Introduction

Soil organic carbon (SOC) plays a dominant role in regulating Earth’s climate through the carbon–climate feedback1,2. Recent studies have suggested that global SOC storage has increased over time3,4, which is promising for mitigating climate warming and the realization of carbon neutrality through ecosystem approaches5. SOC can be separated into two major fractions: particulate organic carbon (POC) and mineral-associated organic carbon (MAOC)6. These two fractions differ in formation mechanisms, functions, and turnover times, and can be envisaged as ‘unprotected’ and ‘protected’ SOC fractions respectively7. Specifically, soil POC is primarily derived from plant materials, is readily accessible to microorganisms and thus, has a relatively short mean residence time in the soil, ranging from years to decades8. In comparison, soil MAOC is derived mainly from microbial byproducts and is associated with clay and silt-sized minerals9, which is therefore protected by mineral chemical bonds and has a longer mean residence time than POC in soil, ranging from decades to centuries10. Furthermore, the relative distribution of unprotected versus protected carbon in soil maintains SOC storage capacity and alter SOC stability in the face of climate change and human activities11,12,13,14. However, temporal changes in the relative distribution of unprotected and protected SOC fractions remain unquantified.

The impacts of human activities on soil POC and MAOC differ14,15,16, influencing SOC storage and its role in climate change mitigation17. For example, decades of forest restoration in Europe have increased soil POC, while MAOC has remained relatively stable18. In north-central China, reduced grazing intensity in grasslands over the past three decades has led to a faster increase in soil POC compared to MAOC19. Similarly, legume intercropping and no-tillage farming, implemented as conservation practices, has promoted faster increases in POC than in MAOC over decadal timescales16. Moreover, conservation agriculture under a ten-year warming treatment not only enhanced SOC storage but also increased the proportion of carbon in protected soil fractions compared to conventional agriculture20. However, little is known about how POC and MAOC respond to climate change and human activities over time across different land types. An integrated global synthesis of the temporal patterns in SOC fractions is still lacking. To guide the development of effective strategies and policies to enhance SOC storage and stability globally, it is important to assess the impacts of human activities on the long-term changing trends of POC versus MAOC.

Here, we evaluated the temporal trends of the relative distribution of global soil POC and MAOC, and investigated the effects of human activities and land management on SOC fractions across different land cover types. To do so, we compiled a global SOC fraction dataset from online databases and published studies. Our dataset spanned a large spatiotemporal gradient, containing 7219 soil samples collected in forests, grasslands and croplands across six continents from 2000 to 2022 (Fig. 1a and Supplementary Fig. 1). Each soil sample was separated into POC and MAOC using a standardized soil fractionation technique, ensuring comparability across samples and time.

Fig. 1: Changes in soil organic carbon fractions from 2000 to 2022.
figure 1

a Geographical distribution of soil samples collected from croplands (n = 3603; red), forests (n = 2149; yellow) and grasslands (n = 1547; blue), and each dot represents a study site. Fitted lines represent the means (± 95% CIs) of the model predictions from the LMMs (Supplementary Tables 1-2) for (b) soil organic carbon (SOC; yellow) and particulate organic carbon (POC; blue) storage and (c) POC to MAOC (mineral-associated organic carbon) ratio (POC:MAOC) in global soils. b–c Each dot represents the data from a soil sample. Source data are provided as a Source Data file.

Full size image

Results

We show that soil POC represented a smaller proportion of total SOC than MAOC: POC accounted for an average of 27.4%; and MAOC for 70.0% in total SOC across all soil samples (n = 7219; Fig.1 and Supplementary Fig. 2). Despite POC’s lower weight, its rate of change was higher than MAOC’s and dominated the global increase in SOC storage. Specifically, soil POC increased by 21.8% from 2000 to 2022 (P < 0.001), while MAOC slightly decreased by 5.3% during the same period (P = 0.163), leading to a net increase of global total SOC storage by 11.0% within the recent 20 years (P = 0.003; Fig. 1b and Supplementary Table 1). The disproportionate changes between SOC fractions led to rapid increases of the POC to MAOC (POC:MAOC) ratio over this period (slope[se] = 0.012[0.002], P < 0.001; Fig.1c). This pattern was confirmed when sample size variation was considered across years (Supplementary Fig. 3).

