Dynamics of Aboveground Carbon Across Karst Terrestrial Ecosystems in China from 2015 to 2021

Abstract

Over the past half-century, environmental degradation and human disturbances have threatened the aboveground biomass carbon (AGC) in China’s karst ecosystems. However, recent ecological programs have led to environmental improvements, leaving it unclear whether China’s karst ecosystems act as an AGC sink or AGC source. In this study, we utilized L-band vegetation optical depth to quantify the dynamics of AGC across the karst regions of China from 2015 to 2021. We observed an increase in AGC density of 0.73 Mg C ha⁻¹ yr⁻¹, suggesting that karst ecosystems in China functioned as an AGC sink throughout the research period. The largest increase in AGC density, 1.29 Mg C ha⁻¹ yr⁻¹, was observed in Central China, indicating an AGC sink capacity stronger than that of other regions. Among the different land-use types, forests played a dominant role, exhibiting the largest net change in AGC density at 1.03 Mg C ha⁻¹ yr⁻¹. Furthermore, using the random forest model, temperature, soil clay content, and altitude were identified as the primary factors driving AGC changes. Our results enhance the understanding of the role of China’s karst terrestrial ecosystem in the global carbon cycle, emphasizing its contribution to the global carbon sink.

1. Introduction

The karst ecosystem covers approximately 12% of the global land area [1], playing a crucial role in the global carbon cycle through aboveground biomass carbon (AGC) [2,3]. Existing studies lack systematic analyses of the spatiotemporal dynamics of AGC and the influence of environmental factors in karst regions [4,5]. Karst regions cover 36% of China’s total land area and are primarily distributed across the southwest, northwest, and the Qinghai–Tibet Plateau [6]. Their unique geological characteristics exert constraints on vegetation growth through differentiated effects on soil formation and water supply, leading to significant regional heterogeneity in vegetation [7,8]. Furthermore, the fragility and sensitivity of karst ecosystems make their carbon sink function more susceptible to environmental changes and anthropogenic disturbances compared to other landform types [9]. In recent years, through ecological restoration projects, the karst ecosystems in Southwest China have exhibited characteristics of carbon sinks [10]. However, the overall role of AGC in karst regions of China as either a carbon sink or a carbon source remains uncertain, and the influence of environmental factors on AGC dynamics is yet to be clearly defined.

Currently, existing studies on the carbon balance in this ecosystem have mainly focused on specific regions, such as the Southwest karst region [10], without encompassing the entirety of China’s karst areas. For instance, based on ecosystem models, Luo et al. estimated an increase in AGC density of 1.74 Mg C ha⁻¹ yr⁻¹ in the karst mountainous regions of Southwest China over the period from 2001 to 2020, indicating a carbon sink [11], while Li et al. estimated that the Nanming River Basin was a carbon source, with a reduction in density of 0.07 Mg C ha⁻¹ yr⁻¹ [3]. Additionally, Lin et al. demonstrated that Guizhou Province experienced a net carbon loss, corresponding to a reduction in AGC density of 0.01 Mg C ha⁻¹ yr⁻¹ from 2010 to 2020 [12]. Qian et al. estimated that Guizhou Province acted as a carbon sink, as indicated by an increase in AGC density of 0.44 Mg C ha⁻¹ yr⁻¹ [13]. Zhang et al., based on optical remote sensing, calculated that the karst region in Northwest Guangxi act as a carbon sink from 1990 to 2005, with an increase in density of 0.70 Mg C ha⁻¹ yr⁻¹ [14]. Using low-frequency passive microwave remote sensing data, Tong et al. estimated that the Southwest China functioned as a net carbon sink, with an increase in density of 0.61 Mg C ha⁻¹ yr⁻¹ [10]. Subsequently, Fan et al. calculated an increase in AGC density of 0.65 Mg C ha⁻¹ yr⁻¹ in Southwest China from 2013 to 2021, indicating that the region functions as a carbon sink [15]. These results not only indicate the fragmented nature of the study areas, but also reveal considerable discrepancies in the findings. Thus, it is necessary to conduct a systematic analysis of the carbon dynamics across the entire karst terrestrial ecosystem.

Numerous environmental factors have been used to analyze their impact on changes in AGC, including climate [16,17], soil [18,19], and topography [20,21]. For example, in tropical and temperate forest ecosystems in China, elevation has the most significant impact on AGC, showing a negative correlation [22]. However, in the Mulun Karst forests of Southwest China, studies have shown that elevation has a direct positive effect on AGC in subtropical forest systems [23]. In the forests of Northeast China, slope has the most direct impact on AGC and exhibits a negative influence [24]. In the Qinghai–Tibet Plateau, precipitation exerts a stronger influence on the spatial distribution of AGC compared to temperature and topographic factors [25]. Although these studies illuminate the impact of environmental factors on AGC in various regions, how these factors drive AGC dynamics in China’s karst regions remains poorly understood.

Previous studies monitoring AGC dynamics have primarily relied on field surveys, model-based inversions, optical remote sensing data, and Light Detection And Ranging (LiDAR) data. Field survey methods provide the most precise AGC estimations, but they are time-consuming, labor-intensive, and unsuitable for covering large areas [26,27]. Model-based inversion methods can provide relatively quick and accurate AGC estimates for specific regions and ecosystems. Yet, the precision of model parameters and the appropriateness of input data play a critical role in determining their effectiveness [28,29]. The method of deriving vegetation indices [30,31,32,33] from optical remote sensing data allows for the extensive and continuous monitoring of AGC dynamics over large areas. However, atmospheric factors like aerosols and clouds constrain optical observations, and they may become saturated in regions with moderate to dense vegetation cover [34]. Since forest height serves as a reliable proxy for AGC, LiDAR observations provide highly accurate data on forest parameters that can be used to estimate global AGC stocks [35,36]. However, it was not until 2019, following the initiation of the Global Ecosystem Dynamics Investigation (GEDI), that a satellite-based LiDAR system specifically designed for estimating vegetation AGC became available [37].

Passive microwave remote sensing at low frequencies (L-bands) is less susceptible to saturation effects, which occur at approximately 400 Mg/ha [38], making it an effective method for AGC monitoring in terrestrial ecosystems [39,40,41]. The low-frequency L-band Vegetation Optical Depth (L-VOD), derived through passive microwave satellite observations, show a strong correlation with vegetation biomass and have become a key method for tracking annual AGC dynamics in terrestrial ecosystems [42,43]. With several years of L-VOD data available, this method enables accurate AGC estimation and facilitates the analysis of AGC variations with time.

Utilizing L-VOD data, this research aims to estimate AGC and their changes across all karst regions in China, including those dominated by limestone and dolomite, from 2015 to 2021. The objectives are to (a) characterize the spatial patterns of AGC, (b) analyze the spatiotemporal variations in AGC, and (c) identify the primary environmental factors driving AGC changes.

2. Materials and Methods

2.1. Study Area

Karst landforms are extensively distributed across China, covering an area of approximately 3.44 × 10⁶ km², which accounts for 36% of the country’s landmass [44] (Figure 1). Drawing from spatial data statistics, the total exposed and buried karst area in China measure $2.06\times {10}^6$ km² and $1.38\times {10}^6$ km², respectively [45]. The karst landforms are found in almost all provinces of China, but are predominantly found in Guangxi, Yunnan, and Guizhou (Figure 1). The karst landforms in these regions cover an extensive area of $5.4\times {10}^5$ km², forming one of the largest continuous karst regions globally [46].

The climatic characteristics in China’s karst regions show significant variation. In the karst regions of South and East China, the climate tends to be warm and humid, with annual mean rainfall between 1000 mm and 2000 mm and annual mean temperatures ranging from 20 °C to 25 °C [47]. In Northern China’s karst regions, the climate is typified by arid and cold weather patterns, with annual precipitation of less than 500 mm and lower temperatures [47]. Additionally, the karst regions in the Qinghai–Tibet Plateau show extremely dry and cold climate conditions, with annual precipitation below 200 mm and temperatures often falling below 0 °C [47].

Due to the diverse climatic conditions, vegetation types in the karst regions exhibit diversity. Sparse vegetation is primarily found in the western region, grasslands dominate the Qinghai–Tibet Plateau, and forests are predominantly distributed in the central and northern areas [48]. Vegetation cover in western is considerably lower than that in the eastern regions [49].

For the purpose of analyzing the temporal and spatial variations in AGC across different karst regions, the research region was segmented into seven distinct regions based on geographical location, natural, and human conditions, including North China (NC), Northeast China (NEC), East China (EC), Central China (CC), South China (SC), Southwest China (SWC), and Northwest China (NWC) (Table A1).

