Evolution Modes, Types, and Social-Ecological Drivers of Ecologically Critical Areas in the Sichuan–Yunnan Ecological Barrier in the Last 15 Years

The ecological barrier is a complex ecosystem that couples the human–nature relationship, and the ecologically critical area is an irreplaceable area with a special value in the ecosystem. Therefore, protecting the ecologically critical area is vital for maintaining and improving regional ecological security. Limited research has been conducted on the evolution of ecologically critical areas, and none of the studies have considered the spatiotemporal heterogeneity of the driving factors for different evolution modes and types. Therefore, this research adopts the ecologically critical index, landscape expansion index, and the random forest model to analyze the pattern, driving factors, and its spatial-temporal heterogeneity to the evolution modes and specific types of ecologically critical areas in the Sichuan–Yunnan ecological barrier area in the last 15 years. The results showed that: (1) the ecologically critical areas in the Sichuan–Yunnan ecological barrier have changed dramatically, with the area reduction being 61.06%. Additionally, the spatial distribution characteristics of the ecologically critical area from north to south include planar, point, and linear forms. (2) The evolution trend of the ecologically critical area is ‘degradation–expansion–degradation’. Spread is the predominant type of expansion mode, whereas atrophy is the predominant type of degradation mode, indicating that the evolution mainly occurs at the edge of the original ecologically critical areas. (3) In general, precipitation, area of forest, area of cropland, and GDP have contributed significantly to the evolution of ecologically critical areas. However, the same driving factor has different effects on the expansion and degradation of these areas. Expansion is driven by multiple factors at the same time but is mainly related to human activities and land use change, whereas for degradation, climate and policy are the main driving factors. The present research aimed to quantitatively identify the evolution modes and specific types of ecologically critical areas and explore the spatiotemporal heterogeneity of driving factors. The results can help decision-makers in formulating ecological protection policies according to local conditions and in maintaining and enhancing the regional ecological functions, thereby promoting the sustainable development of society-economy-ecology.


Introduction
Global climate change has been posing great risks to terrestrial ecosystems [1]. Additionally, the rapid socioeconomic development has caused a series of ecological problems. To safeguard national ecological security, China has built the 'two barriers and three belts' ecological security strategic pattern [2,3]. As a complex ecosystem that couples the

Study Area
The SYEB is part of the Sichuan-Yunnan-loess plateau ecological barrier and is a key component of China's ecological security strategic patterns. The SYEB is located in the Hengduan Mountains and the transition zone from the Yunnan-Guizhou Plateau to the Qinghai-Tibet Plateau, wherein the topography is decreasing from northwest to southeast and the ecological environment is relatively fragile (Figure 1). This region is dominated by plateau mountain temperate and subtropical monsoon climate, with annual precipitation of 500-1400 mm, and the average annual temperature in the region can vary by more than 20 • C [30]. The SYEB has a special ecological status and plays an important role in soil conservation, water conservation, and biodiversity maintenance. cropland evolution and its driving factors through the random forest model from the environmental, socioeconomic, and policy perspectives.

Study Area
The SYEB is part of the Sichuan-Yunnan-loess plateau ecological barrier and is a key component of China's ecological security strategic patterns. The SYEB is located in the Hengduan Mountains and the transition zone from the Yunnan-Guizhou Plateau to the Qinghai-Tibet Plateau, wherein the topography is decreasing from northwest to southeast and the ecological environment is relatively fragile (Figure 1). This region is dominated by plateau mountain temperate and subtropical monsoon climate, with annual precipitation of 500-1400 mm, and the average annual temperature in the region can vary by more than 20 °C [30]. The SYEB has a special ecological status and plays an important role in soil conservation, water conservation, and biodiversity maintenance.  Table 1 shows the names, formats, and sources of the data used in this research. The boundaries were extracted from the China National Ecological Barrier Area dataset, which is produced on the basis of a combination of Google Earth satellite images and vector boundary maps of Chinese counties, etc. According to the land classification system and our research problem, we grouped the data into seven categories, namely cropland, forest, grassland, water, wet land, impervious, and other land. The slope data were generated based on the DEM data. We integrated all data into a 1 km × 1 km grid for a unified analysis.   Table 1 shows the names, formats, and sources of the data used in this research. The boundaries were extracted from the China National Ecological Barrier Area dataset, which is produced on the basis of a combination of Google Earth satellite images and vector boundary maps of Chinese counties, etc. According to the land classification system and our research problem, we grouped the data into seven categories, namely cropland, forest, grassland, water, wet land, impervious, and other land. The slope data were generated based on the DEM data. We integrated all data into a 1 km × 1 km grid for a unified analysis. In this research, we evaluated the ecological significance of each grid and identified the ECA, classified the evolution modes and specific types, and explored its driving factors at the county scale. The overall research framework is shown in Figure 2  acteristics of ECI; (3) determining the expansion mode and degradation mode of ECA evolution during 2005-2010, 2010-2015, and 2015-2019 and classifying the specific types of expansion mode and degradation mode by landscape expansion index (LEI); (4) selecting the explanatory variables according to the research topic and previous research and then putting the dependent variable and the explanatory variables into the random forest (RF) model and running the model; and (5) analyzing the spatiotemporal heterogeneity of the driving factors for evolution modes and specific types.