We further explored the effects of human disturbance intensity on the variation of temporal changes in the relative distribution of SOC fractions. To do so, we generated subsets of fraction data into the following categories: natural vs. planted forests, and ungrazed vs. grazed grasslands. We found that soil POC tended to increase from 2000 to 2022 in both natural forests (slope[se] = 0.006[0.007], P = 0.367) and ungrazed grasslands (slope[se] = 0.028[0.006], P < 0.001), which were less disturbed by humans. In comparison, soil POC decreased in both planted forests (slope[se] = −0.018[0.007], P = 0.015) and grazed grasslands (slope[se] = −0.034[0.012], P = 0.003), which were subjected to high level of human disturbance (Fig. 2 and Supplementary Tables 1-2). In addition, we found that the POC:MAOC ratio increased in natural forests (slope[se] = 0.036[0.006], P < 0.001) and remained stable in ungrazed grasslands (slope[se] = 0.009[0.006], P = 0.180). In both planted forests and grazed grasslands, the POC:MAOC ratio decreased during 2000–2022 (planted forests: slope[se] = -0.018[0.007], P = 0.013; grazed grasslands: slope[se] = −0.062[0.011], P < 0.001; Fig. 2 and Supplementary Tables 1–2). Moreover, among planted forests, soil POC increased with the age of forestation (slope[se] = 0.032[0.005], P < 0.001; Supplementary Fig. 4), supporting the finding that soil POC increased with declining intensity of human disturbance (Supplementary Fig. 5).

Fig. 2: Changes in soil organic carbon fractions in forests and grasslands from 2000 to 2022.
figure 2

Fitted lines are the means (± 95% CIs) of the model predictions from the LMMs (Supplementary Tables 1-2) for changes in particulate organic carbon (POC; blue) storage and POC to MAOC (mineral-associated organic carbon) ratio (POC:MAOC; gray) over time in (a) natural forests, (b) ungrazed grasslands, (c) planted forests and (d) grazed grasslands. Each dot represents data from a soil sample. Solid lines indicate significant changes over time (P < 0.05), and dashed lines indicate non-significant changes (P > 0.05). Source data are provided as a Source Data file.

Full size image

Global croplands are intensively managed by humans and influenced by human management policies, both soil POC and the POC:MAOC ratio remained stable from 2000 to 2022 (POC: slope[se] = 0.001[0.003], P = 0.870; POC:MAOC ratio: slope[se] = 0.002[0.003], P = 0.473; Fig. 3a). We further assessed the changes in the distribution of soil SOC fractions over time in Asian and North American croplands, which represent conventional and conservative agriculture practices, respectively, according to their agriculture-related policies (Supplementary Table 3). We showed that soil POC increased in North America (slope[se] = 0.020[0.005], P < 0.001), but decreased in Asia (slope[se] = −0.020[0.005], P < 0.001) from 2000 to 2022 (Fig. 3b, c). Moreover, soil MAOC increased in North American croplands but decreased in croplands across Asia and other continents (Supplementary Table 1 and Supplementary Fig. 6), suggesting that improved practices such as conservation agriculture could increase soil POC storage as well as MAOC storage in croplands.

Fig. 3: Changes in soil organic carbon fractions in croplands from 2000 to 2022.
figure 3

The fitted lines are the means (± 95% CIs) of the model predictions from the LMMs (Supplementary Tables 1-2) for changes in particulate organic carbon (POC; blue) storage and POC to MAOC (mineral-associated organic carbon) ratio (POC:MAOC; gray) over time in (a) all croplands, (b) croplands in Asia and (c) croplands in North America. Each dot represents data from a soil sample. Solid lines indicate significant changes over time (P < 0.05), and dashed lines indicate non-significant changes (P > 0.05). Source data are provided as a Source Data file.

