Assessment of carbon stocks and influencing factors in terrestrial ecosystems based on surface area Authors: Yang Wang, Min Wang, Jirong Zhang, Yingmei Wu, Yan Zhou Published: 2024 ABSTRACT SummaryThe topography of the border ecological barrier area in southern Yunnan is complex, and utilizing the vertically projected area to estimate carbon stocks may lead to significant errors. This study uses multisource data and multiple models to investigate the spatial and temporal variations in surface carbon stocks and the factors affecting them in the study area. Results show: The difference between the surface area and planar area in the study area is large, and the spatial and temporal changes of surface land use and carbon stock based on this are more significant, showing an inverted V-shape trend in time and a spatial distribution pattern of “high southeast, low northwest”. NDVI and slope were the dominant factors. The results of this study provide a new surface-scale perspective for a deeper understanding of carbon stock and land-use planning in the mountainous region represented by the border ecological zone in southern Yunnan. Introduction The swift advancement of urban development and industrial growth has intensified the release of greenhouse gases like carbon dioxide (CO2), contributing significantly to alterations in the global climate, now a crucial factor impacting the sustainability of ecosystems and human societies.1 The Paris Agreement has established clear goals to achieve a balance between anthropogenic greenhouse gas emissions and removals by the latter half of this century. In the year 2020, the Chinese government declared its objective to cap carbon emissions around 2030 and pursue carbon neutrality by approximately 2060.2 Achieving these dual carbon goals requires enhancing carbon stock and reducing emissions. The International Geosphere-Biosphere Program (IGBP) and the World Climate Research Program (WCRP) emphasize the critical importance of studying carbon stocks in terrestrial ecosystems, which is a primary scientific imperative for maintaining global carbon dynamics equilibrium and mitigating the effects of climate change.3 Land use and land cover change (LULC) serve as the most immediate and pervasive indicators of terrestrial systems, playing a pivotal role in driving fluctuations in carbon stocks across ecosystems,4 altering ecosystem structure, function, and energy cycles, thereby affecting regional carbon stock dynamics.5 Research findings indicate that there has been an observed rise in soil organic carbon levels in the Hexi region between 1980 and 2018, which can be linked to changes in land use practices.6 Another study highlights that the expansion of construction land has negatively impacted regional carbon stock.7 In addition, a study has shown that the loss of soil organic carbon (SOC) change in China due to land-use change is 39.2%. However, it has also increased carbon sinks.8 Thus, investigating changes in land use carbon stocks and their influencing factors is crucial for global carbon balance. Currently, the estimation of terrestrial ecosystem carbon stocks primarily relies on plot inventories,9 IPCC inventory methods,10,11 and model simulations.12,13 However, with the evolution of remote sensing, geographic information systems, and global positioning system technologies, methodologies integrating remote sensing monitoring and various models are increasingly prevalent in the carbon stock estimation process. For instance, Scholars employed a combination of Markov-FLUS and InVEST to project potential variations in carbon stocks in Chongqing by 2035 under scenarios reflecting natural conditions, agricultural safety, and high urbanization levels.14 A framework integrating the Land Use Harmonization (LUH2) dataset, the Patch Generation Land Use Simulation (PLUS) model and the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model was proposed to systematically analyze the spatial and temporal evolutionary characteristics of land use in the YRD region and its impact on carbon stock over the period 1992–2050.15 It has also been shown that forest conservation and regeneration in dam-affected areas drove increases in carbon emissions compared to unaffected areas, while increases in carbon stock were more pronounced.16 While these studies offer theoretical and methodological frameworks for estimating carbon stocks in terrestrial ecosystems, they frequently reduce spatial dimensions to two-dimensional plane space, thus neglecting the spatial heterogeneity resulting from three-dimensional terrain variations.17,18,19,20 Therefore, estimates of carbon stocks in terrestrial ecosystems are predicated on projected rather than actual area. There exists a distinction between projected area (i.e., two-dimensional plane area) and actual surface area, with potentially minor discrepancies in flat or gently sloped areas but substantial differences in mountainous regions.21,22,23,24 China, characterized as one of the largest mountainous countries globally, showcases extensive topographical diversity. Its elevations span from −154 to 8848 m, encompassing a variety of landforms such as plains, hills, mountains, basins, plateaus, and highlands. Spatial terrain analysis reveals that mountains occupy more than 40% of China’s total land area, hills approximately 20%, and plains about 27%. Therefore, over 70% of China’s territory shows three-dimensional characteristics rather than being purely two-dimensional. This substantial coverage of three-dimensional landforms, including mountains and hills, contributes to China’s total surface area exceeding 960 × 104 km2.25 The calculation of surface area can be influenced by varying resolutions of DEM (Digital Elevation Model) data. According to analysis using 90m SRTM DEM data, research determined that China’s land surface area measures 1003.36 × 104 km2, surpassing the two-dimensional planar estimate by about 43 × 104 km2.26 Based on 25m DEM data, a different study determined that the surface area of the Shaanxi section within the Qin Ling Mountains is 7.52 × 104 km2, which represents a 22.04% increase compared to the two-dimensional planar area.27 Clearly, the surface area differs from the two-dimensional planar area. Estimations of carbon stock based on projected area are prone to significant errors. Thus, it is crucial to accurately evaluate carbon stock in terrestrial ecosystems by utilizing the actual surface area of the land. The prominent southern border ecological barrier of Yunnan Province significantly shapes the ecological security framework in Southwest China. Firstly, the region features extensive mountainous and hilly terrain, with a significant elevation difference of 2887 m, steep slopes, and pronounced vertical changes. This results in a substantial gap between its actual surface area and two-dimensional projection. Secondly, its distinctive geographical environment places it at the forefront of carbon stock, biodiversity conservation, and soil erosion control efforts. In recent years, driven by economic development, increasing production demands, lifestyle changes, and unsustainable land practices, the region has experienced ecological degradation and intensified habitat fragmentation, resulting in significant alterations in land use and cover. Therefore, this study focuses on the southern border ecological barrier of Yunnan Province as its study area. Utilizing diverse