEI objectively determines contribution weights based on data characteristics, thereby effectively reducing subjective bias. Furthermore, it enables spatio-temporal visualization of ecological conditions and supports dynamic monitoring of the region’s ecological evolution [8]. However, constructing the RSEI using traditional methods faces multiple challenges, including mass data processing, missing pixel values, cumbersome preprocessing, and computational complexity. The remote sensing cloud computing platform Google Earth Engine (GEE), with advantages such as powerful computational capacity, extensive data archives, real-time updates, and cross-disciplinary applicability, has been widely applied in scientific research [10]. Although some scholars have carried out ecological environment evaluations in regions such as typical mountainous areas [11,12], protected areas [13,14], urban agglomerations [15,16,17], and townships [18,19] using the GEE platform, ecological quality assessments in complex transitional zones remain limited.

To analyze the dynamic evolution and spatial characteristics of RSEI, spatial autocorrelation analysis is one of the core methods for revealing its intrinsic spatial correlation [20,21]. Spatial autocorrelation analysis effectively identifies the spatial clustering of RSEI and clarifies the distribution characteristics of high- and low-value areas. Furthermore, it quantitatively characterizes the synergistic evolution of ecological quality among adjacent spatial units, overcoming the neglect of spatial correlation in traditional statistical methods and providing robust quantitative support for developing differentiated ecological protection and governance strategies. In addition, by combining spatial autocorrelation analysis at different scales [22,23], it can effectively decouple the cross-scale interaction mechanisms between natural elements and anthropogenic activities, revealing the transmission paths between macro policy regulation and micro ecological responses, thereby promoting precision management and sustainable development of the regional ecological environment.

The spatio-temporal dynamics of the RSEI are influenced by both natural geographic drivers and anthropogenic activities. Understanding its evolutionary driving mechanisms is crucial for deciphering regional ecological patterns and formulating science-based protection and restoration strategies. The geographic detector [24] is a statistical method for identifying spatial heterogeneity and uncovering latent drivers. The parameter-optimal geographic detector [25], developed by Wang Jinsong’s team, enhances analytical precision and has been widely applied in fields such as environmental impact factor analysis [26,27] and driving force analysis of vegetation changes [28]. Existing studies, based on varying research objectives, regional characteristics, and data accessibility, have used this model to construct diverse driving factor analysis frameworks [29,30,31,32]. However, current research on the spatio-temporal evolution of the RSEI and its driving mechanisms is mostly limited to a single spatial scale (e.g., grid or administrative units) [33], and the exploration of the linear/nonlinear relationships between the RSEI and driving factors at different scales using the parameter-optimal geographic detector is still insufficient [23,34]. If the dependence on spatial scale is ignored, research conclusions may deviate due to scale effects, leading to misjudgments or omissions regarding the variation patterns of the RSEI and its relationships with ecological and social elements. Therefore, there is an urgent need to construct a multi-scale analysis framework to systematically reveal the scale effects of RSEI evolution and its driving mechanisms, providing a scientific basis for regional ecological governance.

The correlated and coordinated growth of the three ecosystems—mountain, river, and coastal zone—is referred to as the mountain–river–sea regional system. The study area of Southwest Guangxi Karst–Beibu Gulf Coastal Zone has complex and varied topography and geomorphology, with the overall terrain sloping from northwest to southeast, starting from the northwestern Yungui Plateau, gradually transitioning to the south-central low hills, and eventually extending to the southern coastal plain. This forms a long and narrow transition zone with rivers such as the Zuojiang and the Right River flowing through, creating a unique mountain, river, and sea transitional landscape [35]. Currently, research on the ecological evaluation of this unique transitional space remains in its early stages, with most studies focusing on single-indicator analyses. The GEE platform has opened new avenues for ecological evaluation research through its powerful computational capabilities. Therefore, this study uses the GEE platform to construct the RSEI model for the mountain–river–sea regional system based on MODIS data from three periods: 2000, 2010, and 2022. Additionally, correlation analysis and the optimal parameter-based geographic detector were used to explore the spatio-temporal variation characteristics and key socio-ecological drivers of the RSEI at both grid and county/district administrative unit scales. The findings aim to provide robust data support and a theoretical basis for formulating ecological protection and restoration strategies at different scales, ultimately promoting the synergistic development of regional ecological and economic systems.

