Topography Dependence of Terrestrial Litter‐Derived Soil Respiration

ABSTRACT

Carbon (C) released from decomposing plant litter constitutes a major component of soil CO₂ efflux at the land surface, yet its contribution is rarely constrained separately from heterotrophic respiration of stable soil organic matter in global C budgets and Earth system model evaluations. We combined a global dataset with high‐resolution field observations to quantify the contribution of litter‐derived soil respiration (R _s) using litter‐input and litter‐removal experiments, and to project its current distribution and future dynamics with machine‐learning models. On average, litter‐derived R _s accounted for 30.9% of total R _s. Higher contributions were estimated for experiments of shorter duration, highlighting the important role of fast‐cycling C alongside soil organic matter turnover. Litter‐derived R _s varied substantially among ecosystems (grasslands > croplands > forests > wetlands) but did not differ significantly between tropical and temperate climates. Land surface slope exerted a stronger control than climatic or edaphic factors across both mountain and non‐mountain regions, suggesting a pronounced topographic regulation. High‐frequency field measurements further confirmed this pattern, with litter‐derived R _s at mountain ridges being 1.5 times that in valleys. Global projections indicated a higher litter‐derived R _s at low latitudes and greater vulnerability in cold climates under SSP 1–2.6 and SSP 5–8.5 scenarios. These findings demonstrate that litter decomposition is a substantial source of soil CO₂ flux that is strongly controlled by terrain. Accounting for this CO₂ pathway improves our understanding of how landscape heterogeneity influences terrestrial C cycle and enhances future predictions of ecosystem responses to climate change.

Using a global dataset and high‐resolution field observations, we show that litter‐derived soil respiration (R _s) accounts for about 30% of total R _s. This fast‐cycling flux is strongly regulated by topography, with slope exerting a greater influence than climate or soil properties, and future projections indicating increasing vulnerability under climate change, especially in cold climates. Field observations further revealed that litter‐derived soil R _s were 1.5 times higher on ridges than in valleys in mountain forests.

1. Introduction

Soil respiration (R _s) accounts for 60%–90% of terrestrial ecosystem respiration and is a fundamental component in the global carbon (C) budget and a key process represented in Earth system models (ESMs) (Jian et al. 2021; Friedlingstein et al. 2025). R _s is commonly partitioned into autotrophic respiration (R _a) and heterotrophic respiration (R _h) (Wei et al. 2022; Bond‐Lamberty et al. 2024), with R _h arising primarily from the microbial decomposition of plant litter and soil organic matter (SOM) throughout the soil profile. Litter‐derived R _s refers to the component of R _h attributable to the microbial decomposition of plant litter, whereas the remaining component is mainly associated with SOM decomposition. Because litter and SOM differ markedly in their turnover rates, substrate quality, and sensitivity nthe fast‐cycling litter‐derived component. Separately constraining litter‐derived R _s is therefore important for evaluating how models represent rapid C turnover and short‐term ecosystem responses to climate change.

Contemporary land surface models, including the Community Land Model (CLM), explicitly represent litter and SOM as multiple C pools with distinct turnover rates, but model–data comparisons are typically based on aggregated R _h, and independent empirical constraints on litter‐derived fluxes remain limited (Tang et al. 2026). Global C input from litterfall is estimated at approximately 40–80 Pg C yr.⁻¹, about 80% of which is released as CO₂ during progressive decomposition (Lal 2008; Cotrufo et al. 2015). This large but rapidly cycling C flux turns over more rapidly and is more sensitive to temperature than older SOM, but it is rarely constrained independently using field observations for model benchmarking, resulting in substantial uncertainty in estimates of terrestrial C–climate feedbacks (Throop et al. 2025; Zhao et al. 2025). For instance, current TRENDY ensemble simulations are generally evaluated against aggregate R _h, and the lack of field‐based constraints on litter‐derived R _s makes it difficult to diagnose the independent contribution and environmental sensitivity of fast litter‐C turnover (Shao et al. 2013; Tang et al. 2020). Observation‐based quantification of litter‐derived R _s is therefore essential for benchmarking model representations of fast C turnover and improving predictions of terrestrial CO₂ fluxes under climate change.

Climate exerts a strong control on litter‐derived R _s by shaping the biogeographic distribution of plant productivity, litter substrate quality, soil microbial communities and edaphic biogeochemistry conditions. Together, these factors drive variation in litter decomposition across climatic zones and ecosystems. First, temperature and precipitation regulate organic C decomposition by modulating water availability, microbial activity and enzymatic efficiency (Moyano et al. 2012; Hartley et al. 2021). Vegetation productivity regulates litter‐derived R _s by determining the quantity and quality of litter inputs andthereby affecting substrate availability (Bond‐Lamberty et al. 2018). Moreover, empirical evidence shows that litter traits associated with evergreen versus deciduous species and coniferous versus broadleaved species substantially influence the biochemical composition of decomposing litter, including its C/N ratio, lignin content and proportion of readily degradable C (Wu et al. 2025). These litter chemical traits determine substrate accessibility to microbes and decomposition efficiency (Walela et al. 2014), ultimately affecting the contribution of litter‐derived R _s. Meanwhile, soil microbial community is shaped by climate and litter substrate, and differences in microbial functional strategies determine decomposition pathways and efficiency (Glassman et al. 2018). Finally, edaphic variables indirectly regulate litter decomposition by modifying microbial habitats and nutrient supply. Specifically, soil texture and bulk density determine moisture retention and aeration, thereby constraining or facilitating microbial activity in contributing to litter C release (Freeman et al. 2001). Soil nutrient availability further controls microbial C use efficiency across C fractions with distinct molecular structures, ultimately influencing litter‐derived R _s (Yu et al. 2025). In addition, soil pH imposes an important constraint on C release from recalcitrant litter materials by shaping microbial community structure and regulating extracellular enzyme activities responsible for litter breakdown (Huang et al. 2023; Liang et al. 2024).

