Substantially Enhanced Landscape Carbon Sink Due To Reduced Terrestrial‐Aquatic Carbon Transfer Through Soil Conservation in the Chinese Loess Plateau

Soil conservation is of global importance, as accelerated soil erosion by human activity is a primary threat to ecosystem viability. However, the significance and role of soil conservation in reshaping landscape carbon (C) accounting has not been comprehensively integrated in the terrestrial C sink. Here, we present the first integrated assessment of the modified terrestrial C sink and aquatic C transport due to soil conservation for the semiarid Chinese Loess Plateau (CLP), the world's most vulnerable region to soil erosion. We show a surprisingly low terrestrial‐aquatic C transfer that offset the terrestrial net ecosystem productivity by only 7.5%, which we attribute to the effective implementation of soil conservation practices. Despite the highest soil erosion, the semiarid CLP acts as effective C sink at 43.2 ± 22.6 g C m−2 year−1, which is comparable to temperate forest in absorbing atmospheric CO2. Moreover, C burial in reservoirs has created an additional anthropogenic C sink of 2.9 ± 1.1 g C m−2 year−1. Our findings indicate that effective soil conservation can significantly increase landscape C sequestration capacity. The co‐benefits of soil conservation in erosion control and C sequestration have important implications for policy makers in other regions undergoing increasing erosion intensity to pursue environmental sustainability.

. The transport, transformation, and redistribution of terrestrial C along this continuum can change the land C sink capacity within a given region. Ignoring the terrestrial-aquatic C transfer can thus result in erroneous estimates of C storage in terrestrial ecosystems. Yet, a comprehensive study of the magnitude and significance of terrestrial-aquatic C transfer in affecting the landscape C balance remains poorly explored.
Updated estimates of the global land-ocean C exchange show that terrestrial ecosystems are laterally transferring C into inland waters at a rate of 3.0-5.1 Pg C year −1 , which is comparable in magnitude to the global net terrestrial C uptake of 4.1 ± 1.5 Pg C year −1 (Drake et al., 2018;Regnier et al., 2022). Nevertheless, such a large C loss has often been left out of terrestrial C budgeting studies (Butman et al., 2016;Webb et al., 2019). Furthermore, the lateral C transfer flux is likely conservative and future estimates may further increase with more field-based measurements in currently underexplored regions across the globe (Drake et al., 2018;Gomez-Gener et al., 2021). As such, an accurate evaluation of the net landscape-scale C sink capacity must comprehensively consider the corresponding aquatic C transfer flux.
Despite that the global terrestrial C sink strength is disproportionately contributed by the productive tropical forest ecosystems (Poulter et al., 2014;Tagesson et al., 2020), recent studies suggest that semiarid ecosystems play a dominant role in modulating interannual variability of the sink and its long-term trend (Ahlström et al., 2015;Poulter et al., 2014). Semiarid regions are widespread in the world, covering approximately 15% of the Earth's land surface (L. . Compared with tropical ecosystems with large C pools, semiarid ecosystems are typically considered to be small C pools owing to the low productivity and high turnover rates of C pools driven by strong respiration (Keenan & Williams, 2018). The landscape of semiarid regions is primarily composed of grasslands, shrublands, and savannahs, which are particularly vulnerable to climate variability and soil erosion. Accelerated soil erosion has become a global problem due largely to population growth and unwise management of land resources (Borrelli et al., 2017), posing severe economic and environmental impacts. It has been predicted that large quantities of terrestrial C may have been mobilized through erosion and transported into inland waters for subsequent redistribution (Borrelli et al., 2017;Regnier et al., 2022).
Three redistribution pathways of terrestrially derived C in inland waters are downstream export to catchment outlet (i.e., the ocean or downstream receiving river) as dissolved organic and inorganic carbon (DOC and DIC) and particulate organic carbon (POC), gas emissions at the water-air interface as CO 2 and methane (CH 4 ), and organic C burial within the inland water systems (e.g., lakes, floodplains, reservoirs, and wetlands). Although several studies have accounted for aquatic export of terrestrial C (Butman et al., 2016;Duvert et al., 2020;Hutchins et al., 2020), they do not consider all three redistribution pathways and are, thus, inconclusive and even misleading in evaluating the strength of landscape C sink. Further complicating this issue is that human disturbances can significantly alter terrestrial and aquatic C dynamics in direct or indirect ways (Regnier et al., 2013(Regnier et al., , 2022. The available estimates of the terrestrial C sink after accounting for aquatic export are confined to tropical and boreal regions. Integrated views of terrestrial C sink and aquatic C transport fluxes in arid and semiarid areas are very limited, although their significance in modulating global terrestrial C sink has been well documented (Ahlström et al., 2015;Poulter et al., 2014).