data sources, this research combines the InVEST model, random forest model, and geodetector to assess the spatial variations in carbon stock within terrestrial ecosystems from 2000 to 2020, alongside investigating the factors influencing these dynamics. The study aims to provide scientific insights that support the achievement of dual-carbon goals in the ecological barrier along Yunnan Province’s southern border. Results Characteristics of temporal changes in surface land use in the past (2000–2020) During the study period, the surface area of the study area differed significantly from the planimetric area, and the land use types changed drastically. From 2000 to 2020, the total surface area of the study area remained approximately 107968.6 km2 (Table 1), with an increase of 4.10 × 104 km2 compared to the baseline. Woodland and grassland dominate the study area, encompassing 62.06% and 19.89% of the gross land area, respectively. Cropland occupies 16.56% of the land surface, with built-up areas, water bodies, and unused land collectively comprising under 2%. In the last two decades, significant shifts in land use have been observed: grassland has decreased by 2453.64 km2, and unused land has contracted by 7.04 km2. In contrast, woodland and water bodies have expanded, with gains totaling 1270.43 km2 and 574.11 km2, respectively, while cropland and built-up land have also increased but at a slower rate. It is evident that the conservation of woodland and water bodies is prioritized in the southern ecological barrier area alongside economic development, aligning with recent efforts in Yunnan Province to promote ecological civilization.Table 1Changes in surface area from 2000 to 2020YearTypeCroplandWoodlandGrasslandBuilt-up landWater bodyUnused land2000surface area/km217544.89065754.64023938.660368.370333.26028.770Proportion/%16.25060.90022.1700.3400.3100.0302010surface area/km217514.41067663.82021914.390370.630447.64024.160Proportion/%16.23062.69020.3000.3400.4100.0202020surface area/km217880.62067025.07021485.020672.360907.37021.730Proportion/%16.56062.06019.8900.6200.8400.020 Table 1Changes in surface area from 2000 to 2020YearTypeCroplandWoodlandGrasslandBuilt-up landWater bodyUnused land2000surface area/km217544.89065754.64023938.660368.370333.26028.770Proportion/%16.25060.90022.1700.3400.3100.0302010surface area/km217514.41067663.82021914.390370.630447.64024.160Proportion/%16.23062.69020.3000.3400.4100.0202020surface area/km217880.62067025.07021485.020672.360907.37021.730Proportion/%16.56062.06019.8900.6200.8400.020 Changes in surface area from 2000 to 2020 Surface area conversion between land use types The land use transfer matrix delineates the dynamics of land use types in the southern border ecological barrier of Yunnan from 2000 to 2020. It captures the transitions between categories and emphasizes the directional changes in land use dynamics within the study area (Table 2). Analysis of these transitions reveals that cropland predominantly shifts to woodland, grassland, and built-up land; woodland primarily transforms into cropland, grassland, and water bodies; grassland mostly transitions to cropland and woodland; and water body predominantly converts to cropland and forest land. Approximately 54.96% of cropland conversions originate from grassland, while 59.63% of built-up land conversions originate from cropland. Grassland is the primary source of cropland expansion, while cropland is the main driver of built-up land growth. This analysis shows that the extension of built-up lands has led to decreased ecological and cropland areas, highlighting urban development’s impact on the diminishing extents of both agricultural and ecological lands. This transfer pattern stems from rapid population growth and economic development over the past few decades, which have heightened the demand for land resources. Urban sprawl continues to erode cropland, resulting in a decrease in cropland area. Moreover, to alleviate the pressure on food production due to rapid population growth and to adhere to the 1.8 billion mu red line for preserving arable land, there is an ongoing conversion of grassland into cropland, creating a vicious cycle. Additionally, from 2000 to 2020, there was significant conversion between woodland, grassland, and water bodies in the study area, primarily driven by ecological conservation policies and the implementation of certain ecological projects, which have largely stabilized the total area of ecological land.Table 2Surface area transfers by land-use type, 2000–2020(km2)123456TotalTransferred out113742.9201925.880937.920103.740381.1001.28017092.8603349.94021629.41060783.7902992.670193.690142.4100.42065742.3804958.60032081.3504158.96017484.640100.620109.3600.48023935.4106450.770427.14044.30022.440267.5405.7201.170368.320100.770543.26014.1906.8501.070267.8600.030333.26065.40065.6201.9000.3602.1700.50018.22028.77010.550Total17529.71066929.02021444.880668.840906.95021.610107501.00014936.030Transferred in3786.7806145.2403960.240401.300639.0903.38014936.030–Notes: 1, 2, 3, 4, 5 and 6 represent cropland, woodland, grassland, water body, built-up land, and unused land, respectively. Table 2Surface area transfers by land-use type, 2000–2020(km2)123456TotalTransferred out113742.9201925.880937.920103.740381.1001.28017092.8603349.94021629.41060783.7902992.670193.690142.4100.42065742.3804958.60032081.3504158.96017484.640100.620109.3600.48023935.4106450.770427.14044.30022.440267.5405.7201.170368.320100.770543.26014.1906.8501.070267.8600.030333.26065.40065.6201.9000.3602.1700.50018.22028.77010.550Total17529.71066929.02021444.880668.840906.95021.610107501.00014936.030Transferred in3786.7806145.2403960.240401.300639.0903.38014936.030–Notes: 1, 2, 3, 4, 5 and 6 represent cropland, woodland, grassland, water body, built-up land, and unused land, respectively. Surface area transfers by land-use type, 2000–2020(km2) Notes: 1, 2, 3, 4, 5 and 6 represent cropland, woodland, grassland, water body, built-up land, and unused land, respectively. Spatial and temporal variations in surface carbon stocks The overall carbon stock in the study area is high but with significant spatial and temporal heterogeneity. Utilizing the InVEST model, this study computed carbon stock across diverse land use types within the southern border ecological barrier zone of Yunnan. The cumulative carbon stock in the research area amounted to 1197.52 × 106, 1201.51 × 106, and 1195.90 × 106 T in 2000, 2010, and 2020, respectively (Figure 1). Between 2000 and 2020, there was an initial increase followed by a decrease in total carbon stock, leading to an overall reduction of 1.62 × 106 T, representing a decline of 0.14% over the period. Furthermore, the carbon stock per hectare for these respective years was 110.91, 113.32, and 110.74 T/hm2. The pattern of change in carbon stock per hectare mirrors that of total carbon stock, suggesting a declining capacity for carbon stock in the study area. Spatially, carbon stock displays a distinct spatial distribution, characterized by elevated levels in the southeast and diminished levels in the northwest. Regions with high carbon stock are primarily located in the southeastern border areas and specific mountainous regions like Menglian, Mengla, Jiangcheng, and Hekou counties. These regions are characterized by high vegetation cover and low urbanization levels, mainly composed of woodland and grasslands. Conversely, areas with relatively low-carbon stocks are centered in the middle of the study area, exhibiting a discernible pattern of expansion in both directions, especially in counties such as Gengma, Yongde, and Zhenkang. These areas are characterized by dominant land uses of croplands and built-up lands, heavily influenced by human activities, leading to relatively lower levels of carbon stock.Figure 1Carbon stocks in the border ecological barrier area of southern Yunnan from 2000 to 2020(A) Carbon stocks in the border ecological barrier area of southern Yunnan from 2000.