Various responses of global heterotrophic respiration to variations in soil moisture and temperature enhance the positive feedback on atmospheric warming

Abstract

The ratio of soil heterotrophic respiration to total soil respiration (R_h/R_s) is critical for soil carbon pool stability and atmosphere-biosphere exchanges, yet its spatiotemporal dynamics and responses to environmental drivers remain poorly constrained. Here, we estimate global soil and heterotrophic respiration by integrating ground observations with machine learning models. From 1980 to 2022, annual increases reached 0.76 and 0.54 g C m⁻² yr⁻² for soil and heterotrophic respiration, respectively. Globally, R_h/R_s averaged 61.30 ± 0.54% with a decadal rise of 0.13%. Rising soil temperature enhances R_h/R_s, while soil moisture suppresses it, exhibiting stronger global-scale influence. Temperature dominates R_h/R_s regulation in forests and shrublands, whereas moisture controls this ratio in grasslands and croplands. These findings elucidate ecosystem-specific mechanisms governing R_h/R_s dynamics, advancing predictions of soil carbon-climate feedbacks essential for carbon neutrality strategies.

Introduction

The carbon stored in global soils is at least twice that in the atmosphere^1,2. As the largest source of carbon dioxide emissions to the atmosphere in the terrestrial carbon cycle, soil respiration (R_s) releases 68 − 108 Pg C yr^-1, playing a crucial role in regional and global carbon cycling^3,4,5. Consequently, even minor changes in Rs can have great impacts on atmospheric carbon dioxide concentrations^6,7. Although numerous field experiments^8,9,10 and modeling analyses^11,12,13 indicate great changes in R_s in the context of global change, this change may still have uncertainty in areas with fewer observation points², and the magnitude of this long-term change needs to be further investigated in the context of warming¹⁴.

The complexity of soil respiration’s response to global change stems from our incomplete understanding of its component dynamics^6,15. R_s releases carbon dioxide to the atmosphere through heterotrophic respiration (R_h) and autotrophic respiration (R_a). R_h primarily originates from microbial decomposition of litter and soil organic matter, while R_a mainly comes from metabolic activities of plant roots and rhizosphere microorganisms^16,17. Quantifying the relative contributions of R_a or R_h to R_s is crucial for a better understanding of soil C storage, which largely determines the balance between soil carbon inputs and outputs¹⁸. Furthermore, isolating R_h from R_s can also provide insights into soil C stability^19,20. Increases in R_a typically do not generate positive feedback to climate change, as plant respiration often balances with production on the timescale of anthropogenic climate change, whereas R_h does not^21,22. It is well-established that R_h serves as a critical indicator of soil organic carbon (SOC) decomposition and plays a key role in SOC stability¹. Soil microorganisms can convert unprotected carbon into carbon dioxide, leading to soil carbon loss^23,24. A study based on a large amount of observational data found that global R_h/R_s have an upward trend over time, and conducted a systematic analysis of the mechanism of this change, providing valuable insights into soil carbon cycling². However, the spatial heterogeneity of this proportion’s temporal variation is not yet fully understood. Therefore, understanding the dynamic characteristics of heterotrophic respiration and its proportion in soil respiration, as well as their responses to global change, is essential for constructing and accurately predicting terrestrial ecosystem carbon cycle models¹⁹.

The two components of R_s may respond to environmental changes in different ways^19,25. Despite considerable attention given to partitioning R_s into R_a and R_h, quantifying these fluxes under global climate change remains a challenge^26,27. For instance, numerous studies have shown that warming will promote both R_a and R_h, but there is still debate regarding their relative temperature sensitivities^28,29. Some experimental studies have found that R_h is more sensitive to warming than R_a in forest ecosystems^30,31, while others have reported opposite results³². Other studies have suggested that R_a and R_h respond similarly to increased temperature³³. Another important factor influencing R_a and R_h is water availability, but the mechanisms of this influence are also subject to debate³⁴. Existing research indicates that, compared to R_a, increased water availability may promote R_h to a greater extent^35,36, to an equal extent³⁷ or to a lesser extent^38,39. Given the differential responses of R_a and R_h to environmental changes, as well as the potential direct interaction effects², understanding these differences will help reveal the response and feedback mechanisms of R_s to global change^40,41.

