Introduction

Soil is the largest carbon reservoir in terrestrial ecosystems, with soil organic carbon (SOC) stocks (SOCS) of approximately 3,200 Pg, and has a substantial impact on the global carbon cycle1. Even small changes in SOC can significantly affect atmospheric carbon dioxide concentrations, affecting the balance of the global climatic system2. Climatic conditions, human activities, vegetation, and soil physicochemical properties are the main factors influencing changes in soil carbon pools3,4. Particularly in rapidly urbanising areas, the shift from natural to human-managed ecosystems results in increased vulnerability of land to degradation, leading to the loss and release of SOC, which undermines the stability of soil carbon pools5. Urban SOCS (USOCS) account for a significant proportion of global SOCS, and changes in these stocks have important implications for the global carbon balance. Urban areas cover 3% of the global land surface, house 60% of the world’s population, and drive 75% of global CO2 emissions6,7. With the ongoing rapid global urbanisation, the total amount of urban land is likely to increase by a factor of 1.8 to 5.9, and this expansion, accompanied by material production and infrastructure development, is expected to lead to further increases in CO2 emissions8,9,10. Although localised environmental problems associated with urbanisation11, such as heat stress and air pollution, are now well-recognised, the impact of urbanization on regional and global carbon cycles remains uncertain.

Most current studies have focused on specific regions or single scenarios, and there have been few comprehensive assessments of changes in the SOCS under different urbanisation scenarios on a global scale12,13,14,15,16. These studies have focused on wealthy countries (e.g. United States and United Kingdom), with little research on urban soil organic carbon (USOC) in rapidly growing developing countries (e.g. China and India). Even when independent studies have provided clear and important models, obtaining a broad perspective on the effects of climate, vegetation, or anthropogenic factors on SOC is difficult. Although studies have attempted to quantify global USOC density (USOCD)17,18,19, they usually have not used careful differentiation between different stages of urban development, land use patterns, and climatic conditions. For example, cities may exhibit different soil properties and USOCD at different stages of development, which are closely related to urban development, land use patterns, and climatic conditions; however, quantitative analyses of the distribution of these SOCD, the factors controlling them, and their variations remain insufficient at the global level20,21,22,23. This inadequacy limits the comprehensive understanding of SOC changes during urbanisation and hinders the development of effective land management and climate change mitigation strategies. Therefore, filling the gaps in existing studies and providing a more reliable scientific basis than is in the literature is urgently necessary to conduct systematic studies on a global scale, combining different urbanisation scenarios and in-depth exploration of the long-term impacts of urbanisation on SOC reserves.

To improve the understanding of the impacts of future urbanisation on USOC, we estimated the current distribution characteristics of USOC globally by using a random forest model (RF) and then compared the impacts of predicted urbanisation scenarios on the spatial distribution of SOC from the current time to the year 2100 in combination with data on the distribution of future cities globally under RCP 2.6 and RCP 8.5. The specific objectives were to (1) analyse the characteristics of the current spatial distribution of USOC globally and (2) compare the effects of future urbanisation on the spatial distribution of SOCS under two different urbanisation pathways. This study elucidates how urbanisation alters soil organic carbon storage and informs future land use and management to manage future urbanisation and climate change challenges.

Materials and methods

To predict the changes in USOC concentration (USOCC) under the future urban expansion environment, we use the RF algorithm. The steps were as follows: (1) construct the study dataset and divide the training data into a training subset and a test subset in a ratio of 8:2; (2) construct a prediction model based on the established training and test sets; (3) evaluate the prediction model, which is used for cross-validation of the model and comparison of prediction performances; and (4) analyse the model predictions and results.