2.2. Datasets

2.2.1. Land-Use Map

The Climate Change Initiative (CCI) land-use map [50] was used to analyze the AGC dynamics across different land-use types within China’s karst regions. The spatial resolution of this map was 300 m. This product was generated through an unsupervised classification process based on the Glob Cover system, incorporating numerous Earth observation datasets from ESA’s Glob Cover initiative. The overall accuracy of the CCI land cover map was reported as 75.4%, which was calculated using a confusion matrix generated from an independent validation dataset. Compared to other land cover products such as GlobCover, MCD12Q1, and GlobeLand3, the CCI map shows the highest accuracy in the karst regions [51,52].

For this analysis, the 300 m land cover data were aggregated to a 0.25 degree resolution by applying a majority rule. The aggregated CCI land cover map contains various types, including forest, grassland, cropland, and others (Table A2). This study primarily focuses on the land cover types of forest, cropland, grassland, and shrubland, and other land cover types were masked.

2.2.2. L-VOD

VOD is a vegetation parameter that quantifies the attenuation effect of microwaves penetrating the canopy [53]. It is closely associated with vegetation density, biomass, and water content [53]. The L-VOD dataset, derived from passive microwave observations at the L-band frequency (1.4 GHz), was utilized to assess annual AGC changes from 2015 to 2021. We used the L-VOD product (version 2) retrieved from the Soil Moisture and Ocean Salinity satellite (SMOS). This L-VOD product was developed using the SMOS INRA-CESBIO (SMOS-IC) algorithm [38]. The SMOS-IC product provides daily L-VOD data spanning from 2010 to 2021 at a spatial resolution of 0.25 degrees. These L-VOD products have been used to quantify annual AGC changes across various major bioclimatic regions [39,42,54].

To ensure the best data quality, we excluded daily L-VOD pixels that were impacted by frozen ground, water bodies, and steep slopes. All valid daily L-VOD data for every month were averaged to calculate the monthly L-VOD values. Given the unreliability of L-VOD retrieval under frozen or snow-covered conditions (i.e., when surface temperatures are below 0 °C), the mean L-VOD calculated during the months from June to September was used to represent the annual L-VOD.

2.2.3. AGC Benchmark Map

The Saatchi aboveground biomass (AGB) benchmark map was utilized to calibrate the relationship between AGC and L-VOD. This map, developed by Saatchi et al. [36], was generated by combining GLAS LiDAR measurements, which provided critical parameters such as canopy height with spatial data from Landsat, MODIS, QuickSCAT, the Shuttle Radar Topography Mission (SRTM), and the Advanced Land Observing Satellite (ALOS). It was primary designed to estimate AGB in pantropical forests during the early 2000s. The spatial resolution of this dataset is 1 km. In this study, we utilized a current version of this dataset [55], which offers estimates of global AGB for the year 2015.

The Saatchi map was resampled to a spatial resolution of 0.25 degrees using a simple averaging method. Additionally, the biomass density values (Mg/ha) in this map were transformed into carbon density values (Mg C/ha) using a conversion ratio of 0.5.

2.2.4. Environmental Parameters

In karst regions, vegetation growth is affected by the interplay of topographical, climatic, and soil factors [20,21,24]. To analyze the impact of these factors on changes in AGC, nine environmental parameters were selected, including altitude, slope, aspect, precipitation (Pre), temperature (Temp), soil moisture (SM), vapor pressure deficit (VPD), soil organic carbon content (SOC), and soil clay content (SClay).

The altitude data were obtained from the NASADEM_HGT V001 [56] dataset, which is an improved version of the SRTM dataset. This dataset was formed by consolidating auxiliary data from different sources, including ASTER GDEM, ICESat GLAS, and PRISM, to improve its accuracy. The original elevation data, with a spatial resolution of 30 m, was used to derive the slope and aspect through standard terrain analysis methods, and then aggregated to a 0.25 degree resolution by averaging.

The precipitation, temperature, and soil moisture data were derived from the fifth-generation European Centre for Medium-Range Weather Forecasts re-analysis (ERA5) monthly average dataset, which has a spatial resolution of 0.25 degree [57]. Vapor pressure deficit was calculated using the surface pressure, the temperature, and dewpoint temperature, based on the ERA-Interim approach outlined by Yuan et al. [58], as follows:

(1)$VPD=SVP-AVP$

(2)$AVP=6.112\times f_w\times e^{{\displaystyle \frac{17.67T_d}{T_d+243.5}}}$

(3)$SVP=6.112f_w\times e^{{\displaystyle \frac{17.67T_a}{T_a+243.5}}}$

(4)$f_w=1+7\times {10}^{-4}+3.46\times {10}^{-6}{P}_{mst}$

(5)$P_{mst}=P_{msl}{\left({\displaystyle \frac{T_a+273.16}{\left(T_a+273.16\right)+0.0065\times Z}}\right)}^{5.625}$

where $SVP$ and $AVP$ are saturated vapor pressure and actual vapor pressure (Kpa), and $T_a$ is the land air temperature (°C). $T_d$ is the dew point temperature (°C). $Z$ is the altitude (m). $P_{mst}$ is the air pressure (hPa), and $P_{msl}$ is the air pressure at mean sea level (1013.25 hPa). $T_a$, $T_d$, and the $P_{mst}$ were derived from the ERA5 monthly average dataset. The monthly precipitation, temperature, soil moisture, and vapor pressure deficit data were processed into yearly averages using a simple aggregation method.

Soil organic carbon content and soil clay content data measured 30 cm deep were sourced from SoilGrids 2.0, a data-driven project by the International Soil Reference and Information Centre (ISRIC). These data were predicted using a random forest approach within a digital soil mapping framework [59]. To align the resolution of soil clay content and soil organic carbon content data (originally at a resolution of 250 m) with the AGC data, these values were resampled at a resolution of 0.25 degrees, employing the mean value method.

2.3. Methods

The method for calculating yearly AGC from L-VOD follows the approach outlined by Fan et al. [60]. AGC at 0.25 degrees was estimated from L-VOD through an empirical calibration equation (Equation (6)), expressed as follows:

(6)$AGC=a\times VO{D}^b$

where $a$ and $b$ are the optimal parameters. We used Saatchi AGC benchmark map and L-VOD data in 2015 to calibrate Equation (6). The calibrated relationship between L-VOD and Saatchi AGC benchmark map are shown in Figure A1, Appendix A. Subsequently, the calibrated equation was applied to calculate the annual AGC of the karst terrestrial ecosystem in China from 2015 to 2021. To compute the AGC stock, the AGC density of each pixel was multiplied by its corresponding area.

To assess the calibration errors and the uncertainties in AGC estimates, a bootstrap cross-validation method [42] was utilized. We calculated the correlation coefficient ($R$) and root mean square error (RMSE) between the benchmark AGC and the bootstrapped AGC estimates to evaluate the calibration errors. Furthermore, the 95% bootstrap confidence interval for AGC estimates based on AGC benchmark maps was also determined. The results showed that high R (R = 0.66, $p<0.01$) and low RMSE (RMSE = 0.04 Pg C) between the benchmark AGC map and the bootstrapped AGC estimates, as well as the retrieved AGC, display a small 95% bootstrap confidence interval (from 3.44 Pg C to 3.56 Pg C), suggesting that errors caused by calibration are limited.

2.4. Statistical Analysis

To analyze the spatiotemporal variation in AGC, we computed AGC variations across different land cover types and natural geographical regions of China. A one-way analysis of variance (ANOVA) was used to assess whether the differences in AGC among these regions were significant ($p<0.05$). We also used the standard deviation to represent the uncertainties of AGC change for different regions and vegetation types, which was calculated based on the bootstrap cross-validation method (sampling rate = 80 %, iterations = 1000).

2.5. Trend Analysis Methods

Tests for the detection of significant trends in time series can be classified as parametric and non-parametric methods. Parametric trend tests require data to be independent and normally distributed, while non-parametric trend tests require only that the data be independent [61]. In this study, the Mann–Kendall test [62] and Sen’s slope estimator [63], two non-parametric methods, were combined. The Mann–Kendall test was used to determine significance, with the significance levels $\alpha =0.01$ and $\alpha =0.05$. Sen’s slope estimator was used to assess the trend direction and quantify its strength.

2.6. Random Forest Model

Tree-based machine learning models have become widely recognized for their effectiveness in predicting and attributing ecosystem dynamics [64,65]. Random Forest (RF), a popular ensemble learning method built on decision trees, is frequently applied in both classification and regression tasks [66]. This method effectively mitigates overfitting by incorporating randomness while processing multiple decision trees in parallel [67]. Recently, RF has been utilized within the SHAP framework for generating localized interpretations of prediction tree models [68].

