Introduction

Soil organic carbon (SOC) constitutes the largest terrestrial pool of organic carbon (C) (containing 2010 Pg C to a depth of 1 m), which is three times greater than the C stored in vegetation and the atmosphere1,2,3. With the dual benefit of removing atmospheric CO2 and increasing soil quality, a greater net SOC stock offers a promising opportunity for climate mitigation and sustainable food production to achieve Sustainable Development Goals4,5. Hence, an accurate understanding of SOC stock variations and their primary drivers is essential.

China, the world’s largest CO2 emitter, has pledged to achieve C neutrality before 20606. This commitment implies a need to offset anthropogenic CO2 emissions with a net C sink. The dryland agricultural regions of China, covering a land area of 49.4 million km2, play a crucial role in grain production, accounting for more than 52% of the total cultivated area7. Due to low and highly variable annual precipitation and high potential evapotranspiration8, and sparse vegetation cover with low annual productivity9, the capacity of dryland to store large amounts of C and serve as a C sink is often overlooked. For instance, these dryland ecosystems exhibit enormous potential for sequestering organic C, with an estimated annual sequestration rate of 0.5–1.4 Pg C year−1 between the years 1980 to 201410. Considering water is the main limiting factor and principal driver of plant biodiversity in dryland11, the well-documented rise in mean annual temperature (MAT), along with the fact that mean annual precipitation (MAP) in dryland areas has remained largely unchanged, the projected further warming and drying caused the anticipated deterioration in plant productivity, and consequently pose a threat to SOC stock12. While the importance and vulnerability of drylands to climate change impacts are recognized, the question of C storage in drylands and its response to changes in weather patterns11,13 remains unresolved. The pursuit of C neutrality underscores the need to understand how dryland SOC storage will respond to evolving environmental conditions. This understanding is paramount for effective climate change mitigation and the establishment of sustainable agricultural management practices.

A common problem in dryland agriculture is soil infertility14. Since the 1980s, a series of projects and supportive agricultural policies have been implemented in China’s dryland agriculture to improve soil fertility and increase crop yields to meet the demands of an ever-growing population. Numerous studies have demonstrated that agricultural management practices profoundly affect spatio-temporal changes in SOC stock15 because anthropogenic activities can disrupt the balance between C inputs and decomposition within a relatively short period of time. Soil disturbance by tillage has been widely shown to be the primary cause of past SOC declines1, with an estimated global C loss of 0.3–1.0 Pg C year16. Despite numerous studies that have examined the effects of agricultural management practices on SOC stocks, ongoing debates persist due to seemingly contradictory outcomes17,18. The limited and regionally biased research from China’s drylands hampers reliable estimations of their potential as regional and global C sinks.

Topsoil responds rapidly to environmental changes, including shifts in temperature and precipitation19. Therefore, we collected field-measured data from the 0–20 cm soil depth in China’s drylands, including historical (1980s) and repeated measures (2010s) data from northern, northwestern, and northeastern regions, using 721 soil profiles (Fig. S1). This enabled us to identify patterns in SOC stock changes over a 30-year period. We further sampled soils from six long-term field experiments located in China’s dryland to establish direct cause-effect relationships and quantify the specific long-term sequestration efficacy of organic amendments (i.e., straw and manure). Our objectives were: (1) to assess spatial and temporal changes in SOC stock in China’s drylands, (2) to explore the climatic, edaphic, and agronomic drivers influencing SOC stock distribution and storage, and (3) to provide insights for agricultural management and strategies for C sequestration in semi-arid ecosystems.

Results

Regional changes in dryland SOC and TN stocks from 1980s to 2010s

In the 2010s, the SOC stock was highest in the northeast (33 Mg C ha−1), intermediate in the northwest (25 Mg C ha−1), and lowest in northern China (19 Mg C ha−1) (p < 0.001, Fig. 1a). Between the 1980s and the 2010s, the SOC stock increased by 14% in northern China (p < 0.001), while it decreased by 12% in the northeast (p < 0.05, Fig. 1a). In contrast, during the same period, the TN stock consistently increased by 6.5% in all dryland regions of China (p < 0.001, Fig. 1b), while pH in the northeast, northwest, and northern China decreased by 0.38, 0.42, and 0.41 units, respectively (p < 0.001, Fig. S4).

Fig. 1: The changes in soil organic carbon (SOC) and total nitrogen (TN) stocks in the top 20 cm in the dryland regions of China (north, northeast, northwest, and total) in the 1980s and the 2010s.
Fig. 1: The changes in soil organic carbon (SOC) and total nitrogen (TN) stocks in the top 20 cm in the dryland regions of China (north, northeast, northwest, and total) in the 1980s and the 2010s.
Full size image

a SOC stock. b TN stock. Box plots display medians (centerline), first quartile and third quartiles (lower and upper box borders), and standard errors (whiskers). Lowercase letters indicate significant differences between regions and total in the 1980s; uppercase letters indicate significant differences between regions and total in the 2010s; **** presented above box-error bars, p < 0.001; *** presented above box-error bars, p < 0.01. The gray background reflects the distribution of the stock in all three regions. All the data shown in the figures were obtained from the historical 1980s and repeated sampling in 2010s.