Construction of the research dataset

To construct accurate prediction models, we integrated various global datasets covering relevant environmental factors, such as climate, topography, soil, and urbanisation. The study dataset was divided into the following parts: (1) Global urban built-up area data. The dataset of global USOCC observations in urban built-up areas used in this study was collected by conducting a literature survey, primarily from the global USOCC observation dataset constructed by Guo, et al.18. Data were collected from ISI Web of Science (http://isiknowledge.com), Google Scholar (http://scholar.google.com), and China Knowledge Network (CNKI; http://www.cnki.net) to ensure broad representation and coverage of the data. This dataset records USOCC in urban areas on a global scale, covering a wide range of cities and their associated environmental contexts. (2) Climate data. Climate data were from the WorldClim2 database and contained several climate variables such as mean annual temperature (MAT), annual precipitation (MAP), precipitation seasonality (PS, fluctuations in precipitation over 1 year), and temperature seasonality (TS, magnitude of temperature change over 1 year). These data were standardised to ensure comparability across regions. Global climate zonation data were from the Köppen Climate Classification (Kottek et al., 2006); these data were divided into five climate zones: A (tropical), B (arid), C (temperate), D (continental), and E (polar). (3) Topographic data. Topographic data are an important factor affecting soil properties, especially at different elevations, where USOCC may vary significantly. The global elevation data were from the Hole-Filled Shuttle Radar Topography Mission24. (4) For soil data, we collected 20 cm soil bulk weight parameters from the SoilGRIDS database (https://www.isric.org/explore/soilgrids). (5) Future climate and urbanisation data. To investigate the potential impacts of future urbanisation expansion on the USOC, we used the CMIP6 dataset, which integrates Shared Socioeconomic Pathways (SSP) and climate models. SSP1 (RCP2.6, low-emission, sustainable development pathway) and SSP5 (RCP8.5, high-emission, fossil fuel development pathway) were selected as future climate scenarios corresponding to SSP126 and SSP585, respectively. Relevant data were from the EC-Earth3-Veg model and are available in the WorldClim database (https://www.worldclim.org/). The choice of climate scenarios helped analyse USOC changes under different emission pathways. The urbanisation data (i.e. urban land cover ratio) were from the Global Urban Extension dataset constructed by Gao and O’Neill25, which can be downloaded from Dataverse (harvard.edu/dataverse/geospatial_human_dimension.com). The dataset contains detailed projections of future urban land expansion globally, providing spatialised information on the urbanisation process for our analysis. (6) Data pre-processing and uniform spatial resolution. To ensure that all input variables were analysed at the same spatial scale, we unified the spatial resolution of all datasets to 1/8° latitude × 1/8° longitude. This spatial resolution balances the accuracy and computational efficiency of global data. Furthermore, the temporal dimension was standardised to a study unit of years to capture changes in USOC between years.

As shown in the results in Fig. 1, the observed data (377 observations) covered all major ecosystem types and climatic zones worldwide. The main concentrations were in the Americas, Europe, and East Asia. The USOCC is presented in Fig. 1b and c: it varies considerably between cities, with distributions ranging from 3.6 to 101.19 g C kg− 1. Globally, the distribution of USOCC is lowest at low latitudes and increases significantly with increasing north and south dimensions.

Fig. 1
figure 1

(a) Spatial distribution of USOCC (Urban soil organic carbon concentration) databases for tropical (A), arid (B), temperate humid (Cf), temperate summer dry (Cs), temperate winter dry (Cw), boreal humid (Df), boreal summer dry or winter dry (Dsw), and polar (E) climate regions (R version 4.3.0 https://www.r-project.org/). (b) Global frequency distribution of USOCD values. (c) Variation in SOCD data with dimension and 95% confidence interval representation.

Full size image

Construction and evaluation of predictive models

Widely used in ecological research, RF is a bootstrap nonparametric model based on regression trees. The main advantages of RF include (1) managing input samples with high-dimensional features without dimensionality reduction, (2) assessing the importance of each feature in the research problem, (3) obtaining an unbiased estimate of the internal generation error (OOB) during the generation process, and (4) obtaining good results for the default value problem26.

We completed the parameterisation of the RF model by using the RF package in R (version 4.3.0) and set the number of trees to 200; thus, for each tree in the model, a random division of the original samples was performed, a portion of the data was selected as a subset with replacement, and the remaining data was used as the data for cross-validation. We divided the data into training and prediction datasets (ratio of 8:2), in which a 10-fold cross-validation of the model was performed using the training data to check the stability of the model. Finally, we evaluated the prediction results by using the model’s root mean square error (RMSE), coefficient of determination (R²), and other metrics to ensure that the model had high accuracy and robustness for predicting changes in the USOCC in the context of urbanisation. Finally, we predicted the trends of USOCD under different climate scenarios (e.g., SSP1 and SSP5) in 2040, 2060, 2080, and 2100 under the low-emission and high-emission pathways, respectively, to investigate the dynamic distribution characteristics.

