The past and future dynamics of ecological resilience and its spatial response analysis to natural and anthropogenic factors in Southwest China with typical Karst

Abstract

With the global land use/land cover (LULC) and climate change, the ecological resilience (ER) in typical Karst areas has become the focus of attention. Its future development trend and its spatial response to natural and anthropogenic factors are crucial for understanding the changes of ecologically fragile areas to human behavior. However, there is still a lack of relevant quantitative research. The study systematically analyzed the characteristics of LULC changes in Southwest China with typical Karst over the past 20 years. Drawing on the landscape ecology research paradigm, a potential-elasticity-stability ER assessment model was constructed. Revealing the characteristics and heterogeneity of the spatial distribution, annual evolution, and development trend of ER in the past and under different scenarios of shared socioeconomic pathways and representative concentration pathways (SSP-RCP) in the future. In addition, the spatial econometric model was utilized to reveal the spatial effect response mechanism of ER, and adaptive development strategies were proposed to promote the sustainable development of Southwest China. The study found that : (1) In the past 20 years, the LULC in Southwest China showed an accelerated change trend, the ER decreased declined in general, and there was significant spatial heterogeneity, showing the spatial distribution pattern of “west is larger than east, south is larger than north, and reduction in the west was slower than that in the east.” (2) Under the same SSP scenario, with the increase of RCP emission concentration, the area of the lowest-resilience increased significantly, and the area of the highest-resilience decreased. (3) The woodland was the largest contributor to ER per unit area in the Southwest China, and grassland was the main LULC type, which had a prominent impact on the ER of the study area. (4) The average precipitation and the normalized difference vegetation index (NDVI) were significant natural drivers of ER in the study area, and the economic growth, innovation, and optimization of industrial structure contributed to the ER of Southwest China. Overall, the integration of quantitative assessment and multi-scenario-based modeling not only provides new perspectives for understanding the pattern of change and response mechanisms, but also provides valuable references for other typical Karst regions around the world to achieve sustainable development.

Introduction

In recent years, along with the rapid advancement of urbanization, human beings have contributed to significant changes in human-land relations through land use/land cover (LULC) changes, and have triggered many spatial conflicts such as spatial resource mismatch, spatial development disorder, ecosystem imbalance, and social development instability^1,2. Especially in the typical Karst region with exposed rocks, shallow soils, poor water-holding capacity, serious soil erosion, and low recoverability, ecological problems caused by LULC have gradually affected the sustainable development of the region and even threatened regional security³. Ecological resilience (ER) assessment provides a methodology and approach for measuring ecological security, thereby mitigating the typical negative ecological impacts of LULC change, such as ecological balance, soil quality decline, carbon cycle imbalance, water shortage and pollution, and providing a scientific basis for optimizing the ecological security pattern and realizing rational spatial planning of the national territory⁴. Therefore, a comprehensive quantification and understanding of the changing characteristics and development trends of ER in the past and future in Southwest China can help in decision-making for regional sustainability policies.

With the increasingly significant trend of disciplinary integration, more disciplinary methods, such as climatological methods, physical methods, ecological methods, and landscape ecological methods, have been introduced into geographic research, providing opportunities for quantitative research on regional ER^5,6. Among them, the study of landscape ecology “pattern and process” and its interrelationship has absorbed the spatial analysis method in geography and inherited the holistic idea in ecology³, which provides a new comprehensive perspective for regional ER research. However, most current research has focused on past LULC change and ecological resilience rather than future resilience. An accurate understanding of future LULC and ER changes and their coupling is critical.

High-resolution LULC prediction products are key to ensuring the accuracy of scientific analysis and assessment modeling of landscape patterns, while they guarantee the effectiveness of optimized mitigation strategies for assessing multiple scales. The sixth phase of the International Coupled Model Intercomparison Program (CMIP6) provides a richer set of global climate model data for the field of climate change prediction. Its science-based scenarios of shared socio-economic pathways (SSPs) and representative concentration pathways (RCPs) incorporate the impacts of socio-economic development and will provide more reliable results of LULC changes under different development scenarios⁷, and thus provide more valuable data support for ER research.

For ecologically fragile regions, most studies emphasize the impact of natural factors such as precipitation, temperature, and solar radiation on the resilience of terrestrial ecosystems⁸. Also, the response of ER to urban expansion, production patterns, and changes in surface morphology has been widely publicized in recent years⁹. Overall, there is still insufficient discussion of the impact of natural and anthropogenic factors on ER based on an integrated perspective. In previous studies, the Human Footprint Index (HFI) has often been used to represent the influence of anthropogenic factors¹⁰, or use the residuals of natural variables to represent the intensity of human activity¹¹. However, it is difficult for these methods to reveal the influence mechanism of anthropogenic factors in a comprehensive and in-depth manner, which is not conducive to providing reasonable references for the adjustment of the development model. The development of natural and anthropogenic factors, as a comprehensive phenomenon linked to economic and social characteristics, requires reflection through a multi-level system of indicators. Currently, two common research methods can be categorized as linear and nonlinear models. Linear models (multivariate linear regression, Lasso regression, linear discriminant analysis, logistic regression, correlation analysis, etc.) are essentially a method of studying the interactions between linear variables and are usually applied to empirical studies^10,12,13. While providing powerful fitting capabilities, nonlinear models (artificial neural networks, mixed-effects models, logistic growth model, random forests, support vector regression, etc.) face drawbacks such as complexity, computational difficulty, sensitivity to noise and outliers, risk of overfitting, poor interpretation of results, and difficulty in validation^14,15. Considering the connectivity of natural and anthropogenic features, a spatial analysis approach is needed to enable effective reflection of spatial spillover effects. The spatial econometric model improves the general linear regression model based on considering the spatial correlation of the independent and dependent variables, and introduces a spatial weight matrix to make the regression coefficients more precise^5,10.

