On error, uncertainty, and assumptions in calculating carbon dioxide removal rates by enhanced rock weathering in Kantola et al., 2023

Abstract

Kantola et al. (2023) present one of the most detailed studies of carbon fluxes in an enhanced weathering (EW) field trial to date—a necessary step toward moving forward this promising carbon dioxide removal (CDR) approach. However, we have several concerns about the method—developed largely by the research team of the for-profit EW supplier Eion Carbon—used to quantify CDR from EW in this study. Our hope is that highlighting these concerns will help facilitate progress in tracking carbon fluxes in enhanced weathering deployments. Soil-based mass balance approaches have a long history of providing key insights into weathering rates (e.g., Brimhall & Dietrich, 1987; Chadwick et al., 1990). We fully support the development of new soil-based mass balance approaches to track EW and indeed are actively pursuing research on the topic (Beerling et al., 2023; Reershemius et al., 2023). Given the large variety in soil and feedstock compositions, it is likely that no single method will be applicable or feasible in every setting. Hence, we view multiple approaches as complementary rather than competitive. Unfortunately, flaws of the method developed in Kantola et al. (2023) render it unusable without modifications. The authors use a linear regression between ∆[REE]soil and [REE]basalt (Figure 1a) to calculate a sample-averaged basalt application rate and from this a CDR rate. This method is based on a series of flawed assumptions. First, the approach requires that under any amount of basalt dissolution, accumulated REEs from the basalt will result in positive ∆[REE]soil. Second, the approach requires that the accumulation of all REEs in soil must be proportional to [REE]basalt for any specific REE, regardless of the value of the ratio ([REE]basalt:[REE]soil). Both assumptions are, for the dataset presented in Kantola et al. (2023), demonstrably incorrect. For some of the REEs used, [REE]basalt < [REE]soil. In these cases, mixing of basalt into the soil results in the dilution of the REE in the soil + basalt mixture—that is, [REE]soil+basalt < [REE]soil at the point of mixing (Figure 1c). After this, [REE]soil+basalt will increase as feedstock dissolves, but REEs are (assumed to) remain in the soil (Figure 1d). Thus, the method should account for an enrichment in [REE] with fdissolution. Positive Δ[REE]soil(2020–2016) values for elements with [REE]basalt < [REE]soil cannot be explained without this enrichment effect. Additionally, [REE]basalt:[REE]soil varies by REE. As a result, for elements where [REE]basalt > [REE]soil, ∆[REE]soil is not proportional to [REE]basalt as assumed by Kantola et al. (2023), but should be treated as a function of [REE]basalt, [REE]soil, massbasalt, and masssoil. Given these conditions, basalt addition alone cannot result in the strong positive correlation that the authors observe between [REE]basalt and ∆[REE] (Figure 1c). Crucially, the regression slope and correlation are dependent on feedstock dissolution (Figure 1d), which the framework presented in Kantola et al. (2023) does not account for. As such the derived application rate is one of multiple possible solutions from the presented REE data (see also 2 and error bars in Figure 1a). Samples from maize/soybean control plots show negative Δ[REE] for all elements (Figure 1b). Fitting a regression to these data and following the method Kantola et al. (2023) apply to the basalt-treated plots yields a basalt mass loss of 2.3–33.5 kg m−2 (23–335 t ha−1) between 2016 and 2020, depending on REE and depth selection (see also 3). Thus, control plots where no basalt was applied yield a stronger signal for mass loss than the treated plots do for mass gain. The authors could correct for the Δ[REE] in control plots, which would dramatically increase the calculated basalt application and thus CDR rate (Figure 1e–h). Alternatively, the authors could assume that the signal in control plots results from natural variability in field scale [REE], which then should be propagated into uncertainty on their calculated CDR rates. In any case, the control plots clearly yield a [REE] signal for unphysical mass loss that is ostensibly similar in magnitude to the signal purportedly from basalt addition in the treated plots. This is a major issue for the presented CDR estimates that is not discussed or addressed in Kantola et al. (2023). The estimates for basalt application presented in Kantola et al. (2023) are primarily controlled by their choice of which REEs and soil sample depth intervals to analyze. The authors state that to overcome soil variability issues, they use only HREEs (Tb-Lu) in their regression analysis for Miscanthus plots, which results in a better correlation for [REE]basalt and ∆[REE]soil. It is unclear why the authors choose a different set of REEs for the maize/soybean system. Depending on which REEs are considered, a wide range of initial feedstock applications can be calculated using the authors' method (Figure 1e–h). Furthermore, applying the authors' method using only samples from 0 to 10 cm, rather than weighting 0–10 cm:10–30 cm samples by 1:2 as is done in Kantola et al. (2023), the basalt application estimate from the authors' REE approach is significantly lower than that reported in the paper (Figure 1e–h). This should not be the case if, as the authors assume, REEs are not mobilized from the soil. In fact, the data from this study show a decoupling between the presence of basalt in the mixture as evidenced by positive ∆[Mg, Ca] and ∆[REE], possibly through dissolution of REE-bearing phases and reprecipitation further down the soil profile, or background variability in the REE system that is not taken into account (see also 2). Given the wide range of possible derived feedstock application and CDR rates depending on parameter choice, the lack of consistency and criteria for the selection of data for analysis is troubling. Assuming an application rate is not necessarily problematic—at a field scale this is actually a known variable. However, such a cation-only method relies on soil sampling being truly representative of field scale, which Kantola et al. (2023) do not rigorously test or demonstrate. In summary, the approach to estimate initial CDR rates from in situ basalt weathering presented by Kantola et al. (2023) has some likely problematic, unjustified, and erroneous assumptions regarding the effect of basalt and soil mixing and basalt dissolution on REE concentrations, and misinterpretation of depth- and time-dependent signals of major element and REE concentration change. There are also seemingly arbitrary decisions on data selection for analysis and disregard of controls. These must all be robustly addressed before this method is used in other academic studies or as a key part of monitoring, reporting and verification (MRV) toolsets for compensatory offset claims. Nonetheless, as the authors' estimates of sample-averaged basalt application rates are 20.2 and 14.5 kg/m2, respectively, for maize/soybean and Miscanthus, the dissolution rate calculated from the concentration of Mg and Ca in samples is not dissimilar to that which would be calculated given the a priori assumption of basalt application rates of 20 and 15 kg/m2. Thus, the flaws in the method for quantifying EW do not necessarily detract from the overall carbon budget presented in this study. However, it should be noted that the key question in cation-only soil-based mass balance approaches for tracking weathering rates is whether the sampling employed gives a representative look at cation concentrations at field scale, which is not a focus of this study. The deficiencies in the methodology notwithstanding, it is commendable that the authors make all of their data openly available. Such data sharing practices allow rigorous scrutiny of methods and are strongly encouraged in order to allow iterative development of a suite of robust EW monitoring tools that can be widely applied in MRV protocols. Tom Reershemius: Conceptualization; data curation; formal analysis; investigation; methodology; validation; visualization; writing – original draft; writing – review and editing. T. Jesper Suhrhoff: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; validation; visualization; writing – original draft; writing – review and editing. The authors declare that they have no conflict of interest. Data sharing not applicable–no new data generated. Calculations based on data from original publication will be shared upon request.

Topics & Concepts