The environmental factors utilized in the simulation, as described in Section 2.2.4, include elevation, slope, aspect, precipitation, temperature, soil moisture, vapor pressure deficit, soil organic carbon content, and soil clay content. To accurately capture the complex nonlinear relationships within ecosystems and the combined effect of these factors on annual AGC change, the RF algorithm, built on pixel-level data, was selected as the model. By integrating multi-source environmental variables, we simulated AGC effectively. Prior to model training, we conducted a reasonable division of the dataset, using 80% of the data to train the random forest and the remaining 20% for validation. The hyperparameters of the RF model were optimized by introducing the GridSearchCV technique, using the goodness of fit $R^2$ as the evaluation metric, in order to obtain an optimally performing model.

2.7. SHAP Analysis

We utilized the TreeExplainer SHAP framework, introduced by Lundberg et al. [68], to generate localized interpretations of RF models and analyze the effects of environmental factors on AGC. The SHAP method is based on the Shapley value concept from game theory [69], which can be viewed as an improvement over the local interpretable model-agnostic explanations approach [70]. The importance of each feature was determined by calculating the average of its absolute SHAP values, providing a clear measure of their impact. To gain a deeper understanding of how features in the dataset influence the model’s output, a bee swarm plot is utilized to provide a densely informative visualization of the impact of each feature on the model’s predictions. In this context, positive SHAP values indicate an increase in AGC predictions, while negative values indicate a decrease caused by the feature.

3. Results

3.1. Spatial Patterns of AGC

During the years from 2015 to 2021, the mean AGC density estimated using L-VOD in the karst regions of China was 41.07 Mg C/ha. The region with the highest AGC density (61.98 Mg C/ha) is located in South China. Meanwhile, the region with the lowest AGC density (11.23 Mg C/ha) is mainly distributed in North China (Figure 2a).

In terms of latitudinal distribution, AGC density showed a decreasing trend with increasing latitude. The observed AGC density peaked at 93.73 Mg C/ha between 24° N and 25° N, while the lowest density of 7.73 Mg C/ha was approximately at 35° N (Figure 2b). Regarding longitudinal distribution, the highest AGC density of 93.73 Mg C/ha was found between 107° E and 108° E, while the lowest density of 1 Mg C/ha was observed between 88° E and 89° E.

The AGC density distribution showed significant variation across different geographic regions. South China (61.98 Mg C/ha) has the highest average AGC density, followed by Southwest China (50.58 Mg C/ha)) and Central China (48.59 Mg C/ha) (Figure 3a). The lowest AGC density is observed in Northeast China (8.48 Mg C/ha) (Figure 3a).

The spatial distribution characteristics of AGC storage demonstrate a different pattern compared to AGC density. The highest AGC stock was found in Southwest China (2.03 Pg C), followed by South China (0.61 Pg C), and the lowest AGC stock was observed in Northeast China (0.03 Pg C) (Figure 3c).

Among different land-use types, forests exhibited the highest mean AGC density (54.94 Mg C/ha), followed by cropland (46.29 Mg C/ha), and grasslands had the lowest AGC density (22.28 Mg C/ha) (Figure 3b). Similarly, forests also had the largest AGC stock of 1.44 Pg C, and grassland had the smallest AGC stock of 0.30 Pg C (Figure 3d).

3.2. Spatiotemporal Dynamics of AGC

From 2015 to 2021, AGC in the karst regions of China showed a net change of 0.063 Pg C yr⁻¹ (Figure 4a), reflecting a balance between total AGC gains of 0.085 Pg C yr⁻¹ and total AGC losses of 0.022 Pg C yr⁻¹. This result suggests that the terrestrial ecosystem in the karst regions of China served as a net AGC sink during the study period. Rapid AGC increase was observed from 2015 to 2020 (0.063 Pg C yr⁻¹). Note that there was a notable AGC decline from 2020 to 2021 (0.057 Pg C).

Further analysis of AGC change across different land-use types revealed that forests experienced the largest net increase in AGC, with a change of 0.027 Pg C yr⁻¹, accounting for 42.9% of the total increment (Figure 4b). This result suggests that forests play a critical role in carbon sequestration and storage within the karst regions. Meanwhile, croplands showed a net change of 0.015 Pg C yr⁻¹, while grasslands exhibited the smallest net increase in AGC, with only 0.006 Pg C yr⁻¹ (Figure 4b).

From a regional perspective, the Southwest karst region exhibited the largest net AGC increase, reaching 0.039 Pg C yr⁻¹, which accounts for 61.9% of the total AGC increment in the study region (Figure 4c). This result suggests that the AGC changes in the Southwest karst region dominate the AGC changes in the karst regions of China, which is due to the high proportion of karst landforms in the Southwest. Following this, the Central China karst region and the South China karst region showed a net AGC change of 0.012 Pg C yr⁻¹ and 0.008 Pg C yr⁻¹, respectively. And the East China and North China karst regions showed an AGC balance during the study period (Figure 4c).

In terms of AGC density, the net change in AGC density across the study region was 0.73 Mg C ha⁻¹ yr⁻¹. Regionally, the highest net change in AGC density was observed in Central China, reaching 1.29 Mg C ha⁻¹ yr⁻¹, followed by Southwest China with a value of 1.26 Mg C ha⁻¹ yr⁻¹(Figure 5a). In terms of land-use types, forests exhibited the largest net change in AGC density at 1.03 Mg C ha⁻¹ yr⁻¹, followed by croplands at 0.87 Mg C ha⁻¹ yr⁻¹, while grasslands showed the lowest value at 0.46 Mg C ha⁻¹ yr⁻¹ (Figure 5b).

Spatially, about 75% of the study region showed a net increase in AGC from 2015 to 2021 (Figure 6a), with a widespread AGC increase observed in the Southwestern region. In addition, about 25% of the study area showed a net AGC decrease, mainly occurring in Yunnan, Gansu, and Xinjiang (Figure 6a). The trends indicated that significant increases (red, 3) are primarily concentrated in the Central, Southwestern, and Southern regions (Figure 7). Slightly significant increases (blue, 2) are more scattered, appearing sporadically across other karst areas (Figure 7). From a latitudinal perspective, the increase in AGC predominantly occurs between 23° N and 34° N, while the decrease is mainly observed between 35° N and 45° N (Figure 6b).

3.3. Environmental Factors Driving AGC Changes

Temperature, soil clay content, altitude, and precipitation are the four most important environmental factors influencing AGC, whereas soil moisture, vapor pressure deficit, and aspect have relatively minor impacts (Figure 8a). The response functions of the trained random forest model are illustrated through partial dependence plots, which explain the response of AGC to the four most important influencing factors. The effects of precipitation and temperature on AGC are monotonic, with higher values of both variables positively correlating with AGC (Figure 8b,e). At altitudes below 1000 m, the relationship between altitude and AGC is positive, while at altitudes above 1000 m, this relationship becomes negative (Figure 8d). The connection between soil clay content and AGC exhibits a nonlinear pattern. AGC increases slowly at low levels of soil clay content, but as soil clay content approaches 25%, AGC increases sharply; beyond this threshold, AGC growth begins to plateau, eventually exhibiting a gradual declining trend (Figure 8c). This indicates that soil clay content is beneficial for enhancing AGC within a specific range, although its marginal effect diminishes as soil clay content continues to increase.

The SHAP values for temperature indicate that low temperature could lead to greater variability in AGC dynamics, characterized by the extended left tails in the SHAP value distribution (Figure 9). The SHAP values of precipitation exhibit a similar trend (Figure 9). Low soil clay content and soil organic content also induce larger AGC changes (Figure 9). Furthermore, altitude displays a relatively balanced pattern across both high- and low-altitude values (Figure 9).

4. Discussion

4.1. AGC Dynamics from 2015 to 2021

During the study period, the karst terrestrial ecosystem in China had an average AGC density of 41.07 Mg C/ha, closely aligning with the GEDI estimate of 38 Mg C/ha [71]. To further evaluate the accuracy of the AGC obtained from L-VOD, we analyzed the spatial correlation between the AGC derived from L-VOD and the AGC from GEDI during the period from 2019 to 2021. The results show that the AGC from L-VOD demonstrates high accuracy, with a correlation coefficient (R) of 0.798 and a Root Mean Square Error (RMSE) of 16.16 Mg C/ha (Figure A2).

