Bedrock mediates responses of ecosystem productivity to climate variability

The study addresses how temporal climatic variability, rather than only mean climate states, affects terrestrial ecosystem productivity, a key facet of ecosystem resilience to climate change. Specifically, it investigates whether and how bedrock lithology and its weathering products (regolith and soils) mediate ecosystem sensitivity of primary productivity to variability in climatic water deficit (CWD). The central hypothesis is that bedrock controls water-holding capacity via regolith porosity, permeability, and thickness, which in turn alters both the magnitude and sign of productivity responses to drying or warming. The authors also posit that bedrock may influence sensitivity through nutrient supply and potential toxins from weathering. The study aims to quantify lithologic effects on sensitivity globally, disentangle mechanisms via mediating variables (porosity, permeability, soil/regolith thickness), and compare effects across regions where primary constraints differ (water-limited vs energy-limited systems).

Prior work has shown climate variability strongly governs ecosystem functioning and resilience, but most assessments emphasize mean climate change rather than variability. Bedrock influences terrestrial productivity by regulating plant-available water storage and nutrient supply: soluble carbonates can generate high-permeability karst systems with low water retention, whereas deep regolith can store substantial plant-accessible water, especially during droughts. Empirical studies have documented greater interannual climatic sensitivity in some karst regions due to reduced water holding capacity. Bedrock weathering also supplies essential nutrients (P, Ca, Mg, K) and may release toxins (As, Se, Cd), affecting growth; lithologic nutrient content varies (e.g., carbonate and mixed siliciclastic–carbonate rocks tend to be low in phosphate). Ecosystem constraints differ by biome: energy-limited high latitudes and mountains respond positively to warming via longer growing seasons, whereas water-limited systems often show productivity declines with increased CWD; humid tropical forests can exhibit higher productivity during drier years due to alleviation of oxygen limitation and enhanced nutrient mineralization. Previous global analyses linked sensitivity to biodiversity, species niches, successional stage, and topography, but global evaluations isolating lithologic mediation via regolith properties have been limited by data availability on permeability/porosity. Newly available datasets (GLiM, GLHYMPS, regolith thickness maps, TerraClimate CWD, and GIMMS NDVI) now enable a global-scale, mechanism-focused assessment.

Design: Two-stage global analysis at 0.5° resolution over 1982–2013. Stage 1 estimates grid-specific sensitivity of ecosystem productivity (NDVI) to interannual variability in CWD; Stage 2 explains spatial variation in sensitivity using regolith and lithology covariates via weighted Bayesian hierarchical regression with spatial autocorrelation.

Data: (1) Productivity: GIMMS AVHRR NDVI3g (1/12°, 15-day), aggregated to annual mean at 0.5°. Robust pre-processing accounts for atmospheric and volcanic effects. Sensitivity tests also used global GPP (8 km, 1982–2016) and growing-season NDVI. (2) Climate: TerraClimate CWD (1/24° monthly), aggregated to annual mean and rescaled to 0.5°. (3) Explanatory variables: Bedrock lithology (GLiM, 15 first-level hydrolithology classes); regolith porosity and permeability (GLHYMPS); soil and regolith thickness (Pelletier et al. global thickness dataset; soil = upland soils or lowland unconsolidated sediments; regolith = intact weathered layer beneath soils). All predictors standardized (z-scores).

Stage 1 sensitivity estimation: For each land grid, detrend NDVI and CWD time series and fit a Bayesian linear regression NDVI ~ CWD to obtain slope a_i as sensitivity and its uncertainty σ_ai (95% credible interval used to assess significance). Total 49,389 grid regressions.

Regional stratification: Based on sign of estimated sensitivity and biome context, define four regions: Region I (negative sensitivity); Region II (positive sensitivity in energy-limited systems: boreal/taiga, temperate high-latitude/montane forests, montane grasslands/shrublands); Region III (humid tropical and subtropical moist broadleaf forests with positive sensitivity); Region IV (hyper-arid deserts and xeric shrublands with positive sensitivity). This supports region-specific hypotheses about water-holding capacity effects.

Stage 2 explanatory modeling: Weighted Bayesian hierarchical regression on log(|a_i|), weighting each grid by variance (σ_ai^2). Predictors: regolith porosity, regolith permeability (log k), regolith thickness, soil thickness; include region-specific coefficients via hierarchical priors, and random effects for lithology by region (15 GLiM classes). Spatial autocorrelation modeled with a Gaussian process kernel K having an exponential covariance that decays with inter-grid distance, truncated beyond 1000 km for efficiency. Alternative model includes biome type as an additional random effect; results robust to inclusion of biome and to alternate productivity metrics (GPP) and growing-season definitions. Model implementation in JAGS via RJAGS with 20 MCMC chains, convergence assessed by Gelman–Rubin; posterior means and 95% credible intervals reported. Model fit: R^2 = 0.29 for predicted vs observed sensitivities.

Global patterns: Mean global sensitivity of productivity to CWD is −0.00026 (SD 0.0018) NDVI units per 1 mm month−1 CWD (i.e., per 12 mm yr−1). Sensitivity ranges from −0.076 to 0.149, with >99% between −0.01 and 0.01. About 33.3% of grids have statistically significant sensitivity (95% CI excludes 0). Latitude patterns show positive sensitivity clustering in polar and tropical regions.

Negative sensitivity dominates: 63% of grids show negative sensitivity (45% of these significant), mean −0.00093 (SD 0.0014); most significant negatives occur in moderately water-limited biomes (tropical/subtropical dry broadleaf forests; grasslands/savannas/shrublands; temperate grasslands/savannas/shrublands; temperate conifer; Mediterranean; desert/xeric shrublands).