Factors influencing SOC stock: climate, soil properties, and soil management

Between the 1980s and the 2010s, the magnitude of changes in SOC stock (ΔSOC) was greater in all dryland regions of China that had higher initial SOC stock (R2 = 0.13–0.55, p < 0.001, Fig. 2a). However, the magnitude of changes in SOC stock was lower in northern and northeastern China, which had higher initial TN stock. Conversely, in the northwestern region, which had lower initial TN stock, the magnitude of ΔSOC was slightly higher (Fig. 2b). Moreover, in the northern and northeastern China, the magnitude of ΔSOC was positively correlated with the initial soil pH in 1980s (Fig. S5a). Notably, the magnitude of ΔSOC decreased with the C:N ratio in all three regions in the dryland (Fig. S5b).

Fig. 2: Correlations between changes in soil organic carbon (ΔSOC) and soil and climatic factors depending on dryland regions of China (north, northeast, northwest).
Fig. 2: Correlations between changes in soil organic carbon (ΔSOC) and soil and climatic factors depending on dryland regions of China (north, northeast, northwest).
Full size image

a ΔSOC and the initial stock of SOC. b ΔSOC and total nitrogen (TN) in the 1980s. c ΔSOC and annual mean temperature. d ΔSOC and precipitation. e ΔSOC and potential evapotranspiration. f ΔSOC and aridity index. ΔSOC was calculated as the difference in SOC stocks between the 1980s and the 2010s. All regression lines are significant at least at p < 0.001. The vertical arrow indicates the threshold of SOC and TN stocks, above which the changes are negative: SOC stock decreased over 30 years. All the data shown in the figures were obtained from the historical 1980s and repeated sampling in 2010s.

Between the 1980s to the 2010s, the magnitude of ΔSOC decreased in the northeast, correlating with changes in mean precipitation, temperature, and evapotranspiration (Fig. 2). In northwestern China, the magnitude of ΔSOC consistently increased, correlating with changes in mean precipitation and aridity, but showed a negative correlation with temperature (Fig. 2d). In northern China, the magnitude of ΔSOC was positively correlated with increases in both mean temperature and aridity.

The outcomes of the structural equation modeling (SEM) analysis revealed a strong correlation between the magnitude of change in SOC stocks and the initial SOC stock levels in the 1980s, irrespective of specific regions (path coefficients of −0.77, −1.10, and −0.88, respectively). In northern China, fertilizer application bolstered corn and wheat yields, contributing to greater straw return (Fig. 3). Notably, wheat residues had a more pronounced impact on SOC stock changes compared to corn residues (path coefficient = 0.24). Conversely, in the northeast region, despite fertilizer input enhancing corn and soybean yields, straw return did not significantly affect ΔSOC stock. Furthermore, the rise in mean air temperature in the northeast correlated with a reduction in ΔSOC stock (path coefficient = −0.43, Fig. 3b). In the northwest, the greater input of wheat residues was correlated with an increase in ΔSOC stock (path coefficient = 0.34, Fig. 3c).

Fig. 3: Results of structural equation models showing the direct and indirect effects of initial soil properties (soil pH in the 1980s, total N and SOC stocks in the 1980s), climatic factors (mean annual air temperature and precipitation), and management practices (mean annual chemical fertilizer input, crop yield, and crop residue C input) on changes in SOC stock for dryland areas.
Fig. 3: Results of structural equation models showing the direct and indirect effects of initial soil properties (soil pH in the 1980s, total N and SOC stocks in the 1980s), climatic factors (mean annual air temperature and precipitation), and management practices (mean annual chemical fertilizer input, crop yield, and crop residue C input) on changes in SOC stock for dryland areas.
Full size image

a Northern. b Northeastern. c Northwestern. Units: TN and SOC stocks (Mg C ha−1), ΔSOC (Mg C ha−1), MAT (°C), MAP (mm), crop yield (Mg ha−1 yr−1), crop residues C input (Mg C ha−1 yr−1), and chemical fertilizer input (kg ha−1 yr−1). The numbers next to the arrows indicate the standardized path coefficients. The width of arrow is proportional to the strength of the association. R2 indicates the proportion of explained variance. Significance levels are as follows: *p < 0.05, **p < 0.01, and ***p < 0.001. All the data shown in the figures were obtained from the historical 1980s and repeated sampling in 2010s.

The changes in SOC stocks based on long-term field experiments

To illustrate the impacts of agricultural practices on SOC changes, we performed six long-term field experiments, encompassing studies conducted in China’s dryland regions. Based on long-term field experiments, the effect of manure and straw amendment on the changes in SOC stock was positive (1.5–35%) in dryland, except for the effect of straw on SOC change in one field in Northwest (Fig. 4 and Fig. S9). The random forest model indicated that the soil properties of TN, total phosphorus (TP), and available potassium (AK), crop yield explained 75% of the variance on ∆SOC, with the best predictor for the SOC change was TN (Figs. S10–14).

Fig. 4: Annual effect size of fertilization strategies on soil organic carbon contents compared to the sole mineral fertilizers application based on 6 long-term experimental sites in the dryland regions of China from 1991–2004.
Fig. 4: Annual effect size of fertilization strategies on soil organic carbon contents compared to the sole mineral fertilizers application based on 6 long-term experimental sites in the dryland regions of China from 1991–2004.
Full size image

NPKM combination of mineral fertilizers and manure, NPKS combination of mineral fertilizers and straw. CP Changping, ZZ Zhengzhou, GZL Gongzhuling, PL Pingliang, YL Yangling, WQ Wrumqi. NC North China, NE Northeast China, NW Northwest China. Data are presented as mean values ± standard error (n = 3).