Ecological Critical Index
By coupling the ecosystem service function and landscape ecological security, the ECI was constructed to represent the degree of ecological importance at the county scale. ECI is the basis for identifying ECA and is calculated using the following formula: where ECI I is the ecological critical index of the ith grid; MESLI I is the multiple ecosystem services landscape index of the ith grid; and LESSI I is landscape ecological structure security index of the ith grid. MESLI can be used to effectively identify the ability of a region to provide multiple ecosystem services at the same time [31], which is a comprehensive and valuable environmental indicator for identifying cold and hot spots for multiple ecosystem services [32]. It is calculated using the following formula: where i is the type of ecosystem service function; n is the number of ecosystem service function; and x i , max(x i ), and min(x i ) are the observed value, maximum value, and minimum value of the ith ecosystem service function, respectively. According to the functional orientation of the SYEB mentioned in the "Master plan for major projects for the protection and restoration of important ecosystems nationwide", four types of ecosystem services are selected, namely carbon sequestration, water production, soil conservation, and habitat quality. Referring to previous studies, NPP was used to represent the carbon sequestration function, and the water production services, soil conservation, and habitat quality were calculated using the INVEST model [33,34].
LESSI evaluates the security of the landscape structure from three perspectives, namely landscape vulnerability, boundary fragmentation, and landscape type fragmentation [35], and it has been widely used in the landscape ecological security assessment [36,37]. LESSI is calculated using the following formula: 6 of 16 where PD is the patch density in the landscape class, ED is the edge density in the landscape class, and V is the landscape fragmentation. AWMSI is the area-weighted mean shape index; F is fractal dimension; D is division; and α, β, and γ are the weights of AWMSI, F, and D, respectively, which are assigned as 0.5, 0.3, and 0.2, respectively.

Identifying the Evolution Modes and Specific Types of ECA
Traditional landscape indices can only reflect the static spatial distribution of the landscape [38], whereas LEI can quantitatively describe the dynamic change process [39]. The formula for calculating LEI is as follows: where A 0 is the area of the original ECA grid in the buffer zone; A E is the area of buffer zone; and A P is the area of new ECA; the range of LEI values is between 0 and 100. Considering the area of the studied region, we set the buffer radius as 1 km. All ECA grids whose buffers overlap belong to the same patch and have the same LEI.
Through the spatial analysis, we divided the ECA evolution into two modes: expansion and degradation. Grids that were previously identified as non-ECA and later identified as ECA are called expansion. By contrast, grids that were previously identified as ECA and later identified as non-ECA are called degradation ( Figure 3).  According to the definition of LEI [39], we have defined six types of ECA evolution. The expansion mode has three specific types: (1) if the buffer of a new ECA patch consists of non-ECA only (LEI = 0), it is classified as the isolation type, implying that the new ECA patch is spatially discontinuous with the original ECA patch; (2) if the buffer of a new ECA patch consists of non-ECA mixed with original ECA (0 < LEI ≤ 50), it is classified as the spread type, meaning that the new ECA patch is located at the edge of the original ECA patch; and (3) if the buffer of a new ECA patch is mostly occupied by the original ECA (50 < LEI ≤ 100), it is classified as the infilling type, indicating that the new ECA patch is located within the original ECA patch. The degradation mode also has of three specific types, namely departed, atrophy, and disintegration. Their classification rules correspond to the specific type of expansion mode.