(B) Carbon stocks in the border ecological barrier area of southern Yunnan from 2010.(C) Carbon stocks in the border ecological barrier area of southern Yunnan from 2020. Carbon stocks in the border ecological barrier area of southern Yunnan from 2000 to 2020 (A) Carbon stocks in the border ecological barrier area of southern Yunnan from 2000. (B) Carbon stocks in the border ecological barrier area of southern Yunnan from 2010. (C) Carbon stocks in the border ecological barrier area of southern Yunnan from 2020. To offer a more comprehensive understanding of the fluctuations in regional carbon stocks, carbon stock data from two specified periods underwent differencing calculations in ArcGIS 10.5. They were classified into three categories: decrease, increase, and unchanged, thereby revealing the spatial distribution associated with these fluctuations. Overall, from 2000 to 2020, the spatial dynamics of regional carbon stocks show a pattern of substantial aggregation and scattered occurrences. Specifically, most of the study area, amounting to 90.68%, maintained stable carbon stocks. Areas demonstrating an increasing trend in carbon stocks, accounting for 6.79%, were primarily clustered in the southeast, with sporadic occurrences elsewhere. Particularly during the period from 2000 to 2010, regions witnessing decreased carbon stocks were concentrated in counties such as Menglian, Lancang, and Ruili. However, from 2010 to 2020, there was a prominent increase in areas experiencing declining carbon stocks compared to the previous period, alongside a growth in the percentage of areas maintaining consistent carbon deposits. Counties with decreased carbon stocks were concentrated in areas including Longyang and Jinghong (Figure 2).Figure 2Carbon stock changes in the border ecological barrier area of southern Yunnan from 2000 to 2020(A) Carbon stock changes from 2000 to 2010.(B) Carbon stock changes from 2010 to 2020.(C) Carbon stock changes from 2000 to 2020. Carbon stock changes in the border ecological barrier area of southern Yunnan from 2000 to 2020 (A) Carbon stock changes from 2000 to 2010. (B) Carbon stock changes from 2010 to 2020. (C) Carbon stock changes from 2000 to 2020. By employing land use transfer matrices and carbon density data from various land categories, this study evaluates how changes in land use affect carbon stocks in the southern border ecological barrier zone of Yunnan Province (Table 3). Between 2000 and 2020, it was found that changes in land use resulted in a net decrease of 2.76 million metric tons in carbon stocks. The conversion of cropland to urban areas accounted for 85.80% of the carbon lost from cropland. Woodland saw a decline of 17 × 106 T in carbon stocks, with carbon losses from the transformation of grassland to cropland contributing to 70.19% of the decrease in grassland carbon stocks. Furthermore, 85.53% of the carbon loss in water body carbon stocks was attributed to the conversion of water bodies into built-up lands.Table 3Changes in surface carbon stocks from 2000 to 2020(103T)123456Total1–7652.2501285.740−336.360−2071.240−6.5006523.9002−6474.270–−7788.520−1397.610−1339.780−3.760−17003.9203−2853.20010823.810–−464.170−744.280−3.1106759.050487.990319.680103.520–−12.540−2.120496.5305235.130133.48046.6402.340–0.010417.600628.42017.2002.2803.940−0.190–51.650Total−8975.93018946.420−6350.340−2191.860−4168.030−15.480−2755.210 Table 3Changes in surface carbon stocks from 2000 to 2020(103T)123456Total1–7652.2501285.740−336.360−2071.240−6.5006523.9002−6474.270–−7788.520−1397.610−1339.780−3.760−17003.9203−2853.20010823.810–−464.170−744.280−3.1106759.050487.990319.680103.520–−12.540−2.120496.5305235.130133.48046.6402.340–0.010417.600628.42017.2002.2803.940−0.190–51.650Total−8975.93018946.420−6350.340−2191.860−4168.030−15.480−2755.210 Changes in surface carbon stocks from 2000 to 2020(103T) Impact factor significance screening Based on relevant literature regarding factors influencing spatiotemporal variations in carbon stocks, these factors were initially categorized into environmental factors (slope, elevation, NDVI, precipitation, temperature) and anthropogenic factors (population density, proportion of tertiary industry to GDP, average years of education). A random forest model was developed using the R language, whereby 70% of the dataset is randomly allocated for training purposes, while the remaining 30% is reserved for testing (Figure 3). The random seed value was fixed to ensure the reproducibility of results. The parameters were set as follows: mtry = 6 and ntree = 500, while the remaining parameters were set to their default values. Larger values of these indices indicate higher variable importance. By combining the %IncMSE and IncNodePurity methods, the two least important overlapping factors (GDP and average years of education) were excluded, resulting in a selection of more important driving factors (NDVI, slope, elevation, mean annual temperature, mean annual precipitation, population density, and the proportion of tertiary sector GDP).Figure 3Ranking of importance of factors influencing carbon stocks from 2000 to 2020Notes: the abbreviations of RF, T, Tertiary, Edu, and People denote rainfall, temperature, the proportion of tertiary sector GDP, average years of education, population density. Ranking of importance of factors influencing carbon stocks from 2000 to 2020 Notes: the abbreviations of RF, T, Tertiary, Edu, and People denote rainfall, temperature, the proportion of tertiary sector GDP, average years of education, population density. More drastic impacts of environmental factors A single-factor analysis was conducted on the seven factors driving spatial differentiation and temporal variations in carbon stocks in the Southern Yunnan Ecological Barrier Zone, as depicted in Table 4. From 2000 to 2020, carbon stock in the region was influenced by both environmental factors and human activities. Environmental factors have a greater impact on carbon stocks than anthropogenic factors. Among environmental factors, average annual precipitation ranked third in explanatory power, with values of 0.115 in 2000, 0.143 in 2010, and 0.144 in 2020, explaining over 10% of the variance in carbon stock. Additionally, NDVI (0.751–0.754) and slope (0.292–0.295) exerted significant influences on carbon stock, explaining more than 20% of the variance. Importantly, NDVI consistently explained over 70% of the variance across different years, underscoring its dominant role in spatial variations of carbon stock. These findings underscore that from 2000 to 2020, areas rich in vegetation cover, such as forests and grasslands, exhibited higher carbon sequestration potential than land types such as croplands and urban areas. This difference played a significant role in determining how carbon stock is distributed spatially within the Southern Ecological Barrier Zone. The integration