Impact of Climate on the Global Capacity for Enhanced Rock Weathering on Croplands

Abstract

Enhanced rock weathering (ERW) on croplands has emerged as an economically and ecologically promising negative emissions technology. However, estimated total carbon sequestration potential from ERW on croplands and its potential sensitivity to climate conditions requires further understanding. Here we combine 1-D reactive transport modeling with climate model experiments to simulate ERW on ∼1,000 agricultural sites globally. Applying a fixed rate of 10 tons of basalt dust per hectare on these sites sequesters 64 gigatons of CO₂ over a 75-year period; when extrapolated to all agricultural land, ERW sequesters 217 gigatons of CO₂ over the same time interval. However, we find that a significant fraction of applied basalt does not weather even on a multidecadal timescale, indicating the need to optimize application strategies for cost effectiveness. We find that ERW becomes modestly more effective with global warming and predict that the payback period for a given ERW deployment is significantly shorter in hot and humid environments currently coinciding with relatively low per-capita incomes. These results provide strong impetus for investment in agricultural reform in developing economies and highlight an additional potential co-benefit of ERW.

Key Points

• Enhanced rock weathering (ERW) with fixed annual application rates of 10 tons of basalt dust per hectare on 1,000 global cropland sites sequesters 64 gigatons of CO₂ over 2006–2080

• Extrapolated to all croplands, ERW with a fixed application rate of 10 tons of basalt dust per hectare sequesters 215 gigatons of CO₂ over 2006–2080

• ERW is resilient to global climate change but is much more efficient over hot and humid environments

Plain Language Summary

Enhanced rock weathering (ERW) on croplands is a promising negative emissions concept that accelerates natural weathering by amending soils with crushed rock. Our simulations of ERW with a fixed rate of 10 tons of basalt dust per hectare on all global croplands suggest ERW can sequester >200 gigatons of CO₂ over a 75-year period. This suggests that cost and logistical concerns, rather than weathering potential, are likely to be the key limiting factors for large-scale deployment of enhanced weathering. Notably, ERW is resilient to global climate change and becomes more effective with global warming. ERW is moreover far more efficient in hot and humid environments currently coinciding with relatively low per-capita gross domestic product economies. Our study provides strong support for the assertion that ERW represents a resilient carbon capture strategy that is non-competitive for arable land and can foster CO₂ removal at the gigaton scale.

1 Introduction

Anthropogenic emissions of carbon dioxide (CO₂) and other greenhouse gases over the past ∼150 years have contributed to global temperatures increasing by about 1.2°C relative to the preindustrial era (World Meteorological Organization, 2020), with the rate of warming accelerating in recent decades (NOAA, 2021). These trends are concerning given that global warming (a) pushes Earth's climate system toward more frequent and more extreme weather (Baek & Lora, 2021; Baldwin et al., 2019; Cook et al., 2014; Payne et al., 2020; Williams et al., 2020); (b) leads to environmental degradation (including soil, vegetation, and water resource degradation; Allen et al., 2010; Almagro et al., 2017; Burrell et al., 2020; Gonzalez et al., 2010; Lindner et al., 2010; Midgley & Bond, 2015; Zhang et al., 2017); and (c) causes sea levels to rise from thermal expansion of seawater and added meltwater (Mengel et al., 2016; Rahmstorf, 2007). The Intergovernmental Panel on Climate Change (IPCC) has responded by delineating numerous mitigation pathways with the potential to limit global warming to 1.5–2°C above preindustrial levels by 2100 (IPCC, 2018). Importantly, all mitigation pathways stipulate the active removal of atmospheric carbon dioxide (in addition to drastic cuts in emissions) on the order of 100 to 1,000 gigatons (billion tons) within the next century to keep total warming below 1.5°C (IPCC, 2018).