Random Forest Model
The RF model is a data mining method for categorical regression trees proposed by Breiman [40]. RF is an effective classifier that is composed of a set of tree-structured classifiers {h (x, θk), k = 1, ……}, where {θk} represents explanatory, identically-distributed random vectors. With the input of the explanatory variable X, each DT casts a unit vote According to the definition of LEI [39], we have defined six types of ECA evolution. The expansion mode has three specific types: (1) if the buffer of a new ECA patch consists of non-ECA only (LEI = 0), it is classified as the isolation type, implying that the new ECA patch is spatially discontinuous with the original ECA patch; (2) if the buffer of a new ECA patch consists of non-ECA mixed with original ECA (0 < LEI ≤ 50), it is classified as the spread type, meaning that the new ECA patch is located at the edge of the original ECA patch; and (3) if the buffer of a new ECA patch is mostly occupied by the original ECA (50 < LEI ≤ 100), it is classified as the infilling type, indicating that the new ECA patch is located within the original ECA patch. The degradation mode also has of three specific types, namely departed, atrophy, and disintegration. Their classification rules correspond to the specific type of expansion mode.

Random Forest Model
The RF model is a data mining method for categorical regression trees proposed by Breiman [40]. RF is an effective classifier that is composed of a set of tree-structured classifiers {h (x, θk), k = 1, . . . . . . }, where {θk} represents explanatory, identically-distributed random vectors. With the input of the explanatory variable X, each DT casts a unit vote for the most popular class [41].
In this research, we chose the minimum Gini value method as the segmentation criterion. The minimum Gini value of an internal tree node was calculated using the following formula: where p(q|t) represents the probability of the risk class q at node t; u represents the number of classes.
In this study, the decrease in the Gini index at the node split was used to calculate the importance of each index to the result of risk classification. The formula used for calculation is as follows: where m represents the total number of indices, k is the number of texturing trees, t is the number of nodes in each tree, D Grij is the Gini decrease value at the jth node in the ith tree that belongs to the rth index, and Pr is the degree of contribution of the rth index from all available indices.

Determination of Explanatory Variables
According to the previous research and the theoretical analysis, the potential driving factors that could influence the ECA evolution were grouped into two categories: environment condition and socioeconomic development ( Table 2). DEM (x1), slope (x2), temperature (x3), precipitation (x4), area of forest (x5), and area of grassland (x6) were selected as the six explanatory variables to represent environmental conditions from the perspectives of terrain, meteorology, and ecosystem composition. The DEM and slope determine whether the conditions of a region are suitable for ecosystem succession to higher levels. Precipitation and temperature determine ecosystem succession by affecting vegetation growth. Land use reflects the regional ecosystem composition, and we selected the major land use components in the ECA including forest and grassland.
Socioeconomic factors indicate the impact of human activities, with the most important factors being the area of cropland (x7), GDP (x8), population (x9), and nightlight (x10). These four explanatory variables are direct representations of human activity. We chose the distance from impervious (x11) and cropland (x12) to represent ECA location, reflecting the ease with which ECA is affected by these regions. This research could be improved by considering these factors, thus providing a better understanding of ECA evolution.

Spatial-Temporal Pattern of ECA
By counting and plotting the frequency distribution of ECI during 2005, 2010, 2015, and 2019, we observed that the frequency distribution of ECI in each period is similar. By comparing the extraction results of different thresholds, including peak, trough, and inflection point, we finally chose the last peak of the trend line as the extraction threshold (T) of ECA ( Figure 4). Therefore, the grids with ECI > T were identified as ECA, whereas those with ECI < T were identified as non-ECA.   According to the aforementioned identification process, we determined the spatial distribution of the ECA of the SYEBZ in 2005, 2010, 2015, and 2019. The results showed a significantly heterogeneous distribution of ECA ( Figure 5). The ECAs in the northern part of the study area were distributed in a planar form and located mainly in Mianyang, Deyang, Chengdu, Ya'an, Meishan, and Leshan. In the central part, the ECAs were distributed in a pointed form, and located in Ganzi, Liangshan, and Panzhihua. The ECAs in the southern part were distributed in a linear form and located mainly in Lijiang, Nujiang, and Dali.