of environmental factors including, annual average precipitation, slope, and NDVI, reveals that these natural conditions serve as foundational prerequisites for carbon sequestration. Unlike flat plains, the Southern Border Ecological Barrier Zone, which is primarily hilly and mountainous, is less favorable for the expansion of agricultural and urban development. Here, human activities exert minimal influence, and with increasing slope, the land predominantly consists of forests and grasslands with extensive and well-preserved vegetation cover, contributing to substantial carbon stock potential.Table 4Carbon stock single factor detection resultsTypes of driversImpact Factor–q-value–––200020102020Environmental factorsAnnual average precipitation(X1)0.1150.1430.144Annual average temperature(X2)0.0890.0320.060DEM(X3)0.0940.0980.098Slope(X4)0.2920.2930.295NDVI(X5)0.7510.7540.754Anthropogenic factorsPopulation density(X6)0.2010.1570.369Tertiary industry in GDP(X7)0.0670.5760.192Note: People represents population density. Table 4Carbon stock single factor detection resultsTypes of driversImpact Factor–q-value–––200020102020Environmental factorsAnnual average precipitation(X1)0.1150.1430.144Annual average temperature(X2)0.0890.0320.060DEM(X3)0.0940.0980.098Slope(X4)0.2920.2930.295NDVI(X5)0.7510.7540.754Anthropogenic factorsPopulation density(X6)0.2010.1570.369Tertiary industry in GDP(X7)0.0670.5760.192Note: People represents population density. Carbon stock single factor detection results Note: People represents population density. Population density, a critical indicator of human activity, exerts the strongest influence on carbon stock among anthropogenic factors. The influence of this factor grew stronger from 0.201 by 2000 to 0.369 by 2020, suggesting that the inevitable increase in population and urbanization will lead to the expansion of carbon-emitting lands such as built-up and agricultural lands. Therefore, during urbanization processes, it is crucial to enhance efforts to protect carbon stock lands effectively, preventing excessive development and utilization of woodlands, water bodies, and other resources. The proportion of GDP from the tertiary sector shows an increasing fluctuating trend in its impact on carbon stock, with the explanatory power rising from less than 10% in 2000 to nearly 20% in 2020. Overall, the combined spatial and temporal changes in carbon stock arise from the intricate interplay of multiple factors. Effective utilization of Yunnan’s border counties' natural environment and resources, alongside optimized land use, enhanced energy efficiency, and strict adherence to ecological conservation measures, is essential to prevent further environmental degradation. During the period from 2000 to 2020, the analysis consistently revealed nonlinear amplification or synergistic effects in interactions between paired factors without any occurrences of independent or diminishing interactions. This suggests that the combination of any factor with others enhances their collective influence on spatial variations in carbon stock, thereby affirming the intricate interaction of factors that impact the spatiotemporal dynamics of carbon stock. As depicted in Figure 4, across different years, the synergistic interaction x1∩x5—where annual average precipitation interacts with NDVI—consistently showed the highest explanatory power for spatial variations in carbon stock within Yunnan’s southern ecological barrier zone, ranging between 0.980 and 0.989. This suggests that regions experiencing combined effects of annual average precipitation and NDVI may reflect richer carbon stock. The increased precipitation likely promotes the recovery and enhanced coverage of forest and grassland vegetation, thereby contributing to continuous growth in vegetation and soil carbon stock within the research district. The synergy between NDVI and other factors primarily manifests as dual-factor enhancement, reinforcing NDVI’s crucial role in driving spatial variations in carbon stock. Across all years, annual precipitation interacts nonlinearly with other factors, consistently showing enhanced effects compared to individual factors. Taken together, the synergistic impact of interacting factors plays a crucial role in determining spatial fluctuations in carbon stock within the study area, indicating a complex interplay rather than a straightforward additive relationship. Hence, a holistic understanding of how driving factors interact is essential for comprehending their combined impact on spatial variations in carbon stock.Figure 4Influence factor interaction detection results(A) Geodetector interactive detection results for the study area, in 2000.(B) Geodetector interactive detection results for the study area, in 2010.(C) Geodetector interactive detection results for the study area, in 2020. Influence factor interaction detection results (A) Geodetector interactive detection results for the study area, in 2000. (B) Geodetector interactive detection results for the study area, in 2010. (C) Geodetector interactive detection results for the study area, in 2020. Discussion Calculation of surface area and its comparison with plane area Surface area estimation is crucial for accurately assessing carbon balances and land use changes in regions with complex terrain. Previous studies have primarily focused on flat regions, particularly plains, resulting in substantial discrepancies in surface area calculations for mountainous and hilly areas with diverse altitudes, such as the southern border ecological barrier zone of Yunnan Province. The resolution of DEM data and the computational methods employed significantly impact surface area calculations. Previous research has shown that surface area increases with higher DEM resolution under similar land cover conditions, often using DEM data with spatial resolutions of 90m or 30m. Studies have shown that China’s total surface area may expand by 43000 km2.26 In this study, 12.5m DEM data were used to calculate the surface area, building on previous research accurately,21,22 the results showed that the total surface area of the study area increased by 4.10 × 104km2 compared to the planar area, which observed an increase in area with improved data resolution. Thus, land use changes in mountainous areas will inevitably result in greater variability in carbon stock. Gradient distribution effects of carbon stocks in mountains Terrain features exert a noticeable influence on carbon stock, with elevation serving as a critical factor in assessing these variables. Elevation changes influence temperature, thereby affecting vegetation growth, while slope variations alter vegetation cover and biological survival, thereby influencing carbon stock in mountainous ecosystems. This study shows that total carbon stock in the study area follows a curvilinear pattern with varying elevation gradients (Figure 5). This trend primarily arises from favorable hydrothermal conditions observed between 1200 and 2400 m, where forests are densely distributed, predominantly comprising evergreen broadleaf and warm-temperate coniferous forests with high vegetation carbon density and substantial soil carbon content, leading to the highest overall carbon stock. The present finding is like that of previous studies.28 Below 1200 m, the area is mainly composed of cultivated lands, residential and industrial areas, and transportation infrastructure, interspersed