2. Materials and Methods

2.1. Study Area

The Southwest Guangxi Karst–Beibu Gulf Coastal Zone (20°26′–25°07′ N, 104°28′–109°56′ E) is located in the land–sea convergence zone of Southwest China. It comprises three geographical units: the Guangxi karst mountainous area, the Left and Right River basins, and the Beibu Gulf Coastal Zone, collectively forming a transitional mountain–river–sea geographical system (Figure 1). This region includes seven prefecture-level cities—Baise, Chongzuo, Nanning, Fangchenggang, Beihai, Qinzhou, and Yulin—and 50 counties (districts), covering 108,000 km², or 45.80% of Guangxi’s total area. The study area has a subtropical monsoon climate, with an average annual temperature ranging from 21.9 °C to 23.1 °C and average annual precipitation of 1377.8 mm. By the end of 2022, its regional GDP reached 1.259 trillion CNY, and its registered population was 21.9 million, about 50% of Guangxi’s total population. Over decades, the ecologically fragile karst environment of the Southwest Guangxi Karst–Beibu Gulf Economic Zone, coupled with intensive human activities, has exacerbated ecosystem vulnerability. Therefore, conducting a comprehensive evaluation of ecological quality and investigating spatial aggregation patterns and their driving mechanisms are crucial for advancing ecological civilization and achieving high-quality development in the mountain–river–sea transition zone.

2.2. Data Sources

MODIS remote sensing data is a key dataset for this study. MODIS products (MOD09A1, MOD11A2) from the years 2000, 2010, and 2022 were accessed online via the GEE platform (https://developers.google.cn/earth-engine (accessed on 20 May 2024)). First, high-quality images from the vegetation growing season (June to October) in the study area were selected using the filter function. Cloud pixels were masked using the QA band, and water pixels were removed by applying the NDWI index (NDWI > 0.2). The preprocessed images were then normalized using the mean function to calculate the average values, resulting in four ecological indicators: greenness (NDVI), heat (LST), wetness (Wetness), and dryness (NDBSI). Finally, the first principal component was extracted using Principal Component Analysis (PCA) and normalized to generate the annual RSEI raster dataset.

The driving factor datasets include night-time light, population density, GDP, digital elevation model (DEM), slope, aspect, precipitation, and temperature data. Annual datasets for night-time light, GDP, and elevation were sourced from the Resource and Environment Science and Data Platform (http://www.resdc.cn/). Using the DEM data and the slope and aspect tools in the spatial analysis module of ArcGIS 10.8, the slope and aspect datasets were generated. Population density data were obtained from the 1 km resolution population spatial distribution grid in the LandScan dataset (2000–2022) (https://landscan.ornl.gov/). Precipitation and air temperature data were sourced from the National Earth System Science Data Center (https://www.geodata.cn/).

2.3. Research Methods

This study is divided into three major sections. The first part delineates the spatio-temporal evolution of the RSEI within the mountain–river–sea regional system at both grid and county (district) administrative unit scales. The second part employs spatial autocorrelation analysis to investigate the spatial clustering patterns of RSEI variations at both grid and county (district) administrative unit scales. The third part utilizes a geographic detector to investigate the mechanisms underlying the spatial differentiation of RSEI at both grid and county (district) administrative unit scales (Figure 2).

2.3.1. RSEI Construction

In this study, four ecological variables—NDVI, LST, WET, and NDBSI—were extracted from the GEE platform to construct the RSEI model (

Table 1. Formulas and explanations for the four eco-quality indicators NDVI, LST, WET, and NDBSI.

Formula	Remark
$RSEI={\displaystyle \int \left(NDVI ,LST ,WET ,NDBSI\right)}$	where NDVI is the Normalized Difference Vegetation Index; LST is Land Surface Temperature; WET is WETNESS and NDBSI is Normalized Differential Soil Index.
$NDVI=\left(NIR-\mathrm{Re}d\right)/\left(NIR+\mathrm{Re}d\right)$	where NIR is the near-infrared band; Red is the red light band.
$LST\left(°C\right)=LST\left(K\right)-273.15$	Direct use of MOD11A2 LST data in K, to be converted to degrees Celsius °C)
$\begin{array}{l}WET=0.01147p_1+0.2489p_2+0.2408p_3+0.3132p_4\\ +0.3122p_5+0.6416p_6+0.5087p_7\end{array}$	where (i = 1, 2, 3, …, 7) is the reflectivity of each MODIS band.
$NDBSI=\left(BSI+IBI\right)/2$$BSI=\left[\left(S_1+\mathrm{Re}d\right)-\left(Blue+NIR\right)\right]/\left[\left(S_1+\mathrm{Re}d\right)+\left(Blue+NIR\right)\right]$$\begin{array}{l}IBI=\left\{2S_1/\left(S_1+NIR\right)-\left[NIR/\left(NIR/\mathrm{Re}d\right)+Green/\left(Green+S_1\right)\right]\right\}/\\ \left\{2S_1/\left(S_1+NIR\right)+\left[NIR/\left(NIR+\mathrm{Re}d\right)+Green/\left(Green+S_1\right)\right]\right\}\end{array}$	where S1, Red, Green, NIR, and Blue represent the short-wave infrared, red, green, near-infrared, and blue bands, respectively; IBI represents the Index-Based Built-Up Index; BSI represents the Bare Soil Index.