In this study, we evaluate how crucial it is to integrate aquatic C transfer into assessing the terrestrial C sink capacity of the semiarid Chinese Loess Plateau (CLP) that once exhibited the strongest soil erosion intensity in the world (Y. . Unprecedented soil conservation involving vegetation rehabilitation and engineering structures has been devoted to controlling its soil erosion over the past several decades (S. Wang et al., 2016;Zeng et al., 2022). We hypothesize that the terrestrial C sink capacity in this region may have been considerably offset by large terrestrial-aquatic C transfers, and that effective soil conservation can greatly alter the lateral losses and the overall landscape-scale C sink strength. To test these hypotheses, the terrestrial C sink derived from CLP ecosystems and the three flux pathways constituting the aquatic C export during the period 2010−2020 were quantified (see Supporting Information S1 for the selection of the study period), upon which the net landscape C budget was examined. The severe erosion on the CLP probably represents the most pessimistic erosion scenario for other regions currently undergoing rapid environmental changes. Our study provides the first comprehensive assessment on how terrestrial-aquatic C transfers in a semiarid region can affect the net landscape C balance under perturbed conditions, demonstrating the critical role of effective soil conservation in sequestering terrestrial C that would otherwise be released into the atmosphere during fluvial transit.

Study Area
The CLP covers a drainage area of around 623,300 km 2 , mainly in the middle reaches of the Yellow River (Huang He) in northern China (Figure 1). Located in a semiarid region, the average annual precipitation in the CLP is 145-674 mm (average: 464 mm; Fu et al., 2011) with a decreasing trend from the southeast to the northwest. As a result of sparse vegetation cover and highly erodible loess deposits (100−200 m thick), coupled with a long history of anthropogenic disturbances, the CLP suffers from the highest soil erosion rate in the world. The average erosion rate is 6,000−10,000 t km −2 year −1 , and it reaches 30,000 km −2 year −1 in some localities (Hassan et al., 2008; Figure S4 in Supporting Information S1). Starting from the early 1970s, large-scale soil conservation, including ecological management and engineering interventions, has been launched to reduce soil erosion and sediment yielding with substantial achievements (S. Wang et al., 2016). Particularly, the largest ever vegetation restoration program in human history (the "Grain-for-Green" project) initiated in 1999 has further reduced soil erosion rate on the CLP (Fu et al., 2011;. Detailed descriptions are available in Texts S2 and S3 in Supporting Information S1.

Calculation of Terrestrial NEP
We estimated the net ecosystem productivity (NEP) of the terrestrial ecosystems on the CLP based on annual net primary productivity (NPP) and soil heterotrophic respiration (R h ) estimates (NEP = NPP − R h ). The NPP data were downloaded from the data center of the United States Geological Survey. Particularly, the NPP images covering the study area were collected from the MOD17A3HGF Version 6 NPP map (https://lpdaac.usgs.gov/products/mod17a3h-gfv006/) that provides global NPP estimates at 500 m pixel resolution ( Figure S11 in Supporting Information S1). Owing to the interannual variations in NPP, the average NPP result for the period 2010−2020 was adopted to minimize their impact on terrestrial NEP. The R h data were derived from Y. He (2021). For the CLP, the data set reports annual R h for the period 2000−2016 at a spatial resolution of 0.5*0.5°. For the 4 years (i.e., 2017−2020) without R h values, we used the average R h over the period 2010−2016 to represent their R h (Text S6 in Supporting Information S1).

Quantification of CO 2 and CH 4 Emission Fluxes
We measured CO 2 and CH 4 emissions at 122 stream and river sites and 30 lake and reservoir sites on the CLP during the period 2015−2020 (Text S5 in Supporting Information S1). Floating chamber deployments were performed in both dry and wet seasons to account for the seasonal variations in water surface area (SA) and dissolved gas emissions. We separately estimated the CO 2 and CH 4 emission flux in the dry and wet seasons. The wet and dry seasons start from June to September (122 days) and from October to next May (243 days), respectively. Annual flux was calculated by summing the two seasonal emission fluxes. For streams and rivers, the annual emission flux (F total , in Tg C year −1 ) was computed as where, SO is the Strahler order (7 orders in total), areal is the mean areal emission flux (CO 2 unit: mmol m −2 d −1 ; CH 4 unit: μmol m −2 d −1 ) of discrete sampling site-based results for a given SO in the CLP stream network, and SA is the water SA in the dry or wet season (m 2 ).
