Lithosphere as a bottom-up modulator of mountain vegetation resilience

Abstract

Understanding vegetation resilience, defined as the recovery speed of vegetation following external perturbations, is essential for predicting ecosystem stability under climate change, yet the role of lithosphere in mountain systems remains poorly understood. Here, we assess how lithology and tectonics modulate vegetation resilience across global mountains using integrated climate, soil, topography, human footprint, lithology, land-cover, and vegetation datasets. Vegetation type and temperature dominate global patterns, with resilience highest in trees, followed by herbs and shrubs, together explaining 86% of the spatial variability. After accounting for climatic and anthropogenic effects, resilience across vegetation types is strongly mediated by lithospheric pathways. Tree resilience peaks in steep, actively incising terrains underlain by erosion-resistant plutonic and metamorphic rocks, as well as dissolution-prone carbonate rocks. Herb resilience is strongly linked to soil organic carbon enriched on mafic and metamorphic rocks. Shrub resilience is primarily temperature-dependent (93%) but shows reduced warming sensitivity on nutrient-poor substrates. Moreover, resilience peaks at intermediate channel steepness (ksn ≈ 500), suggesting that moderate tectonic uplift promotes vegetation resilience. These findings identify the lithosphere as a fundamental modulator of mountain vegetation resilience and highlight geological constraints on ecosystem stability across heterogeneous landscapes.

Ecosystem resilience is usually linked to climate, soil, and biological factors, while geological influences are often overlooked. This study suggests that lithology and tectonic intensity fundamentally modulate mountain vegetation resilience.

Introduction

Exploring the drivers of vegetation variability is critical for understanding the stability of terrestrial ecosystems^1,2. Previous studies have primarily focused on climatic forcing, disturbance regimes, and vegetation functional traits, yielding insights into how temperature, moisture availability, and biological strategies shape ecosystem responses to perturbations^3–5. Vegetation, the upper skin of Earth’s critical zone, is fundamentally constrained by its lithological substrate^6,7. As Earth System Science increasingly emphasizes the coupling among the biosphere, lithosphere and climate system^8,9, growing evidence suggests that bedrock composition, weathering processes, and tectonic activity exert substantial controls on ecosystem productivity, biodiversity, and biogeochemical cycling^10–13. However, due to still limited connection of geology to ecology¹⁴, the role of the lithosphere as a fundamental dimension of vegetation stability remains largely unexplored.

Vegetation resilience, commonly quantified as the recovery rate of vegetation after perturbations, has emerged as a key metric of vegetation stability^15,16. Advances in satellite observations, ecological theory, and ecosystem modeling have enabled increasingly robust assessments of resilience across broad spatial and temporal scales^17,18. Studies show that rising temperatures, altered hydroclimatic variability, and the intensification of extreme events can weaken vegetation resilience by pushing vegetation closer to physiological thresholds^19,20. Besides, plant functional traits such as rooting depth, hydraulic diversity, and carbon allocation strategies strongly influence the capacity of vegetation to resist and recover from disturbances^21–23. These findings highlight the importance of climate–biosphere interactions in regulating resilience and have promoted resilience-based approaches for assessing ecosystem stability and forecasting ecological tipping points.

Despite these advances, current resilience research has remained largely climate-centric or biocentric, with geological factors frequently overlooked²⁴. Yet emerging evidence indicates that substrate properties and long-term topographic evolution shaped by lithology and tectonic processes can exert first-order controls on water availability, nutrient supply, and ultimately ecosystem functioning^12,25,26. As the mineral source, lithology governs soil development and nutrient availability^27,28. Tectonic uplift and erosion continuously rejuvenate landscapes, modulating mineral release, soil thickness, and hydrological flow, which in turn regulate vegetation rooting strategies and its access to nutrients and water^29–31. The growing recognition of Earth’s surface as a coupled geo–bio system underscores the need to incorporate lithospheric factors into resilience assessments, that has not yet been achieved at the global mountain scale with high climatic sensitivity and geological heterogeneity.

Mountains, as ecologically sensitive and erosion-dominated landscapes, provide a natural laboratory for disentangling lithosphere–biosphere interactions. Mountain vegetation exhibits more prominent responses to climate change compared with lowland ecosystems^32,33, making these regions sentinel zones for detecting shifts in vegetation resilience. At the same time, rapid erosion in mountain belts exposes fresh bedrock and minimizes overprinting by far-traveled sediments, allowing lithological signatures to be expressed more directly than in sedimentary basins³⁴ or fluvial systems where sediment storage obscures lithological signals³⁵. Mountain regions are also global hotspots of chemical weathering²⁹, where tectonic uplift accelerates the release of nutrient-bearing minerals³⁶, thereby reshaping soil development, hydrology, and ultimately ecosystem functioning³⁷. These combined attributes make mountain systems ideal for a triaxial assessment of climate–geology–vegetation interactions, and for revealing how the lithosphere imprints on vegetation resilience.