Processing of USOC data

To estimate and map future changes in USOCD and USOCS in the context of urbanisation and development, we converted USOCC data to USOCD (Mg C ha− 1), calculated according to Eq. 1.

$$SOCD=SOCC \times BD \times D/10$$
(1)

where SOCC is soil organic carbon concentration (g C kg− 1), BD is soil bulk density (g cm− 3), and D is soil thickness (cm).

To ensure the comparability of data from different studies, raw SOCD values were normalised to a depth of 20 cm, according to Eqs. 2 and 3.

$$Y=1 - {\beta ^d}$$
(2)
$${X_{20}}=\frac{{1 - {\beta ^{20}}}}{{1 - {\beta ^{d0}}}} \times {X_d}_{0}$$
(3)

where Y is the proportion of SOCD accumulated from the soil surface to depth d (cm), β is the relative rate at which SOCD decreases with soil depth, X20 is the SOCD in the upper 20 cm, d0 is the original soil depth (cm), and Xd0 is the original SOCD. We used a value of 0.9786 for β, which has been widely used in existing studies, including urban ecosystems studies17,20.

After completing the normalisation of the SOCD, we calculated the USOCS for PgC. The formula is in Eq. 4.

$$USOCS=\frac{{USOCD \times UA}}{{{{10}^6}}}$$
(4)

USOCD is the standardised SOCD in Mg C ha− 1; UA is the urban built-up area in hectares (ha), which was obtained by multiplying the area of each grid point (with a spatial resolution of 1/8°) on a global scale by the urban land cover ratio of the corresponding grid and then summing to obtain the urban area data.

Results

Main influencing factors of USOC

An RF model based on a seven-variable parameterisation predicted global USOCC values at a 1/8° spatial resolution (R2 = 0.53; RMSE = 10.26). The model revealed the order of importance of the variables: PS, MAT, climate type (Köppen), TS, mean annual precipitation (MAP), degree of urbanisation, and topographic elevation (elevation) through changes in mean squared error and nodal purity. The interaction and bias dependence plots revealed the differential effects of different environmental variables on USOCC. As shown in Fig. 2, PS and MAT were the most significant variables affecting USOCC. The strong interaction between PS and MAT suggests that these climatic factors play important roles in regulating carbon fixation and decomposition rates. Specifically, soil moisture fluctuations due to seasonal high precipitation contribute to the accumulation and release of SOC, and higher mean annual temperatures accelerate the mineralisation of organic matter, reducing carbon stocks. The effect of TS on USOCC was also significant, with increasing TS increasing the volatility of carbon stocks, which may be related to the regulation of mineralisation rates by temperature differences. The partial dependence plot further showed that PS and TS explained 37.12% and 20.72% of the variance, respectively, suggesting that seasonal variation had a significant impact on the dynamic balance of carbon. Although the direct effect of urbanisation had a lower variance (6.8%), it showed a significant negative trend with USOCC. Climate type (Köppen) also significantly affected USOCC, especially in humid climates (Cf, Df), where USOCC was significantly higher, possibly due to humid conditions that promote plant growth, which increases organic matter inputs to the soil. These results reveal the complex driving mechanisms of different environmental variables and their interactions with USOCC, highlighting the importance of climatic factors, the natural environment, and urbanisation in maintaining the carbon dynamic balance.

Fig. 2
figure 2

(a-g) shows the biased dependence analysis of environmental variables on USOCC. Blue and purple areas indicate the density of the distribution of the data, with purple representing low density and blue representing high density. The black curve represents the generalised linear regression fit. (h) Relationship plot of importance and interaction of environmental variables, where node size indicates the importance of the variable, thickness and colour depth of the connecting line indicates the strength of the interaction between the variables, ‘Vint’ denotes the interaction of the variables, and ‘Vint’ denotes the importance of the variables.