with shrublands and grasslands, characterized by intensive human activities and relatively lower soil and vegetation carbon content, thus showing weaker carbon stock capacities, 29 in line with findings by Zhang et al. (2016). Above 3000 m elevation, beyond most forest vegetation, surface vegetation mainly consists of cold-temperate coniferous forests, shrub meadows, and alpine moss vegetation, leading to a significant decrease in carbon density and a gradual reduction in carbon stock.30 Therefore, mountainous regions with hilly terrain demonstrate significant potential for enhancing carbon sequestration.Figure 5Changes in carbon stocks at different elevation gradients Changes in carbon stocks at different elevation gradients Importance of surface area in carbon stock estimation The land use area obtained directly from remote sensing images is only the projection of land cover on the horizontal plane. In contrast, in reality, the land surface is not absolutely flat but is three-dimensional and has various undulating topography, especially in areas with dramatic topographic undulations; the difference between the true area of the land surface and the vertically projected area is especially obvious.31,32 If the two-dimensional planar-acquired land use area is directly used in the regional land resource scientific management, plant resource development and utilization, and ecosystem service value, it is likely to cause some degree of error or even erroneous conclusions.33 In view of the current situation that existing studies focus on land use and carbon stock in terms of planar area (vertical projection area), this study attempts to introduce the three-dimensional concept of surface area by taking into account the topographic characteristics of the border ecological barrier area in southern Yunnan, and explore the spatial and temporal changes of carbon stock and the influencing factors in the study area based on the multi-source data from 2000 to 2020. For example, Zhao Ting et al. found that the difference between the surface area and the projected area is mainly dependent on the slope gradient, and that the terrain is relatively flat in the lower elevation areas of the Qin Ling Mountains. At the same time, the slope gradient in the higher elevation areas is gentle, and the difference between the surface area and the projected area is small.27 In this paper we found that the difference between the surface area and the projected area increases with the increase of the slope gradient. Slope class increases, and the area difference becomes more and more obvious. When the slope is less than 60°, the area ratios of different land use types are basically equal, and the land use types have basically no effect on the results of the area ratio, and the slope mainly causes the differences; when the slope is greater than 60°, the differences in the area ratios of different land use types are obvious; when the slope is greater than 80°, the differences are significant (Figure 6). The findings of this paper are consistent with those of the above scholars. While LULC data is one of the key input elements of the Carbon module of the Invest model, previous studies have mostly used the vertical projected area directly to calculate carbon stocks,34,35 which may lead to a large bias in the estimation of carbon stocks. Therefore, in this paper, the real surface area is used to represent the area of each land use type, which can more accurately assess the carbon stock in the border ecological barrier area in southern Yunnan.Figure 6Relationship between surface area to projected area ratio and slope in the study area Relationship between surface area to projected area ratio and slope in the study area Drivers of carbon stock distribution The distribution of carbon stock in the study area is primarily determined by NDVI, which correlates strongly with areas abundant in vegetation cover. These regions demonstrate elevated levels of net primary productivity (NPP), thereby increasing their ability to sequester carbon. The dynamics of carbon stock in response to environmental factors involve a complex interplay of multiple variables.36,37 Additionally, within environmental factors, slope emerges as a crucial factor influencing both carbon stock and density, following NDVI. Slope and other topographic factors act as substrates for vegetation growth, affecting sunlight exposure and hydrothermal conditions, and directly influencing the distribution of vegetation types.38 A decrease in slope often facilitates transitions in vegetation types from forests to cultivated vegetation, thereby contributing to the accumulation of carbon stock. Conversely, in steep mountainous regions with minimal human activity, vegetation cover remains intact, leading to a rapid increase in carbon stock.39 However, climate factors such as temperature and precipitation also significantly influence the carbon sequestration capacity of terrestrial ecosystems. During the period from 2000 to 2020, the study area consistently experienced annual precipitation levels exceeding 800 mm, which provided abundant moisture for vegetation growth. This has stimulated increases in both aboveground and belowground biomass, thereby enhancing carbon stock in terrestrial ecosystems.40 However, temperature’s explanatory power for the spatial distribution of carbon stock is relatively weaker, likely due to moderate seasonal and annual temperature fluctuations in the study area, resulting in lower vegetation sensitivity compared to precipitation. Population density emerges as the predominant factor shaping the spatial distribution of carbon stock. Studies emphasize that population growth drives the expansion of urban and rural areas, leading to ecological land fragmentation and a subsequent reduction in carbon stock capacity.41 In the last twenty years, population concentration has exacerbated congestion effects, intensifying energy consumption and carbon emissions,42 thereby hindering the accumulation of carbon reserves. Furthermore, high population concentration and dense urban construction contribute to the urban heat island effect, presenting challenges in maintaining the equilibrium of carbon stock services. China’s urbanization trend, characterized by faster conversion of urban land compared to population growth,43 has resulted in increased impermeable surfaces and extensive destruction of forest vegetation, leading to a gradual decline in carbon sequestration areas. Another unique advantage of geodetector is the detection of two-factor interactions on the dependent variable.44 Compared to one-factor detection, interaction detection can determine whether there is an interaction between the two factors, as well as the strength, direction, linearity, or non-linearity of the interaction, which can better reveal that an independent variable has a significant effect on a dependent variable. According to the geodetector model, the interaction between environmental factors (elevation, slope, precipitation, and NDVI) and human activity factors (e.g., population density) can lead to a bivariate enhancement effect under certain combination conditions. This means that the combined effect of the two factors on carbon stocks is greater than the sum of their individual effects. This interaction demonstrates the complex dynamics of environmental and anthropogenic factors in determining carbon stock capacity. Higher