Because the gas emission rate from lentic waters is dependent on SA size (Holgerson & Raymond, 2016;Kankaala et al., 2013), we classified the lakes and reservoirs into four size classes, that is, <0.1, 0.1−1, 1−10, and >10 km 2 . The CO 2 and CH 4 emission flux per size class was determined as the product of the SA of reservoirs per size class and the mean areal CO 2 or CH 4 emission flux ( areal ) of each size class. While the areal emission flux varied over the dry and wet seasons, we assumed a constant water SA over the year. The annual emission flux was calculated by aggregating the two seasons, which can be expressed as The emission fluxes from both lotic (streams and rivers) and lentic (lakes and reservoirs) waters were summed to calculate the total emission flux (Text S5 in Supporting Information S1).

Carbon Burial Behind Dams
River systems in the CLP have rather high sediment delivery. The sediment delivery ratio is generally higher than 0.9 and can reach 1 (Hassan et al., 2008;Xu, 1999). This indicates that more than 90% of the sediments entering the CLP fluvial networks can be transported out of the catchment. These rivers have limited floodplains and we assumed that sediment deposition and associated C burial on floodplains are negligible on the decadal scale. Because lakes are primarily located in the northwestern part of the CLP where the erosion rate is lowest, we further assumed that C burial in these lakes is negligible (Text S8 in Supporting Information S1). For C burial estimation, thus, we only considered the amount of C buried in artificial impoundments. Our earlier research on reservoir sediment trapping on the CLP shows an average sediment trapping rate of 588.3 ± 34.6 Tg year −1 by ∼1,800 reservoirs (Ran et al., 2013). Since the early 1970s, approximately 50,000 check dams have been constructed on the CLP to reduce soil erosion (S. Wang et al., 2016). These check dams are generally small (e.g., water SA <0.5 km 2 ) and located close to severe erosion areas ( Figure S14 in Supporting Information S1). In view of the similar functioning of reservoirs and check dams in trapping sediments, we consider them together (called "dams" in this study). We collected 2806 sediment samples from 99 sediment cores (depth range: 1.9-30 m), including 86 check dams and 13 reservoirs, across the CLP to evaluate their C burial rate. Spatial distribution of these dam-based sediment cores is consistent with the huge spatial heterogeneity in soil erosion rate and land-cover use. SOC content in deposited sediments was reported in units of g kg −1 . Considering the homogeneous soil texture and composition throughout the CLP and the overlapping distribution of check dams and reservoirs, we used the average SOC content of the 99 sediment cores to estimate the C burial behind all dams (Text S8 in Supporting Information S1).

Downstream Carbon Export
We first determined the annual flux of DOC, POC, and DIC at the inlet and outlet, and then calculated the total downstream flux by summing the three species (Text S7 in Supporting Information S1). DIC and POC fluxes at the inlet were based on sampling results at the Tangnaihai gauge, which is the closest hydrologic gauge to the inlet and free of damming impacts. In contrast, DOC flux was based on high-frequency sampling (n = 486) at the Jungong gauge during the period 2013−2014 (You & Li, 2021), located ∼100 km upstream of the Tangnaihai gauge ( Figure 1). Calculation of the annual flux of the three C species at the outlet was based on weekly sampling results at the Huayuankou gauge (n = 61) during the year 2018 ( Figure 1). Except DIC and POC at the inlet. all the annual fluxes of the three C species were estimated using the regression model Load Estimator (LOADEST) (Runkel et al., 2004) with daily flow discharge and sediment load data as input parameters. To optimize the calibration and estimation procedures, the Adjusted Maximum Likelihood Estimation method was employed, and the best-fit model was selected according to the Akaike Information Criterion. For the POC flux at Tangnaihai, we estimated it as the simple product of average annual sediment load during the period 2010−2020 and the average POC content in sediments. This may have led to flux errors due to the limited POC content data. Yet, because the sediment load at Tangnaihai accounted for only 8.8% of that at Huayuankou in the same period, we assumed that the impact of the inlet POC export on the lateral POC flux from the CLP is negligible. The DIC flux at Tangnaihai was estimated by using the monthly mean DIC concentration and the corresponding monthly flow.
To constrain the interannual changes in flow regime, the daily average flow discharge and sediment load for the period 2010−2020 were used to determine the downstream C export. The difference in the total C flux between the outlet and the inlet was attributed to be the C exported from the CLP.