Discussion

Spatial patterns of SOC content and storage in dryland agricultural regions

Spatially explicit analyses revealed complex soil organic carbon (SOC) dynamics in China’s dryland agricultural regions. Despite stable national mean SOC stocks (0–20 cm) across three decades (Fig. 1), profound regional heterogeneities emerged. Our findings contrast with higher global dryland SOC estimates (∼34 Mg C ha−1)10,20, suggesting China’s historically intensive farming practices and limited conservation measures may have suppressed C sequestration potential7. Critically, this apparent stability masked opposing dynamics: 65% of monitored sites exhibited net SOC depletion from the 1980s to 2010s, while only 35% showed accumulation (Fig. 2a). This suggested that over half of China’s dryland areas experienced SOC loss. Persistent regional disparities were pronounced, with SOC stocks declining markedly from northeast (32.7 Mg C ha−1) to northwest (25.4 Mg C ha−1), and to northern (19.0 Mg C ha−1) (Fig. 1a). Against this backdrop, northern China emerged as a significant C sink, with topsoil C accumulation rates (0.015 Pg C yr−1) offsetting ∼1.5% of China’s annual CO₂ emissions from 1980 to 201221. This mitigation service is particularly noteworthy given northern croplands’ recent origin from saline-alkaline soils, where vegetation productivity is extremely low and organic C inputs are correspondingly low7,15. This may explain the low level of current farmland SOC in the northern. Remarkably, substantial topsoil TN accumulation (+0.7 Mg N ha−1) in the 2010s compared to the 1980s (Fig. 1b), contrasted with divergent SOC patterns. We attribute this TN enrichment to intensive N fertilization. The sequestered N in the soil amounts to approximately 1.7 times China’s N fertilizer consumption in 2014 (24 Mt, National Bureau of Statistics of China, data.stats.gov.cn), indicating significant soil nutrient enhancement benefits in China’s drylands over the past three decades.

Soil properties controlling the SOC stock in dryland agricultural regions

The decrease in SOC stock was more pronounced in the northeast, while an increasing trend was observed in northern China (Fig. 1a). Three interlinked mechanisms drove this spatial heterogeneity. First, SOC dynamics demonstrated universal dependence on baseline C and N stocks (Figs. 2a, 3). Soils with lower initial SOC and TN content–particularly northern China’s croplands–consistently exhibited greater C sequestration capacity than soils rich in SOC (17.8 Mg C ha−1) and TN (1.8 Mg N ha−1) (Figs. 2a, 4). This reflects their larger C saturation deficit, enabling both higher initial sequestration rates and prolonged accumulation before reaching equilibrium22. Second, the northeast’s distinctly lower topsoil pH (mean 7.73 vs. 7.97–8.03 in northern/northwestern sites; Fig. S4) compounded SOC losses (Fig. S5a). Acidification depleted Ca²⁺/Mg²⁺ cations, degrading aggregate stability23. Subsequent disruption of microaggregates exposed previously protected particulate organic matter to rapid mineralization24,25. Third, in the northeast, characterized by vast areas of black soil with gentle gradients, soil erosion was intense22 and contributed to accelerated decomposition as eroded SOC was detached and transported26. This erosion-mineralization feedback generated net CO₂ emissions. Concomitantly, the region’s inherently SOC-rich soils (∼40 ± 1.6 Mg C ha−1) hosted robust microbial communities that further mineralized native C when organic inputs were insufficient27,28.

Agricultural management practices controlling SOC stock in dryland agricultural regions

Agricultural management practices differentially modulated SOC dynamics, with targeted intensification in northern China driving net accumulation. Tripled synthetic N inputs (e.g., from 45 Mg yr−1 in the 1980s to 153 Mg yr−1 in 2010 in Hebei Province), leading to an 85% increase in maize net primary productivity and a remarkable 151% boost for wheat (Table S1). This amplified plant C input into the soil directly correlating with SOC gains in northern China via linear yield–SOC stock relationships (Fig. 3, Figs. S7, S14). Enhanced crop productivity further increased straw return ranged from 60 kg ha−1 in 1990 to 1240 kg ha−1 in 2010 (Table S1), which strengths straw- and plant-derived C transfer into the soil29, stimulates microbial activity and turnover30. This subsequently triggers the production of microbial necromass31,32, thereby bolstering SOC stocks33. This was evidenced by both from repeated measures between 1980s and 2010s (Fig. 3) and two field experiments conducted in northern China (Fig. 5 and Fig. S10), as well as the positive correlation between the changes in SOC stock and the amount of returned crop residues (Fig. 5). Critically, regional SOC formation efficiency was modulated by residue biochemistry. The higher lignin content of maize-derived compared to wheat-derived straw34, which impedes straw decomposition35, consequently, the contribution of maize-derived straw C to SOC accumulation in northern China is diminished (Fig. 3a). The resultant nutrient recycling cascade further sustained this cycle: straw-derived N/P/K mineralization, as evidenced by higher soil nutrients (particularly N, P, and K) following straw return based on long-term field experiments (Figs. S10–12), stimulated subsequent yields, closing the loop in a self-sustaining C–N co-sequestration spiral. Also, the ongoing expansion of organic fertilizer inputs directly contributes to the restoration of soil structure and enhanced soil fertility36. Indirectly, manure augments SOC sequestration by increasing belowground plant biomass due to improved soil porosity and aeration37. This was supported by the two long-term fertilization experiments conducted in northern China (Zhengzhou, Changping), which showed that the effect size of manure on SOC stock was around 20% relative to their mineral fertilization control (Fig. 5 and Fig. S9). In summary, the agricultural management practices implemented in northern China over the past three decades, including straw return, N fertilization, and manure application, have been instrumental in fostering SOC accumulation and stock.