), which intuitively reflects the harshness of the regional ecological environment [36].

The above indicators were normalized using following formula [37]:

$$NI=\left(I-I_{\min }\right)/\left(I_{\max }-I_{\min }\right)$$

(1)

where NI is the normalized indicator; I is the original indicator; I_min is the minimum value of indicator I, and I_max is the maximum value of indicator I.

The four indicators were combined using the PCA method to obtain the first principal component (PC1). The initial Remote Sensing Ecological Index (RSEI₀) was then calculated by inverting the sign of PC1. Subsequently, RSEI₀ was normalized to obtain the RSEI. The formula is as follows [38]:

$$RSE{I}_0=1-\left\{PC1\lfloor {\displaystyle \int \left(NDVI\right),LST,WET,NDBSI}\rfloor \right\}$$

(2)

$$RSEI=\left(RSE{I}_0-RSE{I}_{\min }\right)/\left(RSE{I}_{\max }-RSE{I}_{\min }\right)$$

(3)

where RSEI_min and RSEI_max are the minimum and maximum values of RSEI after normalization, respectively. RSEI, or the Remote Sensing Ecological Index, ranges from 0 to 1, where values closer to 0 indicate low ecological value, and values closer to 1 indicate high ecological value.

2.3.2. Analysis of Correlation

Spatial autocorrelation analysis was conducted for the study area using both global and local Moran’s I to quantify the correlation between the attribute values of a spatial unit and those of its neighboring units. The formulas are as follows [39]:

$$I={\displaystyle \frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}}\times \left(X_i-\overline{X}\right)\left(X_j-\overline{X}\right)}}{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}\times \frac{1}{n}{{\displaystyle \sum _{i=1}^n\left(X_i-\overline{X}\right)}}^2}}}}$$

(4)

$$I={\displaystyle \frac{{\displaystyle \sum _{i=1,j\ne 1}^n{w}_{ij}\left(X_i-\overline{X}\right)\left(X_j-\overline{X}\right)}}{{\displaystyle \sum _{i=1}^n{\left(X_i-\overline{X}\right)}^2}}}$$

(5)

where I is the global/local spatial autocorrelation value, n is the total number of spatial units, and $\overline{X}$ is the average value of variable X_i in n spatial units; W_ij is the spatial weight matrix, and X_i and X_j are the observed values of locations i and j. The Moran’s I value ranges from [−1, 1]. When I = 0, there is no correlation between the spatial elements. When I is close to 1 and statistically significant, it indicates that the global spatial elements are highly clustered, with strong autocorrelation and high similarity to neighboring regions; when I is close to −1 and statistically significant, it indicates that the global spatial elements are dispersed, with significant differences and large variation from neighboring regions. The significance of the global/local Moran’s I index is tested using the Z-value test [40].

2.3.3. Geographic Detector

The optimal parameter-based geographic detector [41] can automatically determine the optimal classification boundaries and category numbers based on the intrinsic structure and characteristics of the data, enhancing the scientific rigor and analytical accuracy. Accordingly, this study selected eight driving factor indicators from the natural, social, and economic dimensions: elevation (X1), slope (X2), aspect (X3), temperature (X4), precipitation (X5), population density (X6), GDP (X7), and night-time lights (X8). After extracting the attribute values of the dependent variable (RSEI) and each driving factor using ArcGIS 10.8, the GD package in R was employed to perform the optimal partitioning of the driving factors, followed by factor detection and interaction analysis.

The factor and differentiation detectors were used to assess the explanatory power of various drivers on the spatial distribution of RSEI. The impact of each factor was assessed by measuring the q-value, which is calculated as follows:

$$q=1-{\displaystyle \frac{{\displaystyle \sum _{h=1}^L{N}_h{\sigma }_h^2}}{N{\sigma }^2}}$$

(6)

where q represents the degree of explanation of each factor on the ecological quality in the study area. The q-value range is [0, 1]; the larger the q-value, the greater the explanatory power, indicating a stronger influence of the factor on ecological quality, and vice versa, a weaker influence. L is the number of classifications of the dependent variable; N_h and N are the number of samples of the classification h and the global sample, respectively. ${\sigma }_h^2$ and ${\sigma }^2$ are the variances for the classification h and the global sample, respectively. A higher q-value indicates a greater influence of the factor on the RSEI.

Interaction probing further revealed the combined effects of the interactions between different drivers on the target variable RSEI by calculating the interaction effect between two drivers, q (X_i∩X_j), and comparing its value to q (X_i) and q (X_j). The interaction types were categorized as shown in

Table 2. Interaction q-value and single factor.