Long‐term litter manipulation experiments, such as those conducted within the Detritus Input and Removal Treatments (DIRT) network, provide strong evidence for environmental and chemical controls on litter‐derived R _s (Crow et al. 2009; Lajtha et al. 2018; Li et al. 2022). In these experiments, litter is operationally defined as fresh plant residues located on the soil surface or within the organic horizon before being incorporated into mineral soils. Global syntheses show that litter‐derived R _s on the soil surface is more sensitive to environmental fluctuations than respiration derived from older SOM in deeper soil layers (Chen and Chen 2018; Soong et al. 2020; Rocci et al. 2024). Incorporating these insights has improved predictive accuracy across climatic and edaphic gradients. However, previous DIRT experiments and global syntheses have primarily focused on experimental litter manipulation effects or broad climatic and edaphic controls, whereas spatially explicit predictions and fine‐scale terrain‐driven heterogeneity in litter‐derived R _s remain insufficiently resolved. Building on these studies, we integrated global observations, machine learning predictions, and detailed measurements from mountain forests to link global patterns with local evidence of topographic variation in litter‐derived R _s. Nevertheless, most existing models, such as CENTURY, ROTHC, and Biome‐BGC, assume spatial uniformity of soil biogeochemical processes within model grid cells (White et al. 2000; Jenkinson and Coleman 2008; Chakrawal et al. 2020), giving limited attention to how local environmental differences under complex terrain regulate the fractionations of soil CO₂ flux. In site‐level natural ecosystems, spatial heterogeneity occurs under similar climate and vegetation types, suggesting that moisture redistribution, microclimate fluctuation, and substrate differentiation can change litter‐derived R _s on the soil surface at multiple spatial scales (Wang et al. 2017; Kumari et al. 2020; Wei et al. 2022).

Here, we used a global dataset of litter‐derived R _s comprising 2793 in situ observations under natural litter input and litter removal treatments across diverse climatic zones and ecosystems to quantify the proportion of Rs derived from decomposing litter rather than older SOM. By integrating this dataset with high‐resolution climate, topographic, vegetation, and edaphic variables, we trained a machine learning model to evaluate and identify the key factors influencing litter‐derived R _s and to predict changes in its biogeographical distribution under current and future climate scenarios. Moreover, as the topography effect cannot be fully resolved in global datasets, we further established an hourly field monitoring at ridge and valley in a mountain forest to directly measure high‐resolution litter‐derived R _s.

2. Methods

2.1. Global Dataset

We conducted a comprehensive literature search in using Web of Science (http://www.webofknowledge.com), Google Scholar (http://scholar.google.com) and the China National Knowledge Infrastructure (CNKI, http://www.cnki.net) to identify peer‐reviewed studies reporting paired measurement with and without litter input. These measurements were used to calculate litter‐derived R _s. The keywords include “litter OR debris OR residue”, “manipulation OR alteration OR reduction OR removal” and “soil respiration OR soil CO₂ flux OR soil CO₂ efflux/emission/release”. A standardized protocol was applied to avoid publication bias. Only field experiments that measured paired observations of soil CO₂ fluxes under natural and litter removal manipulations were included, and studies involving artificial manipulations such as nitrogen fertilization, warming or burning were excluded. In each study, we extracted the mean soil CO₂ efflux, sample size (n) and standard deviation (SD) or standard error (SE) for both control and litter removal plots. When only SE was reported, standard deviation (SD) was calculated as SD = SE × $\sqrt{n}$. Numerical values presented in figures were digitized using WebPlot Digitizer (version 5.2, https://automeris.io; Free Software Foundation).

2.1.1. Proportion of Litter‐Derived R _s

We used the effect size of litter removal to estimate the apparent litter‐attributable contribution to total R _s based on paired control and litter‐removal treatments, hereafter referred to as litter‐derived R _s. The natural logarithm of the individual effect size (lnRR) was calculated following Hedges et al. (1999), with lnRR and its variance (v) computed using Equations (1) and (2). The weighted mean effect size ($\overline{\ln \mathit{RR}}$) was estimated from individual observations by assigning greater weight to effect sizes with lower variance using the rma.mv function (Viechtbauer and Cheung 2010). Litter‐derived R _s was calculated using Equation (3).

$$\ln RR=\ln \left(\frac{{\mathrm{X}}_{\mathrm{t}}}{{\mathrm{X}}_{\mathrm{C}}}\right)$$ (1)

$$v=\raisebox{1ex}{$S_t^2$}\!\left/ \!\raisebox{-1ex}{$n_t{X}_t^2$}\right.+\raisebox{1ex}{$S_c^2$}\!\left/ \!\raisebox{-1ex}{$n_c{X}_c^2$}\right.$$ (2)

$$\text{Litter}-\text{derived}{R}_s\left(\%\right)=\left[1-\exp \left(\overline{\ln \mathit{RR}}\right)\right]\times 100\%$$ (3)

where $X_t$ and $X_c$ are the mean values, $n_t$ and $n_{\mathrm{c}}$ are the sample sizes, and $S_t$ and $S_c$ are the standard deviations at the treatment (litter removal) and control plots, respectively.