Full size image

Global Distribution of USOCC in 2000

Global USOCC generally showed a distinct latitudinal distribution pattern (Figs. 1C and 3a), with high USOCD concentrated in high-latitude regions of the Northern Hemisphere, especially in Europe, and low USOCD concentrated in some low-latitude regions of Asia, Africa, and South America. As shown in Fig. 3, the global mean USOCD was 62.95 Mg C ha− 1. Per the predicted USOCD data, the estimated global total USOCS was 14.53 Pg C, 2.6% of total global SOC storage at equivalent soil depths (543 Pg C)18. On a national scale, the top 10 countries with the highest USOCD were mostly high-income European countries, with Belgium having the highest USOCD, 128.64 Mg C ha− 1. The 10 countries with the lowest USOCD were mostly in low-income or middle-income low-latitude regions, and Bangladesh had the lowest USOCD, 37.22 Mg C ha− 1, globally. Significant differences in SOCD between countries and regions imply that climatic conditions and natural terrestrial ecosystems are responsible for the differences and that urban sprawl development between countries and regions may affect USOCD.

Fig. 3
figure 3

(a) Distribution of global USOCC projections for 2000 with a spatial resolution of 0.8° (R version 4.3.0 https://www.r-project.org/). (b) Mean USOCD predicted for the top 10 countries globally. (c) The 10 countries with the lowest ranked mean USOCD globally. (b-c) Error lines in b and c indicate 95% confidence intervals.

Full size image

Future temporal and spatial changes in the global USOCD

To investigate the dynamic trends of the global USOCD under future urban expansion scenarios, we parameterised the model by combining relatively stable external environmental factors with data on the spatial and temporal distribution of future urbanisation to predict the spatial and temporal rates of change in the global USOCD from 2000 to 2100. We selected two urbanisation development scenarios, SSP126 (representing sustainable urbanisation development) and SSP585 (representing rapid urban development), which correspond to global radiation intensity reaching 2.6 and 8.5 W m− 2 by 2100, respectively. Figure 4 illustrates the projected changes in the global USOCD under different scenarios. Under SSP126, changes in USOCD were relatively moderate and concentrated in specific regions with accelerated urbanisation, such as China and India in Asia, parts of North America, and parts of Europe. The USOCD in these regions showed a clear decreasing trend; however, the reduction in USOCD was relatively small, and the change in the USOCD was mainly influenced by local urbanisation rather than drastic global changes. By contrast, the changes in USOCD under SSP585 were more pronounced and broader in scope than those under SSP126. Significant reductions in USOCD under the rapid urbanisation scenario are shown in Fig. 4 for the eastern and midwestern United States in North America, for the European regions, and for China and India in Asia. These reductions were not confined to the rapidly urbanising countries; they spread to a wide range of regions, with a significant increase in the area covered. This trend suggests that rapid urban expansion has been accompanied by the transformation of many natural ecosystems, including significant reductions in agricultural land, forests, and wetlands, and that rapid urbanisation has further exacerbated the loss of USOCD. Furthermore, changes in USOCD in different regions largely reflected interregional variability in economic development and urban expansion. For example, between 2000 and 2010, the urbanised area in Europe increased by 15,000 km2, from 144,000 to 159,000 km2, and the urban land area in Africa increased from 46,000 to 63,000 km225. This significant regional variation suggests that the impact of urbanisation expansion on USOCS is not uniform worldwide but is driven by regional patterns of economic development and urban expansion strategies. To further analyse future spatial and temporal changes in USOCD across continents, we differentiated the data by continent and compared the changes in USOCD over a 100-year period under SSP126 and SSP585. The results (Fig. 5) showed that Europe, Asia, and North America had the most significant decline in USOCD under SSP585, and Africa and South America had smaller changes in USOCD than the aforementioned continents. The changes in global USOCD are more dramatic under the high-emission scenario, showing the enormous pressure on land resources and ecosystems that rapid urban expansion causes.