Role Relationship	Compared to the q-Value	Single-Factor Lower Value	Single-Factor Larger Value	Sum of the Two Drivers
Nonlinear weakening	Interaction q-value	<	<	<
Single-factor nonlinear attenuation	Interaction q-value	>	<	<
Bivariate enhancement	Interaction q-value	>	>	<
Independent	Interaction q-value	>	>	=
Nonlinear enhancement	Interaction q-value	>	>	>

“<” indicates a smaller value, “>” indicates a larger value, and “=” indicates an equal value.

.

3. Results

3.1. Analysis of Spatial and Temporal Variations of RSEI at Different Scales

3.1.1. Grid Scale

The ecological quality analysis shows that the average RSEI values for the mountain–river–sea system in 2000, 2010, and 2022 were 0.5491, 0.5615, and 0.6271, respectively, indicating an overall upward trend. The RSEI values were reclassified into five categories, each with intervals of 0.2: poor [0, 0.2), fair [0.2, 0.4), medium [0.4, 0.6), good [0.6, 0.8), and excellent [0.8, 1) (Figure 3). The areas and proportions of each category were then calculated (

Table 3. Area statistics of RSEI level at grid scale.

Level	2000		2010		2022		Area Change 2000–2022
	Area/ km2	Percent/ %	Area/ km2	Percent/ %	Area/ km2	Percent/ %
Poor	8159.08	7.54%	6027.96	5.52%	3933.10	3.64%	−4225.97
Fair	18,374.92	16.98%	19,697.31	18.05%	12,076.40	11.19%	−6298.52
Medium	31,649.69	29.25%	33,977.02	31.13%	27,787.37	0.74%	−3862.32
Good	33,846.31	31.28%	31,795.79	29.14%	37,487.41	34.73%	3641.10
Excellent	16,167.76	14.95%	17,635.16	16.16%	26,655.54	24.70%	10,487.78

). The results revealed that from 2000 to 2022, ecological quality showed a spatial distribution pattern, with superior conditions in the northwest and poorer conditions in the central core, showing significant regional disparities. Ecological quality levels were mainly dominated by the good and medium categories, which together accounted for over 50% of the total area. The medium level was mainly distributed in the center, while the good level was found in the northern and southern regions. The excellent level accounted for 14.95%, 16.16%, and 24.70% of the area in 2000, 2010, and 2022, respectively, and was mainly distributed in the northern and southwestern parts. The fair grade covered a small area, with proportions below 20% across all three years, mainly in the central and southern regions. The poor grade had the smallest proportion, decreasing from 7.54% in 2000 to 3.64% in 2022, mainly concentrated in the central area. From 2000 to 2022, areas of different ecological quality levels underwent significant changes. Areas classified as poor, fair, and medium decreased, while those of good and excellent grades increased.

Based on the ecological quality classification, a difference analysis was performed for three time periods (2000–2010, 2010–2022, and 2000–2022). Positive values indicated improvement in ecological quality, while negative values signified degradation, and zero values indicated no change. The changes were categorized into five classes at intervals of 0.5: strongly worse [−1 to −0.5), slightly worse [−0.5 to 0), stable [0], slightly improved (0 to 0.5], and strongly improved (0.5 to 1], with the area and proportion of each category analyzed. Figure 4 shows that the ecological quality classes in the study area changed over time, reflecting dynamic shifts in the ecological environment. Areas with improved quality formed large continuous clusters, while degraded areas were smaller and more scattered, often appearing as isolated patches. According to

Table 4. Spatial distribution of RSEI level changes at grid scale.

Level	2000–2010		2010–2022		2000–2022
	Area/ km2	Percent/%	Area/ km2	Percent/ %	Area/ km2	Percent/ %
Strongly Worse	3770.32	3.26	1273.48	1.11	542.34	0.47
Slightly Worse	51,250.91	44.26	45,271.26	39.44	40,940.26	35.66
Slightly Improved	56,878.45	49.13	63,638.51	55.43	70,014.59	60.99
Strongly Improved	3879.17	3.35	4616.14	4.02	3300.27	2.88

, ecological quality improved from 2000 to 2022, with areas categorized as slightly improved and strongly improved accounting for 60.99% and 2.88%, respectively. In contrast, strongly worse and slightly worse areas comprised 0.47% and 35.66%, respectively. Overall, the ecological quality of the study area showed an upward trend from 2000 to 2022.

3.1.2. County (District) Administrative Unit Scale

From 2000 to 2022, significant spatio-temporal differences were observed in the ecological quality of the 50 prefecture-level administrative units within the mountain–river–sea regional system (Figure 5). Overall, a gradient distribution pattern emerged, with higher ecological quality in the northwest and lower quality in the southeast. Areas with excellent and good ecological quality were mainly in the northwestern karst mountains, including Mashan, Tianlin, Xilin, and Leye Counties. Regions with medium ecological quality were concentrated in the central core area, dominated by the Nanning Metropolitan Area, where environmental evolution exhibited a concentric structure with peripheral optimization and core degradation. Areas with poor and fair ecological quality were mainly found in coastal regions with intensive human activity, such as Hepu County, Yinhai District, and Tieshan Port District. As shown in

Table 5. Area proportion of RSEI level changes at county (district) administrative unit scale.