altitude areas with lower population densities are less likely to be developed. They are more likely to retain their natural vegetation and, therefore, exhibit greater carbon stock potential.45,46 Limited accessibility to these areas helps to preserve the integrity of their vegetation. Steep slopes away from higher population densities also tend to maintain high levels of vegetation and carbon stocks.47 Regional land-use changes based on human activities, as well as climate change processes in the natural context, directly drive the dynamics of surface vegetation, and the interactive impacts of different drivers should be considered comprehensively in ecological environmental protection and regional low-carbon construction, with differentiated and multifaceted regulatory strategies. Conclusion In view of the current situation that existing studies focus on land use and carbon stock in terms of planar area (vertical projection area), this study attempts to introduce the three-dimensional concept of surface area by taking into account the topographic characteristics of the border ecological barrier area in southern Yunnan, and explore the spatial and temporal changes in carbon stock and the factors affecting it based on the multi-source data for the period of 2000–2020 using the Invest model, the Random Forest Model, and the Geodetector. The results of this paper are as follows.(1)In the study area, there is a big difference between the surface area and the planar area, the surface area is 107968.6km2 and the planar area is 66968.6km2, with a difference of 4.10 × 104km2.The change of land use on the surface is more significant, among which the grassland and the unutilized land surface area decreases significantly, and the surface area of the forest land and the water surface area shows an increasing trend, and the growth rate is higher than that of the arable land and the construction land.(2)The spatial and temporal variations of surface carbon stock in the study area were significant, and the average value of carbon stock was maintained at 1198.31 × 106T, showing an inverted ‘V’-type trend of ‘increasing first and then decreasing later’ in time, and the carbon stock decreased by 1.62 × 106t during the study period, with a decrease of 0.14%. The spatial distribution pattern of ‘high in the southeast and low in the northwest’ is closely related to the spatial distribution of land use types.(3)NDVI and slope were the dominant factors affecting the spatial distribution pattern of carbon stocks, the degree of explanation of carbon stocks by multi-factor interaction was much higher than that by single factor, the interaction between mean annual precipitation and NDVI was the largest, and the environmental factor had a stronger explanatory power than the anthropogenic factor on the spatial distribution of carbon stocks.(4)The gradient effect of carbon stock in the study area is significant, and the carbon stock shows an inverted ‘U’ trend of increasing and then decreasing with elevation, with the peak at 1800 m. (1)In the study area, there is a big difference between the surface area and the planar area, the surface area is 107968.6km2 and the planar area is 66968.6km2, with a difference of 4.10 × 104km2.The change of land use on the surface is more significant, among which the grassland and the unutilized land surface area decreases significantly, and the surface area of the forest land and the water surface area shows an increasing trend, and the growth rate is higher than that of the arable land and the construction land. In the study area, there is a big difference between the surface area and the planar area, the surface area is 107968.6km2 and the planar area is 66968.6km2, with a difference of 4.10 × 104km2.The change of land use on the surface is more significant, among which the grassland and the unutilized land surface area decreases significantly, and the surface area of the forest land and the water surface area shows an increasing trend, and the growth rate is higher than that of the arable land and the construction land. (2)The spatial and temporal variations of surface carbon stock in the study area were significant, and the average value of carbon stock was maintained at 1198.31 × 106T, showing an inverted ‘V’-type trend of ‘increasing first and then decreasing later’ in time, and the carbon stock decreased by 1.62 × 106t during the study period, with a decrease of 0.14%. The spatial distribution pattern of ‘high in the southeast and low in the northwest’ is closely related to the spatial distribution of land use types. The spatial and temporal variations of surface carbon stock in the study area were significant, and the average value of carbon stock was maintained at 1198.31 × 106T, showing an inverted ‘V’-type trend of ‘increasing first and then decreasing later’ in time, and the carbon stock decreased by 1.62 × 106t during the study period, with a decrease of 0.14%. The spatial distribution pattern of ‘high in the southeast and low in the northwest’ is closely related to the spatial distribution of land use types. (3)NDVI and slope were the dominant factors affecting the spatial distribution pattern of carbon stocks, the degree of explanation of carbon stocks by multi-factor interaction was much higher than that by single factor, the interaction between mean annual precipitation and NDVI was the largest, and the environmental factor had a stronger explanatory power than the anthropogenic factor on the spatial distribution of carbon stocks. NDVI and slope were the dominant factors affecting the spatial distribution pattern of carbon stocks, the degree of explanation of carbon stocks by multi-factor interaction was much higher than that by single factor, the interaction between mean annual precipitation and NDVI was the largest, and the environmental factor had a stronger explanatory power than the anthropogenic factor on the spatial distribution of carbon stocks. (4)The gradient effect of carbon stock in the study area is significant, and the carbon stock shows an inverted ‘U’ trend of increasing and then decreasing with elevation, with the peak at 1800 m. The gradient effect of carbon stock in the study area is significant, and the carbon stock shows an inverted ‘U’ trend of increasing and then decreasing with elevation, with the peak at 1800 m. Limitations of the study This research employs diverse data sources and combines the InVEST model with geodetector analysis to examine changes in surface area and carbon stock in the southern border ecological barrier area of Yunnan from 2000 to 2020 (Figure 7). The research evaluates how environmental and human activity factors influence carbon stock dynamics. However, this study has certain limitations that warrant attention in future research: (1) To improve the accuracy of surface area assessments, additional high-resolution remote sensing data and field survey data should be integrated into the study; (2) The carbon density data input into the InVEST model’s carbon module is estimated based on land cover carbon density. Future research could improve the precision of carbon stock results by employing more detailed vegetation and soil classification for estimation.Figure 7Study area location Study area location Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Y.M.W. (2287@ynnu.edu.cn). Materials availability This study did not generate new unique materials. Data and code availability •The data used in this study are all available from public resources that have been appropriately cited within the manuscript.