The study systematically constructed an ER assessment model, identifying and quantifying the natural and anthropogenic factors driving ER, spatial spillover effects, as well as spatial and temporal differentiation characteristics and development trends of ER under different scenarios of future SSP-RCP. Figure 1 presents the flowchart of this study. The objectives of this paper are: (1) to identify the change patterns of LULC in Southwest China, and to provide a new method for assessing ER based on the potential-resilience-stability model by drawing on the research paradigm of landscape ecology; (2) to comprehensively analysis the interannual evolutionary characteristics and heterogeneity of ER in the past of the study area; (3) to emphasize regional variability and geographic dependence of ER, innovatively using a combination of spatial autocorrelation analysis and spatial measurement modeling to identify and quantify specific natural and anthropogenic factors and their spatial spillover effects, and proposing adaptive development strategies to promote sustainable development in Karst regions; (4) to explore the coupled ER and SSP-RCP scenarios, and to study the interannual evolution and change trends, in which to help determine the optimal scenario strategy and to provide valuable references for realizing the sustainable development of the ecologically fragile region.

Figure 1

The flowchart of this study.

Materials and methods

Study area

The typical area of Karst in China is located in the southwest (20°09′N–34°19′N, 97°21′E–117°19′E), centered on the Yunnan-Guizhou Plateau, with an average elevation of 800–1000 m, and a complex and varied terrain¹⁷ (Fig. 2). It includes Guizhou, Yunnan, Sichuan, Chongqing, Guangdong, Guangxi, Hunan and Hubei eight provincial administrative units. Located near the Tropic of Cancer, it has a subtropical monsoon climate, with rain and heat at the same time, an average annual temperature of 18 °C, and annual precipitation of over 900 mm^18,19. Most of the area is found in the subtropical evergreen broad-leaved forest belt, with rich and varied vegetation types and high biodiversity^20,21. At the same time, it is also the water supply area of the Pearl River and the Yangtze River, the water source area of the South-to-North Water Diversion and the Three Gorges Reservoir Area, the regional geographic and ecological position is very important²². However, the ecological problems in southwest China are very prominent, due to the extremely slow rate of soil formation in the Karst region, the thin soil layer²³, and the steep slopes and high mountains, which make it difficult to maintain soil and water. The fragile Karst environment coupled with irrational human activities has led to serious damage to the ecological environment, triggering a series of ecological problems²⁴.

Figure 2

Location and elevation of the study area (a), spatial distribution of Karst in China (b), land use/cover in 2020 and sample unit map (c). The Figure is created using ArcGIS ver.10.6 (https://www.esri.com/).

Data collection

The data used in this research include the vegetation data, the meteorological data, the surface solar radiation data, the soil data, and the land cover types data. The Normalized Difference Vegetation Index (NDVI) and Modified Normalized Difference Water Index (MNDWI) data (30 m × 30 m) came from the National Comprehensive Earth Observation Data Sharing Platform (http://www.chinageoss.org/). Annual average rainfall, annual average temperature, and administrative boundary data of the study area were obtained from the Data Center of Resource and Environmental Sciences, Chinese Academy of Sciences (https://www.resdc.cn/). Soil data (1 km × 1 km) were from the World Soil Database (HWSD) China Soil Data Set (v1.1). Land cover types were derived using the MCD12Q1 version 6 data product (The National Aeronautics and Space Administration, https://lpdaac.usgs.gov/products/mcd12q1v006), which were derived through supervised classification of the Terra and Aqua reflectance data. Data preprocessing was accomplished through multiple remote sensing techniques in conjunction with MRT tools. The Global IGBP LULC projection dataset (1 km × 1 km) was obtained from the Open Data (https://figshare.com/articles/dataset/Global_IGBP_LULC_projection_dataset_under_eight_SSPs-RCPs/20088368/1), which includes projections for land use and land cover changes at five-year intervals between 2015 and 2100 was developed using the Future Land Use Simulation (FLUS) model¹⁶. Taking into account research needs and drawing insights from peer studies^6,7, the 17 land cover types originally classified based on the International Geosphere-Biosphere Programme (IGBP) classification system (https://fluxnet.org/data/badm-data-templates/igbp-classification/) have been reorganized into 8 categories (Table 1). The construction of the Economic-Social data was sourced from the China Statistical Yearbook, the China Science and Technology Statistical Yearbook, the China Third Industry Statistical Yearbook, and the statistical yearbooks of various provinces.

Table 1 Results of land use reclassification.

