Introduction

Human demand for natural resources is rising since population growth and economic development, which leads to major issues such as resource scarcity, ecological degradation, and biodiversity loss1,2,3. Since 1990, due to the combined effects of climate change and human activities, the global net loss of forest area has reached 178 million hectares4. In recent years, a series of policy initiatives aimed at reducing deforestation and enhancing tree cover restoration have been widely promoted to reverse ecological degradation and increase vegetation’s ability to store carbon5,6. Tree cover restoration involves intricate processes between society and the ecosystem7,8, and a thorough analysis can give the scientific foundation for forest management.

Currently, researches on tree cover restoration primarily focused on assessing spatial patterns of forest change9,10 and analyzing the driving factors11,12. The terrestrial ecosystem model demonstrated that in addition to the enhancement effect of environmental factors on vegetation growth, human activities such as urbanization were leading to the earth’s greening13. Recently, many studies have explored the factors affecting tree cover restoration. The environmental factors were recommended as the primary driving factors of vegetation growth12,13,14. For example, water and soil properties have an impact on vegetation15,16. Soil nutrient content was a limiting factor for plant physiology, and has been proven to be related to the photosynthetic rate17. Additionally, it has been suggested that socio-economic factors also have varying effects on vegetation greening. Some studies proved that socio-economic growth would have a negative effect on vegetation greening, while recent studies also recommended that socio-economic development would promote vegetation greening due to the construction of green infrastructure and reduction of negative human interference18,19,20. The social-ecological system promotes a comprehensive assessment of social systems and ecosystems at a holistic level21,22, which is potential for identifying the factors of tree cover restoration23,24. To date, there is no knowledge on the drivers of tree cover restoration based on the perspective of coupled social-ecological system, and the socio-economic effect on tree cover restoration remains uncertain.

As the world’s largest continuous karst zone, South China Karst covers ~1.9 million km2 and hosts over 220 million people, plagued by increased exploitation of natural resources and land degradation25 (Supplementary Fig. 1). Characterized by vegetation degradation and exposed rock, this region is regarded as the main ecologically fragile area in China2,26. Affected by the slow speed of soil formation and thin soil layer in karst landforms, the vegetation in this region is sparse. Under these circumstances, the rapid vegetation growth in South China Karst over the past two decades has attracted the attention of the world. The mechanism of tree cover restoration in this region was complicated, involving ecological engineering, urbanization as well as environmental factors13,15,16. There are karst and non-karst counties in South China Karst, where the lithology and economic development are different, which may lead to diverse impacts on tree cover restoration in karst and non-karst area, yet have not been fully explored. Understanding the social-ecological driving forces in this region, especially in terms of the contrast between karst and non-karst areas, is necessary for tree cover restoration, and provides a way for the coordinated eco-socio-economic development of ecologically fragile areas around the world27,28.

Based on the satellite-based socio-economic and ecological datasets, we explored the forest land change in South China Karst and analyzed the social-ecological factors’ impacts on tree cover restoration in karst and non-karst area. Here, we investigated and reported on (1) tree cover restoration in South China Karst from 2000 to 2020; (2) the difference in social-ecological factors between karst area (KA) and non-karst area (NKA); and (3) the contrast of social-ecological factors’ impact on tree cover restoration between KA and NKA. Here we demonstrated that the slope of tree cover proportion change in KA (0.50) was higher than that of NKA (0.36) from 2000 to 2020. Rural population decrease promoted tree cover restoration in undeveloped counties. The correlation degree of the social-ecological network in KA was lower (47/78) than that in NKA (62/78), showing the social-ecological systems in karst areas were more heterogeneous.