Evolution Modes of ECA
The ECA evolution trend was "degradation-expansion-degradation". From 2005 to 2010, the number of expanded grids was 1537 and the number of degradation grids was 17,305; the direction of ECA evolution mainly showed a degradation trend. The expansion grids were mainly located at the junction of Chengdu, Leshan, and Ya'an, whereas the degradation grids were concentrated in Mianyang, Deyang, Chengdu, and Meishan in the north. From 2010 to 2015, the number of expansion grids was 11,248, the number of degradation grids was 3379, and the direction of ECA evolution mainly showed an expansion trend. The expansion grids were mainly concentrated in Mianyang, Aba, Chengdu, Ya'an, and Meishan in the northern part of the region. The distribution of the degradation grids was more discrete, located mainly in the south of the region. From 2015 to 2019, the number of expansion grids was 2271, the number of degradation grids was 24,335, and the ECA evolution showed mainly the degradation trend. The expansion grids showed mainly a discrete distribution in the central part of the region, and the degradation grids were mainly concentrated in Mianyang, Aba, Ya'an, Meishan, and Leshan in the northern region (Table 3, Figure 6).  The ECA evolution trend was "degradation-expansion-degradation". From 2005 to 2010, the number of expanded grids was 1537 and the number of degradation grids was 17,305; the direction of ECA evolution mainly showed a degradation trend. The expansion grids were mainly located at the junction of Chengdu, Leshan, and Ya'an, whereas the degradation grids were concentrated in Mianyang, Deyang, Chengdu, and Meishan in the north. From 2010 to 2015, the number of expansion grids was 11,248, the number of degradation grids was 3379, and the direction of ECA evolution mainly showed an expansion trend. The expansion grids were mainly concentrated in Mianyang, Aba, Chengdu, Ya'an, and Meishan in the northern part of the region. The distribution of the degradation grids was more discrete, located mainly in the south of the region. From 2015 to 2019, the number of expansion grids was 2271, the number of degradation grids was 24,335, and the ECA evolution showed mainly the degradation trend. The expansion grids showed mainly a discrete distribution in the central part of the region, and the degradation grids were mainly concentrated in Mianyang, Aba, Ya'an, Meishan, and Leshan in the northern region (Table 3, Figure 6).  Figure 6. Spatial-temporal distribution of evolution modes.

Specific Types of Evolution Modes
Among the three specific types of the expansion mode, the predominant type is spread (Figure 7), indicating that expansion mainly starts at the edge of original ECA. In the last 15 years, as for the isolation type, the proportion of grid count increased from 5.70% to 10.45%, and the proportion of patch area increased from 10.30% to 16.75%, indicating that the spatial distribution of the isolation type tends to be concentrated. For the spread type, the proportion of grid count increased from 70.14% to 79.58%, and the proportion of patch area increased from 54.20% to 65.23%, indicating that the spatial distribution of the spread type shifted from concentrated to dispersed. Finally, for the infilling type, the proportion of grid count decreased from 24.16% to 9.97%, and the proportion of patch area decreased from 35.50% to 18.02%, showing that the dispersed characteristic is more obvious for this type. Atrophy is the predominant type among the three specific types of the degradation mode (Figure 7), indicating that degradation also mainly starts at the edge of the original ECA. In the last 15 years, for the departed type, the proportion of grid count increased from 4.92% to 5.77%, and the proportion of patch area increased from 22.54% to 29.74%, indicating that the spatial distribution of departed type tends to be more dispersed. For the atrophy type, the proportion of grid count increased from 86.09% to 92.47%, and the proportion of patch area decreased from 57.21% to 52.05%, indicating that the spatial

Specific Types of Evolution Modes
Among the three specific types of the expansion mode, the predominant type is spread (Figure 7), indicating that expansion mainly starts at the edge of original ECA. In the last 15 years, as for the isolation type, the proportion of grid count increased from 5.70% to 10.45%, and the proportion of patch area increased from 10.30% to 16.75%, indicating that the spatial distribution of the isolation type tends to be concentrated. For the spread type, the proportion of grid count increased from 70.14% to 79.58%, and the proportion of patch area increased from 54.20% to 65.23%, indicating that the spatial distribution of the spread type shifted from concentrated to dispersed. Finally, for the infilling type, the proportion of grid count decreased from 24.16% to 9.97%, and the proportion of patch area decreased from 35.50% to 18.02%, showing that the dispersed characteristic is more obvious for this type.