Fig. 4
figure 4

Future global USOCD percentage changes under two scenarios, SSP126 and SSP585. (a)-(b) USOCD change in 2000 compared with 2040; (c)-(d) USOCD change in 2040 compared with 2060; (e)-(f) USOCD change in 2060 compared with 2080; (g)-(h) USOCD change in 2080 compared with 2100 (R version 4.3.0 https://www.r-project.org/).

Full size image
Fig. 5
figure 5

Distribution of kernel densities for percentage change in USOCD by continent under different climate scenarios. (a-f) Distribution of kernel density estimates of the percentage change in SOCD between 2000 and 2100 for each continent under SSP126. (g-l) Distribution of kernel density estimates of the percentage change in SOCD between 2000 and 2100 for each continent under SSP585. Black dashed line represents the 0% change lin; blue dashed line represents average percentage change for that continent.

Full size image

Relationship between dynamic urban expansion and USOC

The relationship between urbanisation and USOCD showed significant dynamics under future urban expansion scenarios. Figure 6 shows that under SSP126, despite the gradual increase in the urbanisation rate, the global average USOCD remained relatively stable and did not decline significantly. This suggests that under a moderate urbanisation development pathway, the impact of urbanisation on SOC is more limited, and although the urbanisation rate increases steadily over time, the sustainable urbanisation pattern seems to interfere relatively little with soil carbon stocks, and the USOCD level is able to remain relatively balanced at a high level. By contrast, under SSP585, USOCD showed a significant downward trend, in contrast with the rapid increase in urbanisation rates. This trend revealed a negative correlation between urbanisation expansion and the declining USOCD. With the rapid increase in the urbanisation rate, land use changes have increased, and natural ecosystems have been gradually replaced by urban infrastructure and construction land, implying that land cover changes directly contribute to the continuous decline of the USOCD, further suggesting that high-intensity urbanisation will have a sustained reduction effect on SOC.

Although USOCD continued to decline under SSP126 and SSP585, global USOCS showed a steady growth trend (Fig. 7). The increase in USOCS was more significant under SSP585 than under SSP126. Compared with the baseline value of USOCS in 2000 (14.53 Pg C), USOCS was projected to increase by 178.75% by 2100 under SSP585 compared with 66.13% under SSP126. This indicates that the increase in USOCS was significantly accelerated under the high emissions and rapid urbanisation scenario (SSP585). Further comparison of the changes in USOCS in different climatic zones revealed that USOCS was the highest in the temperate humid continental climate (Df climate zone), accounting for approximately 28% of the total global stock. However, an increase in USOCS did not equate to an increase in the total global SOCS. The increase in USOCS was mainly due to the transformation of natural ecosystems into urban sites, which has resulted in the reclassification of carbon stocks from the original ecosystems into USOCS. This process may result in the loss of natural soil carbon stocks; thus, the increase in USOCS is, to some extent, an artefact of carbon transfer.

Fig. 6
figure 6

Trend map of the impact of future urbanisation expansion on global average USOCD. (a) denotes SSP126 and (b) denotes SSP585. Red circles in the box plots are the global mean USOCD; the black straight line is the median. Black stars in the box plots indicate the global average urban land cover ratio for the corresponding year.

Full size image
Fig. 7
figure 7

Trend of urban soil organic carbon stocks (USOCS) over time in different climate zones. (a) USOCS data for each climate zone under SSP126 and (b) USOCS data under SSP585.

Full size image

Discussion

Global carbon balance cycle patterns

This study examined global projections of the distribution and future changes in USOCD by using a comprehensive dataset from Gao and O’Neill25. The global distribution of USOCD was influenced by a combination of factors, among which climatic and natural environmental factors played fundamental regulatory roles on large scales. As shown in Fig. 1, USOCD increased significantly with increasing latitude, which is consistent with the distribution pattern of natural soil carbon stocks27,28,29. At high latitudes, lower temperatures and higher soil moisture contribute to the accumulation of organic matter, as microbial activity weakens, and the decomposition of organic matter slows at low temperatures. Furthermore, vegetation types such as coniferous and deciduous broadleaf forests at high latitudes provide rich inputs of organic matter, which further increases soil carbon stocks. By contrast, in hot and humid areas at low latitudes, the decomposition rate of organic matter is higher, and microbial activity is more vigorous, resulting in relatively lower SOC stocks22,30,31. This phenomenon indicates that climatic conditions are one of the main natural factors affecting the distribution of USOCD. In addition to climatic factors, other natural environmental factors influenced the distribution of USOCD to varying degrees. Seasonal variations in precipitation and temperature significantly affected the accumulation and decomposition of soil C. High PS increases soil moisture volatility, which affects carbon fixation and release, and TS indirectly affects microbial activity and organic matter decomposition rates by altering the magnitude of intra-annual soil temperature fluctuations32,33. The interactions of these climatic variables constitute the natural distribution pattern of USOCD in different regions.