At the site and regional scales, there has been considerable research on R_s and its components’ responses to environmental changes, yet great differences exist among various studies^42,43. However, comprehensive assessments at the global scale are relatively scarce. Due to regional heterogeneity, accurately assessing the dynamic characteristics of R_s and its components at the global scale, as well as their responses to environmental changes, requires a fundamental prerequisite: the availability of big data that matches the global scale. Machine learning methods, which have rapidly developed in recent years, possess unique advantages in analyzing carbon fluxes in complex environments^44,45,46. Therefore, in this study, we collected and compiled extensive ground observation data and multi-source remote sensing data, forming a large database for estimating soil respiration. Additionally, we leveraged machine learning to simulate soil respiration and its components in global terrestrial ecosystems from 1980 to 2022, supported by this big data. On this basis, we analyzed the impacts of changes in hydrothermal conditions on soil respiration and its components using partial correlation analysis, structural equation modeling and variance partitioning analysis. The objectives of this study are: (1) to estimate the spatiotemporal dynamic characteristics of soil respiration and its components at the global scale; and (2) to quantify how soil respiration components respond to changes in hydrothermal conditions in the context of global change. This study represents a step forward in understanding the dynamic characteristics of global soil carbon fluxes and their responses to global change, providing important references for constructing and accurately predicting global terrestrial ecosystem carbon cycle models.

Results

The spatiotemporal dynamic of global terrestrial R_s and R_h

The spatial patterns of global terrestrial R_s and R_h exhibit a gradual decrease from the equator to the poles, aligning with the spatial pattern of global surface temperature (Fig. 1). The annual total of global terrestrial R_s is approximately 102.19 ± 1.97 Pg C yr^-1, with an average value of 826 g C m^-² yr^-1, while R_h amounts to 60.92 ± 1.33 Pg C yr^-1, averaging 493 g C m^-² yr^-1. Higher values are predominantly concentrated near the equator, particularly in the Amazon rainforest, Congo Basin rainforest, and Indonesian rainforest regions. The average coefficient of variation for global terrestrial R_s is approximately 15.60%, with high values primarily observed in India, southern Africa, and high-latitude regions. The average coefficient of variation for global terrestrial R_h is approximately 16.10%, slightly higher than that of R_s, and high values are mainly concentrated in high-latitude regions near 60°N.

Fig. 1: Spatial patterns and coefficients of variation for R_s and R_h.

a Spatial pattern of R_s; (b) Spatial pattern of R_h; (c) Coefficient of variation for R_s; (d) Coefficient of variation for R_h. Insets show histograms of frequency distributions.

The global terrestrial R_s and R_h are showing an overall upward trend (Fig. 2). The regions with the fastest annual growth of global terrestrial R_s are India and Europe, especially India, where the overall annual growth of R_s is above 3.99 g C m^-² yr^-², and the trend is statistically significant. The areas where R_s have decreased are scattered, mainly concentrated in northern South America and Central Africa. The annual growth of global terrestrial R_h is slightly lower than that of R_s. The regions with rapid growth are mostly concentrated in Europe and Southeast Asia. Unlike R_s, the changes in R_h in India are not statistically significant, indicating that the changes in Indian R_s may be mainly caused by R_a.

Fig. 2: Trends in R_s and R_h.

(a) Trend of R_s; (b) Significance of the R_s trend; (c) Trend of R_h; (d) Significance of the R_h trend.

To compare the consistency of our global terrestrial R_s and R_h estimates with observed carbon cycle dynamics, we analyzed their relationship to atmospheric CO₂ growth (Fig. 3). Our R_s estimates exhibited a significant positive correlation with annual atmospheric CO₂ increments (r = 0.57, p = 6.68E-5), slightly surpassing R_h correlations (r = 0.56, p = 7.58E-5), confirming the reliability of RS and Rh interannual variability as supported by observational CO₂ growth data. The results revealed annual growth of 0.93 for R_s, 0.65 for R_h, and a notably lower growth of 0.28 for R_a. Structural equation modeling further demonstrated that R_h amplifies atmospheric CO₂ variability by enhancing R_s, with both R_s and fossil fuel emissions significantly effect CO₂ changes, collectively explaining 41% of its variability (Figure. S5). Variance decomposition analysis attributed 18.10% of atmospheric CO₂ variability to R_s and 22.90% to fossil fuel emissions. While the direct contribution of soil respiration processes to CO₂ growth remains secondary to fossil fuel emissions, the substantial magnitude of global soil carbon stocks underscores their non-negligible influence on atmospheric CO₂ under warming conditions. These findings highlight the critical effect between soil-derived carbon fluxes and anthropogenic emissions in shaping atmospheric CO₂ trends.

Fig. 3: Relationship between R_s, R_h, R_a, and atmospheric CO₂ increment.

a Interannual variation of R_s and atmospheric CO₂ increment; (b) Scatter plot with fitted line of R_s and atmospheric CO₂ increment; (c) Interannual variation of R_h and atmospheric CO₂ increment; (d) Scatter plot with fitted line of R_h and atmospheric CO₂ increment. e Interannual variation of R_a and atmospheric CO₂ increment; (f) Scatter plot with fitted line of R_a and atmospheric CO₂ increment. Anomaly calculations were performed on R_s, R_h, and atmospheric CO₂ increment to eliminate differences in magnitude.