Specific Types of Evolution Modes
Among the three specific types of the expansion mode, the predominant type is spread (Figure 7), indicating that expansion mainly starts at the edge of original ECA. In the last 15 years, as for the isolation type, the proportion of grid count increased from 5.70% to 10.45%, and the proportion of patch area increased from 10.30% to 16.75%, indicating that the spatial distribution of the isolation type tends to be concentrated. For the spread type, the proportion of grid count increased from 70.14% to 79.58%, and the proportion of patch area increased from 54.20% to 65.23%, indicating that the spatial distribution of the spread type shifted from concentrated to dispersed. Finally, for the infilling type, the proportion of grid count decreased from 24.16% to 9.97%, and the proportion of patch area decreased from 35.50% to 18.02%, showing that the dispersed characteristic is more obvious for this type. Atrophy is the predominant type among the three specific types of the degradation mode (Figure 7), indicating that degradation also mainly starts at the edge of the original ECA. In the last 15 years, for the departed type, the proportion of grid count increased from 4.92% to 5.77%, and the proportion of patch area increased from 22.54% to 29.74%, indicating that the spatial distribution of departed type tends to be more dispersed. For the atrophy type, the proportion of grid count increased from 86.09% to 92.47%, and the proportion of patch area decreased from 57.21% to 52.05%, indicating that the spatial Atrophy is the predominant type among the three specific types of the degradation mode (Figure 7), indicating that degradation also mainly starts at the edge of the original ECA. In the last 15 years, for the departed type, the proportion of grid count increased from 4.92% to 5.77%, and the proportion of patch area increased from 22.54% to 29.74%, indicating that the spatial distribution of departed type tends to be more dispersed. For the atrophy type, the proportion of grid count increased from 86.09% to 92.47%, and the proportion of patch area decreased from 57.21% to 52.05%, indicating that the spatial distribution of the atrophy type tends to be more concentrated. Finally, for the disintegration type, the proportion of grid count decreased from 9.00% to 1.76%, and the proportion of patch area decreased from 20.25% to 18.21%, indicating that the distribution of this type tends to be more dispersed.

Verification of RF Model Accuracy
RF models were run with two modes of ECA evolution, three specific types of expansion modes, and three specific types of degradation modes as predictor variables, respectively. The percentage of correct classification results was used to evaluate the accuracy of the RF model. According to Table 4, the highest prediction accuracy was 97.96%, whereas the lowest prediction accuracy was 68.86%, and the average accuracy reached 83.98%. These data suggest that the results of the RF model are credible.

Heterogeneity of Driving Factors to the Evolution Modes and Specific Types of ECA
The importance of explanatory variables to ECA evolution was classified from low to high as follows: I (0-0.5037%), II (0.5038-1.  distribution of the atrophy type tends to be more concentrated. Finally, for the disintegration type, the proportion of grid count decreased from 9.00% to 1.76%, and the proportion of patch area decreased from 20.25% to 18.21%, indicating that the distribution of this type tends to be more dispersed.

Verification of RF Model Accuracy
RF models were run with two modes of ECA evolution, three specific types of expansion modes, and three specific types of degradation modes as predictor variables, respectively. The percentage of correct classification results was used to evaluate the accuracy of the RF model. According to Table 4, the highest prediction accuracy was 97.96%, whereas the lowest prediction accuracy was 68.86%, and the average accuracy reached 83.98%. These data suggest that the results of the RF model are credible.