The dataset includes 2793 pairs of observations with and without litter input from 97 studies published before April 2025. The dataset spans terrestrial sites located from 34.07 °S to 55.50 °N and from 122.10 °W to 133.31 °E (Figure 1A), covering forests, grasslands, croplands, and wetlands. We further categorized the dataset by biome (temperate and tropical), forest type (primary, secondary, and plantation forests), leaf‐shedding strategy (deciduous and evergreen), leaf type (coniferous and broadleaved), and experimental duration (< 6 months, 6–12 months, and > 12 months). Litter‐removal magnitude (< 50% and 50%–100%) was examined in an additional subgroup analysis, and the corresponding results are presented in Figure S1 and Dataset S1.

FIGURE 1.

Site distribution and litter‐derived R _s. (A) Global site distribution, with colors indicating the proportion of litter‐derived R _s in total soil CO₂ emissions. Grey shading denotes global mountain regions based on the GMBA inventory (https://www.earthenv.org/mountains). (B) Field high‐resolution monitoring sites. P1–P4 show the ridge sites, and P5–P8 show the valley sites. (C) Variations in litter‐derived R _s across climatic zones, ecosystems, forest types, leaf habits, leaf types, and experimental durations. Shaded areas indicate data density, and boxplots show medians and interquartile ranges. Different letters indicate significant differences (p < 0.05) among subgroups. Sample sizes are shown in parentheses. R _s: Soil respiration.

2.2. Variables Predicting Litter‐Derived R _s

We extracted climate, topography, vegetation, and edaphic variables with spatial resolution of 500 m to 1 km at each sampling site to assess the environmental constraints on litter‐derived R _s (Table S1).

2.2.1. Climate

To minimize the influence of extreme climate events, this study adopted long‐term mean values of climatic variables. To ensure comparability among study sites, we consistently extracted climate variables from global databases rather than using the values reported in individual publications. Mean annual temperature (MAT) and mean annual precipitation (MAP) were obtained from WorldClim (https://worldclim.org/), while the aridity index and annual evapotranspiration were obtained from Zomer et al. (2022).

2.2.2. Topography

We utilized the ETOPO Global Terrain Model (version 2022) provided by the National Oceanic and Atmospheric Administration (NOAA, USA; https://www.ncei.noaa.gov) to extract site topography variables, including aspect, slope, and elevation.

2.2.3. Vegetation

Vegetation data were obtained solely for net primary productivity (NPP) derived from the MODIS dataset (https://developers.google.com/earthengine/datasets/catalog/modis; MOD17A3HGF.061). We extracted annual NPP values during 2012 to 2021 and used their 10‐year average for subsequent analysis.

2.2.4. Edaphic Factors

Most published studies reported only soil pH or texture, which limited the data availability. Therefore, we retrieved edaphic variables from a global soil dataset developed by Shangguan et al. (2014) including clay content, bulk density, volumetric water content, pH, soil organic C, and total nitrogen. Soil temperature at 0–5 cm depth was obtained from the global database provided by Lembrechts et al. (2022).

2.2.5. Model Training and Selection

To reduce multicollinearity among climate, topographic, vegetation, and edaphic predictors, we calculated variance inflation factors (VIFs) by using the “car” package and iteratively removed variables with VIF > 5 until all remaining predictors met the threshold. Ultimately, 12 key predictors were selected as the most important drivers for predicting litter‐derived R _s.

We used the “caret” package for model training and comparison, testing four algorithms: Bagged CART (Breiman 1996), Boosted CART (Friedman 2001), Cubist (Kuhn and Weston 2023), and Random Forest (Grömping 2009). The dataset was split into training (80%) and testing (20%) subsets using the createDataPartition function. Model performance was evaluated through cross‐validation using R ² and rooted mean squared error (RMSE). For Boosted CART, we tested 10–100 trees, 1–7 terminal nodes, shrinkage rates of 0.01–0.1, and a maximum tree depth of 5. The Cubist model was trained using 1–9 rules, two neighbor settings, and committee sizes of 1, 5, 10, 50, 75, and 100. The Random Forest model was limited to 1000 trees. Based on comparative performance, Random Forest was selected for subsequent analyses (R ² = 0.46, RMSE = 14.85, Table S2). Variable responses were visualized using the “forestFloor” package, and model generalization was further assessed using out‐of‐bag R ² (Figure S2).