Fig. 5: Correlations between changes in soil organic carbon stock (ΔSOC, 0–20 cm) and plant C input depending on the dryland regions of China (north, northeast, and northwest).
Fig. 5: Correlations between changes in soil organic carbon stock (ΔSOC, 0–20 cm) and plant C input depending on the dryland regions of China (north, northeast, and northwest).
Full size image

a Straw C inputs. b Total C inputs (including straw and root residue). ΔSOC was calculated as the difference in SOC stock between the 1980s and the 2010s. All regression lines are significant at least at p < 0.05. (a: y = 6.2 × log(x) – 36; b: y = 7.1 × log(x) – 43). Note that only high C inputs: > 270 kg C ha−1 yr−1 straw, or > 350 kg C ha−1 yr−1 total C can compensate the SOC losses. All the data shown in the figures were obtained from the historical 1980s and repeated sampling in 2010s.

Climate controls on SOC stocks in dryland agricultural regions

The divergent dynamics of SOC stocks across China’s dryland croplands reveal profound climatic controls. Northeastern China–characterized by historically cold, humid conditions that fostered substantial SOC accumulation–exhibited significant C depletion (−3.9 Mg C ha−1, ∼12% of 1980s baseline) despite retaining considerable stocks (33 Mg C ha−1). Indeed, as colder regions such as the northeast warm up (Table S3), the rate of SOC decomposition escalates. The decline in SOC stocks in the northeast stems from heightened microbial turnover, greater substrate utilization by microorganisms resulting in amplified CO2 emissions, and diminished C fixation by plants within the soil38,39. Notably, rising temperatures trigger C loss from the topsoil, predominantly observed in the northeast compared to the north and northwestern regions. This suggested that biophysical processes which control SOC accumulation in cold regions may differ from those in other regions40. This trend was corroborated by the increased loss of organic C pools with increased temperatures (Fig. 2c), particularly at higher latitudes41. The SEM model further highlighted the risk of SOC loss in cold regions under warmer climate conditions (Fig. 3b). Crucially, projection models indicate a temperature-dependent SOC decline of 1.05 Mg C ha−1 °C−1 (Fig. 2c). Based on the region’s 3.5 × 10⁶ km² cropland extent (National Bureau of Statistics data, http://www.stats.gov.cn/), a 1 °C mean annual temperature (MAT) increase would liberate approximately 3.6 Tg C. Considering that global average soil surface temperatures are projected to rise by approximately 2 °C over the next three decades under a business-as-usual emissions scenario, this projection implies that warming might trigger a net loss of around 7.3 Tg C from the topsoil. If, as anticipated, this C is released into the atmospheric pool, the atmospheric CO2 concentration could increase by about 3.6 ppm over this period. These warming-induced SOC losses underscore the critical role of cold conditions in retaining soil C when contemplating biosphere-atmosphere feedbacks.

In contrast, northwestern China exhibited no significant net change in SOC stocks over 30-year study period (Fig. 1a). Our analysis suggests potential gains from increased straw return in this arid region were likely counteracted by several factors. These include amplified runoff stemming from variable precipitation patterns and accelerated glacial melt in the Qilian Mountains attributed to global warming42,43,44. Furthermore, regional practices such as overgrazing, grassland reclamation, and desertification significantly degrade SOC stocks45. The intensification of cultivation and grazing, which leaves soils bare for extended periods, combined with recurring drought and frequent high-wind events, collectively undermined the soil’s capacity to retain organic matter46. Together, these complex and interacting drivers effectively masked any significant net C accumulation in the northwest despite increased plant inputs.

Implications on soil C sequestration in dryland of China

Northern China’s SOC restoration provides a template for global drylands occupying 45% of Earth’s cropland area. Its validation of C-input cascades-where strategic fertilization enables biomass returns that fuel microbial stabilization-offers a replicable model across C-deficient drylands of Central Asia and Mediterranean Europe. Accelerated SOC accumulation in northern China occurred through a self-reinforcing feedback loop: strategic N fertilization initially boosted crop productivity, while subsequent policy-driven straw returns delivered sustained organic inputs. However, diminishing returns emerge beyond cumulative thresholds of 270 kg C ha−1 yr⁻¹ (Fig. 5a). Excessive C inputs may constrain SOC sequestration in soils through stimulating decomposition of cellulose-dominant crop residues, reducing C retention efficiency, and even limiting root growth by stimulating soil acidification20.

Straw incorporation’s climate benefits are persistently hampered by inherent dryland constraints. Low moisture availability suppresses microbial residue transformation, while high solar irradiance intensifies photodegradation47, particularly in arid northwestern regions. Consequently, crop-derived C conversion to stabilized SOC rarely exceeds 0.1 Mg C ha−1 yr−1 despite ≈50% national straw return rates. Furthermore, while enhanced plant productivity and C inputs from higher temperatures offer partial benefits, this gain can be offset by accelerated microbial decomposition48-especially in cooler agroecosystems like the northeast with high inherent SOC thresholds. This decomposition may consequently increase net atmospheric CO₂. Our long-term Gongzhuling field experiment (1990–2003), showing unchanged SOC under straw return partly, corroborates this limitation (Fig. 4 and Fig. S9).