Among various negative emissions technologies (Campbell et al., 2022; Fawzy et al., 2020; Jeswani et al., 2022), enhanced rock weathering (ERW) has high potential for rapid scaling since it can leverage existing infrastructure for its application (Fuss et al., 2014, 2018; Minx et al., 2018). ERW is based on the natural process of chemical weathering, which sequesters atmospheric carbon dioxide through dissolution of silicates or carbonates (via carbonic acid). Natural weathering stores carbon in the oceans as bicarbonate ions for millennia, and even longer when mineral carbonates (produced by calcifying organisms) are buried in marine sediments (e.g., Archer et al., 1997; Ridgwell & Hargreaves, 2007). This process currently removes about ∼1.0 gigatons of atmospheric carbon dioxide per year (IPCC, 2022; Strefler et al., 2018). Although silicate weathering will ultimately balance anthropogenic CO₂ release rates over geological time scales, the natural response time of this process is orders of magnitude too slow to mitigate anthropogenic fossil fuel emissions (Archer et al., 1997). The goal of ERW is to accelerate natural weathering by (a) amending soils with crushed rocks (particularly silicates) and (b) concentrating weathering activity on regions where rock/mineral dissolution is favored (e.g., Beerling et al., 2020; Goll et al., 2021; Taylor et al., 2016).

ERW could in theory be applied to any vegetated area (e.g., Goll et al., 2021). However, logistical accessibility for transport of silicate feedstock significantly improves economic viability. In this respect, ERW on croplands (which are road accessible, in contrast to most forested lands) has emerged as a promising, scalable approach toward CO₂ removal (e.g., Beerling et al., 2020), with ecological co-benefits, such as (a) rejuvenation of depleted soils, (b) improved crop yields (from additional nutrient supply from rock feedstock), and (c) mitigation of ocean acidification (from addition of alkalinity), that make ERW socially desirable (Beerling et al., 2018; Strefler et al., 2018; Taylor et al., 2016). Estimates of carbon capture from ERW on global croplands nevertheless remain sparse, with only one existing study to our knowledge (Beerling et al., 2020). Moreover, the sensitivity of such estimates to regional climate and global climate change has not yet been investigated. Therefore, there is an obvious need for additional model-based and empirical constraints on the potential for ERW to act as a means of large-scale CO₂ removal.

Here, we provide model-based, first-order estimates of the global CO₂ removal potential of ERW on cropland systems. We use a newly developed 1-dimensional reactive transport model (Soil Cycles of Elements simulator for Predicting TERrestrial regulation of greenhouse gases; SCEPTER) to simulate ERW (Kanzaki et al., 2022). This soil biogeochemical model represents an advancement in the complexity of models used to predict enhanced weathering behavior on a global scale. In particular, the model includes comprehensive particle size distribution and specific surface area tracking, regionally specific host soil development, and a transient soil spin-up (see Supporting Information S1 for more details). We combine SCEPTER with global climate model simulations with the dual aims of (a) estimating the total CO₂ sequestration potential of ERW on global croplands and (b) characterizing the sensitivity of this potential to regional climate conditions. Collectively, our efforts provide an illustrative example of the extent to which ERW on Earth's croplands could mitigate global warming in the coming decades. Our modeling framework furthermore provides strong independent support for previous work suggesting that widespread deployment of enhanced weathering on cropland systems has the potential to drive gigaton-scale CO₂ removal annually (Beerling et al., 2020). However, our results also highlight the potential for low weathering efficiency during ERW deployment with coarse-grained feedstock, with significant material remaining undissolved on decadal timescales.

2 Data and Methods

2.1 Identifying 70%+ Cropland Grid Points

We identify cropland sites using a global 1-km consensus land-cover product integrating four remote sensing-derived base products (see Tuanmu & Jetz, 2014 for details). This consensus product provides information on the prevalence of 12 different land-cover classes for every nominal 1-km pixel on the globe (except Antarctica), from which we define cropland to be the cultivated and managed vegetation land-cover classes. We degrade the cropland prevalence from 1-km pixel resolution to 1° grid resolution via averaging, then select grid points where cropland proportion is at least 70% of the land cover in each grid point. This process yields 963 grid points representing global cropland. We estimate the land area of cropland in each 1° grid point by multiplying the percent cropland of each grid point with the grid area, using a 1° latitude = 110.6 km and 1° longitude = 111.3 * cos(latitude) conversion. Our reactive transport model is then applied to these 963 grid points, from which we derive a global ERW estimate.