However, as the stability of natural environmental factors gradually breaks down, levels of economic development and urbanisation begin to have more complex effects on USOCD21,34. We found substantial differences in USOCD across countries, which may be the main reason why urban land use and management led to changes in USOCD. Developed countries tend to adopt finer land management and higher greenspace coverage, which results in relatively high USOCD in these regions. For example, within the same latitude range, Western European countries (e.g. Belgium) had a USOCD of 128.64 Mg C ha− 1, the United States had a USOCD of 94.11 Mg C ha− 1, and China had a USOCD of 63.36 Mg C ha− 1 (Fig. 3). This difference is closely related to the level of urban development and land management patterns in different countries. Developed countries have maintained or even increased the carbon stock of urban soils by implementing rational land use and management, and in developing countries, owing to faster urbanisation, soil carbon stocks may not be effectively protected and managed, and USOCD is therefore lower35,36,37. Therefore, the distribution pattern of USOCD is the result of multiple factors such as climate, natural environment, level of economic development, and urbanisation. Although natural environmental factors play an important and fundamental role in regulating the distribution of USOCD at large scales, changes in the urbanisation process and the level of economic development play a decisive influence at regional scales. The combined effect of these three factors will also be more favourable for the analysis of the global pattern of SOC.

Impact of future urbanisation on USOC under climate scenarios

In this study, we projected changes in the global USOCS under two scenarios: SSP126, sustainable urbanisation development, and SSP585, rapid urbanisation development. These projections showed that the two urbanisation development scenarios had significant impacts on future USOCS, mainly in the rate of change in USOCD and USOCS. First, under SSP126, the change in the global USOCS was more moderate than that under SSP585. They were primarily concentrated in China, India, Asia, North America, and Europe. USOCD decreased in these regions, but the overall change was relatively small (Fig. 5). For example, under SSP126, global USOCD was projected to decrease by approximately 1.19% by 2100. By contrast, under SSP585, the change in global USOCD was more extreme (4.72%) than that under SSP126, with the former having a wider range of regions experiencing significant reduction and greater variability than those of the latter (Figs. 4 and 5). This phenomenon was particularly true in the eastern and midwestern United States in North America, Central and Eastern Europe, and China and India in Asia. Furthermore, there were significant differences in the impact of urbanisation on USOCD in different regions. This study found that continental Europe would experience the greatest global loss of organic carbon, with an average decline of 9.72%. The main reason for this finding is that under future global warming environments, rising temperatures and changes in precipitation patterns will accelerate the decomposition and loss of organic carbon, together with the destruction of natural ecosystems as a result of urbanisation, affecting the accumulation and retention of SOC.