To maximize C sequestration benefits from straw return, farmland management requires integrated amendment strategies to reduce straw-induced GHG emissions. These could include augmenting straw with manure or tailored microbial inoculants to expedite particulate-organic-matter-to-mineral association49,50. Moreover, landscape-scale innovations such as perennial-grain rotations to sustain year-round rhizodeposition-represent priority interventions for deployment51,52. These pathways could transform China’s drylands from transient C sinks into resilient reservoirs, advancing both ‘4 per 1000’ ambitions and SDG-aligned food security priorities while concurrently curtailing N pollution.

Limitations and future work

While inherent soil properties (e.g., texture, mineralogy) critically regulate SOC stabilization pathways, our large-scale resampling campaign prioritized core fertility indicators-SOC, total N, pH, bulk density, and available P and K-due to their direct relevance for agricultural management and regional C accounting. This strategic focus enables robust assessment of broad spatiotemporal SOC trends across China’s drylands. Nevertheless, our findings reveal key mechanistic knowledge gaps demanding targeted investigation. First, future research must resolve mineralogical influences on SOC persistence, particularly in coarse-textured soils (>60% sand content), where limited reactive surface area may constrain organo-mineral association. Explicit quantification of interactions among amendment type (e.g., residue quality), clay mineralogy (e.g., kaolinite vs. smectite reactivity), and SOC fraction dynamics (particulate vs. mineral-associated organic matter) is essential to predict amendment efficacy across diverse pedogenic settings. Second, although our empirical data and structural equation modeling identify vulnerability to moderate warming, projections under high-emissions scenarios (SSP5-8.5) require process-based models integrating microbial temperature responses, soil moisture feedbacks, and C saturation thresholds. Incorporating observed SOC stabilization rates into microbially explicit models (e.g., MEMS, MIMICS) would improve long-term C budget constraints. Third, direct quantification of residue-derived versus native SOC dynamics necessitates δ¹³C natural abundance or pulse-labeling approaches within long-term experiments. These techniques can resolve: (i) the fractional contribution of priming effects vs. direct mineralization to residue-derived CO₂ emissions; and (ii) the relative temperature sensitivity of newly stabilized versus legacy SOC pools.

In summary, the total SOC stock in the top 20 cm of dryland agricultural soils was 26 Mg C ha−1 in the 2010s, exhibiting no difference over the 30-year period from the levels observed in the 1980s. Using a combination of SEM and partial regression analyses, we provide compelling evidence that SOC losses were substantial in dryland soils characterized by high initial SOC and TN contents and lower pH values. Northern China’s successful 3.1 Mg C ha−1 accumulation resulted from management-driven C-N co-sequestration: strategic N fertilization tripled crop productivity, enabling policy-enhanced straw returns that elevated residue inputs >20-fold. These practices initiated a self-sustaining cycle–straw-derived nutrients boosted subsequent yields, amplifying SOC stabilization, overcoming initial SOC/TN deficits. Critically, this management efficacy proved insufficient in the northeast’s warming-vulnerable northeastern black soils, where temperature-driven decomposition (1.05 Mg C ha−1 °C−1) outpaces stabilization, equivalent to decrease of 3.9 Mg C ha−1 compared to the 1980s baseline. Optimizing strategies must recognize biophysical thresholds: residue retention yields diminishing SOC returns beyond 270 kg C ha−1 yr−1 in the north, while photodegradation and moisture limits constrain straw conversion efficiency in the northwest. Northern China’s template–leveraging targeted intensification to overcome historical C deficits–offers a transferable model for global drylands (45% of Earth’s croplands).

Methods and methods

Our approach consisted of two main steps. First, we estimated the differences in SOC stocks between the 1980s, which were the historical data and the 2010s, which we sampled in the field, as well as the effects of climate and soil properties on changes in SOC stocks. Second, we performed 6 long-term field experiments with consecutive sampling on SOC, TN, TP, TK, pH, and yield to establish direct cause-and-effect relationships and quantify the specific long-term sequestration efficacy of organic amendments (i.e., straw and manure) in the dryland of China.

Soil sampling in the 1980s

We obtained topsoil data (0–20 cm) for 611 agricultural sampling sites from the historical Second National Soil Inventory of China conducted in the 1980s (1979–1984) (Fig. S1a). The data collected included sampling location, soil physical and chemical properties, such as soil organic matter (SOM), bulk density (BD), as well as clay, silt and sand content in soils. SOM in the 1980s was measured using the potassium dichromate oxidation method52, and SOC was calculated by multiplying SOM by 0.58 according to the Van Bemmelen factor53.

Soil sampling in the 2010s

To assess changes in SOC stocks in China’s drylands, we conducted three consecutive sampling campaigns between 2011 and 2013, sampling 721 sites in the northern, northwestern, and northeastern regions of the country (Fig. S1b). The selected sites had a broad spatial distribution and were representative of dryland in China. Initially, sampling sites were determined based on historical sampling locations and digitized historical maps, including administrative and soil type maps. Then, a global Positioning System (GPS) navigator was applied to find the sites and conducted soil sampling with a positioning error of less than 10 m. At each sampling site, a square pit measuring 1 × 1 × 0.2 m was excavated, and three undisturbed soil samples were collected and then combined to form a composite sample of 0–20 cm soil depth. All the samplings were conducted in discrete times from mid-June to October of the year when seasonal crops approach harvest and before fertilization for the next growing season. The collected soil samples were air dried, passed through a 2 mm mesh sieve, cleaned manually to remove plant debris, and finally finely ground using a ball mill (NM200; Retsch, Hannover, Germany). SOC and TN contents in the 2010s were measured using the same method as in the 1980s.