The spatiotemporal dynamic of global terrestrial R_h/R_s

Based on the spatiotemporal dynamics of global terrestrial R_s and R_h, we also focused on assessing the spatiotemporal dynamics of the proportion of heterotrophic respiration, denoted as R_h/R_s (Fig. 4). The global terrestrial average of R_h/R_s is approximately 61.30 ± 0.54%, exhibiting a spatial distribution pattern opposite to that of R_s and R_h, with an increase from the equator to the poles. Lower values of R_h/R_s are observed in regions with good vegetation growth, such as the Amazon, central Africa, and South and Southeast Asia, while R_h/R_s values reach up to 70% in central Asia, the western United States, and southern South America. The variation in global terrestrial R_h/R_s is more complex, with higher coefficients of variation in Southeast Asia and India and lower coefficients of variation in high-latitude regions of the Northern Hemisphere. The average coefficient of variation for global terrestrial R_h/R_s is approximately 9.12%. From 1980 to 2022, there was a slight upward trend in global terrestrial R_h/R_s, with a mean increase of 0.13%/10 yr. Notably, R_h/R_s in India exhibited a significant downward trend, while significant upward trends were observed in the Black Sea, northern Caspian Sea, and western United States. We also explored the contributions of R_h and R_a to the changes in R_h/R_s using a contribution decomposition method based on partial derivatives. The results showed that the growth of R_h/R_s was mainly driven by the growth of R_h (with contributions of 61.20% and 38.80% for R_h and R_a, respectively). Although the relationship between R_a and R_h/R_s is negative, R_a has shown a certain growth trend in the past 43 years, which offsets the relationship between R_h and R_h/R_s. However, the growth rate of R_h is higher than R_a (0.65 and 0.28 respectively, with R_h’s growth rate being 2.3 times that of Ra), ultimately leading to an overall upward trend in R_h/R_s over the past 43 years.

Fig. 4: Spatial pattern, coefficient of variation, and trend of R_h/R_s values.

a Spatial pattern of R_h/R_s values; (b) Coefficient of variation of R_h/R_s values; (c) The spatial variation trend of R_h/R_s; (d) The interannual variation of R_h/R_s; (e) The contribution ratio of R_h and R_a to R_h/R_s changes. The small graph in C represents the significance of spatial changes. The contribution ratio in e is estimated by contribution decomposition based on partial derivatives.

We also compared the R_h/R_s values and their trends across five ecosystem types (Fig. 5). In natural ecosystems, R_h/R_s values in different ecosystems were quantified as 59.57% (forest), 61.47% (shrubland), and 61.38% (grassland). The R_h/R_s value was relatively higher in agricultural ecosystems, at 61.76%. In other ecosystem types with sparse vegetation, the R_h/R_s value was significantly higher than in ecosystems dominated by vegetation, reaching 68.81%. Regarding the changes in R_h/R_s values, all ecosystem types except agricultural ecosystems showed an increasing trend. Particularly in forest and shrubland ecosystems, the increase in R_h/R_s values was significantly higher than in other ecosystem types. Conversely, in agricultural ecosystems, R_h/R_s values exhibited a significant decreasing trend (−0.44%/10 yr), with the magnitude of decrease averaging more than twice the magnitude of increase observed in other ecosystems.

Fig. 5: R_h/R_s values and their trends across different ecosystem types.

a Comparison of R_h/R_s value differences across various ecosystem types; (b) Comparison of R_h/R_s trend differences across various ecosystem types. Error bars represent 5.16 times the standard error (±5.16SE). The different letters indicate significant differences among the different ecosystem type (P < 0.05). Given the substantial sample size, conventional confidence multipliers were referenced: 1.96 × SE for 95% confidence level and 2.58 × SE for 99% confidence level. To achieve ultra-conservative statistical assurance, a 5.16 × SE threshold was implemented, corresponding to a 99.9% confidence level.

The impact of global terrestrial soil moisture and temperature on R_h/R_s

Based on the assessment of the spatiotemporal dynamics of global terrestrial R_h/R_s values, we further evaluated how soil temperature and moisture influenced changes in R_h/R_s over the period from 1980 to 2022 (Fig. 6). The results indicate that the increase in soil temperature over the 43-year period had a promoting effect on R_h/R_s, with a global terrestrial average increase of 0.50% in R_h/R_s for every 1 °C increase in soil temperature. This poses a serious threat to the carbon stability of soil carbon pools in the context of global warming. Conversely, an increase in soil moisture over the 43-year period was conducive to a decrease in R_h/R_s, with a global terrestrial average decrease of 0.38% in R_h/R_s for every 1% increase in soil moisture.

Fig. 6: Annual impacts of ST (soil temperature) and SM (soil moisture) on R_h/R_s.

a The effect of soil temperature on R_h/R_s; b The effect of soil moisture on R_h/R_s. Each point represents the global average of s soil temperature, soil moisture and R_h/R_s for the same year, spanning a total of 43 years, with N = 43.