Uncertainty Analysis
We used two independent approaches to evaluate the uncertainty in the terrestrial NEP estimates. We first assessed the uncertainty by using gross primary productivity (GPP) and its empirical relationship with ecosystem respiration (Baldocchi & Penuelas, 2019). A slope of 0.9, indicating a 90% loss of GPP to ecosystem respiration, was used to estimate the terrestrial NEP (Text S10 in Supporting Information S1). In the second approach, we quantified R h by using 677 soil respiration measurements retrieved from the Soil Respiration Database (SRDB V5.0; https://daac.ornl.gov/ SOILS/guides/SRDB_V5.html) ( Table S5 in Supporting Information S1). Furthermore, we employed a Monte Carlo approach to upscale the site-based gas evasion measurements to the catchment scale and to estimate uncertainties. For each simulation, we generated a total of 10,000 iterations for each stream order (7 orders in the stream network) and size class (4 classes in lakes and reservoirs) for both seasons, in which a GHG flux was randomly selected from a normal distribution around their means in each iteration. For stream networks, the water SA for each stream order was also selected from a normal distribution surrounding the mean in each iteration. However, because there is a lack of sufficient CH 4 flux data in lakes and reservoirs with size classes of 0.1-1 km 2 and >10 km 2 , we assume that the standard deviation of CH 4 flux in the 0.1−1 km 2 size class is the same as that in the <0.1 km 2 size class, and the mean and standard deviation of CH 4 flux in the >10 km 2 size class is equivalent to that in the 1−10 km 2 size class.
For downstream exports of DOC, DIC, and POC, we ran a Monte Carlo analysis with 10,000 iterations for each species where a DOC, DIC, and POC flux was randomly selected from a normal distribution around their means in each iteration for both the inlet (Tangnaihai, Figure 1) and outlet (Huayuankou, Figure 1). We determined the propagated error associated with annual total flux estimates for gas emissions (CO 2 and CH 4 ) and downstream export (sum of DOC, POC, and DIC fluxes) with a square root of the sum of squares approach by assuming that the error in individual components is statistically independent. To analyze the uncertainty associated with C burial in reservoirs, we also generated 10,000 Monte Carlo simulations by randomly selecting a SOC content value from a normal distribution surrounding the mean in each iteration. Simulations that led to negative SOC content (373 out of 10,000 simulations) were manually discarded from the error analysis. The uncertainty of C burial in reservoirs was expressed as standard deviation (σ) (Text S10 in Supporting Information S1).
We noted that the employed datasets encompass different temporal intervals. The selection of 2010-2020 as the study period is largely based on data availability and hydroclimatic conditions. Estimation of terrestrial ecosystem productivity involved the entire study period, and thus the results reflected the long-term average. In comparison, estimation of aquatic C fluxes was based on shorter and sometimes disparate temporal intervals. However, in view of the relatively steady flow and sediment transport dynamics during the study period and the stable C concentrations over time (Texts S3 and S7 in Supporting Information S1), we believe that the overall impact of the temporal variance in terrestrial-aquatic C transfer on the net landscape C balance is minimal.

Enhanced Terrestrial Net Ecosystem Productivity
The vegetation cover on the CLP has significantly increased (S. Wang et al., 2016), since the implementation of massive vegetation rehabilitation programmes, particularly the "Grain-for-Green" project launched in 1999, as an effective way to stabilize soils. Many reservoirs and check dams were constructed in the CLP, primarily during the 1970s and 1980s. Using sediment load as a proxy for soil erosion rate, sediment export from the CLP has presented a stepwise reduction from 1,310 Tg year −1 during the period 1950−1969 to 110 Tg year −1 during the period 2000−2020 (Figure 2a). Our estimation of terrestrial C accumulation on the CLP shows that its NPP has been steadily increasing at a rate of 4.1 Tg C year −1 during the period 2000−2020 (Figure 2b). To minimize the effect of interannual variations and to be consistent with contemporary measurements of aquatic C export (Texts S5 and S7 in Supporting Information S1), we used the average NPP estimate of 199.9 ± 14.6 Tg C year −1 for the recent decade (2010−2020) to evaluate the strength of terrestrial C sink. The simultaneous heterotrophic soil respiration (R h ) was estimated at 168.9 ± 1.6 Tg C year −1 , showing a relatively small interannual variation in the same period (Text S6 in Supporting Information S1). Deducting the R h term from the NPP estimates shows that the CLP was a net land C sink in the recent decade (2010−2020), although the sink strength exhibited a pronounced interannual variation ( Figure S13 in Supporting Information S1). On average, the terrestrial NEP on the CLP was 31.0 ± 14.1 Tg C year −1 , with the majority stored in topsoil (Text S6 in Supporting Information S1). This NEP rate corresponds to 49.7 ± 22.6 g C m −2 year −1 when normalized to the landscape area.