2.2.6. Litter‐Derived R _s Predictions Under Current and Future Climates

Land use was obtained from the 2015 ESA CCI‐LC dataset and reclassified into four ecosystem types: forests, grasslands, croplands, and wetlands (Table S3). A Random Forest model was trained using global climate, topographic, vegetation, and edaphic variables to predict the spatial distribution of litter‐derived R _s. Future projections for 2070–2100 under SSP 1–2.6 and SSP 5–8.5 were based on CMIP6 climate data from three ESMs: GFDL‐ESM4, IPSL‐CM6A‐LR, and MPI‐ESM1‐2‐HR (O'Neill et al. 2016; Karger et al. 2021). Future projections were conducted under a consistent climate‐scenario framework, with only MAP updated and other predictors held constant. This provides comparable global estimates and a conservative basis for assessing potential changes in litter‐derived R _s under future climate scenarios. Relative change in litter‐derived R _s was calculated as $\left(R{\mathrm{s}}_{\text{future}}–R{\mathrm{s}}_{\text{current}}\right)/R{\mathrm{s}}_{\text{current}}\times 100\%$. Model uncertainty was assessed by running the Random Forest model 500 times and calculating the coefficient of variation (CV) for each 0.5° grid cell. CV values were summarized in 2° latitude bands to show latitudinal variability (Figure S3). Model residuals showed no systematic bias (Figure S4). Global predictions were further classified by Köppen climate zones to compare spatial patterns and future changes in litter‐derived R _s across major climate regions (Beck et al. 2018; Figure S5).

2.2.7. Topography‐Based Variable Importance Assessment

Topography‐based variable importance assessment. We used the GMBA v2.0 mountain boundary dataset (https://www.earthenv.org/mountains) to classify observational sites into mountain and non‐mountain regions. Separate Random Forest models were fitted for each group using the same climate, topographic, vegetation, and edaphic predictors to compare variable importance based on %IncMSE. This analysis was conducted only to evaluate whether environmental controls differed between mountain and non‐mountain regions. Global spatial prediction was performed separately using a unified Random Forest model trained with all observations. The GMBA classification was then applied to the gridded predictions under current climate conditions and the SSP 1–2.6 and SSP 5–8.5 scenarios. This allowed spatial patterns and future changes to be compared between mountain and non‐mountain regions within a consistent prediction framework.

For each category, we extracted the same climate, topography, vegetation, and edaphic variables used in models. Before model construction, multicollinearity was evaluated using VIF calculated with the “car” package. Predictors with VIF values greater than 5 were removed sequentially until all retained variables had VIF ≤ 5. We fitted separate Random Forest models for mountain and non‐mountain regions using the same parameter settings. Variable importance was quantified using %IncMSE to compare the relative influence of climate, topographic, vegetation, and edaphic predictors between the two topographic categories. This subgroup analysis was used only to assess differences in environmental controls and was not used for global spatial prediction.

2.3. High‐Resolution Field Monitoring of Litter‐Derived R _s

This study was conducted at the Shanghang experimental site in Fujian, China (25°01′ N, 116°28′ E; Figure 1B). The site is a low‐hill subtropical monsoon region (~500 m a.s.l.) with a MAT of 20.1°C and MAP of 1660 mm. The vegetation is naturally regenerated evergreen broadleaved forest dominated by Schima superba and Castanopsis fargesii. Four ridge plots (P1–P4) and four valley plots (P5–P8) were established with > 100 m distances between each plot to ensure spatial independence.

Soil CO₂ fluxes were monitored continuously from January to late April 2025. R _s was measured using automated chambers (TH‐Soi1200) connected to a multi‐channel CO₂ flux system (TH‐6101‐8) and a TS910‐A1‐10 data logger. Two automated chambers (one for control and another for litter removal) were installed on PVC collars with a diameter of 21 cm and a depth of 15 cm at each plot. Control plots remained undisturbed and received natural litter inputs ($R_{\text{control}}$), whereas a 1 × 1 m litterfall trap was installed above each collar in the litter‐removal plots ($R_{\text{removal}}$) to exclude external litter inputs (Figure S6). Each measurement lasted 120 s, during which the first 30 s were used for signal stabilization, and CO₂ fluxes were calculated from the subsequent 90 s by linearly fitting CO₂ concentration (ppm) as a function of time. The online monitoring system recorded CO₂ flux data every 5 min and then produced hourly averages (Dataset S2). Gaps in the time series occurred when valid measurements were unavailable due to weather‐related instrument malfunction during rainfall and lightning events. Litter‐derived R _s was calculated as follows:

$$\text{Litter}-\text{derived}{R}_{\mathrm{S}}\left(\%\right)=\frac{R_{\text{control}}-R_{\text{removal}}}{R_{\text{control}}}\times 100\%$$ (4)

Platinum electrodes were installed within the 0–10 cm soil layer at each sampling site, with reference electrodes placed nearby. Soil redox potential (Eh), soil temperature, and volumetric water content were continuously recorded at 5 min intervals from January to April 2025 using an automated monitoring system. The raw data were quality controlled and aggregated to hourly intervals for subsequent analysis. Measured potentials were converted to standardized values relative to the standard hydrogen electrode based on the reference electrode used.