In this study, we hypothesize that the lithosphere influences mountain vegetation resilience through different pathways among vegetation types across geological settings. To test this hypothesis, we integrate datasets of climate, soil, topography, human footprint, lithology, landcover, and remote sensing-based vegetation to quantify the lithospheric modulation of vegetation resilience across the world’s mountain areas (Fig. 1a, b and Supplementary Fig. 1). Specifically, we first develop a unified global framework to assess vegetation resilience from long-term satellite observations, enabling consistent comparisons across biomes and climatic gradients. We then quantify the partial effects and relative importance of climatic, ecological, and edaphic factors on resilience using generalized additive models (GAMs). Finally, we reveal the pathways through which the lithosphere influences the resilience of different vegetation functional groups after accounting for the spatial effects of climate and human activities (Fig. 1c). This integrative approach links short-term vegetation responses with long-term geological constraints, establishing the lithosphere as a critical but often overlooked dimension of vegetation resilience.

Fig. 1. Conceptual model of lithology-mediated pathways modulating vegetation resilience of various functional groups across mountains.

a The study area of global mountain regions, ranging from 60°S to 60°N. b The lithological distribution across global mountains⁸². Lithological abbreviations are as follows: us unconsolidated sediments, pc pure carbonate rocks, mc mixed sedimentary rocks, ss siliciclastic sedimentary rocks, ev evaporites, mt metamorphic rocks, pa acid plutonic rocks, pi intermediate plutonic rocks, pb basic plutonic rocks, va acid volcanic rocks, vi intermediate volcanic rocks; vb basic volcanic rocks, py pyroclastics. c The conceptual model of this study. Lithology is grouped into four classes: unconsolidated sediments, sedimentary rocks, igneous rocks, and metamorphic rocks. Mountain vegetation consists of three major functional groups: trees, shrubs, and herbs, accounting for 97% of the study area, excluding croplands.

Results

Vegetation resilience across global mountains

Vegetation resilience values (α), estimated from enhanced vegetation index (EVI) during 2001–2020, show strong spatial heterogeneity across the world’s mountain regions (Fig. 2). High-resilience zones (α < 0.3) are primarily concentrated along humid montane regions with forest cover (Fig. 3a), including the Andes, the eastern Himalaya–Hengduan arc, and Maritime Southeast Asia, as well as in high-latitude northern mountains. In contrast, low-resilience zones (α > 0.6) are widespread in arid and semiarid mountain systems, such as the southern Rockies, Atlas, Iranian Plateau, central Andes, eastern and southern parts of Africa, and Great Divide Range. These regions are typically characterized by shrubs (Fig. 3a). Overall, the global pattern indicates that vegetation resilience is generally higher in forest-dominated, moisture-rich mountains and lower in shrub-dominated drylands.

Fig. 2. The global pattern of mountain vegetation resilience and the relative importance of environmental predictors.

Vegetation resilience is measured by the resilience coefficient (α), with lower α corresponding to higher resilience.

Fig. 3. Contrasting resilience and environmental controls among mountain vegetation functional groups.

a Spatial pattern of landcover over global mountains. b Predicted marginal means of α among three major vegetation types after accounting for climatic, edaphic, topographic, and anthropogenic covariates, with error bars representing 95% confidence intervals. c Relative contributions of climatic, soil, topographic, and human factors to resilience across three vegetation types.

Variance decomposition of the GAM reveals strong hierarchical controls on vegetation resilience (Fig. 2, Supplementary Fig. 2, and Supplementary Table 1). Vegetation type (vegtype) explains more than half of the total variance, followed by mean annual temperature (MAT), which accounts for ~30%. Topsoil organic carbon (TOC), river channel steepness (ksn), and climatic water availability (SPEI) provide secondary contributions (3–6%), whereas available water capacity (AWC) and soil thickness (ST) each exhibit minor effects (~1%). Human footprint (HFP) contributes the least to variance (0.1%), reflecting that the anthropogenic disturbance is limited to the spatial heterogeneity of resilience. Together, vegetation functional identity and regional thermal regimes govern the resilience pattern of global mountain vegetation.

Dominant factors of resilience in different vegetation types

Across global mountain regions, vegetation resilience exhibits clear differences among functional groups. After statistically controlling for climatic, edaphic, topographic, and anthropogenic influences, trees exhibit the highest baseline resilience, especially for evergreen broadleaf forests, followed by herbs (woody savannas > savannas > grasslands), whereas shrubs (closed > open shrublands) display the lowest resilience (Fig. 3b and Supplementary Fig. 2h).