Magnitude of terrestrial NEP is highly dependent on climate and land-use management with large NEP in tropical, subtropical, and temperate forests (Erb et al., 2013;G. Yu et al., 2014). In comparison, semiarid biomes composed primarily of grasslands have a considerably lower NEP and may not necessarily act as a perpetual C sink (Smith, 2014), as also revealed by our findings from the semiarid CLP. However, our results also indicate that terrestrial C sink in such biomes can be enhanced through continuing vegetation rehabilitation even over a short period, given the negligible impact of climate change on terrestrial ecosystems in the study period (Text S1 in Supporting Information S1). Specifically, while the terrestrial NEP on the CLP for the period 2000−2010 was estimated at 10−30 g C m −2 year −1 (Keenan & Williams, 2018), it has almost doubled in the recent decade, demonstrating an enhanced C sink capacity of the CLP terrestrial ecosystems in absorbing atmospheric CO 2 . It is reasonable to expect that the terrestrial C sink of the CLP will further increase with continued vegetation restoration efforts.

Aquatic Loss of Terrestrial Carbon
Downstream C export reveals the crucial role of erosion in mobilizing soil C from erodible hillslopes. Although the POC content in dry sediments is typically low, generally in the range of 4-14 g kg −1 due to unproductive soils throughout the CLP (H. Yu et al., 2019), the average POC concentration at Huayuankou gauge, the catchment outlet (Figure 1), was 9.0 ± 4.1 g kg −1 as a result of large sediment load. The corresponding DOC changed from 1.9 to 5.8 mg L −1 with an average of 3.4 ± 0.8 mg L −1 . In comparison, the DIC export was an order of magnitude higher than the DOC and the average DIC was 41.2 ± 8.0 mg L −1 due largely to the high carbonate content in loess soils (Qu et al., 2020) and its rapid dissolution promoted by erosion. Downstream export of DOC, DIC, and POC from the CLP were estimated at 0.07 ± 0.002, 0.47 ± 0.03, and 0.48 ± 0.08 Tg C year −1 , respectively.
Summing up the three components shows a total export of 1.0 ± 0.1 Tg C year −1 , or 1.7 ± 0.1 g C m −2 year −1 if expressed as per unit landscape area. Erosion-induced POC is the largest component of the downstream export, which, together with DIC, accounted for 93% of the total downstream flux.
To estimate the annual flux of CO 2 and CH 4 emissions from CLP inland waters, we first quantified the water SA of streams, rivers, lakes, and reservoirs (Text S4 in Supporting Information S1). For streams and rivers on the CLP, a total length of 102,400 km was classified into seven Strahler orders. The water SA of all streams and rivers varied from 1,072 km 2 in the dry season to 1,206 km 2 in the wet season, accounting for only 0.24−0.27% of the landscape area (Table S1 in Supporting Information S1). In addition, 4068 lakes and 1,813 reservoirs with the SA varying from ∼1,800 m 2 to 210.7 km 2 were identified on the CLP and classified into 4 size classes for emission estimation (Holgerson & Raymond, 2016). The SA of all lakes and reservoirs combined was 1,975 km 2 , of which 62% was contributed by large water bodies (>10 km 2 in size) ( Table S2 in Supporting Information S1). With field-based areal CO 2 and CH 4 evasion fluxes for each waterbody type, the total flux of CO 2 and CH 4 evasion from flowing streams and rivers after accounting for spatial and seasonal variability in areal evasion fluxes and water surface extent was 1.1 ± 0.7 and 0.002 ± 0.0009 Tg C year −1 , respectively. Large rivers played a dominant role in CO 2 and CH 4 emissions, contributing to 61−79% of the total flux ( Figure 3). In contrast, the small contribution of headwater streams (Figure 3) is contradictory to previous studies in tropical and temporal regions that show disproportionately high emissions from headwater streams (Butman & Raymond, 2011;Marescaux et al., 2018). The unexpectedly small contribution is likely because of the low stream network density and shallow streamflow regulated by the dry climate. For standing waters of lakes and reservoirs, the corresponding CO 2 and CH 4 emission flux was 0.2 ± 0.1 and 0.0099 ± 0.0043 Tg C year −1 , respectively, of which 66−70% was degassed from large water bodies (Figure 3). The total emission flux from the CLP inland waters was 1.3 ± 0.7 Tg C year −1 , which is equivalent to 2.1 ± 1.1 g C m −2 year −1 . Notably, CO 2 emission dominated the total flux and represented 99% of the latter, and it remained dominant after considering the global warming potential of CH 4 (Table S3 in Supporting Information S1).