In addition to assessing the interannual impact of soil moisture and temperature on global terrestrial R_h/R_s values, we also explored the spatial response of R_h/R_s values to soil moisture and temperature through partial correlation analysis (Fig. 7). Spatially, after excluding the influence of soil moisture, global terrestrial R_h/R_s values were generally positively correlated with soil temperature, particularly near 45°N, where an increase in soil temperature significantly promoted R_h/R_s values. Unlike soil temperature, soil moisture was generally negatively correlated with R_h/R_s values, especially in India and the region between 30°N and 60°N, where an increase in soil moisture significantly reduced R_h/R_s values. The areas where soil moisture promotes root respiration and soil temperature promotes heterotrophic respiration to dominate R_h/R_s changes account for 47.96% and 52.04%, respectively (Figure. S6). Overall, both from an interannual and spatial perspective, the impacts of soil temperature and moisture on global terrestrial R_h/R_s values are opposite. An increase in soil temperature may promote the R_h/R_s, thereby exerting a positive feedback effect on climate change, while an increase in soil moisture may reduce the R_h/R_s and promote R_a and associated productivity, potentially exerting a negative feedback on climate change.

Fig. 7: Spatial partial correlation of R_h/R_s with soil temperature and soil moisture.

a Spatial partial correlation between R_h/R_s and soil temperature; (b) Pixel statistics of partial correlation between R_h/R_s and soil temperature along latitude; (c) Spatial partial correlation between R_h/R_s and soil moisture; (d) Pixel statistics of partial correlation between R_h/R_s and soil moisture along latitude. The study initially computed partial correlation coefficients (PCC) among soil temperature, soil moisture, and heterotrophic respiration ratio on a pixel-by-pixel basis across the 1980–2022 period. Subsequently, partial correlation coefficients between soil temperature/soil moisture and heterotrophic respiration ratio were systematically calculated using the partial correlation method (Eq. 5), maintaining consistent spatial resolution throughout the analysis.

We also compared the differential impacts of soil moisture and temperature conditions on R_h/R_s values across five ecosystem types (Fig. 8 and Figure. S7). Based on partial correlation analysis, after excluding the influence of soil moisture, soil temperature had varying degrees of promoting effects on R_h/R_s values across all ecosystem types. The promoting effect of soil temperature on R_h/R_s values was significantly higher in shrubland ecosystems than in other ecosystems, followed by other types, grasslands, forests, and agricultural lands in descending order. Conversely, after excluding the influence of soil temperature by partial correlation analysis, soil moisture had varying degrees of inhibitory effects on R_h/R_s values across all ecosystems. The inhibitory effect of soil moisture on R_h/R_s values was significantly higher in agricultural ecosystems than in other ecosystems, followed by grasslands, shrublands, other types, and forests in descending order. In summary, R_h/R_s values in global forests and shrubland ecosystems are primarily influenced by soil temperature, while R_h/R_s values in global grasslands and agricultural ecosystems are primarily influenced by soil moisture.

Fig. 8: Partial correlation statistics between R_h/R_s values and soil temperature/soil moisture across different ecosystem types.

a Partial correlation differences between R_h/R_s values and soil temperature across different ecosystem types; b Partial correlation differences between R_h/R_s values and soil moisture across different ecosystem types. PCC: partial correlation coefficient. Error bars represent 5.16 times the standard error (±5.16SE). The different letters indicate significant differences among the different ecosystem type (P < 0.05). This study initially generated spatial pattern of partial correlations between soil temperature and R_h/R_s throug_h partial correlation analysis. Subsequently, zonal statistics were performed by integrating MODIS land classification data, treating each grid cell as an independent sample. This methodology enabled the calculation of mean partial correlation coefficients, standard deviations, sample sizes, and standard errors across distinct ecosystem types. Given the substantial sample size, conventional confidence multipliers were referenced: 1.96×SE for 95% confidence level and 2.58×SE for 99% confidence level. To achieve ultra-conservative statistical assurance, a 5.16×SE threshold was implemented, corresponding to a 99.9% confidence level.

Discussion

The dynamic characteristics of global R_h and its proportion

In the study of global carbon cycling, R_h, as a key flux connecting carbon exchange between soil and the atmosphere, reveals profound significance for understanding global climate change by uncovering its spatiotemporal dynamic characteristics^2,47. Substantial uncertainties persist in current global estimates of terrestrial R_h, with reported annual totals spanning 37.80–72.30 Pg C yr⁻¹, discrepancies approaching twofold (Figure. S8). Our study estimated the annual terrestrial R_h as 60.10–61.70 Pg C yr⁻¹, clustered within the upper range of existing estimates and demonstrating comparability with previous R_h studies. In addition to comparing in terms of quantity, we also compared the estimated Rh with ecosystem respiration observation data from flux towers and machine learning upscaled Rh products over time (Figure. S9 and Figure. S10). The results showed a high consistency between our estimated Rh and observation data, and a certain comparability with spatial product trends.