Partitioning the vegetation types further reveals the significant differences in the control of vegetation resilience by climate, soil and topography (Fig. 3c and Supplementary Table 1). For trees, resilience is primarily shaped by climate, with MAT and SPEI contributing the largest share of explained variance (31 and 24%, respectively). Geomorphic forcing, represented by ksn, provides the next strongest influence (17%), while soil properties and human activities make comparatively minor contributions. Shrub resilience is governed almost exclusively by MAT (93%), which overwhelmingly dominates its spatial variation. This strong climatic control suggests that shrub-dominated mountain ecosystems exhibit limited buffering capacity against thermal fluctuations. For herbs, resilience is mainly influenced by MAT (60%) and TOC (21%), indicating a combined sensitivity to thermal stress and soil organic matter availability. This vegetation-type dependence underscores the need for differentiated frameworks when assessing the vulnerability of mountain ecosystems under climate change.

Roles of the lithosphere in vegetation resilience

Across mountain ecosystems, vegetation types exhibit strikingly different resilience patterns shaped by underlying lithological conditions. Forest ecosystems display distinct resilience variation among lithological classes after removing climatic and anthropogenic signals (Fig. 4a). Specifically, the violin plots reveal that trees on ev, us, and va lithologies consistently show the lowest adjusted resilience, whereas those on pi, mt, and pc lithologies exhibit comparatively higher resilience. Across all lithological categories, tree resilience increases with elevated ksn (Fig. 4b). This strong positive correlation (R² = 0.79, p < 0.001) suggests that steeper and more actively incising landscapes (Fig. 4c) tend to support more resilient forest ecosystems.

Fig. 4. Lithological roles in the resilience of different vegetation types.

a, b Variation in adjusted tree resilience among lithological classes and its relationship with ksn. Violin plots represent the kernel density distribution of resilience values; center lines indicate medians; white dots denote mean values; vertical lines represent the interquartile range. Each violin represents independent spatial grid cells (n = 5170, 18940, 37546, 30854, 20, 29971, 25471, 3092, 3013, 4917, 8285, 8096, and 3002 for us, pc, mc, ss, ev, mt, pa, pi, pb, va, vi, vb, and py, respectively). c Scatter distribution between median ksn and adjusted tree resilience across lithological types. The dashed line indicates the fitted linear regression (p < 0.001). d Sensitivity of shrub resilience to MAT under different lithologies. Each bar represents independent spatial grid cells (n = 1314, 3358, 3387, 4027, 48, 2926, 1807, 281, 153, 768, 1470, 1372, and 725 for us, pc, mc, ss, ev, mt, pa, pi, pb, va, vi, vb, and py, respectively). e Variation in adjusted herb resilience across lithological types. Violin plot elements are defined in Fig. 4a. Each violin represents independent spatial grid cells (n = 10,334, 39,541, 62,092, 44,273, 435, 45,970, 44,475, 3664, 5163, 9698, 16,853, 13,901, and 8003 for us, pc, mc, ss, ev, mt, pa, pi, pb, va, vi, vb, and py, respectively). f GAM-predicted marginal means of soil TOC across lithological types for herbs after controlling for all other covariates. Bars represent estimated marginal means, and error bars indicate 95% confidence intervals. All resilience values were calculated after removing climatic and anthropogenic signals using GAMs. Lithological abbreviations are defined in Fig. 1b.

Since MAT dominates the resilience in shrubs, the sensitivity of resilience to MAT mediated by lithologies is analyzed using a generalized additive mixed model (GAMM) in Fig. 4d. In general, shrub resilience displays a decreasing trend with rising MAT across all substrates, but the magnitude of decline varies markedly. Shrubs on ev, pc, va, and py rocks experience the relatively weak temperature sensitivity, whereas those on pi, vi, and pa lithologies show a much stronger decline in resilience with warming.

Herb-dominated ecosystems show a distinct lithology–soil pathway of control. Their resilience varies significantly among lithological types (Fig. 4e), and these differences closely mirror lithology-associated variations in soil resource availability (Fig. 4f). Lithologies characterized by higher (lower) soil TOC, such as vb, mt, and ss (ev, us, pc, and va), consistently support higher (lower) resilience, highlighting that soil organic matter acts as a key stabilizing factor for herbaceous ecosystems.

The GAM results also reveal a nonlinear relationship between tectonic forcing and vegetation resilience. Using ksn as a proxy for relative uplift intensity, resilience increases with ksn up to ~500, beyond which it declines (Supplementary Fig. 2f).

Discussion

Difference in mountain vegetation resilience

Mountain ecosystems exhibit a pronounced hierarchy of intrinsic resilience, with trees consistently outperforming herbs and shrubs (Fig. 3b). This global pattern likely reflects fundamental contrasts in ecosystem structure and trait-mediated stress responses. Forest ecosystems possess deep-rooting systems that access relatively stable water sources²¹, large carbon and biomass pools³⁸, and multilayered canopies that buffer hydroclimatic variability and facilitate rapid post-disturbance recovery³⁹. Evergreen broadleaf forests, characterized by greater structural complexity and functional redundancy, exhibit particularly high resilience^40,41 (Supplementary Fig. 2h). Together, these attributes contribute to the generally high resilience observed in forest-dominated mountain ecosystems.