Level	2000		2010		2022		Area Change 2000–2022
	Area/ km2	Percent/ %	Area/ km2	Percent/ %	Area/ km2	Percent/ %
Poor	2784	2.54%	413	0.38%	8563	7.80%	−4225.97
Fair	7682.00	7.00%	13,804	12.58%	20,078	18.30%	−6298.52
Medium	62,915	57.34%	54,168	49.37%	27,402	24.98%	−3862.32
Good	36,335	33.12%	33,676	30.69%	45,808	41.75%	3641.10
Excellent	0	0.00%	7655	6.98%	7865	7.17%	10,487.78

, the proportion of areas with excellent and good ecological quality increased from 33.12% to 48.92% between 2000 and 2022, accounting for nearly half of the study area. Conversely, the proportion of areas with poor and fair ecological quality fluctuated between 9.54% and 26.1%. In conclusion, from 2000 to 2022, the ecological quality of the study area underwent significant changes, with the area of improved regions exceeding that of degraded regions.

3.2. Correlation Analysis of RSEI at Different Scales

To reveal the spatial clustering characteristics of ecological changes in the mountain–river–sea regional system, GeoDa software (https://geodacenter.github.io/zh-Hans/) was employed to analyze the spatial autocorrelation of ecological quality for the years 2000, 2010, and 2022, generating the Local Indicators of Spatial Association (LISA) cluster map (Figure 6). The results showed that the Moran’s I indices for 2000, 2010, and 2022 were 0.806, 0.784, and 0.591, respectively. All values (p < 0.01) were greater than zero, confirming significant spatial clustering of ecological quality in the study area. However, the gradual decline in Moran’s I values suggests a weakening trend of spatial autocorrelation over the study period.

3.2.1. Grid Scale

As shown in Figure 6 and

Table 6. LISA types of RSEI at grid scale.

Type of Spatial Autocorrelation	2000		2010		2022
	Number	Percent/%	Number	Percent/%	Number	Percent/%
High–High	722	17.32	714	17.11	1316	31.57
High–Low	2	0.05	4	0.1	184	4.42
Low–High	4	0.1	3	0.12	70	1.68
Low–Low	748	17.94	726	17.45	1183	28.38
Non-Significant	2693	64.59	2713	65.22	1415	33.95
Total	4169	100.00	4160	100.00	4168	100.00

, the spatial distribution of ecological quality in 2000, 2010, and 2022 was mainly characterized by high–high and low–low clusters, with very few high–low and low–high clusters occasionally distributed across the study area. The area of high–high clusters continued to expand, mainly forming concentrated patterns in the northwestern and southern regions. The proportion of low–low clusters increased from 17.94% in 2000 to 28.38% in 2022, shifting from scattered distributions to a continuous concentration in the south-central region. The number of high–low clusters surged from 2 to 184 units, reflecting a growing divergence in ecological quality between these units and their surroundings. Low–high clusters exhibited notable spatial migration, with only 4 units (0.10%) at the edge of high–high clusters in Fangchenggang City by 2000; by 2010, they had shifted to the northwestern karst mountains and increased sharply to 70 units by 2022, forming a dispersed distribution pattern. Other regions showed little change, remaining non-significant clusters.

3.2.2. County (District) Administrative Unit Scale

To better understand the spatial characteristics of ecological quality in the study area, a local spatial clustering analysis was performed at the county/district administrative unit scale using ArcGIS software (https://www.arcgis.com/index.html). The resulting LISA cluster map (Figure 7) illustrates spatial patterns, with the corresponding cluster type areas and proportions quantified in

Table 7. LISA types of RSEI at county (district) administrative unit scale.

Type of Spatial Autocorrelation	2000		2010		2022
	Area/ km2	Percent/%	Area/ km2	Percent/%	Area/ km2	Percent/%
High–High	8563	7.80%	3558	3.24%	2048	1.87%
High–Low	0	0.00%	0	0.00%	5882	5.36%
Low–High	0	0.00%	2461	2.24%	1871	1.71%
Low–Low	15,547	14.17%	2810	2.56%	5573	5.08%
Non-Significant	85,606	78.03%	100,887	91.95%	94,336	85.99%

.