2.3.1. Metagenomic Sequencing

We conducted metagenomic sequencing on 10 soil samples to explore topography‐induced differences in carbohydrate‐active enzymes (CAZy) gene profiles in ridge and valley soils. DNA extraction and metagenomic sequencing were conducted by Guangdong Magigene Biotechnology Co. Ltd. Microbial total DNA was extracted from eight soil samples collected from the 0–10 cm soil layer. After quality assessment, the DNA was fragmented by sonication and used for library construction. Following library quality control, paired‐end sequencing with a read length of 150 bp was performed on an Illumina NovaSeq 6000 platform to generate raw reads, which were quality filtered by using fastp to remove low‐quality sequences and obtain clean reads. Clean reads were assembled de novo by using MEGAHIT, and open reading frames were predicted with Prodigal. A non‐redundant gene catalog was then constructed by using Linclust. Clean reads were mapped to the non‐redundant gene catalog by BWA (version 0.7.17) to estimate gene abundance. The relative abundance of gene i in sample S (${\mathit{RA}}_i$) was calculated as follows (Qin et al. 2012):

$${\mathit{RA}}_i=\frac{\raisebox{1ex}{$r_i$}\!\left/ \!\raisebox{-1ex}{$L_i$}\right.}{{\sum }_j\raisebox{1ex}{$r_i$}\!\left/ \!\raisebox{-1ex}{$L_i$}\right.}$$ (5)

where ${\mathit{RA}}_i$ is the relative abundance of gene $i$ in $r_i$ sample S, $r_i$ is the counts that gene $i$ can be detected in sample S, and L _i is the length of gene $i$.

To visualize functional differentiation in CAZy gene families between ridge and valley soils, we calculated the log₂ fold change (log₂FC) of mean relative abundance for each CAZy family. Metagenomic CAZy profiles indicate microbial carbohydrate‐degradation potential, not direct enzyme activity or decomposition rates. The absolute log₂FC was used as an effect size metric to weight the flow width in the alluvial diagram, linking CAZy families to their primary substrate categories and enrichment direction. Only families with |log₂FC| > 0.5 were retained to emphasize major functional shifts and improve visualization clarity.

2.3.2. Statistical Analysis

Observations were grouped by climate zone, ecosystem type, forest type, leaf‐shedding strategy, leaf type, experimental duration and litter removal magnitude. Depending on data distribution, the differences among subgroups were tested by one‐way ANOVA with Tukey's HSD post hoc test or Kruskal–Wallis test with Dunn's post hoc comparisons. We also compared monthly litter‐derived R _s between ridge and valley plots using independent t‐tests. All statistical analyses were conducted in R software (version 4.4.2; R Core Team 2024), with statistical significance set at p < 0.05.

3. Results

3.1. Litter‐Derived R _s

The average litter‐derived R _s was 30.9% across climate zones and ecosystems, ranging from 0.13% to 98.3%, with similar values observed in tropical and temperate climates (p = 0.06, Figure 1A). However, litter‐derived R _s differed markedly among ecosystems, with highest values observed in grasslands (59.2%, ranging from 0.6% to 86.5%) but lowest in wetlands (23.9%, ranging from 0.28% to 84.5%). Forest structure and function showed complex patterns that primary forests (30.0%) and plantations (29.8%) showed greater average values than secondary forests (26.8%), and evergreen (31.0%) and broadleaved forests (31.0%) had greater proportions of litter‐derived R _s than deciduous (28.1%) and coniferous forests (28.9%, Figure 1B). However, litter‐derived R _s declined with experimental duration during progressive decomposition (Figure 1C).

3.2. Topography Control of Litter‐Derived R _s

The optimized Random Forest model suggested that the combined effects of climate, topography, vegetation, and edaphic variables effectively predicted litter‐derived R _s at global scale (Figure 2A). By integrating 2793 observations from 84 sites, the Random Forest model identified slope, elevation, and aspect as the most important predictors of litter‐derived R _s, with %IncMSE values of 42.4%, 35.6%, and 32.3%, respectively. These results highlight the strong association between topographic variables and spatial variation in litter‐derived R _s. However, partial dependence plots showed that these variables followed nonlinear increasing and decreasing trends, suggesting strong interactions among these environmental factors in explaining litter‐derived R _s.

FIGURE 2.

Random forest model shows performance and variable importance in predicting litter‐derived R _s. (A) Global patterns of predictor importance (%IncMSE) and model performance, including observed versus predicted litter‐derived R _s (R ² = 0.46; RMSE = 14.85) and partial dependence relationships for significant variables. (B, C) Variable importance and partial dependence responses for litter‐derived R _s in (B) mountain and (C) non‐mountain regions. ADI, Aridity index; AET, Annual evapotranspiration; BD, Bulk density; clay, Soil clay content; MAP, Annual mean precipitation; NPP, Net primary productivity; R _s, Soil respiration; SOC, Soil organic carbon; TN, Soil total nitrogen; VMC, Volumetric water content. Full variable descriptions are provided in Table S1.

3.3. Topographic Divergence Between Mountain and Non‐Mountain Regions

Considering the dominant role of topography in driving litter‐derived R _s in the model, we further compared the variations in litter‐derived R _s between mountain and non‐mountain regions. In mountain regions, the variation in litter‐derived R _s was primarily driven by topography, and aspect, elevation, and slope were the strongest predictors followed by soil bulk density and NPP (Figure 2B). However, in non‐mountain areas, NPP was the primary driver, followed by topography such as slope and elevation (Figure 2C). Partial dependence plots showed that although these predictors exhibited strong nonlinear responses in both regions, their effects differed in opposite directions between mountain and non‐mountain areas, suggesting a topographic dependence of litter‐derived R _s.