We compared the USOCS of different climatic zones. The results demonstrated that the tropical climatic zones showed the least change in USOCS under the two urbanisation scenarios, and the Df climatic zones showed a higher USOCS than that in the tropical climactic zones and a significant increase in USOCS compared with that in the tropical climactic zones. The reason for this is that the influence of climate variables on the spatial pattern of global urban organic carbon is more dominant than that of anthropogenic urbanisation. Tropical climate zones showed the smallest change in USOC, mainly because of their high temperatures and rainfall, high biological productivity, vigorous soil microbial activity, fast decomposition of organic matter, and slow accumulation of SOC, which is maintained in a dynamic equilibrium33. Therefore, even if urbanisation results in land use changes, its impact on the USOCS will be limited. By contrast, although the Df climatic zone had many densely populated and highly urbanised areas, its cold climate and low rate of SOC decomposition, coupled with the large input of organic matter provided by deciduous broadleaf and mixed forests, contributed to the accumulation of SOC. Thus, the shift from natural to managed ecosystems during urbanisation has resulted in a marked increase in USOCS (Fig. 5)38,39. However, although the USOCS in some areas showed an increasing trend, it does not represent an increase in total global SOCS10. This increase in USOCS reflects the phenomenon of carbon transfer during the transformation of natural ecosystems to urban land, whereby carbon stocks in the original natural ecosystems are redistributed to urban soils. With the conversion of high-carbon stock natural ecosystems such as forests and grasslands, the actual total amount of soil carbon stock decreases, and part of the carbon is released into the atmosphere through the conversion process, which adversely affects the global carbon balance40,41,42,43. In summary, the rapid growth of the USOCS in the future reflects that changes in carbon stocks due to urban expansion weaken the soil’s ability to store carbon and lead to an increase in the release stored carbon. Therefore, future urban planning and land management policies need to balance the relationship between urban expansion and the protection of carbon stocks in natural ecosystems while paying attention to the USOCS.

Limitations of predicting USOC

In this study, we estimated the global USOCD and quantified the role of climate, urban development, and topographic factors in explaining its spatial pattern. Other factors may also influence the USOCD, such as physicochemical soil variables, parent material, and city age. However, obtaining data related to other factors, such as the parent materials of urban soils, many of which include materials transported from other locations, is difficult44. Furthermore, the age of a city indicates the length of time over which SOC accumulates because topsoil organic matter is often stripped and lost during urban construction. As a result, a higher USOCD is usually found in older urban soils45. Because a large proportion of newly established urban areas are in developing countries at low and middle latitudes46, the ages of these cities may be spatially correlated with MAT and TS. In our analyses, this potential covariance may have led to an overestimation of the impact of MAT and TS on the spatial patterns in USOCD. Unfortunately, our ability to assess the ages of cities quantitatively was limited by the limited availability of relevant data. Anthropogenic disturbances, such as land use changes, topsoil removal, or soil inputs from elsewhere, are commonly observed during urban expansion, but we were unable to assess their potential impacts in our analyses. The data limitations described may have led to a potential underestimation of anthropogenic impacts on spatial changes in USOCD.

At the same time, there are limitations in our study, i.e., obtaining USOC data from a single measurement point in the greenfield area of the city in order to approximate the USOCD and USOCS for the whole city. Despite the feasibility of this method with limited data, this may lead to an overestimation of the USOCS due to the higher USOCD of the greenfield area. Therefore, we have analysed the potential bias of this method in the discussion and suggested that more USOC data from different areas of the city should be integrated in future studies to improve the accuracy of the estimation.

In summary, future research should aim to collect more comprehensive data to better analyse and validate the effects of urbanisation, climate change and other variables on USOC. In addition, the introduction of more contexts and variables will help to improve the accuracy and reliability of the models, thus providing a more solid scientific basis for urban soil organic carbon management and policy formulation.

Conclusion

In this study, we used 377 urban observations to model and assess spatial and temporal changes in global USOC and its drivers and quantified future changes in USOC by using urban sprawl under the Shared Socioeconomic Pathways (SSP126 and SSP585) model. The results of this study showed that TS and the average annual temperature were the main factors influencing USOC distribution, overriding the effects of factors such as precipitation and elevation. Compared with 2000, there will be an overall global decline in USOCD: a projected decline of 1.19% and 4.72% under SSP126 and SSP585, respectively. This decline is mainly due to the conversion of natural ecosystems (e.g. forests and grasslands) as a result of urban expansion, which reduces SOC inputs to these ecosystems and increases carbon emissions. Despite the decline in USOCD, the global urban organic carbon stock showed an overall increasing trend. Compared with 14.53 Pg C in 2000, USOCSs were projected to increase by 66.13% by 2100 under SSP126 and by 178.75% under SSP585. This increase was due to the shift from natural ecosystems to urban land use, resulting in the reallocation of carbon stocks from the original ecosystems to USOCS. Despite the apparent increase in USOCS, this ‘carbon transfer’ illusion may actually lead to an increase in carbon emissions, affecting the global carbon balance.