Herbaceous ecosystems display intermediate resilience. In contrast to forests, their relatively shallow rooting systems, simple structure, and low biomass constrain their ability to withstand prolonged drought and heat stress⁴². However, high turnover rates and fast compensatory growth support efficient recovery following stress events⁴³. As a result, herbs tend to show moderate resistance but comparatively strong resilience, especially where disturbance regimes are recurrent but not persistent⁴⁴. This intermediate stability is evident in grass-dominated systems of the mid-high-latitude mountains of the Northern Hemisphere, Central Asian ranges such as the Himalaya, Tianshan, and Pamir Plateau, and parts of the Andes and Rockies (Figs. 2 and 3a).

In contrast, shrub-dominated ecosystems exhibit the lowest resilience. Although shrubs are generally adapted to persistent environmental stress, their conservative resource-use strategies are often associated with slow growth, low photosynthetic capacity, and limited phenological flexibility⁴⁵, which constrain recovery following disturbance. These limitations are particularly amplified in arid and semiarid mountain regions⁴⁶, such as the southwestern North America, the Atlas Mountains, the Iranian Plateau, and interior ranges of Australia (Fig. 2). Overall, these biome-scale contrasts underscore that variation in mountain vegetation resilience is primarily governed by ecosystem structure and functional traits⁴⁷.

Divergent lithological pathways modulating mountain vegetation resilience

Trees

Our results reveal a counterintuitive but robust geo-ecological pattern: mountain trees achieve their highest resilience in topographically adverse environments characterized by steep slopes. This pattern can be explained through a continuous lithosphere–topography–rooting strategy–resilience cascade.

Steep slopes, typically associated with sustained tectonic uplift and active erosion⁴⁸, enhance gravitational drainage and reduce water residence time in surface soils, thereby favoring deeper rooting strategies (Supplementary Fig. 3a). To maintain water supply, trees increasingly exploit fractured bedrock water and fissure-stored moisture^49,50, forming hydraulically buffered systems that are less sensitive to short-term climatic variability⁵¹. Deep rooting also reduces the risks of hydraulic failure and carbon starvation under elevated temperatures²³, enhancing resistance to climatic extremes and accelerating post-disturbance recovery^52,53. Meanwhile, active erosion continuously exposes fresh mineral substrates, promoting soil renewal, nutrient supply, and relatively stable TOC levels^29,36 (Supplementary Fig. 4a), further reinforcing tree recovery capacity³⁰ (Fig. 3c).

The persistence of these steep environments is ultimately controlled by lithospheric properties. Mechanically resistant lithologies, such as plutonic and metamorphic rocks, resist denudation and maintain high-relief terrain under tectonic uplift⁵⁴ (Fig. 4c). Carbonate rocks, although chemically weatherable, also sustain steep topography through dissolution-driven valley incision⁵⁵ (Fig. 4c). These settings favor deep rooting and hydraulic stability⁵⁶. In contrast, weakly consolidated sediments and volcanic substrates are more susceptible to rapid weathering and surface collapse, commonly forming low-relief terrains with water-retaining soils that inhibit deep root penetration⁵⁷ (Supplementary Fig. 3b). Such conditions increase dependence on shallow soil moisture and amplify climatic sensitivity, resulting in lower drought resistance and slower recovery following disturbance⁵⁸.

Taken together, these findings support a bottom-up pathway linking lithology, topography, and vegetation resilience. Lithology regulates variations in topographic steepness, topography likely shapes rooting systems through soil hydrology, and ultimately, rooting strategies influence vegetation resilience. This cascading framework highlights the lithology as a fundamental constraint on ecological strategies that underpin resilience in mountain forests.

Shrubs

Lithology also plays a significant role as a modulator in shrub resilience, although its spatial pattern is dominated by MAT variation (Fig. 3c). Shrubs exhibit reduced sensitivity of resilience to warming on the most nutrient-poor and harsh lithologies such as ev, pc, va, and py (Fig. 4d), where low nutrient availability and weak water retention impose persistent resource stress (Supplementary Fig. 5). These constraints select for conservative shrub strategies, characterized by slow-growth, stress-tolerant traits⁵⁹. Such traits confer large hydraulic safety margins and reduced vulnerability to xylem embolism, enabling shrubs to maintain stable water transport under drought and heat stress⁶⁰. As a result, shrubs established on poor substrates exhibit low sensitivity of resilience to warming gradients, since their physiological performance is already adapted to chronic stress conditions. In contrast, fertile igneous substrates generally support finer-textured soils with higher nutrient availability and greater water retention⁶¹, favoring acquisitive strategies associated with faster growth rates⁶². However, these traits reduce hydraulic safety margins and increase susceptibility to hydraulic failure and carbon starvation under thermal stress⁶³, making shrub resilience more sensitive to warming.