Spatial autocorrelation patterns of ecological quality were dominated by high–high and low–low clusters, consistent with grid-scale analysis results. In 2000, high–high clusters concentrated in Xilin County, Longlin Multinational Autonomous County, and Lingyun County. By 2010, these clusters shifted to Lingshan County along the Beibu Gulf coast, potentially linked to the diffusion of ecological pressure gradients driven by the development of the Beibu Gulf Economic Zone. By 2022, only Lingyun County, with greater ecological resilience, retained high–high clusters. Low–low clusters showed a trend of central contraction and southward migration. In 2000, they were mainly distributed across counties (districts) in central-western Nanning City and all of Beihai City. By 2010, their coverage contracted to Long’an County and Tieshangang District. In 2022, they expanded southward to cover Beihai City and Luchuan County. During the study period (2000–2022), counties (districts) such as Jiangnan District, Fusui County, and Qingxiu District in the central region transitioned from low–low clusters to non-significant clusters, indicating improvements in urbanization levels and ecological conditions. From 2000 to 2022, the spatial extents of both high–high and low–low aggregation types decreased, while non-significant clusters remained the most extensive.

3.3. Driver Detection Analysis of RSEI at Different Scales

3.3.1. Grid Scale

The results of the single-factor detection are shown in Figure 8. The p-values for all selected drivers were less than 0.001, indicating significant influence on the spatial distribution of RSEI. From 2000 to 2022, the drivers with the highest q-values were as follows: elevation was the dominant factor in both 2000 and 2022, while GDP showed the strongest explanatory power in 2010. Therefore, elevation and GDP were the key drivers of RSEI’s spatial heterogeneity. Additionally, night-time light and population density consistently ranked among the top four drivers, with relatively strong explanatory power. In contrast, the q-values for slope and aspect remained low, indicating their limited impact on RSEI’s spatial pattern. Precipitation consistently exhibited the weakest explanatory power, likely due to the unique hydrogeological conditions in the karst region of southwestern Guangxi. The well-developed karst landforms and underground rivers promote the rapid infiltration of precipitation through subsurface conduits, leading to minimal surface water retention. Since the humidity index in RSEI primarily reflects surface water content, the correlation between precipitation and RSEI is reduced in this region.

The interaction detection analysis (Figure 9) reveals the multi-dimensional driving mechanisms behind the spatial differentiation of ecological quality. A synergistic enhancement effect was consistently observed among all driving factors, with no mutual independence or weakening. The interaction value between elevation and other driving factors was generally high, confirming elevation as a key determinant of RSEI variability. The dominant interaction combinations from 2000 to 2022 were elevation ∩ GDP, elevation ∩ night-time light, and GDP ∩ night-time light, with explanatory values of 0.469, 0.482, and 0.524, respectively. Notably, when the interaction q-value exceeded 0.5, the combinations of night-time light with elevation, precipitation, and GDP showed the strongest explanatory power, with q-values of 0.535, 0.501, and 0.524, respectively. This indicates that ecological quality dynamics in the mountain–river–sea regional system result from the synergistic interplay of natural and socio-economic factors. Consequently, ecological protection strategies must account for these multi-factor interactions. In areas with high anthropogenic pressure, targeted management measures should be implemented to ensure sustainable ecosystem development.

3.3.2. County (District) Administrative Unit Scale

The single-factor detection results are shown in Figure 10. From 2000 to 2022, various factors influenced RSEI to varying extents. In 2000 and 2010, GDP had the highest explanatory power and was the dominant factor driving changes in RSEI within the study area. The q-value of GDP fluctuated over time, notably dropping to 0.226 in 2010. This decline reflects the slowdown in economic growth after the global financial crisis and the initial impact of ecological protection policies. By 2022, population density surpassed GDP as the primary factor. Among topographic drivers, elevation consistently impacted the spatial differentiation of RSEI, staying in the top three throughout the study period. In contrast, aspect remained at the bottom, indicating its limited influence. Among the climate drivers, precipitation had a negligible impact, with its explanatory power for ecological quality differentiation being minimal. Overall, economic activities—particularly fluctuations in GDP—played a dominant role in shaping the ecological quality patterns throughout the study period. Economic factors had stronger explanatory power for environmental changes than topographic and meteorological drivers.

The results of the two-factor interaction analysis are presented in Figure 11. The interaction types among the influencing factors were mainly categorized as bivariate enhancement and nonlinear enhancement, with no independent relationships observed. In 2000, slope was the dominant factor driving the spatial ecological differentiation of RSEI. Combinations involving aspect and other factors showed high explanatory power, with the interaction q-value for slope ∩ aspect reaching the maximum value of 0.659. In 2010, the interaction q-value for population density ∩ night-time light increased to 0.623, reflecting the combined ecological pressure from accelerated urbanization. Interaction values for aspect and temperature with other factors were generally low, indicating a reduced influence of topographic and climatic factors on RSEI variability. By 2022, the interaction values between night-time lighting and all other factors exceeded 0.57, with the combined explanatory power of night-time light and population density reaching at 0.727. Overall, the combined effects of most interacting factors were more effective at explaining the spatial differentiation of RSEI in the study area than individual factors.