Full size image

Discussion

Our analyzes revealed that changes in unprotected fraction (POC) predominantly influence changes in SOC, despite POC stocks accounting for only half of the protected fraction (MAOC) in the soil carbon pools. Results from our study are consistent with model predictions of increased SOC storage over the past two decades3,21, but suggest that the distribution of SOC among fractions was altered and influenced by human activities. The rising storage of POC and the increasing POC:MAOC ratio were presumably driven by increased plant carbon inputs, enhanced by the fertilization effect of increasing atmospheric CO2 concentration22, and reduced disturbance due to the implementation of regional ecosystem protection practices23. However, microbial conversion of POC to MAOC is limited by the scarcity of mineral binding sites in soils15,24. While global SOC storage is linked to the increased net primary productivity, carbon turnover time has decreased3,25, likely due to the growing proportion of soil POC. Overall, we highlight that the increase in SOC storage was driven by increasing POC, but warn of the declining stability of SOC associated with climate change and human activities.

Tremendous efforts have been made to increase SOC via increasing MAOC, because the storage potential of protected SOC in soils is twice that of unprotected SOC13,26. However, organic carbon stored in the MAOC pool has limitations15. Mineral-protected SOC in soils is mainly formed from organo-mineral interactions with secondary minerals by weathering over long periods, ranging from decades to centuries, prohibiting the accumulation of a substantial amount of MAOC in a short time27,28. Thus, soils with a low proportion of clay and silt-sized mineral are inappropriate targets for boosting MAOC. For example, many types of soils in western Australia and southern Africa are sand-rich with low mineral binding sites, hindering the sequestration of MAOC13. In contrast, soil POC saturation depends on climatic and environmental conditions as well as human activities. Importantly, POC can continue to accumulate even when the MAOC pool is saturated, consequently increasing the total soil organic carbon storage15. Moreover, focusing solely on the enhancement of MAOC can lead to the neglect of soil function loss. We showed that MAOC increased in grazed grasslands, but the total SOC storage remained unchanged due to the rapid decline in POC. Grazed grasslands may suffer from severe soil degradation, where the accelerated weathering of soil minerals and increasing proportion of clay and silt-sized minerals promote the sequestration of MAOC in soils29. As a result, the increase in carbon induced by MAOC may fail to achieve the goal of increasing SOC storage within a short time window, especially to realize carbon neutrality by 2050 and to limit warming to 1.5 °C by 210030.

In comparison, as changes in POC dominate the changes in SOC, organic carbon stored in the POC pool can serve as a promising solution to enhancing SOC. On the one hand, human activities can directly impact on POC, driving rapid declines in both POC and SOC. For example, we showed that declines in POC and SOC were associated with the increasing intensities of human disturbance in planted forests and grazed grasslands. On the other hand, SOC will increase if POC storage continues to grow. For example, forest protection and restoration can increase biodiversity and ecosystem productivity, leading to the increase of soil carbon inputs and substantial increase in forest soil POC31,32,33. Similarly, we showed that reduced grazing intensity can increase soil POC in grasslands (Fig. 2). More importantly, an increase in POC can address soil carbon sequestration and human needs at the same time. Reducing grazing intensity in grasslands can simultaneously maintain grassland production and enhance soil carbon sequestration34. The increase in POC stimulated by protective practices consistently raises the POC:MAOC ratio. Nevertheless, POC can act as a direct precursor of MAOC15,35, with higher POC:MAOC ratios potentially facilitating MAOC formation13. As noted, increased POC not only contributes to SOC storage but also provides a foundation for MAOC formation.