Our results indicate that the net AGC in the karst regions of China increased by 0.063 Pg C yr⁻¹ from 2015 to 2021, which is consistent with previous studies. Xu et al. [72] reported an AGC increase of 0.061 Pg C yr⁻¹ from 2015 to 2019, while Liu et al. [43] observed a similar increase of 0.063 Pg C yr⁻¹ over the period from 1993 to 2012. Additionally, our findings show that the AGC density increased by 0.73 Mg C ha⁻¹ yr⁻¹, which is in close agreement with Xu et al.’s estimate of 0.7 Mg C ha⁻¹ yr⁻¹ [72]. Our estimate for net AGC change in forest regions from 2015 to 2021 was 0.027 Pg C yr⁻¹, closely aligning with the estimate of 0.024 Pg C yr⁻¹ reported by Chen et al. [73]. We also estimated that the AGC density in forests increased by 1.03 Mg C ha⁻¹ yr⁻¹, which is consistent with Xu et al.’s estimate of 1.0 Mg C ha⁻¹ yr⁻¹ [72]. These results suggest that China’s karst terrestrial ecosystem has functioned as a carbon sink in recent years.

We estimated the total AGC loss is 0.022 Pg C yr⁻¹, which is higher than the estimate by Harris et al. of 0.016 Pg C yr⁻¹ [74]. Harris et al. underestimated the carbon losses compared to our findings. This may be due to their focus on carbon loss from stand-replacement disturbances while neglecting the losses caused by forest degradation [75].

Note that our results indicate a decline in AGC in 2021, which is consistent with the observations of Chen et al. [73]. The decline in AGC in 2021 can be primarily attributed to the extreme drought events in the Southwestern karst region [76]. In the spring of 2021, Southwest China experienced a severe drought, resulting in a severe soil moisture deficit, and soil moisture levels in the region did not return to normal until summer [77]. On one hand, the karst region’s soil layer is typically shallow and has poor water retention capacity, with drought events further depleting soil moisture and inhibiting plant growth. On the other hand, the impact of drought on spring phenology can directly reduce summer crop yields and lead to tree mortality [78].

4.2. Key Environmental Factors Influencing AGC Dynamics

Based on our findings, the primary factors influencing changes in AGC include temperature, soil clay content, altitude, and precipitation. The plant growth cycle and photosynthesis are directly affected by temperature and precipitation [48]. Our results show that increases in temperature and precipitation enhance carbon absorption and storage, which aligns with previous studies on forests and grasslands in China [79,80]. Similarly, research by Qian et al. [13] and Liu et al. [81] has demonstrated a positive correlation between AGC and both temperature and precipitation in karst forests, further supporting our conclusions. This suggests that rising temperatures and increased water availability can boost AGC levels in karst regions. Note that excessive temperatures can enhance evapotranspiration, diminish soil moisture, and impede plant growth [82]. As a result, these positive effects may exhibit nonlinear behavior under extreme climate changes, warranting further investigation in future climate scenarios.

The influence of soil clay content on AGC is complex. Our analysis demonstrates that vegetation’s capacity for carbon sequestration grows with increasing soil clay content, provided it is less than 25%. This may be because when soil clay content is below this threshold, high soil clay content usually has better water retention, helping to maintain soil moisture under drought conditions and providing enough water for plaint growth [83,84]. Yet, when soil clay content exceeds this threshold, the stimulatory effect diminishes. This may be attributed to high soil clay content may lead to poor drainage and cause root hypoxia, which inhibits plant growth [85,86,87].

Topographical factors, particularly elevation and slope, influenced the availability of resources like sunlight, water, and soil nutrients [88]. According to our study, the relationship between elevation and AGC exhibits a clear threshold around 1000 m; below this threshold, elevation and AGC are positively correlated, while above it, the relationship becomes negative. This finding agrees with the research by Zeng et al. and Jia et al., which also identified the negative impact of high elevations on carbon sequestration [24,25]. Liu et al.’s study on the Mulun Nature Reserve [23] demonstrated that, from an elevation of 250 m to 1028 m, AGC increases with elevation, further supporting our results. This could be associated with the sharp decline in temperature and soil moisture at higher elevations, which limits plants’ carbon absorption capacity and subsequently reduces AGC storage.

4.3. Limitations and Future Perspectives

The L-VOD data used in this study have a coarse spatial resolution of 0.25 degrees, limiting the ability to estimate AGC dynamics at finer scales [60]. Aggregating data of different resolutions to 0.25 degrees may introduce uncertainties. For instance, resampling the CCI land cover map from 300 m to 0.25 degrees may lead to mixed effects and reduce the accuracy of the analysis. In addition, this study analyzes AGC changes based on the 2015 land cover map and does not take into account changes in land cover types over time, assuming no changes in land-use over the study period. This assumption may introduce uncertainties in the analysis of long-term AGC changes.

The current discussion of AGC drivers does not adequately address the coupling effects between factors. For instance, vapor pressure deficit and soil moisture are typically coupled through interactions between the land and atmosphere. The complexity of these interactions warrants further investigation to better understand their impact on AGC dynamics. Future research should focus on a more detailed analysis of these coupling effects to improve the accuracy of identifying AGC driving factors.

Additionally, different vegetation types and land-use changes may lead to differences in AGC responses across surface types, with comparative analyses between plantation forests and natural forests providing valuable insights for identifying various drivers more accurately. Given the distinct variations in natural conditions and anthropogenic activities across regions, the relative importance of drivers also varies. Future studies should conduct more regional, partitioned analyses to improve the applicability and interpretability of the findings across different regions.

5. Conclusions

This study estimates the AGC storage in the karst terrestrial ecosystem of China from 2015 to 2021 using the L-VOD and systematically analyzes its spatial distribution and spatiotemporal variation characteristics. Our analysis reveals an average AGC density of 41.07 Mg C/ha in the study area, showing significant variability based on latitude, region, and land-use categories. Yearly changes in AGC showed a net increase of 0.063 Pg C yr⁻¹ in China’s karst ecosystem from 2015 to 2021. The Southwestern region contributed the largest share, reaching 0.039 Pg C yr⁻¹. Among different land-use types, forests exhibited the largest increase in AGC stock, reaching 0.027 Pg C yr⁻¹, suggesting that forests are the primary contributors to the enhancement of AGC sinks. For AGC density, the net change in AGC density across the study region was 0.73 Mg C ha⁻¹ yr⁻¹. Central China showed the largest increase of 1.29 Mg C ha⁻¹ yr⁻¹, indicating a stronger carbon sink capacity than other regions. Among land-use types, forests exhibited the largest net change in AGC density at 1.03 Mg C ha⁻¹ yr⁻¹. These findings further highlight that strengthening sustainable forest management can significantly increase regional AGC storage and thereby effectively mitigate climate change.

We analyzed the most significant environmental factors influencing AGC changes, which primarily include temperature, soil clay content, altitude, and precipitation. Future studies should aim to uncover the underlying mechanisms by which these factors interact with one another to influence AGC, particularly under varying climatic conditions and land-use practices in the karst regions.

Author Contributions

Conceptualization, J.S. and L.Y.; methodology, J.S., L.Y., H.F., K.Z., J.-P.W., X.L., C.L., Y.J., D.W. and T.C.; software, J.S., H.F. and L.Y.; formal analysis, J.S. and L.Y.; data curation, J.S., L.Y. and H.F.; Validation, J.S. and L.Y.; Visualization, J.S. and L.Y.; writing—original draft preparation, J.S.; writing—review and editing, J.S. and L.Y.; supervision, L.Y.; funding acquisition, D.W. and T.C. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The CCI land cover map is available in ESA CCI Land cover website at https://www.esa-landcover-cci.org/, (accessed on 20 August 2024); The ERA5 data are available at http://cds.climate.copernicus.eu, (accessed on 25 August 2024).

Conflicts of Interest

Y.J. and D.W. are employed by Fujian Space Carbon Co., Ltd. Their employer’s company was not involved in this study, and there is no relevance between this research and their company. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as potential conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures

Figure 1. The spread of karst landforms within China.

Figure 2. AGC density spatial distribution. (a) Mean AGC density over the 2015–2021 timeframe. (b) Latitude-dependent fluctuations in AGC density. (c) Longitude-dependent fluctuations in AGC density.

Figure 3. Variations in AGC density across different geographic regions (a) and land-use types (b), as well as AGC stock distribution (c,d). Letters a–g indicate that identical letters represent no significant difference, while different letters indicate a significant difference ([Forumla omitted. See PDF.]). Error bars represent the standard deviation (std).

Figure 4. Dynamic changes in AGC from 2015 to 2021. (a) Yearly AGC changes across the study region. (b) Yearly AGC changes across different land-use types. (c) Yearly AGC changes across seven geographical regions. Error bars represent the standard deviation (std).

Figure 5. Regional and land-use-based changes in AGC density in the karst regions of China from 2015 to 2021. (a) Changes across different regions. (b) Changes across different land-use types.

Figure 6. Spatial changes in AGC density in the karst regions of China from 2015 to 2021. (a) Net AGC density change during 2015–2021. (b) Latitudinal variation in net AGC density change from 2015 to 2021.