Results

Tree cover restoration contrast in karst and non-karst area

There were 880,951 km2 of forest land with no land use change from 2000 to 2020, accounting for 45.46 % of the study area (Fig. 1a). Among them, the area was 627,887 km2 in NKA, which was 2.48 times of that in KA (253,064 km2) (Fig. 1b). Moreover, it was found that the area of increased forest land in South China Karst was 77,635 km2, with the area of 53,797 km2 in NKA and 23,838 km2 in KA. The forest land increasing proportion was high in counties in the northeastern and southwestern regions, while it was relatively low in the northwestern region (Supplementary Fig. 2a). The median of forest land increasing proportion in undeveloped counties (3.96%) was higher than that in developed counties (3.28%) (Supplementary Fig. 2b). A total of 47.86% of forest land increased area was converted from grassland and 40.15% was converted from cropland (Supplementary Fig. 3). The area of cropland and grassland converted to forest land in KA were 9817 km2 and 11,258 km2, while that in NKA were 21,352 km2 and 25,897 km2, respectively. It was worth noting that tree cover proportion increased in KA and NKA in the past two decades (Fig. 1c). The slope of tree cover proportion change in KA (0.50) was higher than that of NKA (0.36) from 2000 to 2020 (Fig. 1d). Tree cover proportion in NKA and KA was 27.54% and 27.07% in 2000, and 33.94% and 35.74% in 2020, respectively. The area with tree cover increasing was 652,457 km2, and counties in the southwestern, southeastern, and northern regions showed a trend of tree cover increasing (Supplementary Fig. 2c). The median of tree cover trend in undeveloped counties was higher than that in developed counties (Supplementary Fig. 2d).

Fig. 1: Spatial pattern and statistics of forest land change.
figure 1

a Forest land changes; b area of forest land changes; c tree cover trend from 2000 to 2020; and d tree cover trend in karst area and non-karst area.

Full size image

Difference of ecological factors between karst and non-karst areas

For soil status, spatial pattern of soil organic matter showed a characteristic of being high in the northwestern region and low in other regions, with the average values of 3.92% and 4.04% in karst and non-karst areas, respectively (Supplementary Fig. 4). The soil bulk density showed a feature of being high in the northern and eastern regions, and low in the central and northwestern regions, with the average values of 1.23 g/cm3 and 1.24 g/cm3 in karst and non-karst areas, respectively. The average values for slope and elevation were 9.50° and 1247 m in karst area, with 8.74° and 1319 m in non-karst area, respectively. Spatial distributions of annual average precipitation and temperature were similar, showing a spatial pattern of high values in the southwestern and southeastern regions, and low values in the northwestern region, with the average value of 1187 mm and 15.19 °C in karst area, and 1185 mm and 14.76 °C in non-karst area.

The result of the Kruskal-Wallis test showed that there were differences in ecological factors of different forest land change areas between KA and NKA (Fig. 2). For soil factors, we found that the median of soil organic matter in the area of cropland to forest land in KA were 2.99%, which were lower than that of grassland to forest land in KA (3.28%). The value in NKA was 2.84%, which were also lower than that of grassland to forest land in NKA (3.05%). The median of soil bulk density of different forest land change areas was the same between KA and NKA, with 1.22 g/cm3 in KA and 1.23 g/cm3 in NKA, respectively. The soil organic matter of the forest land change area in NKA were slightly lower than that of KA. For topographic factors, the median of elevation and slope in the area of cropland to forest land in KA (789 m and 5.98°) and NKA (485 m and 5.22°) were lower than that in the area of grassland to forest land in KA (980 m and 7.95°) and NKA (857 m and 8.02°). The elevation and slope of the forest land change area in KA were higher than those in NKA. Besides, the median of annual average precipitation and temperature in the area of cropland to forest land in KA (1310 mm and 20.49 °C) and NKA (1310 m and 21.80 °C) were higher than that in the area of grassland to forest land in KA (1171 mm and 19.62 °C) and NKA (1146 mm and 20.80 °C).

Fig. 2: Comparisons among ecological factors in the forest land change area between karst area and non-karst area.
figure 2

p < 0.01 means there were differences between groups.