An increase in POC storage has also been demonstrated under conservative agricultural management. Compared to natural ecosystems, croplands are intensively managed by humans, making soil carbon storage in these systems highly dependent on management policies and practices36. Notably, in North America, ‘The Soil Conservation and Domestic Allotment Act’ was enacted in 1935, since then many conservative agricultural practices have been implemented, including regenerative agriculture and cover crop planting37. As a result, a significant amount of crop residues has been incorporated into soil, which can be converted to soil POC and increase SOC16,38. Our findings showed that soil POC and SOC in North American croplands have increased rapidly in the recent decades, indicating the success of conservative agricultural practices. More importantly, because soil carbon sequestration processes are gradual and complex39, large-scale enhancements in SOC through the POC pool require long-term policy implementation and broad geographical scale guidance40. The straw return policy has proven to be an effective conservative agricultural practice for increasing soil POC and SOC in croplands38,41. However, the short duration of this policy (<20 years in Asia, Supplementary Table 3) may partly explain the decline in POC and SOC across extensive Asian croplands over the past two decades. Further, an increase in MAOC storage in North America but a decrease in Asia indicated that nitrogen-enriched carbon input can support microbial growth by providing microbes with a stoichiometrically balanced diet, thus resulting in higher microbial carbon use efficiency and ultimately a higher likelihood of carbon stabilization in MAOC under conservative agricultural management7,35,42. In contrast, nitrogen inputs under conventional agriculture with greater disturbance leading to poor soil structure may reduce MAOC stocks20. Therefore, with the extended implementation of conservative agricultural practices, Asia has the potential to become a significant hotspot for boosting SOC in croplands in the future.

We showed that global SOC has been increasing over the recent 20 years, which was dominated by the increases of POC. The observed patterns of increase in POC and SOC are obviously a positive sign, indicating that an increasing amount of carbon has been stored in soils (Fig. 4). We suggest that the increase trend in POC storage has great potential to increase SOC and realize the goal of carbon neutrality within a short time window, challenging the conventional priority of adopting a MAOC-targeted pool to increase SOC storage. Conversely, increasing the proportion of POC comes with declining stability of SOC. However, a higher POC:MAOC ratio indicates that appropriate management potentially enhance POC conversion to MAOC, as global soils generally exhibit MAOC saturation levels below 50%13. Interpretation of our findings within the following context can provide valuable guidelines for improving the efficiency of POC conversion to MAOC. First, reducing human disturbance can improve soil structure and potentially foster a conducive microbial environment for MAOC formation43. Under conditions of nitrogen deposition, improved soil structure can stimulate MAOC formation through enhanced microbial activities11,44. Second, deep-rooted plant species can support MAOC formation by delivering organic carbon to subsurface soils, where mineral binding sites are less saturated13,45,46,47. Third, nitrogen-rich carbon inputs such as found in undisturbed natural forest are particularly effective in supporting the nitrogen-intensive process involved in MAOC formation11,48. Generally, we highlight that POC and MAOC do not necessarily respond in a similar way to climate change and human activities49,50. However, soil POC and MAOC are interdependent in contributing to SOC storage15,35, both of which should be considered in addition to the status and dynamics of SOC.

Fig. 4: A conceptual framework illustrating temporal changes in soil organic carbon fractions across various land types globally, influenced by climate change and human activities over decadal timescales.
figure 4

Temporal variations in soil organic carbon (SOC) fractions and the particulate organic carbon to mineral-associated organic carbon (POC/MAOC) ratio differed significantly across land types, with management practices and the intensity of human activities playing key roles. Compared to undisturbed natural ecosystems, POC declines to varying degrees in planted forests, grazed grasslands, and croplands under management practices such as heavy grazing and conventional tillage. White arrows indicate an increase (upward) or decrease (downward) in the POC/MAOC ratio from 2000 to 2022, while white circles represent no significant change in the ratio (P > 0.05).