Methods

Ecological resilience analysis

Ecological resilience assessment model

The concept of ER introduced by S. Holling in 1973, refers to the persistence of natural systems in response to ecosystem changes caused by natural or human-induced factors²⁵. Gunderson noted that ER also includes the adaptive capacity of ecosystems²⁶. As the meaning of ER has deepened, Nyström and Folke have expanded its practical use in research and provided implementation methods for how ecosystems can be reorganized, rebuilt, renewed, and developed after a disturbance²⁷. Integrating the existing definitions, the ER mainly emphasizes the ability of ecosystems to reorganize, adapt, and sustain themselves under disturbances. Drawing on the pattern-process-function research paradigm of landscape ecology, we constructed an ER assessment model of potential-elasticity-stability in terms of sustainability, restorative, and adaptability.

The potential is an attribute of ecosystems that characterizes their ability to provide services and is an important support for ecosystems to form and maintain the conditions and functions²⁷. Ecosystem Service Value (ESV) is a comprehensive indicator of ecosystem service capacity. Higher ecosystem service values generally represent higher ecosystem potential²⁸. The calculation method is derived from the method proposed by Costanza modified by Xie et al.²⁹. The formula is as follows:

$$ ESV_{k} = \sum\nolimits_{f} {A_{k} \times VG_{k} } $$

(1)

$$ ESV_{f} = \sum\nolimits_{f} {A_{k} \times VG_{kf} } $$

(2)

$$ ESV = \sum {\sum {A_{k} \times VC_{kf} } } $$

(3)

where $ESV_{k}$, $ESV_{f}$ and $ESV$ denote the potential for land use category k the potential for service function f and the total ecosystem potential, respectively. A_k represents the land area of type k, $V{\text{C}}_{kf}$ represents the potential per unit area of service f of type k.

Ecological elasticity, also known as ecological recovery, is the ability of an ecosystem to recover when it suffers damage⁴. The elasticity of LULC types formed naturally is better, and it is easier to restore the original state than the land use types caused by human activities³⁰. The formula is as follows:

$$ R = \sum {A_{k} \times RC_{k} } $$

(4)

where $R$ is the ecological elasticity; $A_{k}$ denotes the land area of type $k$, and $RC_{k}$ is the ecological elasticity coefficient of type $k$ land use type.

The stability of ecosystems indicates the strength of ecosystem adaptability. The contagion index (CONTAG) can fully identify the agglomeration of patches in landscape ecosystems. It mainly reflects the stability of various ecosystems between patches in the system by whether stable ecological processes can be generated between patches³¹. The formula is as follows:

$$ CONTAG = 100 + 100\sum\limits_{i = 1}^{m} {\sum\limits_{k = 1}^{m} {\left[ {p_{i} \left( {\frac{{g_{ik} }}{{\sum\nolimits_{k = 1}^{m} {g_{ik} } }}} \right)} \right]} } \left[ {\ln \left( {p_{i} \left( {\frac{{g_{ik} }}{{\sum\nolimits_{k = 1}^{m} {g_{ik} } }}} \right)} \right)} \right]/2\ln (m) $$

(5)

where $p_{i}$ is the proportion of landscape patches of type $i$ to the total landscape area; $g_{ik}$ is the number of raster edges adjacent to landscape patches of types $i_{{}}$ and $k$; and $m$ is the total number of patch types.

Due to the different units of calculation of potential, resilience and stability, it is necessary to standardize each indicator to [0, 1]. Ultimately, the ER calculation formula is as follows:

$$ ERI = \sqrt[3]{ESV \times R \times S} $$

(6)

where ER is ecological resilience; ESV is potential; R is elasticity and S is stability.

Quantifying ecological resilience change

The gravity model is widely used in the field of geography and land use science to determine the direction of motion and the distance to the center of gravity of the research object³. It can well describe the spatial distribution position of geographical elements in two-dimensional space^3,30. According to the change of spatial position of the gravity center, it can not only reflect the spatial distribution characteristics of geographical elements, but also identify the spatial movement development trend.

Future LULC simulation scenario setting

In 2021, the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) released a new climate model for CMIP6, which developed a set of new emissions scenarios driven by different socioeconomic models—Shared Socioeconomic Pathways (SSPs). It replaces the four Representative Concentration Pathways (RCPs) in CMIP5 and is an important enhancement in the CMIP6 scenario (Fig. 3). The new scenarios in CMIP6 make up for the lack of a strong link between SSPs and RCPs⁶. Many studies have shown that careful consideration of the rationale for particular combinations is essential when integrating RCP and SSP scenarios. For example, some models have found that the level of radiative forcing is achievable only in SSP5, whereas it is not achievable in SSP3⁷. CMIP6 provides the characteristics and rationality of these scene designs. Therefore, this study combined the eight future scenarios contained in the CMIP6 report (divided into two groups according to priority: Tier-1 and Tier-2) to assess the future ER of the study area.

Figure 3

SSP-RCP scenarios framework.

Spatial econometrics models for exploring ecological resilience impact factors

(1) Spatial autocorrelation analysis

Spatial autocorrelation analysis can be used to explore the spatial distribution pattern and correlation of spatial geographic unit attribute values in the study area⁴. The study used the global spatial autocorrelation Moran's I index to explore the overall spatial distribution trend and clustering status of ER in Southwest China. The calculation formula is as follows:

$$ Moran{\text{'}} s \, I = \frac{{\sum\nolimits_{i = 1}^{n} {\sum\nolimits_{j = 1}^{m} {w_{ij} \left( {x_{i} - \overline{x}} \right)} } }}{{\frac{1}{n}\sum\nolimits_{i = 1}^{n} {(x_{i} - \overline{x})^{2} } \sum\nolimits_{i = 1}^{n} {\sum\nolimits_{j = 1}^{m} {w_{ij} } } }} $$

(7)

where $x_{i}$ is the attribute value of the first geographic unit, n is the total number of units, and $w_{ij}$ is the space weight matrix, i = 1, 2, ⋯ n; j = 1, 2, ⋯ m, when unit i and unit j are adjacent, $w_{ij} = 1$.