Based on extensive and widely distributed ground observation data and machine learning models, we estimated the spatiotemporal dynamic characteristics of global terrestrial R_s and R_h, with our results further supported by observed annual increments of atmospheric CO₂ data⁴⁸. Changes in atmospheric CO₂ are not only influenced by fossil fuel emissions but are also closely related to R_h in terrestrial ecosystems⁶. This study demonstrated that global terrestrial R_s and R_h have both shown sustained increases over time, with these trends exhibiting a significant positive correlation (Fig. 5) to the annual growth of atmospheric CO₂, a concerning relationship that may intensify global warming through reinforcing feedback mechanisms⁴⁹. Numerous scholars have conducted various types of controlled experiments at the plot scale to analyze R_h and changes in their proportion of R_s; however, results from different studies show considerable variation, particularly with a lack of quantitative research on the R_h/R_s at the global scale^19,22. Our estimated R_h/R_s was approximately 61.30 ± 0.54% during 1980-2022 (Fig. 6), a result that is currently supported by plot-scale studies in the literature^{2,19,34,43,50,51,52,53}.

Besides quantifying the magnitude and spatial pattern of the global R_h/R_s, assessing its long-term changes is crucial for understanding the stability of global soil carbon pools and the feedback mechanisms of terrestrial and atmospheric carbon cycling⁴⁷. This study also analyzed the trend of the global terrestrial R_h/R_s from 1980 to 2022, showing a significant increasing trend in the R_h/R_s value over the years (0.13%/10 yr), and spatially, the global terrestrial R_h/R_s value generally exhibited an increasing trend with considerable spatial heterogeneity (Fig. 6). Exploring the changes in R_h/R_s can provide us with more understanding of the stability of soil carbon pools⁵⁴. Although the changes in this ratio alone cannot directly reflect the changes and relationships between R_h and R_a, combining the trend and magnitude of R_h changes can provide us with more useful references⁵⁵.

Although many models divide ecosystem respiration into R_h and total autotrophic respiration (i.e., the sum of aboveground and belowground autotrophic respiration)^40,56, understanding of their proportional changes in spatial and temporal variations and their response to environmental changes remains limited. It should be noted that the environmental variables used in estimating R_s and R_h are strictly corresponding to the sampling years of the observed data, so our model can capture the temporal trends of R_s and R_h to a certain extent, although the spatial pattern differences between the two may have an impact on this process. The assessment of dynamic characteristics of the global R_h/R_s in this study not only aids in understanding the feedback between R_s and terrestrial-atmospheric carbon dynamics but also provides valuable insights for optimizing future Earth system models⁵⁷.

The influence of soil moisture and temperature on the R_h/R_s

Hydrothermal conditions, as one of the most important variables regulating R_s, may have different impacts on R_a and R_h^49,58. We found that overall, the spatial pattern of R_h/R_s may be mainly influenced by temperature and precipitation patterns. In areas with better hydrothermal conditions and higher R_s, R_h/R_s are usually lower, which is supported by previous research^54,59. We also found that R_s/R_h showed an increasing trend over time, which may be related to global warming, consistent with the conclusions of reviews or meta-analyses^29,55,60. Increased soil temperature may boost microbial activity in the soil, thereby accelerating their metabolic activities and the R_h^51,61. Results from multiple warming-controlled experiments have shown that increased microbial activity in warmed soils promotes the decomposition of organic matter, leading to approximately 30% more CO₂ emissions compared to control groups^49,62. Furthermore, studies have found that R_h has a higher temperature sensitivity than R_a^8,63, as R_a is typically more affected by GPP and total underground carbon allocation⁵⁷. Therefore, soil temperature increases due to climate warming may further increase the R_h/R_s in the soil, thereby weakening the stability of soil carbon pools and creating a positive feedback loop with rising atmospheric CO₂ levels and global warming.

Unlike soil warming, our study demonstrates that an increase in soil moisture will reduce the R_h/R_s on both interannual and global scales (Figs. 8 and 9). The mechanisms behind this phenomenon are complex and manifest at multiple levels¹⁴. Firstly, changes in soil moisture primarily affect R_a, as the sensitivity of soil roots to moisture is higher than that of microorganisms, as confirmed by multiple controlled experimental studies^50,52,64. Secondly, soil moisture also influences the utilization of nutrients such as nitrogen and phosphorus by R_h and R_a, and there may be interactive effects between soil moisture and nutrient factors like nitrogen or phosphorus that regulate these respiration processes⁴². Our study further reveals that the R_h/R_s in more regions globally is more influenced by soil moisture than by soil temperature (Figure. S6). This may be attributed to the fact that soil moisture not only directly affects R_s and its components but also influences the temperature sensitivity of R_h^10,52.

Fig. 9: Quantification of the effects of soil temperature and moisture on heterotrophic respiration.

a Structural equation modeling evaluates the relationship between atmospheric CO₂, ST, SM, and Rh, Chisq/df=0.433; CFI = 0.993; RMSEA = 0.006; SRMR = 0.015; *: p < 0.05; b Variance decomposition quantifies the contributions of ST, SM, and other factors to Rh variation.