2.2 1-Dimensional Reactive Transport Modeling

SCEPTER is a 1-D reactive transport model that simulates weathering, with options to: (a) enable variable soil mixing regimes (such as natural bioturbation or tilling operated at croplands); (b) track size distributions of soil mineral particles and calculate their surface areas; (c) trace targeted minerals and/or soil organic matter to the soil surface; and (d) allow temporal changes in climatological boundary conditions (such as surface temperature, water infiltration rate and soil water content). Some of our model features (e.g., d above) have already been implemented by previous ERW modeling approaches (cf. Beerling et al., 2020; Kantzas et al., 2022), but others are new, including mechanistic soil mixing, explicit tracking of soil organic matter and soil CO₂ in depth and time, and basalt application in two steps, that is, basalt application experiments branching off from spin-ups that only implement natural weathering without any basalt application (see below for more details, and Kanzaki et al. (2022) for a full description of SCEPTER). We note that while the shrinking-core model in SCEPTER has reproduced dissolution/precipitation of minerals observed in the laboratory, it has not yet been tested against ground basalt applied in the field. In addition, the simulations presented here do not account for passivation effects. Although this should render our conclusion that dissolution efficiency can be low with coarse-grained feedstocks conservative, it may lead to an overestimation of overall capture potential at a given feedstock application rate.

We use SCEPTER on the 963 identified cropland sites around the globe, using each site's unique set of lithological/geomorphological and climate parameters as inputs. We derive lithology data from Hartmann and Moosdorf (2012), soil organic matter data from Hengl et al. (2017), and erosion rate data from Larsen et al. (2014). We derive detailed soil respiration data based on the mass balance of organic matter in soil (using global datasets of soil organic matter weight fractions and erosion rate). The calculation assumes mass balance of soil organic matter within 30 cm between erosion and respiration, the latter of which is formulated with organic matter decay reflecting its temperature dependence (3 of Q₁₀, a factor of increase in the rate by each 10°C temperature increase; Wang et al., 2010), along with the soil porosity (0.5) and particle densities of soil minerals (∼2.6 g cm⁻³) and organic matter (1.5 g cm⁻³) in accord with assumed boundary conditions for SCEPTER (see Kanzaki et al., 2022). Specifically, organic matter rain (g m⁻² year⁻¹) is specified in SCEPTER as f_SOM * (2.6e6 * (1. − f_SOM) + 1.5e6 * f_SOM) * (1.− 0.5) * (w + k_SOM * 0.3 * 0.5), where f_SOM is the weight fraction of soil organic matter, w is erosion rate (m year⁻¹), and k_SOM is a decay constant (year⁻¹), given by k_SOM,ref * 3^(tc – tc_ref)/10 where k_SOM,ref and tc_ref are reference decay constant (1/8 y⁻¹; Chen et al., 2010) and temperature (15°C).

We input local climate parameters relevant for weathering—surface temperature, total liquid runoff, and volumetric soil moisture at 21 cm depth—from the Community Earth System Model version 1 (CESM1), a fully coupled global climate model, at grid points corresponding to the 1° grid of the 963 cropland pixels with all relevant CESM1 outputs re-gridded to the common 1° grid resolution using bi-linear interpolation. SCEPTER experiments that do not apply basalt dust are conducted under natural bioturbation mixing conditions; SCEPTER experiments with basalt dust application are conducted with tilling for cropland grid cells. We note that the grid resolution in CESM1 is potentially coarse compared to expected variability in weathering rates on the field scale. Although we consider our approach appropriate for the goal of providing first-order estimates of weathering rate and efficiency across broad regions of arable land, performing more high-resolution regional (or field-scale) analysis for targeted deployments will be an important avenue for future work.

Prior to any ERW simulations, we initialize SCEPTER by running a 100 Kyr spin-up experiment with local, observed soil parameters and 850–1850 CE millennium averages of surface temperature, total liquid runoff, and volumetric soil moisture at 21 cm depth (note that the climate variables thus have no seasonal cycle); we do this without any basalt dust application and under natural bioturbation mixing conditions. This 100 Kyr initialization is done so that soil mineralogy can evolve from imposed parent-rock lithology and achieve steady state. The millennium averages of the climate variables are calculated from one member (Ensemble Member #2) of the “Last Millennium Ensemble” Project, which forces the CESM1 model with reconstructed estimates of natural and pre-industrial radiative forcing (see Otto-Bliesner et al., 2016 for details). Outputs from the Last Millennium Ensemble are available on ∼1.9° latitude by 2.5° longitude grid resolution.