To further reduce the effects of soil heterogeneity, we made these selected sites fully matched by Kriging interpolation54. We first applied the Kolmogorov-Smirnov test to assess the normality of the data distribution. For spatial interpolation, a semivariogram was developed to explore the spatial dependence in the underlying variable55. Three semivariance models (Gaussian, exponential, and spherical) were tested to understand the relationship between semivariance and distance for SOC content in the 1980s. All semivariograms were optimized to achieve the best fit between models and data. The final form of the exponential model was chosen on the basis of the criterion that the mean errors between the predicted and measured values were closest to zero. Our data from the 1980s were interpolated to a 1 km × 1 km grid for each county. We then calculated the change in SOC by contrasting the actual observations from the resampling campaign in the 2010s with those extracted from the spatial interpolation of the 1980s. We conducted a two-stage Monte-Carlo simulation to account for potential errors from both Kriging interpolation and field sampling54,56. First, to account for errors from spatial interpolation, for each grid, with the Kriging-based estimations of mean and variance, we did a random sampling from the log-normal distribution of SOC stock. Sampling was conducted for both the 1980s and the 2010s, from which we obtained the difference between the two sampling periods. Second, to account for errors from field sampling, based on the sampled values from the previous step, we then randomly drew a certain size of grids and calculated the average of SOC stock for both periods and also for the difference between the two sampling periods. The sampling size was determined by the number of sites investigated during the field survey. The above processes were repeated 10,000 times to estimate the mean value and the corresponding 95% confidence interval.

Soil analysis

Soil pH was measured in water (1:2.5 w/v). SOC (the K2Cr2O7 oxidation-reduction titration) and TN (the Kjeldahl digestion) were determined according to Bao (2000)52. BD was measured using undisturbed samples taken by a standard ring cut with 100 cm3 volume and afterwards soil samples were oven-dried at 105 °C. SOC and TN stocks in 0–20 cm soil were calculated as follows:

$${C}_{{stock}}={SOC}\times {{BD}}_{i}\times 0.2$$
(1)
$${N}_{{stock}}={TN}\times {{BD}}_{i}\times 0.2$$
(2)

where CStock and NStock are the SOC and TN stocks, respectively (kg m−2); BDi is the soil bulk density in the 0–20 cm soil (kg m−3).

For those sites where BD was not measured in the 2010s, we used the equation developed by Song et al.57 for upland soils:

$${BD}=1.377\times {Exp}\left(-0.0048\times {SOC}\right)$$
(3)

Climate and agronomic practice data collection

Meteorological data from 1980 to 2010 in the study area (north, northeast, northwest) were collected from 387 meteorological stations across northern China (as shown in Fig. S2). These data included daily measurements of precipitation, mean annual temperature (MAT), daily solar radiation, and wind speed at two meters above ground.

Evapotranspiration flux (ET0, mm day−1) was calculated using the following formula58:

$${{ET}}_{0}=\frac{0.408\Delta \left({R}_{n}-G\right)+\frac{900{\gamma }}{{T}_{a}+273}{U}_{2}({e}_{a}-{e}_{d})}{\Delta +{\gamma }(1+0.34{U}_{2})}$$
(4)

where Rn and G are the net radiation flux density at the surface and the sensible heat flux density from the surface to the soil (positive if the soil is warming), respectively, expressed in MJ m−2 day−1; Δ is the slope of the saturation vapor pressure versus temperature curve, while γ is the psychrometric constant, and Δ and γ are expressed in kPa C−1; Ta is the daily or hourly air temperature (°C), U2 is the daily or hourly wind speed at 2-m height (m s−1); es and ea are the saturation vapor pressure at air temperature and the actual vapor pressure of the air, expressed in kPa.

Aridity index (AI) is defined as a numerical indicator of the degree of dryness of the climate at a given location and classifies the type of climate in relation to water availability. The higher the aridity index of a region, the greater the water limitation. Increasing aridity represents a higher frequency of dry years over an area58.

$${AI}={MAP}/{{ET}}_{0}$$
(5)

where AI (aridity index = precipitation/evapotranspiration flux) ranges from 0.33 to 0.98.

For each county, the annual crop residue C input (Mg C ha−1) during 1980–2010 was estimated based on the sum of root and straw/stover C inputs from each crop divided by the total cropland area in the corresponding year (Tables S1 and S2). Only wheat, corn, and soybean for each of the 94 counties were considered as sources of crop residue C inputs because yield data for these crops during the last 30 years were readily available from the agricultural census yearbook for each county.

Long-term field experiment

To establish causal relationships and quantify long-term amendment efficacy beyond observational data limitations, we conducted six long-term field experiments evaluating straw and manure effects on SOC stocks (Fig. 4). The first site at Gaoping Agronomy Farm, Pingliang, Gansu Province (35°16′N, 107°30′E; 1254 m elevation), features Calcaric Regosols developed from loess59 under a temperate climate (mean annual temperature 9.8 °C; precipitation 543 mm). Established in 1978 in a winter wheat-summer maize rotation, this experiment employed a randomized complete block design with four replicated treatments: (1) Unfertilized control (CK); (2) Annual mineral NP; (3) NP + annual wheat straw amendment; and (4) NP + annual farmyard manure amendment. Urea served as the N fertilizer at a rate of 90 kg N ha⁻¹ per year, while superphosphate served as the P fertilizer at a rate of 30 kg P ha⁻¹ per year. In the NP + Str treatment, wheat straw at a rate of 3.75 Mg ha⁻¹, was incorporated into the soil before plowing. Additionally, farmyard manure was applied at a rate of 75 Mg ha⁻¹ (wet weight). Full experimental details are available in Liu et al.60 and Shao et al.61.