Figure 7. AGC trends in China’s karst regions from 2015 to 2021. Categories were defined based on the criteria outlined in Table A3, where 1 indicates a non-significant trend, 2 represents a slightly significant increase, and 3 denotes a significant increase.

Figure 8. Impact of environmental factors on AGC: SHAP values and partial dependence. (a) Bar chart displaying these average magnitudes of SHAP values for each environmental factor. (b–e) The marginal effect on AGC of Temp (b), SClay (c), Altitude (d), and Pre (e). The lines illustrate the average response in the random forest model for a specific variable, keeping the other variables at different values.

Figure 9. A collection of bee swarm plots. Each dot’s position along the x-axis indicates the influence of a variable on the RF model’s prediction for a specific sample. When dots overlap at the same x-coordinate, they accumulate to reflect the density of the impact.

Appendix A

Figure A1. Relationships between annual L-VOD in 2015 and the Saathci AGC benchmark map. The fitted curve (blue line) was obtained using Equation (6) in the main text. Each dot represents an individual data point, and asterisks indicate statistical significance with [Forumla omitted. See PDF.].

Figure A2. A comparison between the AGC derived from L-VOD and the AGC from GEDI during the period from 2019 to 2021. The 1:1 line indicates the perfect correlation where both variables have equal values, and each dot represents an individual data point.

Table A1

The seven regions and their included provinces.

Regions	Province
North China	Beijing, Tianjin, Hebei, Shanxi, Neimenggu
Northeast China	Liaoning, Jilin, Heilongjiang
East China	Shandong, Jiangsu, Anhui, Zhejiang, Fujian, Shanghai, Taiwan, Jiangxi
Central China	Hubei, Hunan, Henan
South China,	Guangdong, Guangxi, Hainan, Hongkong, Macau
Southwest China,	Sichuan, Yunnan, Guizhou, Xizang, Chongqing
Northwest China	Ningxia, Xinjiang, Qinghai, Shaanxi, Gansu

Table A2

The land-use types in the aggregated 0.25 degree CCI land cove map.

Land-Use Types	Code	Description
Cropland	10, 11	Rainfed cropland
	20	Irrigated cropland
	30	Mosaic cropland (>50%)/natural vegetation (tree, shrub, herbaceous cover) (<50%)
Forest	50	Tree cover, broadleaved, evergreen, closed to open (>15%)
	60, 61	Tree cover, broadleaved, deciduous, closed to open (>15%)
	70	Tree cover, needleleaved, evergreen, closed to open (>15%)
	80	Tree cover, needleleaved, deciduous, closed to open (>15%)
	100	Mosaic tree and shrub (>50%)/herbaceous cover (<50%)
	160	Tree cover, flooded, fresh, or brakish water
Shrubland	121	Shrubland
Grassland	130	Grassland
Sparse vegetation	150	Sparse vegetation (tree, shrub, herbaceous cover)
Wetland	180	Shrub or herbaceous cover, flooded, fresh-saline, or brakish water
Settlement	190	Urban
Bare area	200, 201	Bare areas
Water	210	Water

Table A3

Trend categories based on a combination of Mann–Kendall Test and Sen’s Slope Estimator.

$\mathsf{\beta }$	${\mathit{Z}}_{\mathit{s}}$	Trend Type	Trend Features
$\mathsf{\beta }>0$	$2.58<{\mathrm{Z}}_s$	4	Extremely significant increase
	$1.96<{\mathrm{Z}}_s\le 2.58$	3	Significant increase
	$1.65<{\mathrm{Z}}_s\le 1.96$	2	Slightly significant increase
	${\mathrm{Z}}_s\le 1.65$	1	Non-significant increase
$\mathsf{\beta }>0$	${\mathrm{Z}}_s$	0	No change
$\mathsf{\beta }>0$	${\mathrm{Z}}_s\le 1.65$	−1	Non-significant decrease
	$1.65<{\mathrm{Z}}_s\le 1.96$	−2	Slightly significant decrease
	$1.96<{\mathrm{Z}}_s\le 2.58$	−3	Significant decrease
	$2.58<{\mathrm{Z}}_s$	−4	Extremely significant decrease

References

1. Liu, C. Biogeochemical Processes and Cycling of Nutrients in the Earth’s Surface: Cycling of Nutrients in Soil–Plant Systems of Karstic Environments, Southwest China; Science Press: Beijing, China, 2009; Volume 33, pp. 698-705.

2. Schlesinger, W.H. Carbon balance in terrestrial detritus. Annu. Rev. Ecol. Syst.; 1977; 8, pp. 51-81. [DOI: https://dx.doi.org/10.1146/annurev.es.08.110177.000411]

3. Li, Y.; Geng, H. Spatiotemporal trends in ecosystem carbon stock evolution and quantitative attribution in a karst watershed in southwest China. Ecol. Indic.; 2023; 153, 110429. [DOI: https://dx.doi.org/10.1016/j.ecolind.2023.110429]

4. Ansari, F.; Narendra, B.H.; Putri, I.A.S.L.P.; Tata, H.L.; Susi Dharmawan, I.W.; Rachmat, H.H.; Suharti, S.; Windyoningrum, A.; Khotimah, H.; Sayektiningsih, T. et al. Forest cover change and its carbon dynamic of the karst area in Bulusaraung, South Sulawesi, Indonesia. For. Sci. Technol.; 2024; 20, pp. 179-193. [DOI: https://dx.doi.org/10.1080/21580103.2024.2343344]

5. Haryono, E.; Widyastuti, M. Potential of Carbon Stocks and Its Economic Values in Tropical Karst Landscape (Case Study in Biduk-Biduk Karst, East Kalimantan, Indonesia). J. Phys. Conf. Ser.; 2019; 1373, 012030.

6. Jiang, Z.; Lian, Y.; Qin, X. Rocky desertification in Southwest China: Impacts, causes, and restoration. Earth-Sci. Rev.; 2014; 132, pp. 1-12. [DOI: https://dx.doi.org/10.1016/j.earscirev.2014.01.005]

7. Pei, J.; Wang, L.; Wang, X.; Niu, Z.; Kelly, M.; Song, X.; Huang, N.; Geng, J.; Tian, H.; Yu, Y. Time series of Landsat imagery shows vegetation recovery in two fragile karst watersheds in southwest China from 1988 to 2016. Remote Sens.; 2019; 11, 2044. [DOI: https://dx.doi.org/10.3390/rs11172044]

8. Qiao, Y.; Chen, H.; Jiang, Y. Quantifying the impacts of lithology on vegetation restoration using a random forest model in a karst trough valley, China. Ecol. Eng.; 2020; 156, 105973. [DOI: https://dx.doi.org/10.1016/j.ecoleng.2020.105973]

9. Brandt, M.; Yue, Y.; Wigneron, J.P.; Tong, X.; Tian, F.; Jepsen, M.R.; Xiao, X.; Verger, A.; Mialon, A.; Al-Yaari, A. et al. Satellite-Observed Major Greening and Biomass Increase in South China Karst During Recent Decade. Earth’s Future; 2018; 6, pp. 1017-1028. [DOI: https://dx.doi.org/10.1029/2018EF000890]

10. Tong, X.; Brandt, M.; Yue, Y.; Horion, S.; Wang, K.; Keersmaecker, W.D.; Tian, F.; Schurgers, G.; Xiao, X.; Luo, Y. et al. Increased vegetation growth and carbon stock in China karst via ecological engineering. Nat. Sustain.; 2018; 1, pp. 44-50. [DOI: https://dx.doi.org/10.1038/s41893-017-0004-x]

11. Luo, D.; Zhou, Z.; Zhang, L.; Chen, Q.; Huang, D.; Feng, Q.; Wu, T.; Wu, L. Evolution and driver analysis of forest carbon stocks in karst mountainous areas of southwest China in the context of rocky desertification management. Catena; 2024; 246, 108335. [DOI: https://dx.doi.org/10.1016/j.catena.2024.108335]

12. Lin, T.; Wu, D.; Yang, M.; Ma, P.; Liu, Y.; Liu, F.; Gan, Z. Evolution and Simulation of Terrestrial Ecosystem Carbon Storage and Sustainability Assessment in Karst Areas: A Case Study of Guizhou Province. Int. J. Environ. Res. Public Health; 2022; 19, 16219. [DOI: https://dx.doi.org/10.3390/ijerph192316219] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36498292]

13. Qian, C.; Qiang, H.; Zhang, G.; Li, M. Long-term changes of forest biomass and its driving factors in karst area, Guizhou, China. Int. J. Distrib. Sens. Netw.; 2021; 17, pp. 127-134. [DOI: https://dx.doi.org/10.1177/15501477211039137]