We follow up these 100 Kyr initializations with seasonal calibrations that run SCEPTER for an additional 86 years representing the 1920–2005 historical interval, again under natural bioturbation mixing conditions and without any basalt dust added. We use monthly climate variables (surface temperature, total liquid runoff, volumetric soil moisture at 21 cm depth) simulated by a 20-member ensemble mean of the CESM1 Large Ensemble (members #2–21; Kay et al., 2015) forced with historical radiative forcing over 1920–2005. In using monthly climate variables, the calibration experiments impose seasonality, which was absent in the 100 Kyr initialization. Specifically, we use the 20-member ensemble mean, which drastically reduces the influence of internal climate variability and thus represents the radiatively forced response of 1920–2005. For some of the 963 sites, we replace negative total liquid runoff values (which are inherent in the climate model output) with very small positive values (10⁻¹⁵ mm/s) so that runoff is always positive even with seasonal fluctuations. Climate outputs from the Large Ensemble are available on ∼1° grid resolution.

After initialization and seasonal calibration, we apply 10 tons of basalt dust per hectare on the 963 cropland sites every year for 75 years representing the 2006–2080 interval. This basalt application rate is designed to yield a conservative estimate for global carbon dioxide removal (CDR) potential, as it is on the lower end of application rates assumed in previous work (Beerling et al., 2020; Goll et al., 2021) but is broadly consistent with conventional rates of ag-lime application (Galinato et al., 2011; Lukin & Epplin, 2003; T. O. West & McBride, 2005). As before, we use local, site-specific observed soil parameters but use monthly climate variables simulated by the CESM1 Large Ensemble (20-members; members #2–21), this time under the Representative Concentration Pathways 4.5 and 8.5 W/m² (RCP4.5 and RCP8.5) scenarios. The two future scenario basalt application experiments are thus branched out of the historical simulations without basalt application. CDR is then estimated based on differences in cumulative carbon fluxes in the basalt application experiments relative to the end of the historical experiment. Again, for some of the sites we replace negative runoff values with very small positive values (10⁻¹⁵ mm/s) so that runoff is always positive even with seasonal fluctuations. ERW experiments with basalt application are conducted under tilling conditions operated at croplands. While appropriate for our purposes, we note that the use of monthly climate variables (as opposed to, for instance, daily variables) may not fully capture natural wet-dry/cool-hot cycles or percolation events. This may in turn lead to under- or overestimated mineral dissolution rates and represents an obvious target for future research (Buckingham et al., 2022; Calabrese et al., 2022; L. J. West et al., 2023).

The above-described procedures successfully simulate ERW on 929 (927) out of the 963 sites (>96% success rate) under the RCP8.5 (RCP4.5) scenario; SCEPTER did not converge for the other 34 (36) sites (representing <4% of locations), mostly due to (a) climate boundary conditions that make convergence untenable during the 100 Kyr spin-up experiment (e.g., zero/NaN values for runoff [mm/s], soil moisture [unitless fraction], or temperature [K]) and/or (b) high assumed organic matter flux (e.g., >3,000 g C/m² year).

Spin-up experiments (which do not have basalt application) and basalt application simulations described above are conducted in the same way as the “pulsed basalt application” example in Kanzaki et al. (2022), with full tracking of particle size distributions of individual minerals; however, tilling is represented by an inversion mixing in this paper (see Equation S1 in Supporting Information S1), rather than as homogeneous mixing. We also track the evolution of solid, aqueous, and gaseous species. Tracked solid species include: albite, anorthite, K-feldspar, forsterite, fayalite, diopside, phlogopite, tremolite, hedenbergite, quartz, goethite, kaolinite, Ca-beidellite, Mg-beidellite, gypsum, calcite, aragonite, dolomite, and one class of soil organic matter; tracked aqueous species include: Ca, Mg, Na, K, Si, Al, Fe(II), Fe(III), and SO₄; tracked gaseous species include: O₂ and CO₂; Fe(II) oxidation to Fe(III) by O₂ is also explicitly simulated as an extra reaction in addition to dissolution/precipitation of the above-listed solid species. Fundamental thermodynamic (e.g., particle density, molar volume/weight, chemical formula, and solubility of solid species; thermodynamic constants for formation of dependent aqueous species) and kinetic (e.g., rate constants of dissolution/precipitation for solid species and extra reactions; molecular diffusion coefficients for aqueous/gaseous species) data are given in Kanzaki et al. (2022).

