A global temperature control of silicate weathering intensity

Chemical weathering of silicate minerals acts as a long-term negative feedback stabilizing Earth’s climate by consuming atmospheric CO2. A core hypothesis posits that higher volcanic CO2 raises surface temperature, which in turn enhances silicate weathering and draws down CO2. While temperature dependence of silicate weathering is evident in lab and small-watershed studies, it is often obscured at large spatial/temporal scales due to covariation with other factors (e.g., precipitation) and limited temperature variability. Consequently, alternative controls such as hydrology (precipitation/runoff), tectonics (relief/erosion), and land surface reorganization have been invoked. Testing these drivers over geologic timescales is hampered by uncertainties in weathering proxies and scarcity of high-quality paleo-records for individual forcing factors. CIA (chemical index of alteration) and WIP (weathering index of Parker) are widely used sediment-based weathering proxies relying on major elements and partly tracking feldspar hydrolysis. CIA increases and WIP decreases with weathering intensity. This study compiles fine-grained modern river sediment geochemistry from six continents (n = 3828) to evaluate catchment-scale controls (20 environmental variables) on silicate weathering intensity and to assess the utility of these indices, especially CIA, as paleoclimate proxies.

Prior work documents temperature dependence of silicate weathering in experiments and small catchments, yet correlations weaken at larger scales due to confounding factors. Competing frameworks emphasize temperature control, precipitation/hydrology control, or combined climate control. Tectonic uplift, erosion, and land surface reactivity have also been proposed to modulate global weathering and the carbon cycle, including hypotheses that late Cenozoic cooling increased land surface reactivity rather than overall weathering flux. However, interpreting paleo-weathering signals is complicated by inconsistent proxy behavior, limited spatial coverage, and mixed climatic signals in global stacks. CIA/WIP have been widely applied but their large-scale climatic sensitivities and the relative roles of temperature vs precipitation vs geomorphology/lithology require validation with extensive modern datasets spanning broad environmental gradients.

Data compilation: 3828 fine-grained river sediment samples (suspended particulate matter, clay fractions, or bedload sieved to fine sizes, e.g., <63 μm) were compiled from peer-reviewed sources and public datasets (e.g., NAWQA, AGDB2, FOREGS, NGSA). Inclusion criteria: availability of major elements (Al, Ca, Na, K, Mg); full digestion or XRF; exclude samples with K/Al > 1; focus on fine fractions to minimize hydraulic sorting and unaltered mineral contributions. CIA and WIP calculations: CIA = Al2O3/(Al2O3 + CaO* + Na2O + K2O) × 100, using McLennan’s procedure to correct CaO to silicate-bound CaO* (phosphate and carbonate corrections). WIP uses weighted molar proportions of Ca, Mg, Na, K with CaO* as the silicate-bound component. Geospatial analysis: For a sub-sample (n = 2989) from small- to medium-sized catchments (upstream area < 10^5 km²), upstream basins were delineated with SRTM DEM (90 m; alternative data > ~60° latitude). Basin-averaged environmental variables (n = 20) were extracted: climate (MAT, MAP, temperature and precipitation seasonality), geomorphology (actual evapotranspiration, sediment yield modeled by BQART, flow length, drainage area, elevation, slope, regolith and soil thickness), lithology (areal percentages of acidic-intermediate, basic, clastic sedimentary, carbonate, metamorphic rocks; rock erodibility index), and land cover (vegetation, tree, ice/snow). Datasets included WorldClim 2 for climate, global soil-water balance for AET, GLiM for lithology, GLC2000 for land cover, and global soil/regolith thickness grids. Sediment yield estimation: BQART model applied using basin metrics to estimate sediment yield where hydrologic data lacked. Statistical analysis: Correlation coefficients (R) between CIA and each environmental variable were computed; significance assessed (p-values). CIA data were also grouped (binned) by MAT (2 °C intervals) and MAP (0.2 m/yr intervals) to assess isolated effects and reduce covariance. Sensitivity tests evaluated MAT–CIA relations across grain-size classes, mineralogical sources (via CIA–WIP slope differences), and dominant lithologies. Feldspar dissolution and modeling: CIA was converted to percentage feldspar dissolved (fdiss) via stoichiometry of feldspar-to-kaolinite hydrolysis (fdiss = 100/(CIA + 2)). Using the empirical MAT–CIA linear relation (derived from binned data, R = 0.99), an empirical MAT–fdiss function was obtained and used in a simplified feldspar weathering model to estimate transient CO2 consumption changes from Ca, Na, K release, normalized to modern silicate weathering CO2 consumption. Arrhenius-based first-order estimates using literature activation energies (36–107 kJ/mol) provided comparison.