Our research results indicate that global soil temperature increases will promote the R_h/R_s, while soil moisture may have an inhibitory effect on this proportion (Fig. 8). Additionally, soil temperature increases may also enhance evapotranspiration, further reducing soil moisture^65,66, which could lead to an overall increasing trend in the R_h/R_s, as observed in our study. This process will further weaken the stability of soil carbon pools in the future and amplify carbon emissions from soil carbon pools to the atmosphere¹⁹. Soil temperature, as an important factor affecting R_h and also influenced by greenhouse effect, may form a positive feedback loop with global warming, exacerbate future global changes, and pose new challenges to carbon neutrality goals (Fig. 9 and Figure. S5).

Comparison of differences in the R_h/R_s across various ecosystem types

When exploring the complexity of the global carbon cycle, different ecosystem types exhibit great variations in R_h and its proportion due to their unique vegetation composition, soil characteristics, and climatic conditions⁶⁷, which has become key to understanding ecosystem functions and services. This study systematically compared and analyzed the dynamic characteristics of the R_h/R_s in forests, shrubs, grasslands, agricultural and other types of ecosystems, aiming to reveal the similarities and differences in carbon cycling mechanisms among different ecosystems.

Our results indicate that forest ecosystems exhibit the lowest R_h/R_s compared to other types (Fig. 7). This finding may be attributed to the high biodiversity and well-developed underground root networks in forest ecosystems⁶⁸, which promote the relative enhancement of R_a (primarily generated by plant root activities) and, to some extent, reduce the R_h/R_s (primarily generated by the decomposition of organic matter by soil microorganisms)⁶⁹. Furthermore, forest ecosystems typically possess stronger drought resistance and water retention capabilities, maintaining higher soil moisture, which not only favors the continuous activity of underground vegetation but also further promotes the R_a and its contribution to R_s⁷⁰.

In contrast, agricultural ecosystems, due to their frequent and intense anthropogenic management activities such as harvest, irrigation, and fertilization⁷¹, exhibit a unique trend in the R_h/R_s, namely a significant decreasing trend (Fig. 6). This study also found that the R_h/R_s in agricultural ecosystems has a lower sensitivity to soil temperature but is significantly more inhibited by soil moisture than in other ecosystems (Fig. 8). This phenomenon can be explained by the effective alleviation of water stress in agricultural fields through management practices such as irrigation, ensuring sufficient soil moisture⁷², which in turn maintains high vitality of underground vegetation, promotes the R_a, and relatively reduces the R_h/R_s⁷³.

These findings not only deepen our understanding of carbon cycling mechanisms in different ecosystems and reveal the complex relationship between ecosystem types and the R_h/R_s but also provide a solid scientific basis for formulating precise and effective carbon management strategies. Future research should further explore the dynamics of carbon cycling in different ecosystem types under the dual influences of climate change and human activities⁵⁵, as well as how these changes affect global carbon balance and ecosystem service functions, in order to provide more comprehensive strategic support for addressing climate change challenges.

Conclusion

This study elucidates the key dynamics governing the R_h/R_s and its implications under global change. Leveraging global observational data (1980–2022), we reveal a significant multi-decadal increase in R_h/R_s, rising at 0.13% per decade, with contemporary R_h contributing 61.30 ± 0.54% of global R_s and exhibiting pronounced spatial heterogeneity—manifesting as lower ratios in tropical forests and higher values in mid-to-high latitudes. Critically, ecosystem-specific drivers emerged: soil temperature dominantly regulates R_h/R_s in forests and shrublands, while soil moisture is the primary controller in grasslands and croplands. Globally, soil moisture exerted influence over a larger spatial extent than temperature, with a 1% increase in moisture decreasing R_h/R_s by 0.38%, compared to a 0.50% rise in R_h/R_s per 1 °C warming. These findings challenge the assumption that temperature-driven feedbacks universally dominate soil carbon-climate interactions, underscoring the underappreciated role of moisture, particularly in non-forested ecosystems. The accelerating R_h/R_s trend signals growing vulnerability of soil carbon stocks, potentially elevating atmospheric CO₂ beyond current projections. Future efforts must prioritize long-term monitoring of moisture-sensitive ecosystems and integrate biome-specific drivers of R_h/R_s into Earth system models to refine carbon-climate feedback predictions. Our results highlight the urgency of incorporating the complex interplay of soil temperature and moisture on respiration components into climate mitigation strategies, as stabilizing soil carbon demands dual adaptation to warming and shifting hydrological regimes. This mechanistic understanding is critical for addressing climate-carbon feedback uncertainties and advancing sustainable ecosystem management.

Methods

Data source

The study area encompasses global land between 55°S and 75°N latitude, spanning all longitudes from 180°E to 180°W. Owing to the lack of vegetation data used to construct the model in areas with extremely low vegetation coverage, such as polar areas and deserts, and their low contribution to global carbon cycle, these areas were excluded to ensure the integrity of the study.