Geological “benignness” therefore does not necessarily enhance shrub stability; instead, it amplifies climate sensitivity by promoting acquisitive but vulnerable strategies, whereas persistent edaphic stress triggered by poor-nutrient rocks filters for conservative strategies that stabilize shrub ecosystem responses. These findings highlight the critical role of lithology in determining soil conditions and shaping functional strategies that underpin shrub resilience and sensitivity to temperatures (i.e., dominant factor of resilience) across global mountains.

Herbs

Herbs exhibit a resilience pathway distinct from that of trees and shrubs, relying primarily on TOC-mediated resource buffering rather than topographic or substrate stress (Figs. 3c and 4e, f). High TOC improves soil water retention, nutrient availability, and overall resource stability under climatic fluctuations⁶⁴. This buffering effect is particularly critical for vegetation characterized by shallow roots, low biomass, and fast turnover⁶⁵, as it compensates for the lack of deep water access and structural persistence. Consequently, TOC-rich soils reduce the exposure of herbs to short-term climatic variability and facilitate rapid post-disturbance regrowth. Accordingly, highly resilient herbs are concentrated in cold mid- to high-latitude mountains of the Northern Hemisphere (Figs. 2 and 3a), where low decomposition rates favor long-term organic matter accumulation (Supplementary Fig. 4b).

Lithology fundamentally modulates this TOC-mediated resilience pathway. Parent materials derived from igneous and metamorphic rocks, particularly easily weathered mafic lithologies, release more minerals into soils than sedimentary rocks (Fig. 4f). These mineral inputs not only directly enhance nutrient availability for vegetation⁶⁶, but also promote organo–mineral associations that stabilize organic matter⁶⁷. In contrast, sedimentary rocks generally provide limited mineral inputs, which constrain the stabilization and long-term buildup of TOC despite locally higher nutrient contents in some siliciclastic substrates⁶¹. Unconsolidated sediments, although often characterized by high porosity and water availability, are highly susceptible to erosion and mass wasting⁶⁸. This instability disrupts soil development, reduces nutrient retention, and limits organic matter accumulation, thereby weakening the resource buffering capacity required for herb resilience. Through its regulation of mineral supply and substrate stability, lithology therefore constrains TOC formation and persistence, ultimately shaping the resilience of herbaceous ecosystems.

Dual effects of tectonic uplift on vegetation resilience

Vegetation resilience in mountain landscapes follows a threshold-dependent response to tectonic uplift (Supplementary Fig. 2f). At moderate uplift levels, enhanced erosion significantly promotes continuous bedrock weathering and nutrient release (Supplementary Fig. 4a), while simultaneously fostering deeper rooting strategies in forests (Supplementary Fig. 3a), which improve access to subsurface moisture and buffer climatic variability^11,29. In this regime, lithospheric forcing strengthens vegetation resilience by coupling mineral supply with hydrological stability. Beyond the threshold, however, rapid incision and steep slopes suppress regolith development and intensify geomorphic disturbance⁶⁹. Under these conditions, erosion exceeds soil production, causing resilience to decline. Tectonic forcing, therefore, shifts from a resource-enhancing to a disturbance-dominated driver of ecosystem resilience.

The position of this threshold is further modulated by lithology, reflecting its impact on the efficiency of resource production versus loss. Mechanically weak lithologies (e.g., unconsolidated sediments and volcanic rocks) are highly susceptible to erosion and soil depletion, leading to an earlier shift toward disturbance-dominated conditions at relatively low ksn values (Supplementary Fig. 6). In contrast, resistant lithologies (e.g., plutonic and metamorphic rocks) maintain topographic integrity and support sustained regolith development under stronger uplift, thereby shifting the resilience threshold toward higher ksn values. These contrasting responses indicate that identical tectonic forcing can produce divergent ecological outcomes depending on substrate properties. Tectonic forcing, therefore, acts as a conditional driver of vegetation resilience, with its net effect emerging from the balance between resource generation and geomorphic disturbance under varying lithological constraints.

Overall, our findings demonstrate that mountain vegetation resilience is an emergent property of lithosphere–biosphere coupling rather than a direct response to climate alone. Through its regulation of topography, hydrological buffering, soil development, and nutrient supply, the lithosphere exerts a foundational bottom-up influence on vegetation resilience across global mountains. This integrated framework helps explain why resilient trees are preferentially associated with mechanically resistant lithologies, shrub resilience remains relatively insensitive to warming gradients on harsh substrates, and herb resilience is high on mafic rocks. These cascading controls are synthesized in Fig. 5, which integrates the pathways and relative contributions of key factors shaping resilience across vegetation types.

Fig. 5.

Synthesized framework illustrating the bottom-up influencing paths of lithosphere on resilience across mountain vegetation types.