The emitted CO 2 and CH 4 are primarily sourced from soil respiration (i.e., soil CO 2 and CH 4 ) and decomposition of organic matter by microbes within the water column, with the relative importance of soil respiration depending heavily on the magnitude of hydrologic connectivity (Battin et al., 2023;Hotchkiss et al., 2015). For the CLP, degradation of organic matter in soils contributed 26% to the total CO 2 emission flux while 22% to the total CH 4 emission flux (Text S5 in Supporting Information S1). The relatively low impact of soil respiration is consistent with the low contribution of headwaters to total CO 2 and CH 4 emission flux, as discussed above. Check dams and reservoirs have long been considered an effective way to control soil erosion and riverine sediment transport (Hassan et al., 2008;S. Wang et al., 2016). Our previous estimation of sediment transport shows that the average sediment trapping by check dams and reservoirs on the CLP is 792.3 ± 36.6 Tg year −1 (Text S8 in Supporting Information S1), contributing to 66% of the total net sediment load reduction on the CLP since 1970 (Figure 2a). While the organic C content in the deposited sediments (referred to as SOC thereafter) exhibited small spatial variations due likely to the homogeneous soil property across the CLP (Text S1 in Supporting Information S1), significant vertical changes were observed from surface to the bottom of the deposited sediments. The average SOC content decreased rapidly from 3.3 ± 1.5 g kg −1 in the top 0.5 m sediment layers to 2.1 ± 1.1 g kg −1 in the deeper sediment layers (Figure 4). The overall SOC content is 2.2 ± 1.2 g kg −1 (Figure 4). Accordingly, the buried C was estimated at 1.8 ± 0.7 Tg C year −1 , equivalent to 2.9 ± 1.1 g C m −2 year −1 when normalized to the landscape area. In comparison, the global C burial rates in lakes and reservoirs varies from 0.4 to 1.9 g C m −2 year −1 with an average rate of 1.1 g C m −2 year −1 (Mendonça et al., 2017). The significantly higher C burial rates on the CLP demonstrate the combined effects of exceptionally strong soil erosion, rapid sediment trapping, and frequently low dissolved oxygen in bottom waters, which have efficiently preserved the buried C from mineralization (Clow et al., 2015;Quadra et al., 2020).
The net terrestrial-aquatic transfer for the CLP was estimated at 4.1 ± 1.0 Tg C year −1 by summing the three aquatic export components. Carbon burial by check dams and reservoirs is the largest component, accounting for 43% of the total flux and being higher than gas emissions by 27%, due likely to the high soil erosion rates and the efficient burial of C after mobilization from eroding sites. A small percentage of inland water-air interface area in this semiarid region is another possible explanation for the low evasion flux as widely observed in the global context (Mendonça et al., 2017), although the areal emission rates are consistent with those reported in other climatic regions (Butman et al., 2016;Duvert et al., 2020;Raymond et al., 2013). In comparison, downstream export of C laterally leaving the CLP accounted for only 25% of the total terrestrial-aquatic transfer flux. From a mass balance perspective, our results suggest that the CLP aquatic continuum acts more like an active reactor that has processed a majority of the terrestrially derived C on an annual basis.

Net Landscape Carbon Balance
Subtracting the terrestrial-aquatic C transfer flux from the terrestrial NEP result shows a net landscape C balance of 26.9 ± 14.0 Tg C year −1 for the entire CLP. This rate corresponds to 43.2 ± 22.6 g C m −2 year −1 when expressed as per unit landscape area ( Figure 5). Terrestrial-aquatic transfer represents a small proportion of the terrestrial primary productivity, offsetting only 13% of the terrestrial NEP. Carbon burial due to sediment trapping alone has relocated around 6% of the terrestrial NEP during the study period. Because a primary purpose of check dams and reservoirs in the CLP is for sediment interception, the high C burial rate due to rapid trapping of eroded soil reflects greatly reduced post-depositional mineralization of deposited organic C that would otherwise be decomposed during downstream transport within the aquatic systems. It is worth noting that the terrestrially derived C burial by check dams and reservoirs may not necessarily represent a net C loss at the landscape scale. Compared with hillslope soils where C typically has a short turnover time, especially for erosion-prone regions, the buried terrestrial C through sediment trapping can more effectively escape mineralization and be thus removed from the short-term C cycle (Anderson et al., 2020). Therefore, rather than a simple change in storage location, the C burial by check dam and reservoir trapping can be considered as an anthropogenic C sink (Mendonça et al., 2017).