For gas and aqueous species, upper boundary conditions are given by fixed concentration values equivalent to the chemical composition of the atmosphere (0.21 and 10^−3.5 atm for O₂ and CO₂, respectively) and rainwater (assumed here as pure water for simplicity), respectively; the lower boundary is assumed to be impermeable for diffusive flux. Soils are assumed to have the solid phase composition of an unaltered rock at depth: the lower boundary conditions for solid species are given as fixed concentrations, which are obtained from lithology data. The simulated soil column is 50 cm and divided into 30 layers, and the initial porosity for spin-up is assumed to be 0.5 regardless of location. Although a coarser discretization of the soil column may suffice, finer discretization is still desirable given that basalt may accumulate on topsoil when mixing is inefficient. Organic matter is applied continuously at the surface at a rate described above and mixed to a 25 cm depth via natural bioturbation (Fickian mixing: see Kanzaki et al. (2022) for detailed formulation) along with other solid species (listed above), whether basalt is additionally applied or not. When applied, basalt is assumed to have the same mineralogical composition as those adopted by Beerling et al. (2020) and Kanzaki et al. (2022) (Table S1 in Supporting Information S1), with 0.37 g CO₂ as the maximum capture capacity per 1 g basalt. For our “fine-grained” basalt feedstock, initial particle size distributions are defined as a function of particle radius, equivalent to the sum of 4 log-normal distributions centered at 1, 2, 5, and 20 μm, all with 0.2 log unit standard deviations (f_basl(r) ∝  exp(−(log r − log r_i)²/0.4), where f_basI(r) is the initial number of particles as a function of particle radius r in meters and r_i = 1 × 10⁻⁶, 2 × 10⁻⁶, 5 × 10⁻⁶, and 20 × 10⁻⁶ for i = 1, …, 4). The surface roughness is accounted for by using a radius function from Navarre-Sitchler and Brantley (2007) that is also adopted in Beerling et al. (2020) and Kanzaki et al. (2022). Resulting initial surface areas are, for example, 6.6 and 6.5 m²/g for anorthite and fayalite, respectively, of the basalt. The p80 value of our fine-grained basalt feedstock (the diameter for which 80% of the total basalt mass is made up from particles with a smaller diameter) is calculated to be 100 μm (Figure S3 in Supporting Information S1). Parameter values for our “coarse-grained” basalt are defined by the sum of 2 log normal distributions centered at 0.5 and 5 μm all with 0.5 log unit standard deviations, resulting in a p80 of 1,220 μm and, for example, 0.18 and 0.12 m²/g surface areas for anorthite and forsterite, respectively. When applied, the basalt is added at the surface and mixed to 25 cm depth by tilling only during the first 0.1 fraction of each year. During the rest of the year soil is left to be mixed to the same depth by natural Fickian bioturbation together with other solid species. Temperature, soil moisture and net water flux are assumed to be depth-independent and are fixed for long-term spin-up but are varied monthly for historical runs and basalt application experiments where the above climate factors are forced by outputs from CESM1.

2.3 Carbon Capture Estimates

The addition of basalt to croplands increases the capacity of land to act as a net CO₂ sink through two primary pathways: (a) the conversion of atmospheric CO₂ to dissolved CO₂ species (mostly ) and the subsequent loss of dissolved CO₂ to rivers and oceans via water flux; and (b) the precipitation of carbonates (mostly calcite; CaCO₃) in soil. Beerling et al. (2020) calculated net CO₂ capture as 0.86 times the advective flux change of dissolved CO₂ species plus carbonate precipitation within soil, where the factor of 0.86 accounts for the return of CO₂ to the atmosphere after dissolved CO₂ species arrive at the oceans. For simplicity, we follow this principle and define CO₂ capture as net change in the CO₂ diffusive supply, subtracting out changes in CO₂ supply from soil respiration (that are not caused by basalt addition) and atmospheric CO₂ returned from the ocean (0.14 times the advective flux change of dissolved CO₂ species). That is, total CO₂ capture is equivalent to the change in diffusive CO₂ supply minus the change in CO₂ supply from soil respiration plus 0.14 times the change in advective flux of dissolved CO₂ species. However, we note that this is done primarily for comparative purposes, and that the amount of downstream carbon leakage in river/stream systems and from the coastal ocean subsequent to a given amount of initial CDR through ERW may vary from this (Kanzaki et al., 2023).

2.4 Extrapolation to Grid Points Below the 70% Cropland Threshold

The agricultural land area identified with 70%+ cropland grid points represents only a fraction of total global agricultural land area. In order to scale our results for 70%+ cropland area to overall global cropland we use a nearest neighbor interpolation method in which the total carbon capture estimate of the nearest 70%+ cropland grid point is used to fill in cells below the 70% cropland threshold. We then weight these values by the percent cropland of each cell to derive a total cropland ERW estimate.