(2) Spatial measurement models

The spatial measurement model solves the problem of spatial dependence that cannot be solved by linear regression analysis, and the spatial lag model (SLM), spatial error model (SEM), and spatial Durbin model (SDM) are all commonly used models^11,32. The spatial Durbin model is a more general model of the first two models that incorporates endogenous interactions (WY) and exogenous interactions (WX)³³. Among them, θ is the coefficient of the exogenous interaction effect. If θ = 0, it is used as the SLM model. If θ = − ρβ, it is the SEM model. The significance of θ represents the strength of the spatial interaction of the explanatory variables. However, θ cannot be used as the basis for determining the existence of spatial spillover effects. Therefore, it is necessary to estimate the direct and indirect effects of explanatory variables to determine the spatial spillover effects^32,33. The relevant calculation formulas are as follows:

$$ Y = X\beta + \lambda W\mu + \varepsilon ,\varepsilon - N(0,\delta^{2} ) $$

(8)

$$ Y = \rho Wy + X\beta + \varepsilon ,\varepsilon - N(0,\delta^{2} ) $$

(9)

$$ Y = \rho Wy + X\beta + Wx\theta + \varepsilon ,\varepsilon - N(0,\delta^{2} ) $$

(10)

where Y is the dependent variable; X is the independent variable; β is the prognostic parameter; λ is the spatial autocorrelation coefficient, which measures the spatial dependence between the samples; W is the n × n order weight matrix; $Wu$ is the spatial error term; ε is the vector of random errors. ρ is the spatial autoregressive coefficient parameter affected by the W matrix; $Wy$ is the spatial lag term.

Results

Spatial extent and diverging patterns of LULC in southwest China

From 2000 to 2020, about 166,000 km², 8.57% (with an annual growth rate of 0.42%) of the study area experienced at least one change in LULC type (Fig. 4), and the spatial distribution of LULC types changed was relatively uniform. Among them, the distribution of Sichuan and Yunnan in the western part of the study area is relatively concentrated, and the proportion of LULC types that have changed was 11.63% and 11.22%, respectively. Guizhou, Hubei, and Hunan, which are located in the central and northeastern parts of the study area, were sparsely distributed. The proportion of LULC types that changed was 5.56%, 5.92%, and 5.46%, respectively.

Figure 4

Spatial distribution of land use type change from 2000 to 2020. The Figure is created using ArcGIS ver.10.6 (https://www.esri.com/).

The main LULC types in the study area were grassland, woodland, and cultivated land, which accounted for more than 95% of the total area (Fig. 5). From 2000 to 2020, the study area experienced an increasingly rapid LULC transformation, which was characterized by many-to-many conversion between different LULC types. It is worth noting that the mutual conversion of woodland and grassland, grassland and cultivated land play an important role in these transformations. The main trend of LULC changed from 2000 to 2005 was the conversion of cultivated land to grassland, with an area of about 6324.04 km². In comparison, the LULC conversion from 2005 to 2010 was more intense, with the transformation of woodland to grassland covering an area of 14,724.14 km² and the transformation of grassland to cultivated land covering an area of 12,479.74 km². From 2010 to 2015, the conversion of grassland to woodland and the conversion of cultivated land to grassland became the main types of LULC transformation, with roughly equal areas of 14,999.92 km² and 14,115.38 km², respectively. From 2015 to 2020, the conversion of grassland to cultivated land and woodland had the largest areas, with 13,820.91 km² and 11,676.59 km², respectively.

Figure 5

Chord diagram of LULC changed in southwest China from (a) 2000–2005, (b) 2005–2010, (c) 2010–2015, and (d) 2015–2020. The Figure is created using Origin 2021 (https://www.originlab.com/).

Spatial–temporal characteristics of ecological resilience

Based on the classification results of the natural breakpoint method of ER⁶, we categorized ER in the study area into five levels (I–V): lowest (< 0.215), lower (0.215–0.326), medium (0.326–0.495), higher (0.495–0.657), and highest (> 0.657). The study area was dominated by lower-resilience from 2000 to 2020, with a trend of decreasing and then increasing (Fig. 6). Both the lowest-resilience and lower-resilience areas showed a slow increase, with the area proportion stabilizing and the rate of growth easing after 2010. The lowest-resilience was mainly distributed in the north-central, northeastern, and southeastern parts of the study area, located in the Sichuan Basin and the plains of the middle and lower reaches of the Yangtze River, which increased from 8.06 to 13.28% during the study period, the LULC types dominated by construction land and cultivated land. The lower-resilience and medium-resilience were located in the northwestern and central parts of Sichuan Province, mainly concentrated in the Yunnan-Guizhou Plateau, and the LULC type was dominated by grassland and cultivated land. The area proportion of medium-resilience was over 21% in all the study periods, but showed a slow decline, from 23.24 to 21.85% during the study period, mainly in the southwestern part of the study area. The highest-resilience areas continued to decline, with the area proportion reaching its lowest in 2007 at 12.45%. And the higher-resilience areas declined steadily before 2010 and have risen slowly since then. In terms of spatial distribution, the higher-resilience and the highest-resilience were mainly distributed in the western and southeastern parts of the study area, concentrated in the Hengduan Mountains, and the LULC type was dominated by woodland and waters.