14. Zhang, M.; Wang, K.; Liu, H.; Zhang, C.; Duan, Y. Spatio-temporal variation of vegetation carbon storage and density in karst areas of Northwest Guangxi based on remote sensing images. Chin. J. Eco-Agric.; 2013; 21, pp. 1545-1553. [DOI: https://dx.doi.org/10.3724/SP.J.1011.2013.30142]

15. Fan, L.; Dong, G.; Frappart, F.; Wigneron, J.-P.; Yue, Y.; Xiao, X.; Zhang, Y.; Tao, S.; Cao, L.; Li, Y. et al. Satellite-Observed Increase in Aboveground Carbon over Southwest China during 2013–2021. J. Remote Sens.; 2024; 4, 0113. [DOI: https://dx.doi.org/10.34133/remotesensing.0113]

16. Poorter, L.; van der Sande, M.T.; Thompson, J.; Arets, E.J.M.M.; Alarcón, A.; Álvarez-Sánchez, J.; Ascarrunz, N.; Balvanera, P.; Barajas-Guzmán, G.; Boit, A. et al. Diversity enhances carbon storage in tropical forests. Glob. Ecol. Biogeogr.; 2015; 24, pp. 1314-1328. [DOI: https://dx.doi.org/10.1111/geb.12364]

17. Jiao, C.; Yu, G.; He, N.; Ma, A.; Ge, J.; Hu, Z. Spatial pattern of grassland aboveground biomass and its environmental controls in the Eurasian steppe. J. Geogr. Sci.; 2017; 27, pp. 3-22. [DOI: https://dx.doi.org/10.1007/s11442-017-1361-0]

18. Quesada, C.A.; Phillips, O.L.; Schwarz, M.; Czimczik, C.I.; Baker, T.R.; Patiño, S.; Fyllas, N.M.; Hodnett, M.G.; Herrera, R.; Almeida, S. et al. Basin-wide variations in Amazon forest structure and function are mediated by both soils and climate. Biogeosciences; 2012; 9, pp. 2203-2246. [DOI: https://dx.doi.org/10.5194/bg-9-2203-2012]

19. Ma, W.; Yang, Y.; He, J.; Zeng, H.; Fang, J. Above-and belowground biomass in relation to environmental factors in temperate grasslands, Inner Mongolia. Sci. China Ser. C Life Sci.; 2008; 51, pp. 263-270. [DOI: https://dx.doi.org/10.1007/s11427-008-0029-5]

20. Khan, F.; Hayat, Z.; Ahmad, W.; Ramzan, M.; Shah, Z.; Sharif, M.; Mian, I.A.; Hanif, M. Effect of slope position on physico-chemical properties of eroded soil. Soil Environ.; 2013; 32, pp. 22-28.

21. Liu, R.; Pan, Y.; Bao, H.; Liang, S.; Jiang, Y.; Tu, H.; Nong, J.; Huang, W. Variations in soil physico-chemical properties along slope position gradient in secondary vegetation of the hilly region, Guilin, Southwest China. Sustainability; 2020; 12, 1303. [DOI: https://dx.doi.org/10.3390/su12041303]

22. Ding, Y.; Zang, R. Determinants of aboveground biomass in forests across three climatic zones in China. For. Ecol. Manag.; 2021; 482, 118805. [DOI: https://dx.doi.org/10.1016/j.foreco.2020.118805]

23. Liu, L.; Zeng, F.; Song, T.; Wang, K.; Du, H. Stand Structure and Abiotic Factors Modulate Karst Forest Biomass in Southwest China. Forests; 2020; 11, 443. [DOI: https://dx.doi.org/10.3390/f11040443]

24. Jia, B.; Guo, W.; He, J.; Sun, M.; Chai, L.; Liu, J.; Wang, X. Topography, Diversity, and Forest Structure Attributes Drive Aboveground Carbon Storage in Different Forest Types in Northeast China. Forests; 2022; 13, 455. [DOI: https://dx.doi.org/10.3390/f13030455]

25. Zeng, N.; Ren, X.; He, H.; Zhang, L.; Zhao, D.; Ge, R.; Li, P.; Niu, Z. Estimating grassland aboveground biomass on the Tibetan Plateau using a random forest algorithm. Ecol. Indic.; 2019; 102, pp. 479-487. [DOI: https://dx.doi.org/10.1016/j.ecolind.2019.02.023]

26. Mutanga, O.; Adam, E.; Cho, M.A. High density biomass estimation for wetland vegetation using WorldView-2 imagery and random forest regression algorithm. Int. J. Appl. Earth Obs. Geoinf.; 2012; 18, pp. 399-406. [DOI: https://dx.doi.org/10.1016/j.jag.2012.03.012]

27. Seidel, D.; Fleck, S.; Leuschner, C.; Hammett, T. Review of ground-based methods to measure the distribution of biomass in forest canopies. Ann. For. Sci.; 2011; 68, pp. 225-244. [DOI: https://dx.doi.org/10.1007/s13595-011-0040-z]

28. Tadese, S.; Soromessa, T.; Bekele, T.; Bereta, A.; Temesgen, F. Above Ground Biomass Estimation Methods and Challenges: A Review. J. Energy Technol. Policy; 2019; 9, pp. 12-25.

29. Somogyi, Z.; Cienciala, E.; Mäkipää, R.; Muukkonen, P.; Lehtonen, A.; Weiss, P. Indirect methods of large-scale forest biomass estimation. Eur. J. For. Res.; 2007; 126, pp. 197-207. [DOI: https://dx.doi.org/10.1007/s10342-006-0125-7]

30. Maraun, D.; Widmann, M. The representation of location by a regional climate model in complex terrain. Hydrol. Earth Syst. Sci.; 2015; 19, pp. 3449-3456. [DOI: https://dx.doi.org/10.5194/hess-19-3449-2015]

31. He, B.; Li, X.; Quan, X.; Qiu, S. Estimating the Aboveground Dry Biomass of Grass by Assimilation of Retrieved LAI Into a Crop Growth Model. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2015; 8, pp. 550-561. [DOI: https://dx.doi.org/10.1109/JSTARS.2014.2360676]

32. Zheng, G.; Chen, J.M.; Tian, Q.J.; Ju, W.M.; Xia, X.Q. Combining remote sensing imagery and forest age inventory for biomass mapping. J. Environ. Manag.; 2007; 85, pp. 616-623. [DOI: https://dx.doi.org/10.1016/j.jenvman.2006.07.015] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17134821]

33. Barrachina, M.; Cristóbal, J.; Tulla, A.F. Estimating above-ground biomass on mountain meadows and pastures through remote sensing. Int. J. Appl. Earth Obs. Geoinf.; 2015; 38, pp. 184-192. [DOI: https://dx.doi.org/10.1016/j.jag.2014.12.002]

34. Myers-Smith, I.H.; Kerby, J.T.; Phoenix, G.K.; Bjerke, J.W.; Epstein, H.E.; Assmann, J.J.; John, C.; Andreu-Hayles, L.; Angers-Blondin, S.; Beck, P.S.A. et al. Complexity revealed in the greening of the Arctic. Nat. Clim. Chang.; 2020; 10, pp. 106-117. [DOI: https://dx.doi.org/10.1038/s41558-019-0688-1]

35. Baccini, A.; Goetz, S.; Walker, W.; Laporte, N.; Sun, M.; Sulla-Menashe, D.; Hackler, J.; Beck, P.; Dubayah, R.; Friedl, M. Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nat. Clim. Chang.; 2012; 2, pp. 182-185. [DOI: https://dx.doi.org/10.1038/nclimate1354]

36. Saatchi, S.S.; Harris, N.L.; Brown, S.; Lefsky, M.; Mitchard, E.T.; Salas, W.; Zutta, B.R.; Buermann, W.; Lewis, S.L.; Hagen, S. Benchmark map of forest carbon stocks in tropical regions across three continents. Proc. Natl. Acad. Sci. USA; 2011; 108, pp. 9899-9904. [DOI: https://dx.doi.org/10.1073/pnas.1019576108]

37. Chang, Z.; Fan, L.; Wigneron, J.-P.; Wang, Y.-P.; Ciais, P.; Chave, J.; Fensholt, R.; Chen, J.M.; Yuan, W.; Ju, W. et al. Estimating Aboveground Carbon Dynamic of China Using Optical and Microwave Remote-Sensing Datasets from 2013 to 2019. J. Remote Sens.; 2023; 3, 0005. [DOI: https://dx.doi.org/10.34133/remotesensing.0005]

38. Wigneron, J.-P.; Li, X.; Frappart, F.; Fan, L.; Al-Yaari, A.; De Lannoy, G.; Liu, X.; Wang, M.; Le Masson, E.; Moisy, C. SMOS-IC data record of soil moisture and L-VOD: Historical development, applications and perspectives. Remote Sens. Environ.; 2021; 254, 112238. [DOI: https://dx.doi.org/10.1016/j.rse.2020.112238]

39. Mermoz, S.; Réjou-Méchain, M.; Villard, L.; Le Toan, T.; Rossi, V.; Gourlet-Fleury, S. Decrease of L-band SAR backscatter with biomass of dense forests. Remote Sens. Environ.; 2015; 159, pp. 307-317. [DOI: https://dx.doi.org/10.1016/j.rse.2014.12.019]

40. Imhoff, M.L. Radar backscatter and biomass saturation: Ramifications for global biomass inventory. IEEE Trans. Geosci. Remote Sens.; 1995; 33, pp. 511-518. [DOI: https://dx.doi.org/10.1109/TGRS.1995.8746034]

41. Boitard, S.; Mialon, A.; Mermoz, S.; Rodríguez-Fernández, N.J.; Richaume, P.; Salazar-Neira, J.C.; Tarot, S.; Kerr, Y.H. Above ground biomass dataset from SMOS L band vegetation optical depth and reference maps. Earth Syst. Sci. Data Discuss.; 2024; 2024, pp. 1-28.