Full size image

Methods

Data Compilation

In this study, we systematically reviewed multiple online sources and collected a comprehensive set of soil organic carbon (SOC) fraction data. According to the search strings “soil AND carbon AND ((“particulate organic matter” OR “particulate organic carbon” OR “POM” OR “POC”) OR (“mineral associated organic matter” OR “mineral associated organic carbon” OR “MAOM” OR “MAOC”) OR (“light fraction” OR “LF”) OR (“heavy fraction” OR “HF”))”, we searched for original papers containing soil carbon fraction data using Google Scholar (https://scholar.google.com/) and the China National Knowledge Infrastructure (CNKI, https://www.cnki.net/) in October 2022. In addition, we tracked newly published studies using the email alert tool of Google Scholar (https://scholar.google.com/scholar_alerts) to identify relevant studies in both the Chinese and English languages until December 2023. In each selected article, we conducted a comprehensive review of the references cited to ensure the inclusion of all relevant sources. At this stage, we had 1826 articles for further review and data extraction.

We manually checked each publication, and retained those containing (1) information of land cover types (i.e., forest, grassland or cropland); (2) management types, where forests were divided into natural (naturally generated forests without obvious disturbances) and planted forests51, grasslands were divided into ungrazed (sparsely grazed and natural) and grazed (grazed by livestock or human) grasslands52,53; (3) sampling years and locations; (4) SOC fraction data collected in field sites, excluding measurements of soils from incubation treatments in chambers or indoors and sites receiving experimental treatments (warming, elevated CO2, and precipitation manipulation); (5) SOC fraction data collected from surface soils with average sampling depths between 0 and 40 cm, which support the majority of plant roots and belowground net primary productivity in soils54,55, are active in terms of nutrient cycling and maintenance of the structure and are highly influenced by climate change and human activities56; and (6) SOC fraction data collected using standardized physical fractionation techniques (see below). In total, we retained 456 articles for data extraction. In addition, we searched online databases and public reports for SOC fraction data following the above criteria, including datasets and reports from European Soil Data Center (ESDAC), Australia’s National Soil Carbon Research Program (SCaRP) and others (Supplementary Table 4). Duplicated data sources were removed. We conducted a nearest neighbor distance analysis to identify and remove spatial outliers—data points that were sampled only once and were far from other sample points57.

We only included soil fraction data that used the standardized physical fractionation techniques to ensure comparability across studies6,58,59,60,61. In the standardized physical fractionation techniques, soil was sieved to 2 mm or less and dispersed using one of the following methods to ensure adequate breakdown of the aggregates: sodium hexametaphosphate (SHMP, Na6[(PO3)6]), sodium polytungstate (Na6[H2W12O40]), sodium iodide (NaI), sonication, or shaking with glass beads. Particulate organic carbon (POC) was defined as particles in the SOC fraction with densities less than 1.6 to 1.85 g cm−3 or diameters greater than 20–63 μm. Mineral-associated organic carbon (MAOC) was defined as the fine fraction (i.e., diameters < 20–63 μm) or the heavy fraction (densities > 1.6–1.85 g cm−3), which is the counterpart to POC. When POC and MAOC were split into different components via combined size and/or density fractionation methods, the fractions were summed to obtain the total MAOC and POC.

Relevant data were obtained directly from text, tables or from figures using WebPlotDigitizer version 4.6 software (https://automeris.io/WebPlotDigitizer). In our study, stocks of soil carbon were expressed in units of organic carbon mass per unit area of land to a specified soil depth and mass (i.e., Mg ha−1). If SOC, POC, and MAOC stocks were not reported directly in the original source, we further extracted the bulk density from the source, and calculated SOC, POC, and MAOC stocks by multiplying the respective soil carbon concentration by the bulk density and sampling depth using Eq.162. Missing soil bulk density data were obtained using the Regridded Harmonized World Soil Database v1.2 (RHWSD version 1.2, downloaded on July 8, 2023; https://doi.org/10.3334/ORNLDAAC/1247) based on the latitudes and longitudes of the sampling site following previous studies34,63.

$${{OC}}_{{storage}}={{OC}}_{{concentration}}\times {BD}\times {Depth}$$
(1)

where OCstorage is SOC, POC or MAOC stocks of the investigated soil layer (Mg ha−1), OCconcentration is the concentration of organic carbon in the investigated soil layer (%), BD is the bulk density of the investigated soil layer (g cm−3) and Depth is the depth of the respective soil layer (cm).