Figure 6

Spatial distribution of ER levels (a–e) and changes in the proportion from 2000 to 2020. The Figure is created using ArcGIS ver.10.6 (https://www.esri.com/).

The gravity center of ER in the study area had gradually shifted to the west over the past 20 years, with the ER reduction in the west was slower than that in the east. Among them, the longest moving distance was 14.447 km in 2005–2006 (Fig. 7). Both the gravity centers of the lowest-resilience and lower-resilience showed a tendency to move to the east, with 2006–2011 being the period of growth in the moving rate and 2015–2019 being the main period of moving. The gravity centers of medium-resilience and higher-resilience as a whole moved to the north, with a more balanced average annual moving rate relative to the other resilience levels. The gravity center of the highest-resilience gradually shifted to the south, and the moving rate was faster than other resilience levels, in which the moving rate continued to increase from 2014 to 2020, and the moving distance reached 64.657 km in 2019–2020. The above change in the gravity centers of the levels indicated a faster rate of ER degradation in the eastern part of the study area mainly due to the gradual increase in the proportion of both the lowest and lower levels.

Figure 7

The annual progressive changes of the gravity center of different levels of ER from 2000 to 2020. (a) Lowest; (b) Lower; (c) Medium; (d) Higher; (e) Highest; (f) Entirety. The blue bar chart shows the average annual moving distance of the gravity center of different levels of ER.

Prediction of ecological resilience under eight SSP-RCP coupling scenarios

The lowest-resilience areas in the Southwest China showed first an upward and downward trend (SSP1-RCP1.9, SSP1-RCP2.6, SSP2-RCP4.5) or reached a stabilization stage (SSP5-RCP8.5, SSP3-RCP7.0, SSP4-RCP6.0) in the last 20 years and in the future 8 SSP-RCP prediction scenarios, with 2050 being the turning point of this change (Fig. 8). Over the past 20 years, there had been an overall decrease in the area proportion of lower-resilience. For the eight RCP-SSP prediction scenarios, the order of lower-resilience areas, from lowest to highest, was SSP4-RCP3.4 < SSP5-RCP3.4 < SSP1-RCP2.6 < SSP1-RCP1.9 < SSP5-RCP8.5 < SSP2-RCP4.5 < SSP3-RCP7.0 < SSP4-RCP6.0. Except for the SSP4-RCP3.4, SSP5-RCP3.4, and SSP5-RCP8.5 scenarios, the areas of medium-resilience decreased progressively in the past and the prediction scenarios. The order of decreasing areas of medium-resilience in the different prediction scenarios was SSP4-RCP3.4 < SSP3-RCP7.0 < SSP1-RCP2.6 < SSP1-RCP1.9 < SSP2-RCP4.5 < SSP4-RCP6.0 < SSP5-RCP3.4 < SSP5-RCP8.5. The higher-resilience areas showed a decreasing trend over the last 20 years, with the areas in the eight SSP-RCP scenarios in a steady state before 2060 and a fluctuating decrease after 2060, with the most significant rate of decrease in RCP3.4-SSP4. The highest-resilience areas showed a gradually decreasing pattern in the past, but the SSP4-RCP3.4, SSP1-RCP1.9, and SSP1-RCP2.6 scenarios showed a significant increase after 2050.

Figure 8

Relative change in various ecological resilience levels under different SSP-RCP scenarios (with 2000 as the reference baseline). (a) Lowest; (b) Lower; (c) Medium; (d) Higher; (e) Highest.

Analysis of factors influencing ecological resilience

Model validation and analysis results of the spatial Durbin model

The study first performed a diagnosis of multicollinearity for all explanatory variables, which showed that the variance inflation factor (VIF) for each explanatory variable was less than 10, excluding factorial multicollinearity interference (Table S1). Next, the spatial autocorrelation of the data was examined by Moran’s I index with the ArcGIS10.6 platform, which indicated that the spatial distribution of ER in the study area showed obvious spatial agglomeration. Therefore, when analyzing the influencing factors of ER, the traditional linear regression model was no longer applicable due to the limitations of their linear relationship assumptions, and a spatial measurement model needed to be used. Meanwhile, the analysis showed that the significance of the geographic neighborhood weight matrix (weighting according to whether geographic units are adjacent to each other or not) was significantly better than that of the geographic distance weight matrix (weighting based on the actual distance between geographic units). After selecting the appropriate spatial weight matrix (Table 2), the Lagrange multiplier (LM) test, corrected Hausman test, and likelihood ratio (LR) test were conducted respectively (Table S2). Based on the test results we found that the spatial Durbin model does not allow degeneration into a spatial error model or a spatial lag model, and therefore the spatial Durbin model was used for the subsequent analysis.

Table 2 Moran’s I index of ecological resilience. ***, ** and * indicate the significance levels of 1%, 5% and 10%, respectively.