•All custom code can be available on request from the lead contact.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •The data used in this study are all available from public resources that have been appropriately cited within the manuscript. The data used in this study are all available from public resources that have been appropriately cited within the manuscript. •All custom code can be available on request from the lead contact. All custom code can be available on request from the lead contact. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments We would like to express our sincere gratitude to the editors and reviewers who have put considerable time and effort into their comments on this paper. This study was supported by the Yunnan Province Basic Research Key Programs (202401AS070037); the “Yunnan Revitalization Talent Support Program” in Yunnan Province (XDYC-WHMJ-2022-0016, XDYC-QNRC-2022); and Yunnan Philosophy and Society Innovation Team (2014cxp02). Author contributions Conceptualization, Y.W. and. Y.M.W; Methodology, Y.W. and M.W.; Software, Y.W.; writing—original draft, Y.W., M.W., and J.R.Z.; writing—review and editing, Y.M.W.; Investigation, J.R.Z.; Formal analysis, M.W.; Data curation, Y.Z.; Software, Y.W. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited data30m Land useData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/12.5m ALOS DEMData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/Boundary of the study areaData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1kmNDVIData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1km GDPData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1km PeopleWorldPop’s global population density mapshttps://www.worldpop.org/1km TemperatureThe National Earth System Science Data Centerhttps://www.geodata.cn/1km Precipitationthe National Earth System Science Data Centerhttps://www.geodata.cn/Educational years per capitaYunnan Provincial Bureau of Statisticshttps://stats.yn.gov.cn/Proportion of tertiary industry in GDPYunnan Provincial Bureau of Statisticshttps://stats.yn.gov.cn/Software and algorithmsArcgis10.5ESRIhttps://desktop.arcgis.com/zh-cn/desktop/index.htmlR ProjectOpen-source softwarehttps://www.r-project.org/GeodetectorOpen-source softwarehttp://www.geodetector.cn/index.html REAGENT or RESOURCESOURCEIDENTIFIERDeposited data30m Land useData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/12.5m ALOS DEMData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/Boundary of the study areaData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1kmNDVIData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1km GDPData Center for Resources and Environmental Sciences, Chinese Academy of Scienceshttps://www.resdc.cn/1km PeopleWorldPop’s global population density mapshttps://www.worldpop.org/1km TemperatureThe National Earth System Science Data Centerhttps://www.geodata.cn/1km Precipitationthe National Earth System Science Data Centerhttps://www.geodata.cn/Educational years per capitaYunnan Provincial Bureau of Statisticshttps://stats.yn.gov.cn/Proportion of tertiary industry in GDPYunnan Provincial Bureau of Statisticshttps://stats.yn.gov.cn/Software and algorithmsArcgis10.5ESRIhttps://desktop.arcgis.com/zh-cn/desktop/index.htmlR ProjectOpen-source softwarehttps://www.r-project.org/GeodetectorOpen-source softwarehttp://www.geodetector.cn/index.html Method details Study area The southern ecological barrier zone of Yunnan Province is in southwest China (97°31′E−104°17′E, 21°08′N-25°52′N), bordering Myanmar, Vietnam, and Laos. It includes six prefectures (cities) and 24 counties (cities, districts), such as Dehong, Baoshan, Lincang, Pu’er, Xishuangbanna, and Honghe, acting as a vital ecological security barrier for China. It has a tropical climate with abundant rainfall and diverse vegetation types, with a forest coverage rate of 72.86%. The region represents a crucial carbon stock location along the Yunnan border. The terrain is predominantly composed of mountainous hills and river valley landscapes, characterized prominently by mountainous features (Figure 7). Thus, it is well-suited for analyzing carbon stock in mountainous land use. Data description and processing This study incorporates diverse datasets, such as land use remote sensing monitoring data and administrative boundary data sourced from the Resource and Environment Science Data Center, each possessing a spatial resolution of 30 m. These datasets include six primary classifications and 26 secondary classifications. The primary classifications include cropland, woodland, grassland, water body, built-up land, and unused land. The remote sensing classification accuracies for different land types are as follows: cropland and built-up land maintain accuracies of at least 95%, grassland, woodland, and water bodies achieve accuracies of at least 90%, and unused land achieves a classification accuracy of at least 85%, which fulfills the study’s criteria. The study period from 2000 to 2020 was selected based on research needs and practical requirements. Land use data were clipped for each period with ArcGIS 10.5. Subsequently, the 26 secondary classifications were merged into six categories: cropland, woodland, grassland, water bodies, built-up land, and unused land. The DEM data from the Resource and Environment Science Data Center’s ALOS (Advanced Land Observing Satellite) feature a spatial resolution of 12.5 m, characterized by horizontal and vertical accuracies within ±12.5 m. Carbon density data for diverse land categories are accessed via the China Ecosystem Research Network Data Center (http://www.nesdc.org.cn). This study explores the determinants impacting carbon stock within the Yunnan-South Border Ecological Barrier Zone, leveraging insights from pertinent literature.48,49,50,51,52,53,54,55 This study explores the impact of five environmental factors and four human activities on carbon stock within the study area. Elevation and slope information is derived from 12.5m DEM data from the designated provider. NDVI data are retrieved from the National Earth System Science Data Center (http://www.geodata.cn/) and processed using the maximum value composite technique, integrating peak monthly NDVI readings from January to December at a spatial resolution of 1km. Annual average temperature and precipitation datasets are acquired from the National Earth System Science Data Center (https://www.geodata.cn/) at the same 1km resolution. Population density statistics originate from WorldPop’s global population density maps (https://www.worldpop.org/), with time-specific datasets extracted in the ArcGIS platform, also at a 1km resolution. GDP figures are sourced from the "China GDP Spatial Grid Dataset" provided by the Chinese Academy of Sciences Resource and Environmental Data Center (https://www.resdc.cn/) at the 1km resolution. Data concerning average years of education per capita and the proportion of tertiary industry to GDP are obtained from the Yunnan Statistical Yearbook (https://stats.yn.gov.cn/). In this paper, all raster data were resampled to 12.5m using ArcGIS to ensure the consistency of the findings. Surface area calculation In this study, an estimate of the surface area is derived from the slope and direction of the slope within the image element based on