Emission of CO 2 and CH 4 from inland waters constitutes a crucial component of the global C cycle (Raymond et al., 2013). For the semiarid CLP, the water-air gas emissions played a disproportionately small role in mediating the landscape C balance. The flux of CO 2 emission offset only 4.2% of the terrestrial NEP, while the CH 4 emission was negligible (<0.04%). Expressed as per unit landscape area, the emission rate of 2.1 ± 1.1 g C m −2 year −1 is within the range reported for boreal landscapes while considerably lower than those observed in tropical and temperate regions that generally exceed 10 g C m −2 year −1 (Butman & Raymond, 2011;Campeau et al., 2014;Duvert et al., 2020;Hutchins et al., 2020). Small water surface extent due to low network drainage density for gas emissions is probably the primary control on the low emission. The absence of wetlands which are important hotspots of CH 4 emissions (Holgerson & Raymond, 2016;Webb et al., 2019) has further limited the magnitude of gas emissions. Expressed as CO 2 equivalents, CH 4 emissions accounted for only 8% of the annual total gas emission flux in the CLP, in stark contrast to wetland-rich regions where CH 4 emission could represent 25−34% of total C emission (Campeau et al., 2014;Duvert et al., 2020).
As a consequence of rapid SOC trapping by check dams and reservoirs, downstream export of riverine C played a negligible role in the net landscape C balance for the CLP, offsetting ∼3% of the corresponding terrestrial NEP. This surprisingly low reduction of terrestrial C storage by downstream transport may have indeed been overestimated in carbonate-rich regions with potentially large C inputs from rock weathering. We acknowledge that the laterally-transferred C is not solely of terrestrial origin (Hotchkiss et al., 2015). A notable component is the DIC export that constitutes 46% of the downstream flux. In addition to decomposition of organic C during fluvial transit, the impact of dissolution of carbonate-rich loess soils is likely substantial (Qu et al., 2020). However, given that DIC flux can only offset the annual NEP by 1.5%, it is reasonable to assume that whether to consider its biospheric source will not substantially affect the overall reduction of terrestrial NEP due to terrestrial-aquatic C transfer. Furthermore, in-stream production of organic C by aquatic plants is also rather limited owing to greatly reduced light penetration as evidenced by high turbidity and flow velocity (Hu et al., 2015).
If the C burial in the constructed check dams and reservoirs is regarded as an anthropogenic C sink and excluded from the terrestrial-aquatic transfer, the aquatic C loss (sum of downstream export and gas evasion) would only reduce the terrestrial C balance by 7.5% of NEP. This relatively small impact of aquatic C transfer in offsetting terrestrial NEP reflects the contribution of effective soil conservation, which has profoundly modified the terrestrial and aquatic C dynamics. Were soil conservation not implemented, at least 25% of the terrestrial NEP in the Figure 5. Net landscape carbon budget for the Chinese Loess Plateau. Net ecosystem productivity represents CO 2 fixation by terrestrial ecosystems. Carbon burial represents the sequestered C behind all dams, including reservoirs and check dams. All fluxes are expressed as per unit landscape area, with arrow sizes proportional to rate of flux. CLP (12.4 g C m −2 year −1 ) would have been lost to aquatic transport (Butman et al., 2016;Hutchins et al., 2020;Webb et al., 2019). The net landscape C balance of 43.2 ± 22.6 g C m −2 year −1 is comparable to the C sink capacity of temperate forest (H. He et al., 2019;Keenan & Williams, 2018). Our analysis with integrated terrestrial and aquatic C accounting suggests that the semiarid CLP can act as persistent C sink as the integration of the terrestrial-aquatic C transfer flux does not significantly offset terrestrial NEP. However, it must be pointed out that previous work on aquatic C accounting (e.g., Butman et al., 2016;Duvert et al., 2020;Webb et al., 2019) has rarely accounted for all three flux components with C burial, POC transport, or CH 4 emission generally ignored from estimation. Such incomplete accountings may have introduced considerable biases to the aquatic transfer flux, thereby leading to errors when assessing the net landscape C budget. To our best knowledge, this research represents the first attempt to integrate all three aquatic C pathways in assessing landscape-scale C sink strength.
In addition to SOC, soil carbon pool is also composed of soil inorganic carbon (SIC), which has recently received widespread attention in arid and semiarid soils due to its large accumulation. Prior estimates show that the SIC stock in arid and semiarid regions is substantial, usually 2−10 times greater than the corresponding SOC stock (Batjes, 2006;Tan et al., 2014). Soil inorganic carbon dynamics are controlled by a multitude of complex processes, such as land-use changes and intensive farming which are widely performed on the CLP. For example, these activities have been shown to increase both SOC and SIC stocks in arid and semiarid regions, and the increasing SOC stock can simultaneously enhance SIC accumulation due to the formation of pedogenic carbonates (X. Wang et al., 2015). Therefore, arid and semiarid soils could act as an efficient C sink by accumulating SIC. For the semiarid CLP, unfortunately, only a handful of studies on its SIC dynamics have been conducted. A preliminary finding is that its SIC has only been slightly enriched during the erosion-transport-deposition process (e.g., Tong et al., 2020;Wang et al., 2019;Yuan et al., 2023), indicating a small potential of C sink. However, considering the huge amount of the mobilized SIC on the CLP, it is predicted that the accumulation of SIC through the formation of pedogenic carbonates is likely substantial, although it is difficult to accurately quantify it due to data scarcity. Taking into account this ignored term, the strength of landscape C sink would be even stronger than we estimated here.