By comparing the adjusted R² and Loglikelihood, the results estimated by the spatial Dubin model under time-period fixed effects have better goodness of fit and higher accuracy (Table 3). The results showed that the spatial autoregressive coefficient (rho) was significant at the 5% level, accounting for 0.348, indicating that the explanatory variables have positive spatial spillover effects on themselves. It means that the change of ER in Southwest China can significantly affect the neighboring regions, and the result again verified the necessity of adding spatial parameters when analyzing ER. In addition, both population aggregation and urban expansion significantly inhibited the enhancement of ecological resilience, while there were also significant spatial spillover effects. To further analyze the impact of each factor on ER and provide a more detailed explanation, the study continued to decompose their spatial effects.

Table 3 Spatial Durbin model regression results. ***, ** and * indicate the significance levels of 1%, 5% and 10%, respectively; Std. Err. in parentheses. (X₁: NDVI, X₂: MNDWI, X₃: Average temperature, X₄: Average precipitation, X₅: Soil organic carbon, X₆: Economic development, X₇: Urban expansion, X₈: Industrial structure, X₉: Innovation development, X₁₀: Population aggregation).

Decomposition of spatial effects

As there is a large amount of complex interaction information between adjacent regions, to analyze the spatial effects more intuitively, this study decomposed the spatial effects of the explanatory variables on ER into direct, indirect, and total effects (Table 4). The direct effects of all explanatory variables passed the significance test and the majority of the indirect effects of the explanatory variables passed the significance test. The direct, indirect, and total effects of NDVI and MNDWI were all positive, with the direct effect being significantly positive at the 5% level and the total effect being significantly positive at the 5% level. The higher the above two indices are, the less the original land surface is disturbed by humans³⁰. In particular, the integrity of ecological spaces such as woodland, grassland, and shrubland is crucial for ER. The direct and total effects of average temperature, average precipitation, and soil organic carbon were significantly positive, while the indirect effects were negative and insignificant. The above elements are important material bases for the material and energy cycles of ecosystems, in which areas with high soil organic carbon content are more conducive to ecosystem sustainability³¹.

Table 4 Decomposition of space effects. ***, ** and * indicate the significance levels of 1%, 5% and 10%, respectively; Std. Err. in parentheses.

The direct, indirect, and total effects of the urban expansion and population agglomeration were all negative, and the above factors not only directly changed the LULC conditions, but also largely caused irreversible damage to the ecosystem. The direct and total effects of economic development, innovation development, and industrial structure were significantly positive, while the indirect effects were negative and insignificant. Increased levels of economic development and innovation development mean that the government has greater financial capacity and technical means to promote the implementation of ecosystem protection and restoration programs, which can help restore ecosystem functions and enhance ER. Existing research shows that rationalization of industrial structure can better allocate production resources and reduce excessive interference with the natural environment^7,34. However, Unreasonable industrial structures may produce more serious industrial pollution and more industrial waste, which can have a serious negative impact on their ecosystems and may even have a diffuse effect. Finally, this study further replicated the above model using the geographic distance weight matrix and the results showed that there was no significant change in the level of significance of the explanatory variables, suggesting that the constructed model has good robustness.

Discussion

Spatial–temporal heterogeneity of ecological resilience in southwest China

The ER of Southwest China showed significant spatial and temporal heterogeneity, with the spatial distribution being “west is larger than east, south is larger than north, and reduction in the west was slower than that in the east,” which is consistent with the conclusion of previous studies that high-value resilience zones are generally located in the mid-altitude areas^35,36. From the perspective of regional differences, the high-value areas of ER were mainly concentrated in the border areas of Kunming and Yuxi, Lincang and Baoshan in Yunnan Province, and Nanning in Guangxi Province, with less anthropogenic disturbance. The low-value areas were mainly distributed in the alpine areas of northwestern Sichuan. From the perspective of climatic differences, most of the high-value ER was found to be located in mild areas and hot-summer and warm-winter regions, with sufficient water and heat conditions, high temperatures and rainfall throughout the year, suitable environmental conditions, and extremely rich biodiversity³⁷. In contrast, the low-value ER was concentrated in the hot-summer and cold-winter regions, where winter temperatures were low and unfavorable for plant growth, and most of them were grassland, with relatively homogeneous plant types and low biodiversity, the regulation and restoration ability of the ecosystem is relatively poor³⁸. Overall, the estimation results of this study are generally consistent with similar studies^39,40.

We predicted and summarized patterns of interannual change for eight SSP-RCP scenarios. The eight patterns have two obvious characteristics: first, each level of ER showed fluctuating non-cyclical changes, with an overall decrease in the highest-resilience and an increase in the lowest-resilience as a whole. The second is that under the same SSP scenarios, an increase in RCP emission levels leads to a significant increase in the area of the lowest-resilience and a decrease in the highest-resilience. For example, under the SSP4-RCP6.0 and SSP5-RCP8.5 scenarios, the lowest-resilience area was 1.83 and 2.95 times that of the SSP4-RCP3.4 and SSP5- RCP3.4 scenarios, respectively, and the highest-resilience area under the SSP4-RCP6.0 scenario was 0.15 times that of the SSP4-RCP3.4 scenario.