4. Discussion

4.1. Characteristics and Causes of Spatial and Temporal Evolution of RSEI at Different Scales

The study employed the RSEI model to conduct remote sensing monitoring and ecological analysis of the mountain–river–sea regional system at different scales from 2000 to 2022. The results consistently showed RSEI values exceeding the moderate ecological quality threshold (0.5) for all study periods, with a significant upward trend overall. This finding closely aligned with the Ministry of Ecology and Environment’s evaluation based on the Ecological Quality Index (EQI) [42], further confirming the credibility of the RSEI results.

Analysis at multiple scales revealed that the ecological quality in the study area followed a spatial differentiation pattern: mountain ecological barrier → central low-lying zone → coastal ecological fragile zone. Ecological quality in the karst mountainous areas of northwestern Guangxi remained relatively stable, primarily due to high terrain and land use dominated by forest and grassland, which sustained high greenness and humidity indices [43]. The central area, characterized by population agglomeration and economic activity, has dual attributes as both an urban expansion zone and an agricultural transition zone. In urban cores like Qingxiu District and Xixiangtang District, rapid land expansion led to ecological space fragmentation, resulting in persistently low ecological quality. In suburban areas like Liangqing District, Hengzhou City, and Binyang County, ecological quality gradually improved, facilitating the diffusion of higher-grade conditions toward the urban periphery. However, in major grain-producing regions like Binyang County and Hengzhou City, long-term nonpoint source pollution [44] from chemical fertilizers kept ecological quality at a medium level, despite policies aimed at protecting cultivated land. The southern region, characterized by low elevation, is a coastal industrialized zone. Rapid urbanization has driven significant land use changes and increased industrial pollution. In conjunction with port economic development and the degradation of mangrove ecosystems, these factors have led to declines in greenness and humidity indices, resulting in a fragile ecological pattern marked by alternating phases of development and restoration. In the future, it is essential to adopt a strategy centered on “zone-specific precision policies and regional collaborative governance” to balance industrial transformation with the release of ecological benefits.

4.2. Characteristics of RSEI Spatial Clustering Divergence at Different Scales

The ecological quality of the mountain–river–sea regional system demonstrated significant spatial clustering characteristics at different scales. High–high clusters were predominantly located in the northwestern karst mountains, and their stability correlated positively with high vegetation cover, abundant precipitation, well-developed river systems, and limited human activity in the region. Low–low clusters were concentrated in central urban agglomerations and southern coastal zones, with their expansion closely linked to rapid urbanization and the concentration of port industries.

In 2010, Lingshan County exhibited high–high clusters at the county (district) administrative unit scale. However, at the grid scale, some eastern grids showed degradation. In both 2000 and 2022, Lingyun County was identified as a high–high cluster region at the county (district) administrative unit scale, but the central part of the county displayed non-significant clusters at the grid scale. By 2022, Tiandeng County exhibited a small proportion of high–low clusters, although the county was classified as a high–low cluster at the county (district) administrative unit scale.

The results of the scale-based analysis indicate that each scale has distinct advantages and limitations in revealing the spatial aggregation characteristics of RSEI. The grid scale captures subtle spatial variations and local aggregation patterns of RSEI, while the county (district) administrative unit scale more clearly highlights spatial distribution trends and changes of RSEI across different administrative regions. Using a multi-scale approach to analyze the ecological quality changes of the mountain–river–sea regional system provides a more comprehensive understanding of regional ecological conditions. It also enhances the ability to identify local issues, improving the accuracy and reliability of the study. This approach offers a scientific basis for optimizing the ecological security pattern through coordinated land and sea development, ultimately contributing to the sustainable development of the regional ecological environment.

4.3. Response Analysis of RSEI Drivers at Different Scales

The results of the geodetector analysis at different scales reveal significant differences in the key factors influencing the ecological quality of the mountain–river–sea regional system. At the raster scale, elevation is the dominant driver, consistent with the findings of Yu et al. (2022) [45] in the Huaihe River Basin, suggesting that topographic factors predominantly govern local-scale ecological patterns. At the county (district) administrative unit scale, GDP emerges as the primary determinant, aligning with numerous studies exploring the relationship between the economy and ecology at the administrative unit scale [46,47].

Interaction analysis highlights the strong explanatory power of the combination of GDP and night-time light at both scales. While the interaction value at the raster scale (0.524) is lower than at the county (district) administrative unit scale (0.686), it remains higher than the contribution of individual factors. This finding supports the conclusions of Zhao et al. [48], who demonstrated the effect of multi-factor interaction enhancement on the ecological environment. A further comparison reveals that at the grid scale, the synergistic effect of topography and human activity dominates, with the interaction between elevation and night-time light in 2022 exhibiting the strongest explanatory power for changes in RSEI (0.638). At the county (district) administrative unit scale, however, socio-economic factors take precedence, and the explanatory power of the interaction between population density and night-time light peaks at 0.712.