Full size image

To further explore the mechanism of the ecological factors’ impact on forest land in different karst area, we extracted limestone and dolomite area. In general, the area of limestone area (100,300 km2) was larger than that of dolomite area (36,900 km2). The area of forest land increased by 4771 km2 and 1914 km2 in limestone and dolomite areas in the past two decades, respectively (Supplementary Fig. 5). There were obvious differences in soil bulk density between the two forest land change ways in limestone and dolomite areas (Supplementary Fig. 6). In addition, we compared the differences in annual temperature and precipitation, and the results showed that compared to limestone areas, the temperature was a little higher, and the precipitation was obviously lower in the dolomite areas28.

Difference of socio-economic factors between karst and non-karst areas

Taking economic level and rural population into account, the counties were split into four groups: developed and undeveloped counties with rural population decrease and increase (Fig. 3a). 67.13% of counties had a net decrease of the rural population, including 144 karst counties and 342 non-karst counties during the past two decades (Fig. 3b). Among them, 56.25% of karst counties and 48.54% of non-karst counties showed a coexistence of poverty and rural population decrease. The developed counties with rural population decrease were mainly concentrated near urban agglomeration. In contrast, undeveloped counties with rural population decrease were mainly concentrated in KA in Hunan, Hubei, Guangxi, and Sichuan provinces, where the climatic condition was suitable for vegetation growth, with the precipitation of more than 1100 mm. There were 238 counties with a net rural population increase, including 44 karst counties and 194 non-karst counties, and 51.68% counties were developed counties.

Fig. 3: Spatial pattern and statistics of socio-economic factors.
figure 3

a Spatial pattern of socio-economic factors; b statistics of socio-economic factors in karst and non-karst areas. The number of counties was marked in the bar chart.

Full size image

The impact of the rural population on forest land change was also closely related to the economic development mode, and the correlation between economic development and rural population change was significantly negative (R = −0.5, Fig. 4). There tended to be a larger area dedicated to returning cropland to forest land in undeveloped counties with rural population decrease, while in counties with rural population increase, a larger area dedicated to returning grassland to forest land could be observed. The area of increased forest land was large in undeveloped counties with rural population decrease, with tree cover proportion increasing in 43.93% of the area (Supplementary Fig. 7). It indicated that rural population decrease in undeveloped areas could greatly promote forest land increasing16.

Fig. 4: Relationships between forest land change and changes in rural population and economic development.
figure 4

The population change is the ratio change of rural population and total population within the county, indicating the change magnitude of rural population; and economic development is divided into two groups with the median value of night light density (DN = 17.6).

Full size image

Comparison of social-ecological factors’ impact on tree cover between karst and non-karst areas

Tree cover proportion was associated with a variety of social-ecological factors, including human influence, climate factors, land use factors, soil status, and topographic factors (Fig. 5). It was found that tree cover proportion was positively correlated with soil organic matter and slope, and was negatively correlated with cropland proportion, rural population and economic development in both KA and NKA. However, it was worth noting that tree cover proportion was correlated with temperature (R = −0.33) in KA, but not correlated in NKA. Besides, soil bulk density (R = −0.16), elevation (R = 0.18) and village density (R = −0.32) were obviously correlated to tree cover proportion in NKA, but not in KA.

Fig. 5: Social-ecological networks with the core of tree cover.
figure 5

a In karst area; b in non-karst area; c correlation degree of social-ecological networks; d hub in synergy networks; and e hub in trade-off networks.

Full size image

According to the relationships among the variables mentioned above, the correlation degree of the social-ecological network in KA (47/78) was lower than that in NKA (62/78) (Fig. 5a, b), showing that the social-ecological systems in karst areas were more heterogeneous. Furthermore, the correlation degree of social-ecological factors in NKA was higher than that in KA (Fig. 5c). Among them, the correlation degree of soil organic matter, slope, village density, and elevation was the highest in the NKA, which were connected to all the other social-ecological factors. In KA, the correlation degree of forest land proportion was 11, which was higher than the other factors, with soil bulk density for the lowest. According to the results of synergy networks, the slope and soil organic matter in NKA had the greatest impact on the synergistic improvement among the social-ecological factors, while the forest land proportion contributed most on the synergy networks in KA (Fig. 5d). In terms of trade-off networks, it was worth noting that the cropland proportion and slope in both KA and NKA had great effect on the social-ecological networks (Fig. 5e).