The data we used are mainly divided into two categories. One is the input data used for model construction, which are used to estimate R_s and R_h. The other is the R_s and R_h observed data (Figure. S1), Annual growth data of atmospheric CO₂ observed and flux tower observation data, which are used to compare and evaluate the reliability and consistency of the estimation results of this study. The data used to build the model include observational data such as R_s (9874 ground observations) and R_h (1406 ground observations)‌, as well as environmental variables such as ‌annual mean temperature and annual precipitation‌. This study collected global R_s and R_h observation data based on peer-reviewed papers, inventory data released by national forestry departments, and publicly available databases of institutions. In addition, we have established 12 data collection and cleaning standards and quality control conditions to ensure the reliability of the collected databases and observation results (see supplementary materials). This study has expanded and improved the existing soil respiration database (mainly the fifth edition of the Global Soil Respiration Database)⁷⁴, and shared this data with the academic community. In order to evaluate the spatial representativeness of the constructed observation database, this study combined MODIS land cover type products to compare the area ratios of different land cover types globally with the ratios of our constructed observation data in different land cover types. The two have high consistency, indicating that the observation data used has high spatial representativeness (Figure. S2). Detailed information and sources of these data can be found in Table 1.

Table 1 Constructing and validating data for R_s and R_h models

This study employed MODIS land classification product (MCD12C1) with a 0.05-degree spatial resolution and annual temporal composites to categorize ecosystems into forest, grassland, shrubland, cropland, and other types based on the International Geosphere-Biosphere Programme classification scheme. To ensure temporal consistency and cross-year comparability, invariant pixel selection was rigorously applied to identify pixels retaining unchanged ecosystem classifications throughout the 2001–2022 period, which constituted the fundamental spatial units for subsequent ecosystem delineation.

Estimation of R_s, R_h and R_h/R_s

We adopt the approach of first estimating R_s and then deriving R_h from it to obtain more reliable results than directly estimating R_h (Figure. S3). The advantage of this method is that: (1) It considers the high correlation between R_s and R_h, ensuring high accuracy and reliability of R_h estimation results. (2) Relieve the problem of global R_h observation data being far lower than R_s, and enable the use of as much observation data as possible to provide more information for the model. (3) Fully considering the spatial heterogeneity of the relationship between R_s and R_h, as well as the regulatory effects of other environmental factors on the relationship between the two, to avoid the shortcomings of a simple linear regression model that directly simplifies the fixed relationship between R_s and R_h.

The input variables for estimating the R_s and R_h in random forest can be seen in Fig. 1 and Table S1. In this study, the ntree parameter of both models was set to 1500 to ensure a sufficient number of trees to stabilize the prediction results. Meanwhile, with 7 input variables and 6 mtry parameters, 6 variables are randomly selected during node splitting in each tree, further increasing the diversity and robustness of the model. It should be noted that the time of the environmental variables used in the model construction process strictly corresponds to the sampling year of the observation data, rather than using long-term mean data to drive the model. Therefore, the constructed random forest model can capture the temporal trends of R_s and R_h to a certain extent.

Observation-based evaluation and uncertainty assessment

In this study, coefficient of determination (R²), root mean square error (RMSE), and percent bias (PBIAS) were selected as indicators for constructing the models and verifying the model accuracy, as they have been widely used to evaluate model applicability and robustness⁷⁵.

$${{\rm{R}}}^{2}=\frac{{[{\sum }_{{\rm{i}}=1}^{{\rm{n}}}\left({\rm{Oi}}-{\rm{O}}\right)\left({\rm{Pi}}-{\rm{P}}\right)]}^{2}}{{{\sum }_{{\rm{i}}=1}^{{\rm{n}}}\left({\rm{Oi}}-{\rm{O}}\right)}^{2}{{\sum }_{{\rm{i}}=1}^{{\rm{n}}}\left({\rm{Pi}}-{\rm{P}}\right)}^{2}}$$

(1)

$${\rm{RMSE}}=\sqrt{\frac{1}{n}{\sum }_{i=1}^{n}{\left({\rm{Oi}}-{\rm{Pi}}\right)}^{2}}$$

(2)

$${\rm{PBIAS}}=\frac{{\sum }_{i=1}^{n}({\rm{Pi}}-{\rm{Oi}})}{{\sum }_{i=1}^{n}{\rm{Oi}}}\times 100 \% $$

(3)

where Oi is the i-th observation value, O is the average of the observation values, Pi is the i-th estimated value, P is the average of the estimated values, and n is the number of observation data. The R² values of the two random forest models at the station observation scale are above 0.7, as shown in Figure. S4. After estimating R_s and R_h, we calculated the heterotrophic respiration ratio R_h/R_s using the following formula:

$${R}_{h}/{R}_{s}=\frac{{R}_{h}}{{R}_{s}}\times 100 \% $$

(4)

Statistical analysis

The Theil–Sen trend and Mann–Kendell test are widely employed in meteorology, ecology, and environmental research. They are nonparametric test methods. We used the Theil–Sen trend analysis and the Mann–Kendell trend detection methods to study the temporal and spatial trends of global R_s and R_h. The specific calculation process is presented in the Supplemental Material.