Conclusions and Implications
While the terrestrial-aquatic C losses from land to aquatic systems have been increasingly realized to represent a mostly overlooked but nonnegligible component of terrestrial C budgets (Butman et al., 2016;Regnier et al., 2022), our results confirm that these lateral C losses are relatively insignificant in a large semiarid region. The surprisingly low fraction (7.5%) of terrestrial C capture lost via aquatic export does not align with previous large-scale net landscape C flux estimates. For example, for the southwestern United States with climate and terrestrial C capture capacity analogous to the CLP, a substantially larger fraction (i.e., 20−45%; Butman et al., 2016) of the terrestrial NEP has been found to be lost as aquatic export. We argue that effective soil conservation on the CLP has significantly modified its landscape C redistribution, accumulating large amounts of fixed atmospheric C within the CLP landscape that, if without the soil conservation efforts, may have returned to the atmosphere during aquatic transport. Our radiocarbon ( 14 C) analysis shows considerably old 14 C age of 5515-7990 years BP for laterally transported sediment POC and 574−3805 years BP for evaded CO 2 , with relatively younger C released in the wet season (Text S9 in Supporting Information S1). Comparing the 14 C age of the evaded CO 2 against that of the sediment POC further confirms that, except for modern recently fixed C, pre-aged soil organic C has also been mobilized through erosion, and a considerable portion of it must have been mineralized and released into the atmosphere during fluvial transit (McCallister & del Giorgio, 2012). The significantly older 14 C age of the evaded CO 2 is contrary to published studies which show modern organic matter as being the dominant source of evaded CO 2 (Leith et al., 2014;Mayorga et al., 2005). Thus, effective soil conservation aimed at reducing erosion has important implications for C cycling over longer time scales (hundreds to thousands of years).
Restoring degraded terrestrial ecosystems through continued vegetation rehabilitation coupled with improving soil fertility has created an increase in the anthropogenic C sink by 6.6 g C m −2 year −1 (inferred from the slope in Figure 2b). Together with C burial behind dams, implementation of diverse soil conservation practices has generated an additional anthropogenic C sink of 9.5 g C m −2 year −1 , which is equal to 22% of the net landscape C balance. Our findings underscore that effective soil conservation provides co-benefits for erosion control and C sequestration in the context of sustainable catchment management. While erosion reduction after soil conservation has been well investigated (Borrelli et al., 2017;S. Wang et al., 2016), the simultaneous landscape C sequestration capacity remains poorly understood. Particularly, it is worth noting that the C burial rate of 2.9 ± 1.1 g C m −2 year −1 is ∼3 times as high as the global average rate of C burial in lakes and reservoirs (1.1 g C m −2 year −1 , Mendonça et al., 2017). Because previous studies were generally conducted in low-to-moderate erosion regions (Anderson et al., 2020;Clow et al., 2015;Mendonça et al., 2017), future efforts to more accurately account for C burial in high-erosion regions are needed to refine global or regional estimates of C sequestration in lakes and reservoirs.
Soil erosion has been projected to increase worldwide driven by climate change and intensive land-use/land-cover changes such as expanding cropland, deforestation, and overgrazing Wuepper et al., 2020). Developing economies will experience the greatest increase due to unsustainable land management and increasing demand for agricultural products and urban development. Even so, the exceptionally high erosion rates on the CLP are not likely to be observed at large scales in other regions of the world and probably represent the upper end of future erosion predictions (Borrelli et al., 2017). Compared with other regions, although current soil erosion on the CLP remains at relatively high levels with considerable sediment input into rivers, the efficient engineering structures (i.e., check dams and reservoirs) can rapidly trap the eroded sediment and C from hillslopes and deposit them within the system. This is especially true for check dams which are primarily designed to intercept sediment with an efficiency of ∼100%. For other regions highly susceptible to accelerated soil erosion, check dams and reservoirs should be greatly promoted as an effective way to reduce aquatic C transfer and increase C sequestration on the landscape scale.
Our analyses provide integrated perspectives for policy makers in the rest of the world undergoing increasing soil erosion to strategize their policy decisions toward C neutrality and environmental sustainability. We conclude that soil conservation via vegetation rehabilitation and engineering structures can substantially reduce lateral C movement and, simultaneously, enhance the landscape C sequestration capacity. A greater scientific understanding of this dual biogeochemical role that effective soil conservation efforts contribute is particularly important in the context of sustainable land management and climate change mitigation.