Discussion

The social-ecological system enhances comprehension of the relationship between humans and nature through an integrated perspective29,30. Tree cover restoration involves different stakeholders, and social-ecological factors31,32. To give forest land management a scientific foundation, the mechanism of forest land change must be investigated from a social-ecological perspective.

As for ecological factors, soil status, topographic factors, and climate factors of forest land change area were explored in both KA and NKA. It was found that the soil organic matter in NKA were greater than that in KA (Supplementary Fig. 4). Since carbonate rock made up most of the KA, the rate of soil formation was sluggish and the soil cover was thin33. The vegetation can only absorb nutrients and water from the barren topsoil layer, which makes it hard to support the vegetation growth34,35. Ecological engineering will be carried out in areas where the ecological environment of KA is suitable for the survival of forest36. While in NKA, the land unsuitable for growing crops was returned to forest land10, and thus the soil organic matter of forest land change area in NKA was always low. In terms of topography, the elevation and slope of the forest land change area in KA were greater than that of NKA, indicating the characteristics of strong fluctuation of the bedrock surface37. The ways of forest land change were impacted by the slope. In regions with low elevation and gentle slope, the new forest land was primarily converted from cropland, whereas in areas with high altitude and steep slopes, the new forest land was primarily converted from grassland. There were regular droughts despite sufficient precipitation in the KA because of the soil’s low capacity to hold water and its quick evaporation9,38.

The effect of rural population increase and decrease on regional greening was discussed in this study. Area with rural population decrease was primarily centered around urbanized areas (Fig. 3). It was worth noting that the area of cropland and grassland to forest land in developed counties with rural population decrease was smaller (Fig. 4). Although surrounding rural population from these areas have been drawn to the city to work, the cropland and an aspect of the agricultural livelihood have remained39. The rural population who lives close to the provincial capital or popular tourist destinations could go to the city for work and come home to work on their farms during the busy farming season, then affects the forest land change16 (Supplementary Fig. 7). The area of cropland to forest land in undeveloped county with rural population decrease was relatively large, and the possible reason was that a large number of rural farmers go out to work, leading to relatively more cropland abandonment. Moreover, many of these areas were mountainous, with relatively poor farming conditions39.

The erosion of limestone leads to the formation of underground fissures and pipes, which creates an environment where vegetation roots can absorb water and grow40,41. For dolomite areas, the dissolution mainly occurs in the topsoil layer and the soil thickness is thin, which is hard to support the vegetation growth34,42. Therefore, forest land change in limestone areas was often considered to be more effective43. In areas with a high proportion of karst landforms, the high cropland proportion and rural population density, would create a hidden risk for ecological degradation and slow economic growth. To restore the tree cover in these areas and enhance the resilience of natural ecosystems, it is vital to focus on forestation in the sloping farmland of NKA and the gentle slope area (compared with the steep slope area) of KA. Besides, we should pay more attentions to local characteristics while designing tree cover restoration projects in KA. We also discovered that both rural population decrease and economic development could affect tree cover restoration, especially in KA. The main approach of tree cover restoration in KA was to reduce farming and encourage rural population to transition to urban areas for jobs39. Developing regional economies, directing farmers’ livelihood transformation, and promoting a change in the way of life of rural residents, are all essential ways for tree cover restoration44.

Data and methods

Forest land change analysis

The Grain for Green Program was started in 2002 in South China Karst. It was aimed to return cropland to forest land and planted forest and grass on barren land. This program was designed to prevent land degradation and protect downstream ecosystems38. The forest land in 2000 and 2020 was extracted from GlobeLand30 products, with a spatial resolution of 30 m45. As an essential vegetation structure parameter46, tree cover data was derived from the Terra MODIS Vegetation Continuous Fields (VCF) product (MOD44B v061) with a spatial resolution of 250 m. The VCF product provided a continuous and quantitative portrayal of land surface cover with improved spatial details, which was widely used in forest monitoring47,48.