We used linear models in the R, lme4 package, to test the effects of soil temperature and soil moisture on R_h/R_s at the interannual variation level. We performed partial correlation analysis to investigate the relationship between soil temperature and soil moisture and R_h/R_s at the spatial grid level. When discussing the impact of soil temperature on R_h/R_s, soil moisture will be excluded as a control variable. The specific formula for calculating partial correlation coefficient is as follows:

$${r}_{12,3}=\frac{{R}_{12}-{R}_{13}\times {R}_{23}}{\sqrt{(1-{R}_{13}^{2})(1-{R}_{23}^{2})}}$$

(5)

where ${r}_{12,3}$ represents the partial correlation coefficient between variable 1 and variable 2, adjusted for the influence of variable 3. The absolute value of ${r}_{12,3}$ ranges from 0 to 1. ${R}_{12}$ denotes the Pearson correlation coefficient between variable 1 and variable 2. The formula for calculating the Pearson correlation coefficient is as follows:

$$R=\frac{{\sum }_{i=1}^{n}({x}_{i}-\bar{x})({y}_{i}-\bar{y})}{\sqrt{{\sum }_{i=1}^{n}{({x}_{i}-\bar{x})}^{2}}\sqrt{{\sum }_{i=1}^{n}{({y}_{i}-\bar{y})}^{2}}}$$

(6)

where $R$ represents the correlation coefficient, with an absolute value between 0 and 1. ${x}_{i}$ and ${y}_{i}$ are the values of variables x and y for the i-th observation, respectively, and $\bar{x}$ and $\bar{y}$ are the average values of variables x and y, respectively.

One-way ANOVA was used to analyze the differences of R_h/R_s in different ecosystem types. Significant differences (p < 0.05) between R_h/R_s in different ecosystem types were detected using the least significant difference (LSD) test. This study also evaluated the degree of fluctuation of estimation results over time using the coefficient of variation:

$${\rm{CV}}=\frac{{STD}}{{MEAN}}\times 100 \% $$

(7)

where ${\rm{CV}}$ represents the coefficient of variation, ${STD}$ and ${MEAN}$ are the standard deviation and mean values of variables between 1980 and 2022, respectively.

We also used a contribution decomposition method based on partial derivatives to quantify the contribution of R_h and R_a to the interannual variation of R_h/R_s, which is suitable for long-term time series data and particularly for analyzing the contribution quantification of interannual variation scales⁷⁶.

This study also employed structural equation modeling (SEM) to investigate the effects of global atmospheric CO₂ concentration changes, soil temperature, and soil moisture on global terrestrial heterotrophic respiration, as well as the influences of global terrestrial soil respiration and heterotrophic respiration on atmospheric CO₂ concentration dynamics. The SEM framework was constructed and performed using the lavaan package in R. The reliability of the SEM was evaluated through the following indices: Chisq/df, Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR).

Reporting summary

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

Conclusion

This study elucidates the key dynamics governing the R_h/R_s and its implications under global change. Leveraging global observational data (1980–2022), we reveal a significant multi-decadal increase in R_h/R_s, rising at 0.13% per decade, with contemporary R_h contributing 61.30 ± 0.54% of global R_s and exhibiting pronounced spatial heterogeneity—manifesting as lower ratios in tropical forests and higher values in mid-to-high latitudes. Critically, ecosystem-specific drivers emerged: soil temperature dominantly regulates R_h/R_s in forests and shrublands, while soil moisture is the primary controller in grasslands and croplands. Globally, soil moisture exerted influence over a larger spatial extent than temperature, with a 1% increase in moisture decreasing R_h/R_s by 0.38%, compared to a 0.50% rise in R_h/R_s per 1 °C warming. These findings challenge the assumption that temperature-driven feedbacks universally dominate soil carbon-climate interactions, underscoring the underappreciated role of moisture, particularly in non-forested ecosystems. The accelerating R_h/R_s trend signals growing vulnerability of soil carbon stocks, potentially elevating atmospheric CO₂ beyond current projections. Future efforts must prioritize long-term monitoring of moisture-sensitive ecosystems and integrate biome-specific drivers of R_h/R_s into Earth system models to refine carbon-climate feedback predictions. Our results highlight the urgency of incorporating the complex interplay of soil temperature and moisture on respiration components into climate mitigation strategies, as stabilizing soil carbon demands dual adaptation to warming and shifting hydrological regimes. This mechanistic understanding is critical for addressing climate-carbon feedback uncertainties and advancing sustainable ecosystem management.