The fluctuating change pattern of ER in Southwest China was fully consistent with the non-cyclicality of the subtropical monsoon in which it was located, and the spatial distribution of the major natural factors such as precipitation and rainbands were all characterized by obvious non-cyclicality. Therefore, the ecosystem in the study area can effectively exclude the cyclical interference of climatic, which is of great significance for the protection and enhancement of ER in Southwest China. Regarding the impact of climate change on ER, our results emphasized that the lowest-resilience areas increased and the highest resilience decreased as the concentration of RCP emissions increased under the same SSP scenario. Also, the results of the Durbin model suggested that climate change was the main factor influencing ER in Southwest China. Related studies proved that the decrease in ER means more drastic climate change and higher environmental pressure, leading to an increase in ecosystem risk^7,41. In addition, by comparing the cumulative areas of ER levels under different scenarios, the results suggested that the SSP2-RCP4.5 pathway may be the best choice for the sustainable development of Southwestern China⁷. This is because this pathway has a certain operability in the current technical conditions and economic development level, the areas of lowest-resilience accumulated in the next 100 years at the smallest, which can help the ecosystems in Southwest China to adapt to future climate change.

Impact of factors on ecological resilience

The study results indicated that average precipitation and NDVI were the main positive natural drivers of ER in the study area. This is generally consistent with the results of related studies in other regions³⁰, but Southwest China has its special characteristics. Soil water accounts for only 1/100,000 of the total hydrosphere water and 0.05% of the total freshwater reserves, which is easily overlooked but can affect the evolution of lives throughout Southwest China⁴². Influenced by the subtropical monsoon climate, Southwest China is rich in precipitation and tends to be better able to maintain the water needed for ecosystem functioning^2,42. However, the frequent occurrence of engineered water scarcity is common⁴³. It largely restricts the growth of vegetation, with a persistent loss of its cooling effect, energy imbalance, and a series of climate extremes in recent years⁴⁴. This has dealt a blow to the already fragile ecosystems in the typical Karst region and threatened the survival and development of human beings, but this serious problem lacks sufficient attention.

Economic development, industrial structure, and innovative development were important positive factors to the ER of Southwest China. How to adhere to the two bottom lines of development and ecology has become the focus of today's research, but there is a lack of research in this area. The practice has proved that combining ecological restoration with industrial development, industrial restructuring, and livelihood improvement can effectively enhance the sustainability of ecosystem development. However, attention should be paid to innovative development within the carrying capacity of the ecological environment, rational allocation of agriculture-industry-service industry (tourism), perfecting primary industry planting and production, deepening secondary industry processing and production, and developing tertiary industry, culture and tourism integration, to ultimately establish an effective paradigm for the coordinated development of karst industries^45,46.

Responses and implications for future land-system adaptive management

Different LULC changes affect ecosystem landscape composition, processes, functions, and patterns⁴⁷. Therefore, land managers and policymakers need to develop effective management responses to offset the adverse effects of LULC changes on ER (Fig. 9). As the main LULC type with high-value ER, the area of woodland decreased by 10,063.716 km² from 2000 to 2020, while the average value of ER decreased to 0.927 times of the original one. The orderly promotion of urban–rural population relocation and the strengthening of ecological space restoration and management are of great significance for the sustainable development of the ecological environment in Karst areas^34,48,49. Shrubland, a concentrated LULC type with high-value ER second only to woodland in the study area, increased its average value in 2020 to 1.154 times that of 2000, but it is worth noting that shrubland was small in size in the study area and often overlooked in existing studies. Historically, shrubland has been poorly conceptualized and often categorized within the framework of forest management²⁸. Therefore, most studies simply categorize it as a subtype of woodland or grassland, ignoring its unique characteristics. The distribution and coverage of shrublands in the Southwest region decreased rapidly, from 145.862 to 117.753 km² during the study period, mainly due to frequent human activities that led to the replacement of secondary or invasive shrublands to break the structure of primary vegetation types⁵⁰. Related studies have shown that shrubland ecosystems play a crucial role in the cycling of terrestrial ecosystems^50,51. Therefore, the landscape ecological characteristics of the shrubland should be fully studied to explore its resource attributes and optimize its management methods.

Figure 9

Changes in the ecological resilience index of different land types in southwest China from 2001 to 2020 (with 2000 as the reference baseline).

The grassland was the largest LULC type in the study area. The area increased by 10,317.696 km² during the study period, but the average value of ER in 2020 decreased to 0.971 times that in 2000. Since the twenty-first century, ecological restoration projects such as returning farmland to grassland and comprehensive management of rocky desertification in typical Karst areas have greatly promoted the increase of grassland and made important contributions to alleviating and controlling rocky desertification. Studies have shown that the grassland area in the typical Karst area is increasing while the vegetation coverage is deteriorating, which seriously threatens the sustainable development of the ecosystem^52,53. For the restoration of grassland coverage in the Southwest region, first of all, the vegetation types and species that are compatible with the lithological background and climate change should be selected for ecological restoration to be carried out by the situation. There are great differences within different topographies in the typical Karst region⁵⁴, and it is recommended to further develop ecological programs based on environmental characteristics according to the peaks and depressions type, troughs and valleys, plateaus and canyons Karst landforms. Secondly, the ecological restoration of typical Karst areas should consider the water storage capacity of the corresponding weathering layer and thus select the appropriate vegetation. In addition, attention should be paid to the protection of existing natural woodland and cultivated land resources to provide better human well-being, rather than short-term green expansion⁵⁵.