This topographic dependence identified by the model was further corroborated by high‐resolution field observations (Figure 3A,B). From January to April 2025, litter‐derived R _s averaged 40.9% in ridge plots and 24.7% in valley plots. Monthly values in ridge positions remained consistently higher (44.26%, 33.39%, 37.11%, and 47.71%) than those monitored at valley (25.01%, 31.30%, 27.38%, and 27.32%) during January to April, respectively. CAZy gene profiles differed between ridge and valley soils in complex terrain, with several glycoside hydrolase families showing higher relative abundance in ridge soils than in valley soils (Figure 3C).

FIGURE 3.

High‐resolution litter‐derived R _s monitoring and associated CAZy functional profiles in ridge and valley soils. (A) Hourly litter‐derived R _s from January to April 2025. (B) Monthly litter‐derived R _s. Asterisks denote significant difference between ridge and valley sites (***p < 0.001). (C) Sankey diagram shows substrate‐specific enrichment of CAZy gene families between ridge and valley soils based on log₂ fold changes in abundance. C, cellulose; CE, carbohydrate esterase; CQ, cellulose and chitin; GH, glycoside hydrolase; H, hemicellulose; L, lignin; P, pectin; PSC, pectin, starch, and cellulose; Q, chitin; S, starch; R _s, soil respiration.

3.4. Predicted Litter‐Derived R _s Under Current and Future Climates

Global predictions indicated that litter‐derived R _s was an average of 30.9% at current climate status (Figure 4A). Hotspots occurred in tropical regions, including northeastern Amazonia, central Africa, and Southeast Asia, whereas high latitudes such as Alaska, northern Canada, northern Europe, and Siberia showed consistently low proportions. Overall, litter‐derived R _s declined with latitude, except for a localized high‐value band in middle‐ to high latitudes in the Southern Hemisphere (20–45 °S), including southern Chile, eastern South Africa, and southeastern Australia. Under the SSP 1–2.6 scenario, a small increase in litter‐derived R _s was mainly confined to middle‐ and high latitudes, while low‐latitude tropics showed little change (Figure 4B). Under the SSP 5–8.5 scenario, litter‐derived R _s increased substantially, with a global mean relative increase of 3.87%. Several regions exhibited 10%–30% increases, forming a broad strengthening belt across western North America, Alaska, northern Canada, Scandinavia, and Siberia, with a similar increase in middle latitudes in the Southern Hemisphere. However, declines in litter‐derived R _s also occurred in humid tropical forests (Amazon, Congo, and Southeast Asia) and in some high‐latitude areas (eastern Alaska, northern Canada, and parts of Siberia, Figure 4C). Mountain areas are more sensitive to climate change, with croplands and grasslands exhibiting greater increases but wetlands greater declines (Figure S7).

FIGURE 4.

Global distribution of litter‐derived R _s under current and future climates. (A) Present spatial pattern of litter‐derived R _s and its latitudinal trend. (B, C) Projected relative changes in litter‐derived R _s during 2070–2100 under the (B) SSP 1–2.6 and (C) SSP 5–8.5 scenarios, together with their corresponding latitudinal patterns. Values are gridded at 0.5° resolution. Latitudinal summaries were calculated using 2° latitude bands; solid lines show band means, and shaded areas represent mean ± 2 SE. The gray areas have no available data. R _s, soil respiration.

4. Discussion

Litter‐derived Rs accounted for nearly one‐third of total R _s (30.9%), which is higher than the estimate of 21.8% reported by Zhu et al. (2023). R _h from surface soils (including that derived from both litter and older SOM), has been estimated to contribute approximately 70% of the total soil CO₂ flux. The substantial contribution of decomposing litter at the soil surface suggests that litter decomposition constitutes an important component of heterotrophic CO₂ emissions (Nissan et al. 2023). Interestingly, topography exerted strong control over variation in litter‐derived R _s in both mountain and non‐mountain regions, surpassing the traditionally recognized dominance of climate and highlighting the strong dependence of litter‐derived R _s on topographic conditions. This result is consistent with previous local‐scale studies. For example, Tian et al. (2019) found that R _s increased from valleys to ridges in a subtropical forest, primarily because topography altered soil temperature and moisture, vegetation cover, and soil organic C fractions. Our high‐resolution field monitoring also showed that litter‐derived R _s at ridges was 1.5 times that in valleys within a montane forest, further confirming strong topographic control over litter‐derived R _s. Together with the global analysis, this finding indicates that the topographic effect is not confined to individual montane forests but may extend across broader spatial scales.