The scale effect influences the identification of driving factors and their explanatory power [49,50]. Generally, the county (district) administrative unit scale exhibits stronger explanatory power than the raster scale. For example, elevation was a core driver of ecological divergence at the county (district) administrative unit scale in 2000 and 2022, whereas its influence fluctuated at the raster scale, dropping from 0.480 in 2000 to 0.113 in 2010, before recovering in 2022—though it still remained below the county (district) administrative unit scale value for the same period. Moreover, the interaction value of population density ∩ night-time light in 2022 reached 0.7268 at the county (district) administrative unit scale but was only 0.465 at the raster scale. This discrepancy arises from the smoothing effect [51] in administrative data aggregation, which amplifies the role of socio-economic factors and reduces the explanatory power of natural factors at the local scale.

4.4. Innovations and Shortcomings

Previous studies have confirmed the applicability of MODIS remote sensing data in mesoscale studies [52,53]. Based on this, the study selected MODIS remote sensing data as the source and constructed the RSEI model using the GEE platform to explore spatio-temporal variations in ecological quality across different scales in the mountain–river–sea system from 2000 to 2022. PCA results indicated that the first principal component (PC1) contributed over 60% in 2000, 2010, and 2022, with NDVI and WET positively correlated with RSEI, while LST and NDBSI were negatively correlated with RSEI. These results are consistent with actual conditions, suggesting that the model is applicable to other ecological transition zones with similar terrain and climate conditions. However, the dominance and interaction of individual RSEI components may vary across different areas. For instance, in arid–semi-arid transition zones, the dryness and heat components may play a more prominent role in determining ecological degradation, whereas in humid mountainous zones, greenness and wetness tend to dominate ecological quality assessment. Therefore, transferring the model to other regions should be accompanied by contextual adjustments, such as incorporating region-specific ecological indicators or optimizing data sources to enhance the accuracy and reliability of assessment results.

MODIS remote sensing data were used to construct the RSEI model to assess the ecological quality of Guangxi’s mountain–river–sea system. This methodology effectively overcomes the limitations of unidimensional assessments and addresses the temporal lag of conventional statistical data. However, MODIS data has certain limitations, including relatively low spatial resolution and vulnerability to cloud and fog interference. To improve the accuracy and reliability of ecological assessments, future studies could incorporate higher-resolution remote sensing data through comparative analysis or data fusion. Given that ecological quality changes are influenced by multiple factors, future studies should integrate multisource big data reflecting regional natural and anthropogenic factors—such as land use, soil types, national policies, industrial structures, settlement patterns, and transportation networks—to enable a more comprehensive assessment of ecological quality dynamics. Failing to account for land use changes in the ecological quality assessment process may introduce biases, and these biases could pose risks to ecological management and decision-making.

Additionally, grid scale selection impacts the study’s results [23]. Within the same or across different study areas, results at various scales tend to exhibit scale-dependent spatial autocorrelation and heterogeneity. This study used only the 1 km × 1 km grid scale. Future research should assess outcomes at multiple grid resolutions to identify the optimal spatial scale for transitional regions, enhancing the precision and robustness of ecological assessments.

5. Conclusions

This study constructed the RSEI using GEE and MODIS remote sensing data to assess the spatio-temporal evolution of ecological quality in Guangxi’s mountain–river–sea system from 2000 to 2022. Spatial autocorrelation analysis and geographical detector models were used to systematically investigate the spatio-temporal correlations and driving factors of habitat quality variations at both grid and county levels. The key findings are summarized below:

(1) The average RSEI in the mountain–river–sea system increased from 0.549 in 2000 to 0.627 in 2022. Overall, ecological quality showed a consistent improvement. Spatially, ecological quality exhibited a clear northwest–southeast gradient, with better conditions in the northwest and poorer conditions in central urban cores and coastal areas. Temporally, both ecological degradation and enhancement co-existed, but improved areas outnumbered degraded ones.

(2) The spatial distribution of ecological quality changes in the mountain–river–sea system showed significant clustering at different scales. High–high clusters were mainly located in the karst mountain regions of the northwest, while low–low clusters were concentrated in central urban agglomerations and southern coastal areas. From 2000 to 2022, both cluster types expanded at the grid scale, but contracted and fragmented at the county (district) scale.

(3) At different spatial scales, the interactions of RSEI in the mountain–river–sea system were dominated by nonlinear and bivariate enhancement effects, with no independent or weakening interactions. Elevation, night-time light, and population density had the strongest explanatory power for RSEI’s spatial differentiation, while precipitation and aspect had weaker explanatory power. At the grid scale, elevation ∩ GDP dominated in 2000, while elevation ∩ night-time light became the dominant driver in both 2010 and 2022. At the county (district) scale, slope ∩ GDP dominated in 2000 and 2010, but shifted to population density ∩ night-time light in 2022.