3 Results

We identify cropland sites (defined as where cropland proportion is at least 70% of land cover; see Methods) in every continent except Antarctica, with high density of cropland over the American Midwest and southern regions of Canada's Prairie Provinces (North America); Argentina and Brazil (South America); Western Africa and Ethiopia (Africa); eastern Europe and Russia (Europe); India, Thailand, and eastern China (Asia); and southwestern Australia, Victoria, and southern New South Wales (Australia). These cropland sites total 929 grid points, which collectively equal 7,407,423 km² of cropland. Total cropland area across all cells is equivalent to 24,694,492 km².

Using SCEPTER, we apply a fixed rate of 10 tons of basalt per hectare annually under two RCP scenarios (Figure 1) and explore two grain size distributions (here referred to as “fine,” for which the grain size at which 80% of the material passes sieving (p80) is 100 μm, and “coarse,” for which p80 = 1,220 μm). This application rate is within the bounds of current limestone soil amendments in many regions (Galinato et al., 2011; Lukin & Epplin, 2003; T. O. West & McBride, 2005) and the relative grain sizes are comparable to waste fines from aggregate production (Adilson et al., 2016; Bullock et al., 2021). Figure 2 shows the total amount of CO₂ captured under both the RCP4.5 and RCP8.5 scenarios, with total CO₂ captured defined as the net change in CO₂ diffusive supply accounting for (a) CO₂ from soil respiration independent of basalt added and (b) an assumed advective CO₂ returned to the atmosphere from the ocean (Figure S1 in Supporting Information S1; see Methods). ERW simulations under both global warming scenarios show considerable heterogeneity in the rate of CO₂ captured, with regional carbon uptake differences far outweighing differences across the RCP4.5 and RCP8.5 scenarios. Most cropland regions over North America (concentrated around the “corn belt” in the American Midwest), South America, and Europe/Russia, for instance, show carbon capture rates of ∼1 ton of CO₂ per hectare per year (tCO₂ ha⁻¹ year⁻¹) in both global warming scenarios, while carbon capture rates over Australia are less than ∼0.6 tCO₂ ha⁻¹ year⁻¹. Cropland grid points over Africa, Thailand, eastern China, and India show higher carbon capture rates of ∼1.2–1.6 tCO₂ ha⁻¹ year⁻¹.

Figure 1

(Top) The location of cropland sites across the globe where SCEPTER was successfully applied (n = 929). The total cropland area accounted for by these grid points is provided on the lower left corner (weighted by the percentage of agricultural land). (Bottom) Percentages of croplands across the globe (including both above and below the 70% agricultural threshold). The total cropland area is provided on the lower left corner (weighted by the percentage of agricultural land).

Figure 2

(Top) Total enhanced rock weathering (ERW) under the (left) RCP4.5 and (right) RCP8.5 scenarios. Simulations are performed with a spatially uniform basalt application rate of 10 tons per hectare annually with a uniform fine basalt grain size (p80 = 100 μm). (Bottom) Additional total ERW under the RCP8.5 scenario relative to the RCP4.5 scenario, expressed as a percent increase.

Despite relatively minor impacts from climate change, ERW on croplands depends strongly on regional climatic conditions. Croplands in tropical latitudes and India, for instance, are hotspots of carbon capture over the 2006–2080 interval (Figure 3), likely because they receive significant amounts of solar radiation due to their proximity to the equator and ample precipitation from the Intertropical Convergence Zone and the Indian Monsoon, respectively. Notably, there is significant spatial correlation between total carbon capture estimates for the simulated sites (as shown in Figure 2) and climatological surface temperature over the 2006–2080 interval under the RCP8.5 scenario (r = 0.51, p < 0.01) (r = 0.48, p < 0.01 for RCP4.5). Similarly, there is robust spatial correlation between total carbon capture estimates and climatological total liquid runoff (r = 0.57, p < 0.01) (same for RCP4.5). Climatological soil moisture values do not correlate as strongly in space with carbon capture fluxes over the 2006–2080 interval (r = 0.25, p < 0.01) (same for RCP4.5). Our results are nevertheless consistent with expectations that hot and wet environments will favor rock/mineral dissolution (e.g., Beerling et al., 2020; Goll et al., 2021; Taylor et al., 2016).