Implications for ecosystem management and global change adaptation

Our findings provide actionable insights for ecosystem management by demonstrating that vegetation resilience in mountain regions is jointly constrained by vegetation type, climate, and bottom-up effects of the lithosphere. These interacting controls imply that effective management strategies must move beyond climate-based approaches and instead account for the underlying lithosphere–biosphere coupling.

Management priorities should be differentiated across broad vegetation and lithological settings (Supplementary Table 2). Forest ecosystems generally exhibit high intrinsic resilience, particularly in tectonically active regions under moderate uplift and resistant lithologies such as plutonic and metamorphic rocks, where deep rooting and sustained resource renewal support long-term stability. In such settings, low-intervention strategies that preserve geomorphic–ecological connectivity are most effective. By contrast, on erosion-prone lithologies (e.g., unconsolidated sediments and volcanic rocks) or under excessive uplift, shallow soil development and limited rooting depth constrain water access; here, management should prioritize hydrological stabilization, including improving soil water retention and minimizing slope disturbance to prevent soil loss. Shrub-dominated systems are strongly constrained by temperature and exhibit heightened sensitivity to climatic warming, especially on relatively nutrient-enriched substrates (e.g., igneous and siliciclastic sedimentary rocks). In these systems, management should first reduce additional stressors, including overgrazing, land degradation, and vegetation fragmentation, to preserve microclimatic buffering and ecosystem resilience. Herbaceous ecosystems are strongly dependent on TOC-mediated resource buffering and are therefore most vulnerable under nutrient-poor lithologies (e.g., evaporites, unconsolidated sediments, pure carbonates, and acid volcanic rocks). Maintaining soil organic matter and reducing erosion are essential for sustaining ecosystem recovery capacity under climate change.

Such lithology-dependent resilience suggests that climate risks in mountain ecosystems are spatially heterogeneous even within similar biomes and climatic zones⁷⁰. The space-for-time perspective is particularly important for mountain regions, where pronounced geological heterogeneity can lead to contrasting ecosystem responses over short spatial scales. Incorporating lithological and tectonic constraints into ecosystem management frameworks may therefore improve regional adaptation strategies and resilience forecasting under future climate change.

Despite these advances, several uncertainties and methodological limitations should be acknowledged. First, limitations in data representation may obscure underlying spatial heterogeneity. Lithology is represented as categorical classes that may mask within-unit variability, while remotely sensed vegetation indices and derived resilience metrics are affected by sensor noise, atmospheric correction, and spatial aggregation, particularly in heterogeneous mountain terrains. Second, scale mismatches between observation and process may influence the inferred relationships despite the same data resolution. Key ecological processes governing resilience, including microtopographic heterogeneity and rooting depth, operate at fine scales, whereas variables such as ksn and soil properties represent spatially aggregated states, potentially smoothing localized controls. Third, the GAM framework captures nonlinear associations but does not explicitly resolve causal mechanisms or temporal feedbacks. The results, therefore, represent equilibrium-like patterns based on spatial variation (i.e., space-for-time substitution) rather than true temporal dynamics, and caution is required when extrapolating these results to future climate dynamics.

Future work incorporating field observations, high-resolution datasets, and process-based dynamic models will help resolve fine-scale heterogeneity and disentangle the coupled effects of tectonics, lithology, climate, and soil development on ecosystem resilience. Despite these challenges, our results highlight the critical but often overlooked role of lithosphere-related factors in shaping ecosystem resilience through their effects on surface processes, topography, and soil properties. This study, therefore, provides a foundation for incorporating lithospheric variables into ecological prediction and management frameworks across global mountains.

Methods

Vegetation resilience

To quantify vegetation resilience under short-term climatic perturbations, we modeled vegetation dynamics using a first-order autoregressive framework that links current vegetation anomalies to both contemporaneous climate forcing and legacy effects⁷¹. Specifically, standardized EVI anomalies were expressed as a function of antecedent vegetation state, moisture conditions, and temperature variability:

$${\mathrm{EVI}}_t=\alpha \times {\mathrm{EVI}}_{t-1}+\beta \times {\mathrm{SPEI}}_t+\gamma \times {\mathrm{MAT}}_t+{\epsilon }_t$$ 1

where EVI_t denotes the standardized EVI anomaly at time t, EVI_t−1 represents the lagged vegetation anomaly capturing temporal memory, SPEI_t is the standardized drought index at a three-month timescale, and MAT_t is the standardized temperature anomaly. The error term ε_t accounts for unexplained variability. All variables were standardized to ensure comparability among coefficients. Resilience is inversely related to the strength of temporal autocorrelation: systems that recover rapidly from disturbances exhibit weak dependence on past states, whereas slow-recovering systems retain strong memory. Accordingly, vegetation resilience is quantified as the coefficient of the lagged EVI anomaly (α) during the period 2001–2020. Lower values of α indicate faster recovery and greater resilience, while higher values reflect prolonged disturbance effects and reduced system stability.