• Strong global temperature control: Mean annual temperature (MAT) shows the strongest correlation with CIA among all variables (R = 0.60; p < 0.001), with CIA increasing monotonically with MAT for MAT > 0 °C. Temperature explains more variance than precipitation or geomorphic/lithologic factors.
• Weak and non-monotonic precipitation effect: MAP vs CIA exhibits variable trends depending on MAP range (decrease at 0.2–0.8 m/yr; increase at 0.8–1.4 m/yr; near-constant above ~2 m/yr), with low correlations within fixed MAT bins, indicating regional, competing effects of fluid supply vs erosion-induced reduction of reaction residence time.
• Latitudinal pattern: CIA decreases with latitude; high values (80–100) prevail in tropics/subtropics (<10°), while high-latitude (>60°) CIA approaches fresh rock values (~50–60). North America and polar samples show somewhat higher CIA at high latitudes, likely due to differing forcings (mass wasting/glacial processes).
• Subordinate roles for other factors: Land cover and lithology show weak correlations with CIA (|R| ≤ 0.20; modeled source-rock CIA vs sediment CIA R = 0.21). Geomorphic metrics generally weak (|R| ≤ 0.28), except a moderate positive correlation with soil thickness (|R| = 0.49), consistent with clay-enriched thick soils during intense weathering. Modeled sediment yield (BQART) correlates poorly with CIA, suggesting decoupling of physical erosion and chemical weathering intensity at the global scale of these data.
• Empirical MAT–CIA relation: Binning CIA by 2 °C MAT intervals yields a robust linear relation (R = 0.99) across grain sizes, mineralogical sources, and lithologies. Global-average CIA predicted from GSAT = 14 °C is 73.5, matching prior estimates (71.6–75.5).
• Paleotemperature reconstruction: Using ΔMAT = ΔCIA/1.02 (from the MAT–CIA slope), ten paleo-records spanning PTB, PETM, MMCO, LGM, and HCO show ΔMAT from CIA agreeing within uncertainty (offset <1–3 °C) with independent biomarker/pollen reconstructions; distributions are statistically indistinguishable (paired t-test p > 0.05). Estimated uncertainty ~3 °C.
• Nonlinear temperature forcing on feldspar dissolution: Empirical MAT–fdiss relation indicates diminishing sensitivity at higher MAT, interpreted as depletion of more reactive plagioclase and dominance of orthoclase at advanced weathering stages.
• Carbon cycle implication: A simplified feldspar weathering model driven by MAT–fdiss predicts a 28% increase in transient CO2 consumption for a +3 °C warming relative to modern, within the 17–59% range estimated by Arrhenius kinetics.
• Hypothesis supported: Results support increased land surface reactivity during late Cenozoic cooling due to a higher proportion of reactive plagioclase available for weathering at lower temperatures.

The analysis demonstrates that, when covariation is minimized via global coverage and binning, temperature exerts the primary control on silicate weathering intensity in fine-grained sediments, as captured by CIA. Precipitation effects are contingent and can be overridden by erosional processes, explaining contradictory regional observations. The robust MAT–CIA relation enables first-order quantitative paleotemperature reconstructions from siliciclastic archives where other proxies may be unavailable, especially in deep time. The inferred nonlinear MAT–fdiss relationship, attributed to shifting feldspar reactivity (plagioclase to orthoclase dominance with increasing temperature), implies stronger temperature–weathering feedback at lower temperatures. Incorporating the empirical MAT–fdiss into carbon cycle models links surface temperature to weathering-driven CO2 consumption without relying solely on theoretical kinetics, yielding responses consistent with Arrhenius expectations but with concave behavior reflecting mineralogical changes. These findings strengthen the climate-weathering feedback framework and provide a pathway to reconcile disparate paleo-weathering records via evolving land surface reactivity.

This study compiles the largest global dataset of fine-grained river sediment geochemistry to evaluate controls on silicate weathering intensity. Temperature is identified as the dominant global driver, with precipitation and geomorphic/lithologic factors acting regionally and secondarily. A robust empirical MAT–CIA relation (R = 0.99) enables quantitative estimates of paleotemperature changes from sedimentary CIA with ~3 °C uncertainty, corroborated by independent biomarker/pollen records for major climatic events. Transforming CIA to feldspar dissolution reveals a nonlinear temperature dependence consistent with depletion of reactive plagioclase at higher temperatures and supports enhanced land surface reactivity during late Cenozoic cooling. A simple feldspar weathering model predicts a ~28% increase in weathering CO2 consumption for +3 °C warming, aligning with kinetic expectations. Future work should expand high-resolution, provenance-constrained fine-grained sediment records, refine regional baselines for CIA, integrate Mg-bearing silicates, and couple the empirical MAT–fdiss relation with full carbon cycle models across deep time.

• Proxy biases: CIA can be affected by grain-size variability, hydraulic sorting, and changes in sediment provenance; application requires fine fractions (silt/clay), minimal provenance change, and adequate integration timescales (10^3–10^4 years).
• Site-specific baselines: The intercept of the MAT–CIA relation may vary regionally (geology/lithology), making absolute MAT reconstructions uncertain; relative changes (ΔMAT from ΔCIA) are more robust.
• Calcium correction: Standardized CaO* correction (McLennan) may under/overestimate CIA in specific lithologies or weathering regimes, though biases likely average out globally.
• Environmental covariates: Despite global coverage and binning, residual covariance (e.g., relief, extreme precipitation events, glacial processes) can locally deviate CIA from MAT trends, especially at MAT ≤ 0 °C.
• Sediment yield and load partitioning: The CO2 consumption model assumes feldspar export mainly in fine fractions; significant coarse load contributions could bias estimates (likely small in many basins but variable).
• Gridded datasets and basin delineation: Uncertainties in DEMs, lithologic maps (GLiM), and land cover products, especially in heterogeneous terrains and high latitudes, may affect basin-averaged variables.
• Exclusion of Mg-silicate contributions in CO2 model leads to conservative CO2 consumption estimates.