The second site on the Yangling Toutao Plateau, Shaanxi Province (34°16′49″N, 108°03′54″E; 525 m elevation) features Haplic Luvisols derived from loess62 under a warm-temperate monsoon climate (mean annual temperature 13 °C; precipitation 550–600 mm, concentrated July–September; evaporation 993 mm; frequent winter-spring droughts). Initiated in 1990 for winter wheat-summer maize rotation, this experiment used 196 m² plots in a randomized block design with four treatments: (1) Unfertilized control (CK); (2) Annual mineral NPK (urea: 165 kg N ha⁻¹ wheat, 187 kg N ha⁻¹ maize; superphosphate: 57.6 kg P ha⁻¹ wheat, 24.6 kg P ha⁻¹ maize; potassium sulfate: 68.5 kg K ha⁻¹ wheat, 77.8 kg K ha⁻¹ maize); (3) NPK + biennial straw (wheat straw: 4.5 Mg ha⁻¹, maize straw: 3.7 Mg ha⁻¹ dry weight, incorporated pre-plowing); and (4) NPK + annual cattle manure (40 Mg ha⁻¹ dry weight pre-wheat sowing). Each treatment was replicated three times. Basal NPK applications preceded wheat sowing (early October), while maize received topdressing 30 days post-emergence. Wheat received supplemental irrigation (2–3 events at 90 mm/application); maize irrigation was rain-adjusted using equivalent volumes.

The third site at the Anningqu Experimental Station, Urumqi City, Xinjiang (87°25'58” E, 43°58'23” N; 553 m altitude) features gray desert soil developed on loess-like material59 under an arid desert climate (mean annual temperature 5–7 °C; precipitation 150–200 mm; 200-day frost-free period). Initiated in 1990 with initial topsoil properties (0–20 cm: organic matter 15.2 g kg⁻¹, total N 0.868 g kg⁻¹, total P 0.667 g kg⁻¹, pH 8.1), the experiment utilized 468 m² plots in a three-year rotation: winter wheat—cotton (introduced 2009)—maize (single annual harvest with post-harvest fallow). Four treatments were replicated three times: (1) Unfertilized control (CK); (2) Annual mineral NPK (urea: 241.5 kg N ha⁻¹, superphosphate: 138 kg P ha⁻¹, potassium sulfate: 61.9 kg K ha⁻¹); (3) NPK + annual straw amendment (7.7 Mg ha⁻¹ dry weight incorporated pre-plowing); and (4) NPK + annual sheep manure amendment (60 Mg ha⁻¹ dry weight applied before sowing).

The fourth experiment in Changping District, Beijing (40°12’N, 116°12’E; 44 m elevation) features Eutric Cambisols developed on cinnamon fluvo-aquic soil59, under a temperate climate (mean annual temperature 11 °C; precipitation 600 mm). Initiated in 1990 for winter wheat-summer maize rotation on plots of 200 m², baseline topsoil (0–20 cm) properties were: organic matter 12.31 g kg⁻¹, total N 0.805 g kg⁻¹, total P 0.687 g kg⁻¹, pH 8.22. Four fertilization treatments, each replicated three times, were implemented: (1) Unfertilized control (CK); (2) Annual mineral NPK (urea: 150 kg N ha⁻¹, superphosphate: 75 kg P ha⁻¹, potassium sulfate: 37.5 kg K ha⁻¹); (3) NPK + annual maize straw amendment (3.0 Mg ha⁻¹ dry weight, incorporated pre-wheat plowing); and (4) NPK + annual farmyard manure (22.5 Mg ha⁻¹ dry weight applied pre-wheat sowing).

The fifth site in Zhengzhou, Henan Province (34°47’N, 113°39’E; 91 m elevation) features as Fluvo-Aquic Soil, derived from a stratified parent material of Yellow River alluvium interstratified with loess-derived sediments59, with a humid temperate climate (mean annual temperature 14.4 °C; precipitation 1480 mm). Established in 1990 for winter wheat-summer maize rotation on 400 m² plots, four replicated treatments with each replicated three times were applied: (1) Unfertilized control (CK); (2) Annual mineral NPK (urea: 165 kg N ha⁻¹, superphosphate: 82.5 kg P ha⁻¹, potassium sulfate: 82.5 kg K ha⁻¹); (3) NPK + annual maize straw amendment (7.5 Mg ha⁻¹ dry weight incorporated pre-wheat plowing); and (4) NPK + annual farmyard manure amendment (30 Mg ha⁻¹ dry weight applied pre-wheat sowing).