Analysis of influencing factors

The GAM was applied to identify the dominant environmental controls on vegetation resilience, which is well-suited for capturing nonlinear and threshold responses commonly observed in ecological systems⁷². Vegetation resilience, quantified as the recovery metric α, was used as the response variable. A suite of climatic, edaphic, topographic, human-related, and lithological predictors was considered, including MAT, SPEI, TOC, ST, AWC, ksn, HFP, and vegtype (categorical variable). These variables are treated as time-invariant boundary conditions over the study period, and their role is evaluated through their spatial heterogeneity in modulating the temporal dynamics of vegetation^54,73. No substantial multicollinearity was detected among these continuous predictors (variance inflation factors <2). We evaluated the proportion of explained deviance and adjusted R² to quantify model fit. The statistical significance of smooth terms was assessed based on approximate p values. The GAM framework can be expressed as:

$$\mathrm{f}\left(\mathrm{E}\left({\alpha }_i\right)\right)={\beta }_0+{\sum }_n{\mathrm{s}}_n\left(X_{in}\right)+{\sum }_m{\beta }_m{Z}_{im}+{\epsilon }_i$$ 2

where α_i denotes vegetation resilience at location i; β₀ is the intercept; s_n(X_in) represents smooth functions of the n-th predictor X_i; Z_im is a dummy variable for the m-th categorical variable, and β_m is the corresponding regression coefficient relative to the reference level; ε_i is the residual error term. Thin-plate regression splines were used to estimate the smooth terms, with smoothing parameters selected via restricted maximum likelihood to avoid overfitting. The GAM-based attribution approach assumes additive and smooth relationships among predictors, potentially under-representing complex interactions among climate, soil, and biotic processes.

Partial dependence curves derived from the fitted GAM were used to visualize the partial effects of individual predictors on vegetation resilience. This approach can identify nonlinear sensitivities and thresholds in responses across environmental gradients. For categorical predictors, predicted marginal means were calculated to represent the average predicted responses associated with each category level while holding other variables constant. The relative contribution of individual predictors was quantified using a variance-based decomposition of the GAM term-wise fitted values. For each term, the variance of its partial predictions was calculated as a percentage of the summed variances across all model terms, yielding a comparative measure of predictor importance:

$$C_k=\mathrm{Var}\left({\mathrm{f}}_k\left(X_k\right)\right)/{\sum }_{i=1}^n\mathrm{Var}\left({\mathrm{f}}_n\left(X_{in}\right)\right)\times 100\%$$ 3

where f_k(X_k) represents the predicted value of the k-th term, either the smooth function s_k(X_k) for a continuous variable or the parametric factor effects for a categorical variable; and C_k is the relative contribution of the variable.

Lithological impacts

GAMs were used to remove the nonlinear spatial signals of climate variation (MAT and SPEI) and human disturbance (HFP) for trees and herbs to reveal the underlying lithological effects modulating vegetation resilience. Lithology-specific resilience levels were then evaluated among different lithological classes under equivalent climatic and other disturbance conditions. In shrub-dominated systems, where temperature emerged as the primary climatic driver of vegetation resilience, we focused on evaluating whether lithology modulates the sensitivity of resilience to temperature using a generalized additive mixed model (GAMM). Specifically, lithology was introduced as a random slope on the temperature term, allowing the strength of temperature sensitivity to vary among lithological units. Random-effect estimates were extracted from the fitted GAMM to quantify lithology-specific deviations in temperature sensitivity, thereby revealing how geological substrates modulate the climatic responses of shrub resilience. The GAMM can be expressed as:

$$\mathrm{f}\left(\mathrm{E}\left({\alpha }_{ij}\right)\right)={\beta }_0+b_{0j}+b_{1j}\times {\mathrm{MAT}}_{ij}+{\sum }_n{\mathrm{s}}_n\left(X_{ijn}\right)+{\epsilon }_{ij}$$ 4

where α_ij denotes shrubland resilience at location i within lithology unit j; β₀ is the global intercept; b_0j and b_1j represent lithology-specific random intercepts and random slopes for temperature; the smooth function s_n(X_ijn) captures nonlinear effects of other covariates; ε_ij is the residual error term. Lithology-specific deviations in temperature sensitivity were then quantified by extracting the random slope estimates for temperature, that is b_1j.

Data

The datasets used in this study consist of mountain boundary, climate, soil, topography, human activity, landcover, lithology, and remote sensing-based vegetation datasets. All datasets were resampled into the 0.05° resolution, ranging from 60 °S to 60 °N. Given that ecological processes governing vegetation resilience often operate at finer spatial scales than those resolved here. In mountainous terrain, sharp gradients in microclimate, soil properties, and topography can occur over short distances, potentially leading to sub-grid variability that cannot be fully captured by coarse-resolution data. As a result, our findings should be interpreted as representing large-scale patterns rather than local-scale mechanisms.