Positive sensitivity: 37% of grids positive (13.3% significant). Of positive areas: 63% in energy-limited regions (Region II), 20% in deserts/xeric shrublands (Region IV), 14% in humid tropical regions (Region III). Regional mean sensitivities (all grids): Region II 0.0011 (SD 0.0017), Region III 0.0010 (SD 0.0032), Region IV 0.00022 (SD 0.00025). Max absolute sensitivity observed: 0.149. In energy-limited regions, higher CWD (warmer conditions) tends to increase productivity; in humid tropics, increased CWD (typically reduced precipitation) associates with higher productivity; in hyper-arid deserts, positive but much smaller magnitude sensitivity consistent with observed greening.

Effects of porosity and permeability: Their influence is region-dependent and significant. One SD increase in regolith permeability or porosity changes sensitivity by:

• Region I (negative sensitivity): +3% (permeability) and +7% (porosity) relative to regional mean magnitude (i.e., amplifies declines under high CWD).
• Region II (energy-limited): −12% (permeability) and −13% (porosity) (dampens warming-driven productivity increases).
• Region III (humid tropics): −12% (permeability) and −6% (porosity) (dampens positive response to drying; lower permeability/porosity areas respond more strongly to drying).
• Region IV (hyper-arid positives): −11% for both (dampens positive response to CWD).

Effects of thicknesses: One SD increase in thickness changes sensitivity by:

• Region I: Soil −17%, Regolith −6% (both reduce sensitivity; thicker profiles buffer variability).
• Region II: Soil +18%, Regolith +5% (thicker profiles enhance warming-driven increases, likely by alleviating seasonal water limitation).
• Region III: Regolith +34% (strong enhancement of response to drying); soil effect not statistically significant.
• Region IV: Regolith +12%, Soil +7% (enhances positive response, potentially linked to CO2 fertilization interacting with water storage).

Residual lithology effects after controlling for mediators: No significant lithology random effects in Region I; significant in Regions II–IV. Patterns:

• Energy-limited (II): Sedimentary rocks (mixed, carbonate, siliciclastic, unconsolidated) dampen positive sensitivity; pyroclastics and acid plutonic rocks intensify sensitivity; plutonic rocks overall linked to greater productivity increases under warming.
• Humid tropics (III): Sedimentary rocks increase sensitivity to drying.
• Hyper-arid (IV): Sedimentary rocks dampen positive sensitivity; pyroclastics and acid plutonic rocks intensify it.

Findings demonstrate that both the sign and magnitude of ecosystem sensitivity to climatic water deficit vary widely and are systematically mediated by subsurface properties derived from bedrock. Most land shows negative sensitivity—drying or warming reduces productivity—yet positive sensitivity occurs in energy-limited high-latitude/montane systems (warming lengthens growing seasons), humid tropics (moderate drying alleviates oxygen limitation, enhances decomposition/mineralization, and increases light), and some hyper-arid deserts (positive trends likely confounded with CO2 fertilization and multi-decadal precipitation changes). The strong, region-specific effects of regolith porosity and permeability indicate that reduced water-holding capacity amplifies sensitivity where ecosystems are water-limited, while greater storage (lower permeability, lower porosity) buffers or modifies responses in energy-limited and humid regions. Thickness of soils and intact regolith is a major mediator: thicker profiles buffer variability in water-limited regions but enhance positive responses in energy-limited and humid tropical systems by modulating wetness duration and storage dynamics. Residual lithology effects imply additional mechanisms beyond water storage, plausibly involving nutrient supply limitations and geochemical differences among rock types (e.g., low P in carbonate/mixed sedimentary rocks may constrain growth responses). Overall, regolith properties account for substantial variability (effect sizes up to ~30% change in sensitivity per SD change in thickness), and the spatial model reproduces a significant fraction of observed sensitivity (R^2 = 0.29). These results extend prior work by explicitly quantifying bedrock-mediated controls on sensitivity at global scale and highlighting that subsurface critical zone properties are integral to ecosystem resilience under climate variability.

The study shows that bedrock lithology, via its weathering products (regolith and soils), significantly mediates global ecosystem productivity sensitivity to climatic water deficit. Two-thirds of terrestrial ecosystems exhibit negative sensitivity, while positive sensitivity occurs in energy-limited, humid tropical, and hyper-arid regions with distinct mechanisms. Regolith porosity, permeability, and thickness, as well as soil thickness, exert strong, region-specific controls consistent with their regulation of water-holding capacity; lithology-specific residual effects point to additional geochemical mechanisms. These insights underscore the importance of incorporating subsurface critical zone properties into assessments of ecosystem resilience to climate variability and change. Future work should (1) more explicitly account for anthropogenic land use (e.g., croplands), (2) disentangle nutrient-mediated mechanisms across lithologies, (3) integrate other covariates such as biodiversity and topography at comparable scales, and (4) refine causal inference by controlling for confounding drivers and leveraging additional independent productivity datasets and in situ observations.

• Anthropogenic influence: Agricultural areas were not explicitly removed; human management can dampen variability, potentially underestimating sensitivity in affected grids.
• Causality: Inferences about mechanisms (e.g., nutrient mediation by lithology) are based on statistical correlations; causal relationships remain speculative.
• Confounding in arid regions: Positive sensitivity in hyper-arid regions may reflect co-variation with rising atmospheric CO2 (fertilization) and multi-decadal precipitation changes, complicating attribution to CWD alone.
• Model fit: The explanatory model accounts for a moderate portion of variance (R^2 = 0.29), indicating additional unmodeled drivers (e.g., biotic composition, fine-scale topography) likely contribute.
• Data constraints: Global datasets were upscaled to 0.5° and annual means; sub-grid heterogeneity and seasonal dynamics may reduce sensitivity detection in some regions.