Topography can shape litter‐C release during surface decomposition by influencing water redistribution, litter production, and soil C substrate differentiation. Steeper slopes may reduce litter retention through gravity‐driven transport. Better‐drained soils on ridges may also enhance the microbial functional potential to decompose plant structural carbohydrates, such as cellulose and hemicellulose (Shigyo et al. 2022). Several CAZy families encoding glycoside hydrolases were more abundant in ridge soils, suggesting greater microbial potential for carbohydrate degradation (Figure 3C; Zhou et al. 2012). Terrain‐driven hydrothermal variations constitute another key mechanism shaping spatial heterogeneity in litter‐derived R _s. Faster water percolation in coarse‐textured soils characterized by low clay content in steep terrain intensifies wetting–drying fluctuations, generating strong pulses in microbial activity that accelerate litter decomposition (Kateb et al. 2013). Accordingly, ridge soils experience greater redox variability, which is conducive to microbial activity (Figure S8), and steep slopes also exhibit efficient heat exchange at the soil‐atmosphere interface, amplifying temperature and moisture fluctuations at the soil surface. These rapid wetting and drying transitions stimulate microbial activity and enhance microbial access to litter surfaces, thereby increasing C release from decomposing litter (Metz et al. 2023). By contrast, low‐slope areas maintain persistently higher soil moisture levels (Figure S9), which constrain aerobic microbial activity and slow litter decomposition, thereby favoring the persistence of undecomposed litter. Although aspect and elevation explain less variance than slope, aspect affects the surface microclimate through differences in solar exposure and associated photodegradation (Gutiérrez and Vivoni 2013; Austin et al. 2016; Nyman et al. 2018), whereas elevation governs temperature and moisture dynamics, as well as vegetation changes along altitudinal gradients (Contosta et al. 2016; Hua et al. 2024).

However, topographic effects on litter‐derived R _s differ between mountain and non‐mountain regions. In mountain terrain, topographic constraints, such as slope, aspect and elevation jointly regulate gravity‐driven litter redistribution, exposure to surface microclimatic conditions, oxygen diffusion through soil pores and heat exchange efficiency (Yimer et al. 2006; Bennie et al. 2008; Sidari et al. 2008). Because litter is frequently redistributed downslope, decomposition in mountain regions is less constrained by C inputs and more limited by the ability of microbial communities to access and process redistributed litter substrates. This explains why aspect and elevation exert strong controls in mountain ecosystems, whereas the effect of NPP is weaker. By contrast, non‐mountain areas experience less geomorphic disturbance, allowing litter to accumulate and decompose in local surface soils. Higher NPP provides greater C inputs to sustain microbial metabolism and C release from litter (Huang et al. 2021), and slope and elevation explain less R _s variability than NPP. These striking topography contrasts suggest that the environmental controls on litter decomposition depend strongly on the geomorphic context and that the underlying mechanisms are not globally uniform. Plant functional traits also strongly control litter‐derived R _s. Evergreen broadleaved litter show higher contributions than deciduous and coniferous litter because their litter substrate is more decomposable. By contrast, coniferous litter has lower C quality, which reduces microbial C use efficiency and microbial contributions to CO₂ release from decomposing litter (Feng et al. 2024). Mixed litter in primary forests with highly diverse species also show a higher contribution of litter‐derived R _s than that in secondary forests due to more mixed litter materials and more developed decomposer communities (Liu et al. 2017).

Litter‐derived R _s also shows high climate‐ and ecosystem‐dependence. Tropical climates show a higher proportion of litter‐derived R _s (34%, Figure S5), forming pronounced hotspots in the Amazon Basin, Central Africa and Southeast Asia. This spatial heterogeneity is the combined effects of substantial plant C input, rapid litter turnover and high microbial activity driven by warm and humid climates (Hao et al. 2025; Sayer et al. 2024). It is noteworthy that short‐term litter removal experiments included in this study may overestimate litter‐derived R _s because soil microbes preferentially assimilate labile substrates of fresh litter with high C use efficiency and energy source (Cotrufo et al. 2013). However, in long‐term experiments, recalcitrant substrate limitation induces microbial decomposition shifts toward more persistent SOM, reducing the relative contribution of litter‐derived R _s in total soil CO₂ emissions (Sheldrake et al. 2017; Feng et al. 2022). This timing difference supports our previous hypothesis that CO₂ release from decomposing litter is much different with old SOM with lower persistence time and higher temperature sensitivity in contributing to R _h. Increasing experimental evidence has found that undecomposed litter materials on soil surface during early decomposition periods release more labile C, and this is uncoupled with the progressive degradation of structural carbohydrates at later periods (Soong et al. 2015), suggesting that experimental duration could create a bias in estimating litter‐derived R _s. However, it also highlights the appropriacy for partitioning of R _h sourced from litter on soil surface and old SOM across soil profile.