In total, we obtained SOC and SOC fraction data from a total of 7219 soil samples collected between 2000 and 2022 across six continents (no data from Antarctica), including 2130 samples from forests (with 1038 soil samples from natural forests and 1053 from planted forests), 1475 from grasslands (with 1044 from ungrazed grasslands and 428 from grazed grasslands), and 3614 from croplands (with 1533 soil samples from Asia, and 736 soil samples from North America).

Statistical analysis

We used linear mixed models (LMMs) to assess the temporal changing trends of soil carbon fractions and soil carbon storage during 2000–2022. To account for the variability in soil carbon storage caused by variations in the soil sampling depth and geographical site, we included the average soil sampling depth and continent as random intercepts in the LMMs. Furthermore, the area of each land cover type is an important factor in global soil carbon estimates, with cropland accounting for 12% of terrestrial land area64, grassland accounting for 41%52 and forest accounting for 31%65. We included the land cover type as a random intercept weighted by its relative area of the total terrestrial land surface. We applied LMMs to the SOC, POC, MAOC and POC to MAOC ratio. In each model, the fixed predictor effect was the year in which soil sampling was conducted; average sampling depth, land cover type and continent were treated as crossed random intercepts. Before the analyzes, the SOC, POC, MAOC and POC to MAOC ratio data were loge-transformed because of the high skewness of the data distribution. Each model was constructed with a Gaussian error distribution with an identity link using the ‘lme4’ package v1.1-3166 in R v4.2.267.

To account for the variation of sample size across years (Supplementary Fig. 1e), we conducted additional sensitivity analyzes with randomly sampled subsets of the data. Specifically, we randomly sampled 100 soil samples within each year and refitted all the models with the data subsets. The random-sampling procedure was repeated 999 times, and the coefficient estimates were extracted for comparison to the results of the models using all the soil samples. Moreover, the removal of spatial outliers (Supplementary Table 5), the selection of linear models (Supplementary Table 6), and the application of soil fractionation methods (Supplementary Table 7) do not alter the conclusions of our study.

To assess the effects of human disturbance and management on temporal trends in the soil carbon fraction, we re-conducted above analyzes with a subset of soil samples within each management type of each land cover type. For each subset of soil samples within each group, we applied LMMs to SOC (loge-transformed), POC (loge-transformed), MAOC (loge-transformed) and POC:MAOC (loge-transformed), with a Gaussian error distribution with an identity link using the ‘lme4’ package v1.1-3166 in R v4.2.267. In each model, the fixed predictor effect was the year of soil sampling; the sampling depth and continent were treated as crossed random intercepts.

Compared to forests and grasslands, croplands are fully managed by humans and influenced by human management policies. We thus further categorized croplands on different continents to assess the effects of human management on the temporal trends of the cropland soil carbon fraction. In this study, we focused on croplands in Asia (n = 1533) and North America (n = 736) because of the substantial sample sizes available over time from these two continents and that they are typical cases of the implementation of conventional and conservative agriculture practices, respectively (Supplementary Fig. 1 and Supplementary Table 3). For soil samples collected from North American and Asian croplands, we applied LMMs to SOC (loge-transformed), POC (loge-transformed), MAOC (loge-transformed) and POC:MAOC (loge-transformed) data respectively, with a Gaussian error distribution with an identity link using the ‘lme4’ package v1.1-3166 in R v4.2.267, and average sampling depth and country as a random intercept.