The study is limited to the erosion-dominated mountainous landscapes without the sedimentation effect, where the topographic relief is defined as more than 200 m within a 2.5′ cell based on a 30”-resolution SRTM DEM⁷⁴. We used the robust global mountain system derived from the Global Mountain Biodiversity Assessment (GMBA) Mountain Inventory v2.0⁷⁵, which covers 12.3% of the global terrestrial area.

Climatic conditions were characterized using mean annual temperature (MAT; Supplementary Fig. 1a) and the standardized precipitation–evapotranspiration index at a 3-month timescale (SPEI-03; Supplementary Fig. 1b). MAT was calculated from monthly maximum and minimum temperature datasets from CRU-TS 4.09 downscaled with WorldClim 2.1 at a spatial resolution of 2.5 min^76,77. SPEI-03 data were obtained from the global SPEI database at a spatial resolution of 0.5°⁷⁸, which integrates precipitation and potential evapotranspiration and captures short-term hydroclimatic variability.

The chemical and physical characteristics of the topsoil (0 − 30 cm), organic carbon (TOC; Supplementary Fig. 1c), and available water capacity (AWC; Supplementary Fig. 1d) were obtained from the Harmonized World Soil Database version 2.0 (HWSD v2.0) with a 1-km resolution. The soil thickness (ST; Supplementary Fig. 1e) offers the high-resolution estimates of the permeable layer thickness above bedrock at a global 30” (~1 km) grid⁷⁹, in which invalid soil values (<0) were filtered prior to analysis.

The river channel steepness index (ksn; Supplementary Fig. 1f) is a metric of river properties scaling with the relative rock uplift rate and erosion⁸⁰. In this study, the global normalized 500-m resolution ksn map, obtained from a global DEM covering mountain regions up to 60°N^48,54, was used to reflect the strength of tectonic uplift over mountainous areas.

Anthropogenic disturbance was characterized using the Human Footprint Index (HFP; Supplementary Fig. 1g), obtained from the global Human Footprint database⁸¹. The dataset integrates population density, land transformation, infrastructure, and accessibility, and was used at 1-km resolution (year 2009) to represent spatial heterogeneity in human pressure across mountain regions.

Landcover information (Supplementary Fig. 1h) was obtained from the MODIS MCD12C1 product for the year 2010, available from the NASA LP DAAC database. The dataset provides globally consistent landcover classification based on the IGBP classification system at a spatial resolution of 0.05°, and was used to identify the distribution of major vegetation types across mountain regions.

Mountain lithological variation (Fig. 1b) was characterized using the global lithological map (GLiM)⁸², which classifies surface lithology into 13 major lithological categories. The GLiM dataset was assembled from existing regional geological maps and represents global surface rock types with 1,235,400 polygons. Lithological information was obtained from the GLiM database and used to quantify spatial heterogeneity in geological substrate across global mountain regions.

The monthly enhanced vegetation index (EVI) was extracted from the MODIS MOD13C2 products with a 0.05° global resolution for a 20-year period from 2001 to 2020. These quality-controlled EVI values were used to analyze the vegetation resilience over the global mountainous domain. The rooting depth was obtained from the global 5-km rooting depth dataset for plant functional types⁸³.

Reporting summary

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

Author contributions

H.Z. and X.Z. conceived and designed the work. H.Z. collected data, analyzed, and drafted the original manuscript. X.Z., J.Z., X.F., and P.Y. contributed to reviewing and editing the paper. H.Z. and X.Z. secured funding for the study. All the authors read and approved the final version of the paper.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

H.Z. and X.Z. disclose support for the research and publication of this work from the National Natural Science Foundation of China [grant numbers 42301094 and 41972192].

Data availability

The data generated in this study have been deposited in a public repository and are available via Zenodo at 10.5281/zenodo.20519497. The data used in this study are publicly available from third-party sources, including Mountain boundary: 10.48601/earthenv-t9k2-1407; Mean annual temperature: https://www.worldclim.org/data/monthlywth.html#google_vignette; Standardized precipitation–evapotranspiration index: https://spei.csic.es/spei_database; Topsoil organic carbon and available water capacity: https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v20; Soil thickness: https://daac.ornl.gov/cgi-bin/dsviewer.pl?ds_id=1304; Channel steepness index: 10.3929/ethz-b-000411632; Human footprint index: 10.5061/dryad.052q5; Land cover: https://lpdaac.usgs.gov; Global lithological map: 10.1594/PANGAEA.788537; Vegetation index: https://lpdaac.usgs.gov/products/mod13c2v061; Rooting depth: https://zenodo.org/records/15361409.1.

Code availability

The codes used in the present study are freely available online at 10.5281/zenodo.20519531.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74875-3.