Figure 3

Total CO₂ captured over the 2006–2080 interval under the RCP8.5 scenario. Simulations are performed with a spatially uniform basalt application rate of 10 tons per hectare annually with a fine (p80 = 100 μm) uniform basalt grain size. Carbon capture rates for grid points where cropland is 70% or less of total land area are estimated using the rate of the nearest 70%+ cropland. Total carbon captured over each grid point is weighted by the cropland percentage of that grid point.

We next calculate the total amount of CO₂ captured over the 2006–2080 interval over the simulated cropland cells by region (see Figure 2 for demarcations), weighting the rate of CO₂ capture by cropland percentage (Table 1; note Figure 2 shows rates without cropland weighting). ERW on the 70%+ cropland cells can capture 63.4 and 64.5 gigatons of CO₂ (GtCO₂) under RCP4.5 and RCP8.5 scenarios, respectively, which translates to a global average area-normalized CDR rate of 1.14 tCO₂ ha⁻¹ year⁻¹ and 1.16 tCO₂ ha⁻¹ year⁻¹. Consistent with the above results, this analysis highlights the importance of regional heterogeneity in ERW potential. In our simulations, India alone, with its large cropland area and high ERW rates, accounts for 32% of total carbon captured globally. Despite lower ERW rates, Europe/Russia have a disproportionate amount of cropland area and thus account for another 24%–25%. Lastly, Africa accounts for 8%–9% of total carbon captured despite only accounting for a minor portion of total global cropland area due to very high ERW rates.

Table 1. (Top) Total Amount of CO₂ Captured Across 70%+ Croplands by Country/Continent (See Figure 2 for Demarcation of Regions) Over the 2006–2080 Interval Assuming Fine Basalt Grain Size (p80 = 100 μm) and a Spatially Uniform Annual Basalt Application Rate of 10 Tons per Hectare Under the (Left) RCP4.5 and (Right) RCP8.5 Scenarios

70%+ croplands only	RCP4.5 total over 2006–2080 (Gton)	RCP8.5 total over 2006–2080 (Gton)
N. America	5.77	5.81
Europe/Russia	15.77	15.73
S. America	2.99	3.07
Africa	5.29	5.59
Australia	1.25	1.23
India	20.15	20.75
Thailand	2.36	2.39
China	8.12	8.18
Other	1.72	1.77
Total	63.4	64.51

All croplands	RCP4.5 total over 2006–2080 (Gton)	RCP8.5 total over 2006–2080 (Gton)
N. America	23.35	23.26
Europe	36.89	36.73
S. America	29.02	28.93
Africa	39.6	40.63
Oceania	3.95	3.92
Asia	81.52	82.85
Global	214.81	216.79

• Note. (Bottom) Same as (top) but for CO₂ captured across all croplands (estimated using nearest neighbor carbon capture rates of grid points meeting the 70% cropland threshold) by continent over the 2006–2080 interval.

Agricultural land area identified with our 70%+ cropland threshold represents only a fraction of the total global agricultural land area. We therefore attempt to provide a first-order estimate for global carbon capture across all agricultural lands globally using nearest neighbor interpolation (Table 1; Figure 3). Globally, we estimate ERW on croplands can capture as much as ∼215 GtCO₂ over the 2006–2080 interval (Table 1), with substantial regional heterogeneity in capture potential. Africa and Asia alone, for instance, demonstrate ERW carbon capture potentials of up to 40.6 and 82.9 GtCO₂, respectively, over the simulation period. Oceania in contrast shows minimal ERW potential (a total of ∼4 GtCO₂). We note that this global potential of ∼215 GtCO₂ sequestered over the 2006–2080 interval is well above the lower end of the range stipulated by IPCC mitigation pathways to limit global warming to 1.5°C (i.e., 100–1,000 gigatons by 2100). Though a rough approximation, our estimates thus support the notion that ERW on global croplands has the potential to play an integral role in any comprehensive negative emissions strategy, especially when considering that feedstock processing and application rates have not been optimized in our simulations.

Finally, we characterize the sensitivity of ERW to basalt grain size. To do this, we use SCEPTER and apply a fixed rate of 10 tons of coarse-grained basalt sand (p80 = 1,220 μm) per hectare annually (Figure 4). As with fine-grained basalt application, additional global warming from the RCP8.5 scenario relative to RCP4.5 scenario has only a modest influence on ERW with coarse-grained basalt application: global carbon capture rates with coarse-grained basalt application are 0.81 tCO₂ ha⁻¹ year⁻¹ and 0.83 tCO₂ ha⁻¹ year⁻¹, under the RCP4.5 and RCP8.5 scenarios, respectively. However, overall carbon capture is, on average, ∼70% of that achieved in our “fine-grained” simulation even after decades of weathering.