Heterogeneity of Driving Factors to the Evolution Modes and Specific Types of ECA
The importance of explanatory variables to ECA evolution was classified from low to high as follows: I (0-0.5037%), II (0.5038-1.   The Chinese government has proposed a 'new urbanization' policy to alleviate the ecological pressure caused by rapid urbanization; therefore, the impact of urban expansion first increased and then decreased. In addition, due to the implementation of the Cropland Balance Policy, cropland will gradually move out with urban expansion. To explore the differences in the effects of each variable between ECA expansion and degradation, we used three expansion types or three degradation types as predictor variables. As shown in Figure 8, the length of the rectangle reflects relative importance. The solid rectangles and dashed rectangles represent the relative importance of the explanatory variables for three specific types of expansion mode and three specific types of degradation mode, respectively. During 2005-2010, the explanatory variables with greater relative importance to the type of expansion were nightlight (x10) and distance from impervious (x11), whereas the explanatory variable with greater relative importance to the type of degradation was precipitation (x4). During 2010-2015, the explanatory variables with greater relative importance to the type of expansion were area of forest (x5) and area of cropland (x7), whereas the explanatory variables with greater relative importance to the type of degradation were area of grassland (x6) and distance from impervious (x12). During 2015-2019, the explanatory variables with greater relative importance to the type of expansion were area of forest (x5), area of cropland (x7), nightlight (x10), and distance from impervious (x12), whereas the explanatory variables with greater relative importance to the type of degradation were temperature (x3) and precipitation (x4). Overall, forest ecosystems have abundant species and a complex network structure with good stability. Thus, large areas of forest are conducive to ECA expansion. As an artificial ecosystem, the agroecosystem is closely and intricately linked to the socioeconomic sphere of human beings. Under the Cropland Balance Policy, economic development and urban expansion have contributed to the outward expansion of cropland, leading to a negative impact on ECA expansion. However, for ECA degradation, climate factors including temperature and precipitation are more significant. Climate change is aggravating the imbalance of water and heat distribution and thus promoting ECA degradation.

Characteristics of the ECA Evolution Modes and Types
The spatial distribution characteristics of ECA from north to south include planar, point, and linear forms. In the northern part of the study area, ECA is mainly distributed in the marginal mountain areas of the Sichuan Basin. The warm and humid climate has contributed to the increased diversity of animal and plant resources in the area; therefore, the spatial pattern of ECA is planar. The central part of the study area is the middle section of the Hengduan Mountains, where the soil is barren, vegetation is sparse, and the ecological environment is fragile. It is also difficult to form a concentrated and contiguous habitat patch there, so the spatial pattern of ECA is a point pattern. In the southern part of the study area, the Jinsha, Lancang, and Nu rivers flow in parallel, wherein the biological communities are enriched, and the ECA is distributed linearly along the sides of the valley.
ECA has undergone dramatic changes in the last 15 years, and its evolutionary trend is degradation-expansion-degradation. The socioeconomic development owing to rapid urbanization inevitably causes different degrees of damage to the ecosystem [42], thereby causing ECA degradation. To alleviate the ecological pressures of rapid development, China has implemented various extensive ecological projects to protect and expand its forests [43]. Central and Southwest China is one of the key regions for afforestation, and the forest cover has increased significantly in a short time period, which is the main reason for the ECA expansion in the study area in 2015. However, due to the low stability of single species afforestation, forest plantation is more dependent on human management [44]. Therefore, planted forest ecosystems are more susceptible to degradation than natural ecosystems.
The competition between the expansion and degradation processes is most intense at the edge of the ECA, which often represents the junction or overlapping area between different ecosystems. This is the transition zone of the community where species penetrate each other and is the area with obvious environmental gradient changes [45]. Therefore, ECA has a profound edge effect. The edge effect interpenetrates climate, vegetation, and landscape in the area and undergoes gradient mutations, leading to increased environmental heterogeneity [46] and making ECA edges vulnerable to degradation. However, due to edge effects, ecological engineering can enhance the overall ecological benefits. Therefore, the government should implement different policies to promote expansion and curb degradation according to the spatial distribution characteristics and evolution pattern of ECA, which can help in enhancing regional ecological functions and maintaining regional ecological security.