42. Fan, L.; Wigneron, J.-P.; Ciais, P.; Chave, J.; Brandt, M.; Fensholt, R.; Saatchi, S.S.; Bastos, A.; Al-Yaari, A.; Hufkens, K. et al. Satellite-observed pantropical carbon dynamics. Nat. Plants; 2019; 5, pp. 944-951. [DOI: https://dx.doi.org/10.1038/s41477-019-0478-9] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31358958]

43. Liu, Y.Y.; van Dijk, A.I.J.M.; de Jeu, R.A.M.; Canadell, J.G.; McCabe, M.F.; Evans, J.P.; Wang, G. Recent reversal in loss of global terrestrial biomass. Nat. Clim. Chang.; 2015; 5, pp. 470-474. [DOI: https://dx.doi.org/10.1038/nclimate2581]

44. Zhou, G.; Huang, J.; Tao, X.; Luo, Q.; Zhang, R.; Liu, Z. Overview of 30years of research on solubility trapping in Chinese karst. Earth-Sci. Rev.; 2015; 146, pp. 183-194. [DOI: https://dx.doi.org/10.1016/j.earscirev.2015.04.003]

45. Jiang, X.; Dai, J.; Zheng, Z.; Li, X.J.; Ma, X.; Zhou, W.; Lei, Q. An overview on karst collapse mechanism in China. Carbonates Evaporites; 2024; 39, 71. [DOI: https://dx.doi.org/10.1007/s13146-024-00986-x]

46. Hu, T.; Peng, J.; Qiu, S.; Dong, J.; Hu, Y.N.; Lin, Y.; Xia, P. Social-ecological heterogeneity drove contrasting tree cover restoration in South China Karst. Commun. Earth Environ.; 2024; 5, 484. [DOI: https://dx.doi.org/10.1038/s43247-024-01641-y]

47. Lu, Y.; Liu, Q.; Zhang, F.E. Environmental characteristics of karst in China and their effect on engineering. Carbonates Evaporites; 2013; 28, pp. 251-258. [DOI: https://dx.doi.org/10.1007/s13146-013-0158-1]

48. Ma, B.; Jing, J.; Liu, B.; Xu, Y.; Dou, S.; He, H. Spatiotemporal variation of net primary productivity influenced by climatic variables in the karst area of China. Geocarto Int.; 2022; 38, pp. 1-20. [DOI: https://dx.doi.org/10.1080/10106049.2022.2129845]

49. Ge, W.Y.; Deng, L.Q.; Wang, F.; Han, J.Q. Quantifying the contributions of human activities and climate change to vegetation net primary productivity dynamics in China from 2001to 2016. Sci. Total Environ.; 2021; 773, 145648. [DOI: https://dx.doi.org/10.1016/j.scitotenv.2021.145648]

50. ESA. Land Cover CCI Product User Guide Version 2. Tech. Rep. Available online: https://maps.elie.ucl.ac.be/CCI/viewer/download/ESACCI-LC-Ph2-PUGv2_2.0.pdf (accessed on 5 August 2024).

51. Yang, Y.K.; Xiao, P.F.; Feng, X.Z.; Li, H.X. Accuracy assessment of seven global land cover datasets over China. ISPRS J. Photogramm. Remote Sens.; 2017; 125, pp. 156-173. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2017.01.016]

52. Liu, P.; Pei, J.; Guo, H.; Tian, H.; Fang, H.; Wang, L. Evaluating the Accuracy and Spatial Agreement of Five Global Land Cover Datasets in the Ecologically Vulnerable South China Karst. Remote Sens.; 2022; 14, 3090. [DOI: https://dx.doi.org/10.3390/rs14133090]

53. Zhong, S.; Fan, L.; De Lannoy, G.; Frappart, F.; Zeng, J.; Vreugdenhil, M.; Peng, J.; Liu, X.; Xing, Z.; Wang, M. et al. Quantitative assessment of various proxies for downscaling coarse-resolution VOD products over the contiguous United States. Int. J. Appl. Earth Obs. Geoinf.; 2024; 130, 103910. [DOI: https://dx.doi.org/10.1016/j.jag.2024.103910]

54. Carreiras, J.M.; Melo, J.B.; Vasconcelos, M.J. Estimating the Above-Ground Biomass in Miombo Savanna Woodlands (Mozambique, East Africa) Using L-Band Synthetic Aperture Radar Data. Remote Sens.; 2013; 5, pp. 1524-1548. [DOI: https://dx.doi.org/10.3390/rs5041524]

55. Carreiras, J.M.B.; Quegan, S.; Le Toan, T.; Ho Tong Minh, D.; Saatchi, S.S.; Carvalhais, N.; Reichstein, M.; Scipal, K. Coverage of high biomass forests by the ESA BIOMASS mission under defense restrictions. Remote Sens. Environ.; 2017; 196, pp. 154-162. [DOI: https://dx.doi.org/10.1016/j.rse.2017.05.003]

56. Buckley, S.M.; Agram, P.S.; Belz, J.E.; Crippen, R.E.; Gurrola, E.M.; Hensley, S.; Kobrick, M.; Lavalle, M.; Martin, J.M.; Neumann, M. et al. NASADEM User Guide (Version 1); NASA Jet Propulsion Laboratory: California Institute of Technology: Pasadena, CA, USA, 2020; Available online: https://lpdaac.usgs.gov/documents/592/NASADEM_User_Guide_V1.pdf (accessed on 20 August 2024).

57. Hersbach, H.; Bell, B.; Berrisford, P.; Hirahara, S.; Horányi, A.; Muñoz-Sabater, J.; Nicolas, J.; Peubey, C.; Radu, R.; Schepers, D. et al. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc.; 2020; 146, pp. 1999-2049. [DOI: https://dx.doi.org/10.1002/qj.3803]

58. Yuan, W.; Zheng, Y.; Piao, S.; Ciais, P.; Lombardozzi, D.; Wang, Y.; Ryu, Y.; Chen, G.; Dong, W.; Hu, Z. Increased atmospheric vapor pressure deficit reduces global vegetation growth. Sci. Adv.; 2019; 5, eaax1396. [DOI: https://dx.doi.org/10.1126/sciadv.aax1396]

59. Poggio, L.; de Sousa, L.M.; Batjes, N.H.; Heuvelink, G.B.M.; Kempen, B.; Ribeiro, E.; Rossiter, D. SoilGrids 2.0: Producing soil information for the globe with quantified spatial uncertainty. Soil; 2021; 7, pp. 217-240.