The sixth experiment in Gongzhuling, Jilin Province (42°30’N, 124°48’E; 220 m elevation) features Typical Black Soils on Quaternary loess-like sediments63 under a cold-temperate climate (mean annual temperature 4–5 °C; precipitation 550 mm). Established in 1990 for continuous summer maize (single annual cropping with post-harvest fallow) on 400 m² plots, baseline topsoil (0–20 cm) properties were: organic matter 13.2 g kg⁻¹, total N 1.4 g kg⁻¹, available P 11.79 mg kg⁻¹, pH 7.6. Four fertilization treatments, each replicated three times, were implemented: (1) Unfertilized control (CK); (2) Annual mineral NPK (urea: 165 kg N ha⁻¹, superphosphate: 82.5 kg P ha⁻¹, potassium sulfate: 82.5 kg K ha⁻¹); (3) NPK + maize straw amendment (7.5 Mg ha⁻¹ yr⁻¹ dry weight, shredded to 5–10 cm and incorporated during side-dressing); and (4) NPK + cattle manure amendment (23 Mg ha⁻¹ yr⁻¹ dry weight applied post-harvest).

Among the above six field experiments, WQ, YL, PL located in northwestern, GZL located in northeastern, while ZZ and CP belong to northern China. Soil samples were taken from each field experiment following the maize harvest in September 2003. Samples were taken from five random locations in each treatment at 0–20 cm, and mixed thoroughly to create one composite soil sample. At the same time, soil bulk density samples were obtained at 10 cm in each treatment using circular cutting rings (100 cm³). In total, 54 composite soil samples were analyzed (3 fertilization treatments × 3 replicates × 6 field experiments). After collection, samples were combined, mixed, and manually processed to remove visible roots and stones and then air-dried to assess the contents of SOC, TN, TP, TK, as well as pH. The effect size of manure and straw was calculated as follows:

$${Effect\; size}=({{SOC}}_{{sampled}}-{{SOC}}_{{initial}})/{{SOC}}_{{initial}}\times 100 \%$$
(6)

where SOCsampled and SOCinitial represent the SOC stock in 2003 and the time before starting the experiment, respectively.

Data analysis

For the data obtained from historical data in 1980s and repeated measures in 2010s, statistical analyses were performed using the open-source software R (version 4.4.2; R Core Team, 2019). The normality of the data was checked using the Shapiro–Wilk’s test, and log-transformation was used if data were not normally distributed (Fig. S3). To evaluate the significance of the changes in SOC, a paired t test was first used to assess the difference in SOC stock at 0–20 cm soil between the 1980s and the 2010s. Second, we used one-way analysis of variance (ANOVA) to test the difference in SOC stock among regions (north, northeast, and northwest). Tukey’s honestly significant difference (HSD) test was used for multiple comparisons at p < 0.05. Finally, linear correlation was performed to examine the correlations between ΔSOC and ΔET0, ΔAI, ΔMAT, and ΔMAP. We employed multiple regression to test whether ΔET0, ΔAI, ΔMAT, and ΔMAP, and ΔTN, ΔpH, were significant predictors for the effects of original SOC and climatic variables on SOC changes (ΔSOC) across the 721 sampling sites.

Linear and non-linear regression analysis was performed to examine the correlations between the ΔSOC stock or content and the initial pH, SOC and TN stocks, as well as soil C: N in 1980s, MAT, MAP, MET, MAI, input of fertilizers and straw using the “stats” package. Although using initial SOC stocks as a predictor variable may lead to the analysis being susceptible to any regression to the mean artifacts64, disentangling the effect of regression to the mean from the true underlying relationship between ∆SOC and the initial SOC is possible given the appropriate statistical approach. Here, we applied the correction from Blomqvist (1977) to rectify the bias65. The relationships of ΔSOC stock with the straw C input and crop residue C input were predicted by a logarithmic regression model, which was based on a logarithmic function y = a × log(x) + k, where y is the predicted ΔSOC stock, x is the fitted variable (straw C, residue C), a and k are the regression coefficients. The estimation of the coefficients in the model was determined by the nonlinear least-squares (NLS) method using the “gslnls” package. Spearman’s rank correlation coefficients were analyzed to assess the correlation between the ΔSOC and the initial soil properties, climatic factors, as well as agronomic managements.

We then used structural equation modeling (SEM) to assess the impacts of controlling factors including initial soil properties (soil pH in the 1980s, total N and SOC stocks in the 1980s), climatic factors (MAT and MAP), and management practices (mean annual chemical fertilizer input, crop yield, and crop residue C input) on ΔSOC stock from dryland areas in northern, northeastern, and northwestern China, respectively, using the “lavaan” package. During the SEM analysis, we hypothesized that soil pH and TN stock in the 1980s, crop yield, and fertilizer input have an indirect effect, and SOC stock in the 1980s, crop residue C input, MAT, and MAP have a direct effect on ΔSOC stock, respectively, based on the correlation analyses among all variables. The fitness of the SEM was evaluated using the χ2 goodness-of-fit test and the root mean square error.

For the obtained data from long-term field experiments, statistical analyses were performed using the open-source software R (version 4.4.2). The normality of the data was checked using the Shapiro–Wilk’s test, and log-transformation was used if the data were not normally distributed. Two-way repeated ANOVA was applied to evaluate the effects of organic amendments and time on crop yield, SOC, TN, TP, AK, and pH. One-way ANOVA was conducted to evaluate the influences of organic amendments on the determined soil parameters at the same time point. If significant differences were found, the least significant difference (LSD) analysis was conducted. Random forest modeling was used to identify the most important predictor variable that drives the changes of SOC in the six long-term field experiments, using the “randomForest” package. The relative importance of each predictor variable was quantified based on its percentage increase in the mean squared error (MSE). The higher MSE% values of the predictor variables indicate greater importance.

Reporting summary

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