Climate change amplifies litter‐derived R _s, particularly in humid tropical regions. Warming and increased snowmelt in middle‐ and high‐latitudes extend the frost‐free period and improve soil water availability, further accelerating microbial decomposition of plant litter (Lund et al. 2012; Yi et al. 2018). This proportional response from litter decomposition aligns with the well‐known climate sensitivity of R _s. However, the enhanced response in high‐latitudes but constrained response in low‐latitudes suggests a potential poleward shift in the spatial biogeography of soil CO₂ emissions under future climate, with increased risks of transforming the terrestrial C sink into a potential source. However, litter‐derived R _s in response to climate change is ecosystem dependent (Figure S7). Water‐saturated and oxygen‐limited wetlands show a marked reduction in litter‐derived R _s (−5.76% globally and −14.9% in mountains), and increased precipitation and rising groundwater levels may cause excessive water saturation and oxygen limitation in soil pores (Han et al. 2018; Salimi et al. 2021). By contrast, grasslands and croplands show consistent increases, particularly in mountain regions (7%–10%), which have stronger sensitivity because of faster warming and greater variability in wetter conditions (Pepin et al. 2025). These results highlight a differentiated management strategy that wetlands and other high‐risk ecosystems should be prioritized to maintain soil C stability, while vegetation restoration and improved soil management in mountain croplands and grasslands can buffer the sensitivity of soil CO₂ release to climate forcing. Our results also provide parametrized evidence for refining the spatial representation of R _h that is derived from decomposing litter and stable SOM in ESMs and improving fractional predictions of terrestrial soil CO₂ fluxes.

Our results provide important insights into litter‐derived R _s, but several limitations should be addressed in future studies. First, although the global synthesis provides broad insights into litter‐derived R _s, tropical regions remain underrepresented. This sampling imbalance should be considered when interpreting global patterns, and more tropical field measurements are needed to improve future syntheses. Second, litter‐derived R _s in this study was estimated from litter‐removal experiments, which provide a practical and comparable approach for synthesis across diverse ecosystems. Our estimates therefore represent the apparent litter‐attributable contribution to total R _s. Future studies combining litter manipulation with isotopic tracing could further distinguish CO₂ release directly derived from litter decomposition from indirect treatment effects. Finally, the ridge and valley monitoring was conducted from January to April 2025 and mainly captured winter–spring dynamics. These high‐frequency site‐level observations provide complementary process‐based evidence for local microtopographic regulation of litter‐derived R _s, while longer‐term and seasonally comprehensive monitoring would further evaluate the persistence of these topographic patterns across annual and interannual scales.

5. Conclusions

Globally, litter‐derived R _s is estimated to account for an average of 30.9% of total R _s, representing an important but previously poorly quantified component of heterotrophic CO₂ emissions from terrestrial soils. Topography is a dominant driver of spatial variability in litter‐derived R _s across both mountain and non‐mountain regions, whereas NPP also contributes substantially to this variability in non‐mountain regions. Evidence from hourly field monitoring further reveals significant variation in litter‐derived R _s across complex terrain, reinforcing the strong control of topography over CO₂ release from decomposing litter. Under 5–8.5 scenario, litter‐derived R _s is projected to increase globally, despite a substantial decline in wetlands and marked increases in grasslands and croplands, particularly in mountain regions. These contrasting responses suggest that R _h in mountain regions exhibits greater sensitivity to climate change. Overall, these findings provide a new perspective for distinguishing R _h derived from undecomposed litter from that derived from older SOM in ESMs. Explicitly distinguishing these sources may improve the representation of litter‐derived R _s and reduce uncertainty in estimates of the terrestrial C cycle, particularly across extensive mountain regions globally.

Author Contributions

Jing M. Chen: methodology, writing – review and editing. Josep Peñuelas: writing – review and editing. Helena Vallicrosa: writing – review and editing. Xingzhou Huang: methodology, writing – review and editing. Fuzhong Wu: writing – review and editing. Xiangyin Ni: conceptualization, funding acquisition, methodology, writing – review and editing. Mianhai Zheng: writing – review and editing. Yingying Zhao: writing – review and editing. Andi Li: methodology, writing – review and editing. Siyi Tan: investigation, data curation, writing – original draft, methodology, writing – review and editing. Jingjing Zhu: project administration.

Funding

This work was supported by National Key Research and Development Program of China, 2023YFF1305500; National Natural Science Foundation of China, 32471834, 32022056; Fujian Provincial Natural Science Foundation, 2025J011012.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: Supporting Information.

Figure S1: Litter‐derived R _s under various litter removal manipulation intensities.

Figure S2: Out‐of‐bag (OOB) error as a function of the number of trees in the random forest model.

Figure S3: Global litter‐derived Rs uncertainty.

Figure S4: Residual plot of the Random Forest model in predicting litter‐derived Rs. Each point represents the difference between observed and predicted values across the prediction range. Rs: soil respiration.

Figure S5: Litter‐derived Rs across major Köppen climate zones under current and future climate scenarios.

Figure S6: Field monitoring of litter‐derived Rs at ridge and valley.

Figure S7: Projected changes in litter‐derived Rs across ecosystems.

Figure S8: Hourly changes in soil redox potential (Eh) at ridge and valley.

Figure S9: Hourly changes in soil temperature and moisture at ridge and valley.

Table S1: Descriptions of all variables used in the models.

Table S2: Model performance based on the best predictor variables.

Table S3: Correspondence between the IPCC land category used for change detection and the LCCS legend used in CCI‐LC classes.

Table S4: High‐resolution field monitoring sites.

Data Availability Statement

The data that support the findings of this study are available in Zenodo at https://doi.org/10.5281/zenodo.21264949. The global gridded datasets used in this study as environmental covariates are available in the public domain (see Table S1 for a full list). Code availability: All the code used in the analyses presented in this paper is available via https://doi.org/10.5281/zenodo.21264949 in the code folder.