Figure 4

(Top) Total enhanced rock weathering (ERW) under the RCP8.5 scenario for (left) fine uniform basalt grain size (p80 = 100 μm) and (right) coarse uniform basalt grain size (p80 = 1,220 μm). All simulations are performed with a spatially uniform basalt application rate of 10 tons per hectare annually. (bottom) Difference in total ERW across the two experiments.

4 Discussion

Using SCEPTER and a series of climate model experiments, we have simulated ERW by applying 10 tons of fine-grained basalt sand and silt per hectare per year on 963 agricultural sites across the globe. These 963 sites alone can sequester ∼64 GtCO₂ over the 2006–2080 interval. From these results, we extrapolate to first-order that ERW can sequester ∼215 GtCO₂ over the same interval when applied to all cropland locations worldwide. Collectively, our results provide strong support for the notion that ERW on global croplands can foster CO₂ removal at gigaton scale.

Although there are large regional differences in weathering rates linked to local climates, future global warming has a minor effect on global average CDR rates (∼2% increase with a high emissions scenario). This contrasts with soil organic carbon storage, which becomes significantly less stable under higher temperatures and may become a carbon source with continued global warming (Crowther et al., 2016; Davidson & Janssens, 2006; Nottingham et al., 2019; Zhao et al., 2021). Despite small influences from climate change, ERW is nevertheless sensitive to regional climate variability. Notably, ERW is more efficient—for example, more basalt is weathered over a given time interval—in hot and wet environments concentrated around the tropics currently coinciding with economically developing regions (Figures 2 and 4). Ramp-up speed for ERW after initial basalt application is in general faster in regions of relatively low per-capita income (Figure S2 in Supporting Information S1), with shorter time to reach full-capacity sequestration rates. This supports the notion that there is strong potential for ERW in developing economies, even though precision agriculture is uncommon and the infrastructure needed for large-scale basalt deployment is not currently in place. Therefore, our results provide strong impetus for investment in agricultural reform in developing economies, which would represent an additional co-benefit of ERW for regional stakeholders.

Finally, we find that basalt grain size is an important parameter for CDR effectiveness and unit cost of CDR deployment ($/ton of CO₂ captured), with coarser grained basalt feedstock achieving substantially less cumulative CDR than fine-grained basalt feedstock. With coarser grained basalt, large portions of the applied basalt are likely to remain unweathered even after decades within a soil. The potential for persistence of significant amounts of basalt feedstock in a soil column highlights one potential problem with pursuit of ex ante carbon credit schemes prior to extensive data-model comparison. Similarly, these results highlight the pressing need for feedstock grain sizes and application rates to be optimized for local conditions in order to promote both greater CO₂ capture efficiency and more favorable techno-economics in the global North than the global South. Although our estimate of the theoretical potential of enhanced weathering in croplands is very likely conservative, our work nevertheless provides additional support (see Beerling et al., 2020) for the assertion that ERW represents a scalable, resilient carbon capture strategy that is non-competitive for arable land. Overall, this bolsters the case that there is tremendous potential, from a geochemical point of view, for enhanced weathering to foster carbon dioxide removal at scale. Therefore, the geochemical weathering potential (which is at the gigaton-scale) and hard infrastructure (e.g., farm equipment needed for soil amendments) are less likely to limit the scaling of ERW than gaps in the structure of carbon marketplaces and the nascent state of standards for monitoring, reporting, and verification.

Acknowledgments

This work was supported by the Department of Earth and Planetary Sciences of Yale University. SHB, JML, and NP would like to acknowledge support from the Yale Center for Research Computing (YCRC). YK and CTR would like to acknowledge support from the Partnership for an Advanced Computing Environment (PACE) at the Georgia Institute of Technology. SZ would like to thank the support from the Yale Center for Research Computing.

Open Research

Data Availability Statement

All CESM1 model data are publicly available through the Casper cluster (/glade/campaign/cesm/collections/cesmLME/CESM-CAM5-LME/) or through the Climate Data Gateway (https://www.cesm.ucar.edu/community-projects/lme/data-sets). The version of the SCEPTER code used in this study is available at (Kanzaki, 2022).