the raw DEM data. This method is efficient, straightforward, and easily comprehensible. The slope and slope direction of each image element can be extracted from the 12.5m DEM data in the ArcGIS platform. The slope is used to calculate an adjustment factor for the flat area of the image element. If the slope is flat, the adjustment factor is equal to 1, and therefore, the surface area is equal to the planar area. As the slope increases, the adjustment factor equals to(Equation 1)AdjFactor=1/Cos(Slope) Then, assuming that the cell size is defined as the length along one edge of the cell and assuming that the slope is measured in degrees, the surface area can be calculated as follows:(Equation 2)SurfaceArea=∑i=1nC2/Cos(Si×(π/180)) This paper adopts the approach of Chuai Xiaowei to calculate the land surface area,25 with the formula as above. Where C represents pixel size; Si represents the slope of the pixel i. InVEST model The InVEST model incorporates multiple sub-modules, encompassing biodiversity, carbon stock, and soil conservation, enabling a thorough assessment of ecosystem services at the regional level.56,57 This research utilizes the InVEST framework’s carbon sequestration module to quantify the storage of carbon within the study area. The model categorizes carbon stock in ecosystems into four main pools: carbon stored aboveground, belowground, in soil, and from deceased organic matter.58 The equations used to estimate the overall carbon stock within the ecological barrier along the southern border of Yunnan Province are as follows:(Equation 3)Ctotali=∑j=1nAij×(Caj+Cbj+Csj+Cdj) In Equation 3, Ctotali denotes the total carbon stock for region i; Aij represents the area of land use type j within region i; Caj, Cbj, Csj, and Cdj denote the respective carbon densities of the four major carbon stock modules in the land use type j; n denotes the number of land use categories. Carbon density calibration Carbon density is a vital factor in the InVEST carbon stock model, demonstrating fluctuations due to climate, soil properties, and land use classifications. Variations across regions resulting from these factors may affect the accuracy of carbon density estimations and thereby influence research findings. Thus, accurately assessing carbon stock within a region requires calibration. Both aboveground and belowground biomass carbon densities show positive correlations with annual rainfall and average annual temperature, whereas soil organic carbon shows a positive relationship specifically with annual rainfall. Drawing from methods proposed by other experts,59,60,61 this research suggests adjustments to carbon densities based on integrating refinements from related scholars,62,63 as well as proposing specific values of carbon intensity for the whole country and Yunnan Province.14,64,65,66 The specific formula is detailed below:(Equation 4)CSP=3.3968P+3996.1(Equation 5)CBP=6.798e0.05P(Equation 6)CBT=28T+398where, CSP means soil carbon density (t/hm2) derived from mean annual rainfall (P); P stands for annual average rainfall (mm); T represents mean annual temperature (°C); CBP and CBT respectively indicate biomass carbon density (t/hm2) derived from mean annual rainfall and temperature. Yunnan Province and the Southern Border Ecological Barrier Zone have mean annual temperatures of 16.89°C and 16.24°C and mean annual precipitation of 1470.89mm and 1427.39mm, respectively. Comparing these values provides the adjustment coefficient, which when multiplied by Yunnan Province’s carbon density data, yields the carbon density information for the research area. The specific formula is detailed below:(Equation 7)KBP=C1BP/C2BP(Equation 8)KB=KBP·KBT(Equation 9)KS=C1SP/C2SP In the formula, KBP stands for biomass carbon density data adjusted by long-term average rainfall; KBT represents biomass carbon density coefficient adjusted by long-term average temperature factor; KB denotes the ratio of long-term average rainfall to long-term average temperature, which produces the correction factor for biomass carbon density; KS is the correction factor for soil carbon density derived from the ratio of study area annual average rainfall to Yunnan Province annual average rainfall; C1 and C2 denote data pertinent to the study area and national statistics (Table S1). Finally, the carbon density correction data from Table S1 and the surface area calculated from Equation 2 were input into the InVEST model to obtain the final carbon stock data. Random forest model Breiman’s random forest model, a machine learning algorithm grounded in decision tree principles, generates numerous decision trees using random samples and subsets of features from the input dataset.67,68,69 It aggregates the outcomes of these trees using random sampling of data and features to forecast the output variable. In comparison to other machine learning algorithms, the model does not require assumptions about data following specific probability distributions or being generated by models. It is characterized by strong interpretability, high tolerance for data outliers and noise, and effective avoidance of issues like variable collinearity and overfitting. In this research, the R programming language was employed to develop a random forest model, systematically evaluating the significance of explanatory factors in delineating crucial determinants affecting carbon stock in terrestrial ecosystems. This approach aims to elucidate the complex relationships between these factors and terrestrial ecosystem carbon stock, as well as to identify the factors that exert a more significant influence on terrestrial ecosystem carbon stocks. Geodetector model Geodetector employs spatial variability to uncover key drivers of carbon stock, encompassing modules for detecting factors, interactions, risks, and ecological dynamics.44,46,70 The factor detector within geodetector evaluates the extent to which each individual factor X explains spatial variations in the target variable Y. Meanwhile, the interaction detector assesses how pairs of factors collectively influence the target variable Y, encompassing various kinds of interactions such as nonlinear synergies. The metric q serves as a crucial indicator for conducting both factor and interaction detections, formulated as follows:(Equation 10)q=1−∑h=1LNhσh2/Nσh2 In the equation: h represents the stratum (partition) for either the dependent variable Y or independent variable X; Nh and N denote the number of units within stratum h and across the entire region, respectively. Additionally, σh2 and σ2 correspond to the variances of the variable Y within the stratum h and the entire region. The q value ranges between 0 and 1; a higher q value indicates a more pronounced explanatory influence of the factor on the spatial distribution of the target variable, whereas a lower value signifies a weaker influence. Quantification and statistical analysis In this study, we analyzed the data using R language, origin2020 and ArcGIS10.5. ArcGIS10.5 was used to resample, and coordinate transform all the data to unify the resolution and coordinate system of the data. Land use data were cropped and reclassified using ArcGIS10.5 to obtain land use data for the study area. Calculate the surface area using the ‘raster calculator’ in ArcGIS10.5. Random forest model was constructed using R language to screen the important influence factors. Statistical mapping was carried out using origin2020.