Social-ecological factors

We used several variables, including human influence, land use factors, topographic factors, soil status, and climate factors to evaluate the impact of socio-economic and ecological factors on forest land change.

For soil status, soil aeration and water transport capacity are affected by the soil bulk density, which in turn affects vegetation growth49. In addition, soil organic matter is an important limiting factor for plant physiology and has been experimentally proven to be related to the photosynthetic rate of plants50. Thus, we used the product from the National Science Qinghai Tibet Plateau Data Center, including soil bulk density as well as soil organic matter as soil factors.

For human influence, we chose the rural population, economic development and village density. These three factors are the main measurement factors reflecting social and economic changes and have been widely used as standardized measurements of human impacts on ecosystems51. The population data and urban boundary data52 were used to calculate the rural population per county. The population data of 1 km spatial resolution was derived from World Pop. The urban boundary data was used to eliminate the urban area of the county, and then we used it to calculate the rural population of the county. Besides, the nighttime light data of 500 m spatial resolution was used to characterize the economic development53. The maximum value of the mean nighttime light intensity from 2000 to 2020 was calculated in each county to represent the overall economic development. Counties with nighttime light intensity above-average were defined as the developed counties, with the others for undeveloped counties54,55. As for village density, we calculated the number of villages per square kilometer.

Topography is an important factor affecting vegetation growth and socio-economic activities and the difference in photosynthetic response caused by barometric pressure depends on altitude. We calculated the altitude and slope of the terrain using the Shuttle Radar Topography Mission.

For land use factors, we selected the proportion of grassland, forest land, and cropland. As the primary vegetation types in the region, they are essential to vegetation greenness. Land use factors were also extracted from GlobeLand30 products45.

For climate factors, we selected the annual average temperature and precipitation. The climate data of 1 km spatial resolution was derived from Peng et al.56.

We used the county as the basic analysis unit, and the county with the proportion of karst landforms over 50% was defined as the karst county9,38.

Tree cover trend analysis

The trend of the tree cover change was obtained using the Mann–Kendall method, a nonparametric estimate technique that is unaffected by data outliers. To evaluate the trend’s significance, the Z-test was computed. In this investigation, the significance threshold of α = 0.05 was employed. When Z ≥ 1.96, the sequence had a rising trend; when Z ≤ −1.96, it displayed a falling trend. The sequence exhibited no significant trend when −1.96 < Z < 1.96.

Significance test

The Kruskal–Wallis test is a commonly used nonparametric method to test whether variables have statistically differences between groups, which was usually used to explore vegetation change and its drivers57,58. It was used to contrast the various ecological factors’ effects on forest land change. The data was preprocessed using normal distribution test.

Social-ecological network analysis

The social-ecological network approach has the potential to analyze the internal factors’ correlations of the social-ecological system59,60. The construction process is as follows: (1) the definition of social-ecological dependencies of a particular study; (2) the definition of social-ecological nodes; and (3) the definition of the links between social-ecological factors. We used the social-ecological network to explore the social-ecological factors’ impact on tree cover in 2020. The ‘network’ and ‘visNetwork’ packages in R software were used to explore the network.

Network metrics can be used to provide integrated measures of the relationships between social-ecological factors, and to determine how these relationships change between ecosystems61. Since there were positive and negative correlations, separate networks were built for the synergies and trade-offs. We chose degree and hub as the key network metrics. The strength of factors relevant to components of the ecosystem was represented by correlation degree. The node with the largest trade-offs (or synergies) for the components was represented by the hub. Hub was calculated as follows:

$${{D}}_{{i}}={{L}}_{{i}}* {{pcc}}_{{i}}/(n-1)$$
(1)

where Di is the hub of node i, Li is the number of negative (or positive) links of node i, and pcci is the sum of the absolute correlation coefficient of node i with other nodes, and n is the number of nodes. The positive correlation coefficient is set to be 0 in trade-off networks, with 0 for negative correlation coefficient for synergy networks.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.