60. Fan, L.; Wigneron, J.-P.; Ciais, P.; Chave, J.; Brandt, M.; Sitch, S.; Yue, C.; Bastos, A.; Li, X.; Qin, Y. et al. Siberian carbon sink reduced by forest disturbances. Nat. Geosci.; 2022; 16, pp. 56-62. [DOI: https://dx.doi.org/10.1038/s41561-022-01087-x]

61. Gocic, M.; Trajkovic, S. Analysis of changes in meteorological variables using Mann-Kendall and Sen’s slope estimator statistical tests in Serbia. Glob. Planet. Change; 2013; 100, pp. 172-182. [DOI: https://dx.doi.org/10.1016/j.gloplacha.2012.10.014]

62. Hamed, K.H. Exact distribution of the Mann–Kendall trend test statistic for persistent data. J. Hydrol.; 2009; 365, pp. 86-94. [DOI: https://dx.doi.org/10.1016/j.jhydrol.2008.11.024]

63. Sen, P.K. Estimates of the regression coefficient based on Kendall’s tau. J. Am. Stat. Assoc.; 1968; 63, pp. 1379-1389. [DOI: https://dx.doi.org/10.1080/01621459.1968.10480934]

64. Green, J.K.; Ballantyne, A.; Abramoff, R.; Gentine, P.; Makowski, D.; Ciais, P. Surface temperatures reveal the patterns of vegetation water stress and their environmental drivers across the tropical Americas. Glob. Chang. Biol.; 2022; 28, pp. 2940-2955. [DOI: https://dx.doi.org/10.1111/gcb.16139] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35202508]

65. Wang, Z.; Lai, C.; Chen, X.; Yang, B.; Zhao, S.; Bai, X. Flood hazard risk assessment model based on random forest. J. Hydrol.; 2015; 527, pp. 1130-1141. [DOI: https://dx.doi.org/10.1016/j.jhydrol.2015.06.008]

66. Belgiu, M.; Drăguţ, L. Random forest in remote sensing: A review of applications and future directions. ISPRS J. Photogramm. Remote Sens.; 2016; 114, pp. 24-31. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2016.01.011]

67. Parmar, A.; Katariya, R.; Patel, V. A Review on Random Forest: An Ensemble Classifier. Proceedings of the International Conference on Intelligent Data Communication Technologies and Internet of Things (ICICI) 2018; Coimbatore, India, 7–8 August 2018; Lecture Notes on Data Engineering and Communications Technologies 2019; pp. 758-763.

68. Lundberg, S.M.; Erion, G.; Chen, H.; DeGrave, A.; Prutkin, J.M.; Nair, B.; Katz, R.; Himmelfarb, J.; Bansal, N.; Lee, S.-I. From local explanations to global understanding with explainable AI for trees. Nat. Mach. Intell.; 2020; 2, pp. 56-67. [DOI: https://dx.doi.org/10.1038/s42256-019-0138-9]

69. Lundberg, S. A unified approach to interpreting model predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS 2017); Long Beach, CA, USA, 4–9 December 2017; pp. 4766-4775.

70. Rodríguez-Pérez, R.; Bajorath, J. Interpretation of machine learning models using shapley values: Application to compound potency and multi-target activity predictions. J. Comput.-Aided Mol. Des.; 2020; 34, pp. 1013-1026.

71. Dubayah, R.O.; Armston, J.; Healey, S.P.; Yang, Z.; Patterson, P.L.; Saarela, S.; Stahl, G.; Duncanson, L.; Kellner, J.R. GEDI L4B Gridded Aboveground Biomass Density, Version 2; ORNL DAAC: Oak Ridge, TN, USA, 2022.

72. Xu, L.; Saatchi, S.S.; Yang, Y.; Yu, Y.; Pongratz, J.; Bloom, A.A.; Bowman, K.; Worden, J.; Liu, J.; Yin, Y. Changes in global terrestrial live biomass over the 21st century. Sci. Adv.; 2021; 7, eabe9829. [DOI: https://dx.doi.org/10.1126/sciadv.abe9829]

73. Chen, Y.; Feng, X.; Fu, B.; Ma, H.; Zohner, C.M.; Crowther, T.W.; Huang, Y.; Wu, X.; Wei, F. Maps with 1 km resolution reveal increases in above- and belowground forest biomass carbon pools in China over the past 20 years. Earth Syst. Sci. Data; 2023; 15, pp. 897-910. [DOI: https://dx.doi.org/10.5194/essd-15-897-2023]

74. Harris, N.L.; Gibbs, D.A.; Baccini, A.; Birdsey, R.A.; de Bruin, S.; Farina, M.; Fatoyinbo, L.; Hansen, M.C.; Herold, M.; Houghton, R.A. et al. Global maps of twenty-first century forest carbon fluxes. Nat. Clim. Chang.; 2021; 11, pp. 234-240. [DOI: https://dx.doi.org/10.1038/s41558-020-00976-6]

75. Yu, L.; Fan, L.; Ciais, P.; Xiao, J.; Frappart, F.; Sitch, S.; Chen, J.; Xiao, X.; Fensholt, R.; Chang, Z. et al. Forest degradation contributes more to carbon loss than forest cover loss in North American boreal forests. Int. J. Appl. Earth Obs. Geoinf.; 2024; 128, 103729. [DOI: https://dx.doi.org/10.1016/j.jag.2024.103729]

76. Liu, Y.; Hu, Z.-Z.; Wu, R.; Yuan, X. Causes and predictability of the 2021 spring southwestern China severe drought. Adv. Atmos. Sci.; 2022; 39, pp. 1766-1776. [DOI: https://dx.doi.org/10.1007/s00376-022-1428-4]

77. Zhang, G.; Su, X.; Ayantobo, O.O.; Feng, K. Drought monitoring and evaluation using ESA CCI and GLDAS-Noah soil moisture datasets across China. Theor. Appl. Climatol.; 2021; 144, pp. 1407-1418. [DOI: https://dx.doi.org/10.1007/s00704-021-03609-w]

78. Song, L.; Li, Y.; Ren, Y.; Wu, X.; Guo, B.; Tang, X.; Shi, W.; Ma, M.; Han, X.; Zhao, L. Divergent vegetation responses to extreme spring and summer droughts in Southwestern China. Agric. For. Meteorol.; 2019; 279, 107703. [DOI: https://dx.doi.org/10.1016/j.agrformet.2019.107703]

79. Zeng, W.-H.; Zhu, S.-D.; Luo, Y.-H.; Shi, W.; Wang, Y.-Q.; Cao, K.-F. Aboveground biomass stocks of species-rich natural forests in southern China are influenced by stand structural attributes, species richness and precipitation. Plant Divers.; 2024; 46, pp. 530-536. [DOI: https://dx.doi.org/10.1016/j.pld.2024.04.012] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39280971]

80. Chen, X.; Luo, M.; Larjavaara, M. Effects of climate and plant functional types on forest above-ground biomass accumulation. Carbon Balance Manag.; 2023; 18, 5. [DOI: https://dx.doi.org/10.1186/s13021-023-00225-1]

81. Liu, L.; Yang, H.; Xu, Y.; Guo, Y.; Ni, J. Forest Biomass and Net Primary Productivity in Southwestern China: A Meta-Analysis Focusing on Environmental Driving Factors. Forests; 2016; 7, 173. [DOI: https://dx.doi.org/10.3390/f7080173]

82. Zhang, W.; Xi, M.; Liu, H.; Zheng, H. Low sensitivity of net primary productivity to climatic factors in three karst provinces in southwest China from 1981 to 2019. Ecol. Indic.; 2023; 153, 110465. [DOI: https://dx.doi.org/10.1016/j.ecolind.2023.110465]

83. Xu, X.-L.; Ma, K.-M.; Fu, B.-J.; Song, C.-J.; Liu, W. Relationships between vegetation and soil and topography in a dry warm river valley, SW China. Catena; 2008; 75, pp. 138-145. [DOI: https://dx.doi.org/10.1016/j.catena.2008.04.016]

84. Bordin, K.M.; Esquivel-Muelbert, A.; Bergamin, R.S.; Klipel, J.; Picolotto, R.C.; Frangipani, M.A.; Zanini, K.J.; Cianciaruso, M.V.; Jarenkow, J.A.; Jurinitz, C.F. et al. Climate and large-sized trees, but not diversity, drive above-ground biomass in subtropical forests. For. Ecol. Manag.; 2021; 490, 119126. [DOI: https://dx.doi.org/10.1016/j.foreco.2021.119126]

85. Ali, A.; Yan, E.-R. Functional identity of overstorey tree height and understorey conservative traits drive aboveground biomass in a subtropical forest. Ecol. Indic.; 2017; 83, pp. 158-168. [DOI: https://dx.doi.org/10.1016/j.ecolind.2017.07.054]

86. Fensham, R.J.; Butler, D.W.; Foley, J. How does clay constrain woody biomass in drylands?. Glob. Ecol. Biogeogr.; 2015; 24, pp. 950-958. [DOI: https://dx.doi.org/10.1111/geb.12319]

87. Gessler, A.; Schaub, M.; McDowell, N.G. The role of nutrients in drought-induced tree mortality and recovery. New Phytol.; 2016; 214, pp. 513-520. [DOI: https://dx.doi.org/10.1111/nph.14340]

88. Boerner, R.E.J. Unraveling the Gordian Knot: Interactions among vegetation, topography, and soil properties in the central and southern Appalachians. J. Torrey Bot. Soc.; 2006; 133, pp. 321-361. [DOI: https://dx.doi.org/10.3159/1095-5674(2006)133[321:UTGKIA]2.0.CO;2]