Understanding the Spatiotemporal Heterogeneity of Driving Factors
Overall, the explanatory variables identified as having great importance in ECA evolution are precipitation (x4), area of forest (x5), area of cropland (x7), and GDP (x8). Precipitation is an important factor in ecosystem succession, and climate change has been causing significant damage and increasingly irreversible losses to the ecosystem [47]. To relieve ecological pressure, China has implemented various ecological projects such as afforestation and reforestation projects since the 20th century [48]. In the "Outline of National Ecological Fragile Area Protection Plan", the SYEB mainly belongs to the southwest mountainous ecologically fragile area with interlocking agriculture and animal husbandry, where the future development direction needs to strictly return farmland to forests and close hillsides for afforestation. Especially in the "National Main Functional Area Planning", it is further clarified that the SYEB needs to focus on strengthening the functions of soil erosion control and natural vegetation protection, which have contributed to the greening of China [49,50]. GDP represents the economic output of a region and is the primary representation of the intensity of human activity [51]. Therefore, these explanatory variables have strong impacts on ECA evolution.
The explanatory variables identified as having the least effect on ECA evolution are population (x9) and nightlight (x10). This is because the ECA is located far from the city center and thus is sparsely populated and dimly lit. Additionally, the new urbanization policy [52] promoted by the Chinese government requires economic intensification and ecological livability; therefore, the economic output of the region no longer depends only on the size of the population, and economic development is gradually becoming green and sustainable.

Policy Implication
Scientific ecological policies can promote the sustainable development of the region. Therefore, we propose site-specific policies to protect ECA based on the spatiotemporal pattern of ECA, spatiotemporal heterogeneity of driving factors to ECA evolution, and differences in the drivers of expansion and degradation.
The edges of ECA represent the intersection between natural and artificial ecosystems and are the areas where anthropogenic disturbances compete most fiercely with natural ecological restoration; hence, expansion and degradation occur mainly at the edges of ECAs. These areas require the construction of ecological isolation zones, delineation of ecological red lines, and other measures to control the intensity of interference due to human activities. The afforestation and reforestation projects have contributed significantly to increasing the forest area in the study area, which had a positive effect on ECA expansion. However, the importance of human management in this area cannot be ignored [50]. Therefore, the continuous management and monitoring of ecological projects are necessary to maintain the ecological functions of planted forests.
Specifically, the northern part of the SYEBZ is the Sichuan Basin, which has the highest population density and the strongest economic activity. The ECA in this region has a planar distribution and is heavily degraded. We suggest that in the future, urban development boundaries should be strictly controlled, and high-standard cropland construction should be implemented to alleviate the ecological pressure of urban and cropland expansions. Coordinating the contradiction between economic development and ecological protection is essential to curb ECA degradation. The central part of the SYEBZ is the middle section of the Hengduan Mountains, which is an ecologically fragile area with fewer human activities and a high degree of preservation of the original ecological environment. The ECA in this region has a point distribution. To promote ECA expansion in the region, a system of nature reserves should be built to improve ecological containment functions and the local climate. The southern part of SYEBZ, wherein three rivers flow in parallel, has a fragile ecological environment and poor socioeconomic development. The ECA in this region has a linear distribution. In this region, the concept of ecological priority and green development should be strictly implemented in the future. With ecological connotation as the premise, the advantages of water and heat conditions in the high mountain valley should be utilized, and ecological utilization-based agriculture should be vigorously developed.

Conclusions
This study constructs ECI to identify ECAs in the SYEBZ during 2005, 2010, 2015, and 2019. Then, the LEI was used to delineate the specific types of ECA expansion and degradation modes during three periods, namely 2005-2010, 2010-2015, and 2015-2019. Finally, the RF model was used to explore the spatiotemporal heterogeneity of drivers of ECA evolution modes and specific types.
The overall trend of ECA evolution is degradation-expansion-degradation. Among the three expansion types, namely isolation, spread, and infilling, the predominant type is spread. Similarly, among the three degeneration types, namely departed, atrophy, and disintegration, the predominant type is atrophy. This indicates that the evolution of ECA mainly occurred at the edge of the original ECA. The drivers having a great importance in ECA evolution are mainly precipitation (x4), area of forest (x5), area of cropland (x7), and GDP (x8), indicating that ECA expansion is driven by various factors but is primarily associated with land use change and anthropogenic factors. However, for ECA degradation, climate factors including temperature and precipitation are more significant. These conclusions are vital to the protection and management of the national ecological barrier zone and the promotion of coordinated regional socio-economic-ecological development.