The ER of cultivated land has relatively increased substantially, with the average value in 2020 being 1.110 times that of 2000. However, it should be noted that the fragmentation of land parcels in Karst mountainous areas and the sharp contradiction between people and land have led to the expansion of agriculture to slopes, resulting in the cultivated land being dominated by sloping cultivated land⁵². Taking Guizhou as an example, the results of the Third National Land Survey showed that it has 85.06% of sloping cultivated land and a land reclamation rate of 25.73%, much higher than Jiangxi (18.5%) and Fujian (10.8%), which were also pilot ecological civilization zones. In terms of cultivated land retention, Guizhou province was 23.81%, higher than neighboring provinces such as Sichuan (12.95%), Yunnan (14.83%), Guangxi (18.43%), Hunan (18.74%)^46,48. Therefore, it can be seen that the task of cultivated land retention and basic farmland protection in Southwest China is heavy, which is not in line with the actual situation. For the enhancement of the ER of cultivated land, firstly, under the premise of ensuring that China's basic farmland protection area is not reduced, the proportion of sloping cultivated land above 25° should be reduced in an orderly manner through coordinated transfer. Secondly, the structural adjustment of cultivated land should be organically combined with ecological migration and land remediation, and support should be increased to effectively consolidate the results of ecological restoration in China. In addition, formulate policies and regulations, strengthen publicity and education, carry out scientific planning, strengthen supervision and management, and improve the guarantee system, to ultimately realize the sustainable development of agriculture.

For the average ER of waters, there was a slight decrease between 2000 and 2020. The waters of Karst regions are extremely sensitive to external environmental disturbances, and because the frequency of surface water and subsurface interaction is much higher than that of non-Karst regions, pollutants are prone to migration and diffusion^37,56. In addition, the stratification structure and pollution pattern of Karst artificial deep-water lakes differ from that of natural shallow lakes, with poor self-purification ability of external pollutants and easy leakage⁴⁰. Therefore, pollution management is quite important in Karst waters ER restoration. The biggest difficulty in the management of water pollution in Karst is that the multi-scale transport and transformation mechanism of pollutants in the surface–subsurface dichotomy is not clear⁵⁶. In this regard, it is urgent to strengthen the research on the mechanism of surface-underground water compound pollution in Karst, and to establish a synergistic technology system for the prevention and control of surface-underground water pollution suitable for Karst. Further optimization of water quality monitoring and early warning systems in Karst basins is also crucial for Karst waters pollution control, and we should study and develop anti-seepage technologies for Karst artificial lakes as soon as possible, and regularly carry out quantitative prediction and assessment of seepage in Karst reservoirs and assessment of seepage in lakes, to ensure the safety of waters. The ER of bare land and permanent ice and snow decreased significantly, which was 0.622 and 0.793 times that of 2000 in 2020, respectively. However, fortunately, the areas of bare land decreased significantly, which decreased by 3406.485 km² from 2000 to 2020, mainly due to the conversion to construction land. The practice has proved that this has greatly alleviated the pressure of LULC in Southwest China and has practical reference significance. The ER of construction land was relatively stable, with a small increase of 1.023 times in 2020 compared to 2000, and the area increased by 2291.287 km² during the study period. It was found that vegetation growth in urban environments generally increased by 1.8 times, and the vegetation enhancement index in highly urbanized areas converged to 0.22^57,58. Therefore, for the improvement of ER construction land, governmental departments must formulate a more comprehensive green space network system.

Analysis of uncertainty and prospects for research

There are still some limitations in the research, including data error and modeling inaccuracy. First, the low spatial–temporal resolution of the data inevitably introduces estimation errors. Secondly, the ecosystem in the study area also shows significant seasonal and regional changes, making it difficult to comprehensively capture. In addition, the Global IGBP LULC prediction dataset used in this study relies on future LULC models for global simulation, which may not fully explain the specific geological environment, economic structure, land use practices, and lifestyles in Southwest China. Therefore, future research should consider combining the unique geological environment, hydrological environment, and economic structure practices of Southwest China to improve the number of LULC categories, spatial and temporal resolution, and model simulation accuracy. These questions are worthy of further exploration in future research.

Conclusions

In this study, firstly, we systematically analyzed the spatial and temporal changes of LULC in Southwest China over the past 20 years, and constructed an ER assessment model to reveal the characteristics and heterogeneity of the spatial distribution, interannual evolution, and development trend of ER in the past as well as under different SSP-RCP scenarios in the future. Meanwhile, through the in-depth examination of spatial autocorrelation effects, a spatial measurement model was used to validate the spatial response mechanism of ER in Southwest China. The study reveals that the study area experienced an increasingly rapid LULC transformation from 2000 to 2020, the ER showed non-cyclical fluctuations, characterized by a gradual increase in the area of the lowest-resilience and a decrease in the highest-resilience, and the ER reduction in the west was slower than that in the east. There were significant differences in ER changes in different SSP-RCP scenarios, under the same SSP scenario, the areas of lowest-resilience were significantly increased with increasing RCP emission concentration. The SSP2-RCP4.5 may be the most appropriate choice to promote the sustainable development of ER in Southwest China. Woodland and grassland had a prominent impact on the ER of the study area and deserved focused attention. The ER in Southwest China was mainly related to its special geological environment. Meanwhile, development and innovation play a positive role in strengthening the ER and should be continuously emphasized.