Accounting for uncertainty in carbon fluxes: Towards the integration of Bayesian approaches in enhanced weathering

Quantifying carbon dynamics during enhanced weathering has been challenging, given spatial and temporal soil heterogeneity, complex biogeochemical interactions, and limitations in measurement resolution. For example, intensive and repeated sampling campaigns over the duration of several years are required to detect changes in soil organic carbon (SOC) stocks given that a comparatively small shift needs to be detected in a large pool with high spatial heterogeneity. Thus, low signal-to-noise ratios often prevent the detection of shifts, as EW induced changes in SOC stocks are often below the natural variability of SOC. Given the high uncertainty associated with both, SOC sequestration estimates but also CO₂ removal estimates, the integration of Bayesian approaches into modelling efforts could be particularly valuable. Bayesian modelling frameworks offer a powerful tool, explicitly integrating experimental observations with mechanistic understanding and prior knowledge, while quantifying uncertainty across all model components, thereby also accounting for soil heterogeneity.

In our presentation we will illustrate the advantages of Bayesian analyses via a model developed for soil CO₂ efflux partitioning (in rhizosphere respiration and SOM decomposition). Soil CO₂ flux partitioning is an important tool to inform EW effects on organic C dynamics that requires a sequence of calculations and the combination of different data sources. Several sources of uncertainty arise which often remain unaccounted for in conventional partitioning approaches (e.g. the regression used during the determination of the isotopic signature of the soil CO₂ efflux extrapolates far beyond measured data, isotopic fractionation due to physical and/or biological processes, CO₂ originating from soil carbonates, CO₂ removal due to enhanced weathering, etc.). A Bayesian model allows the integration of such diverse datasets with different structures flexibly while explicitly accounting for variability in the data and for sources of uncertainty. Given its probabilistic framework, outputs are expressed as probability distribution rather than point estimates, therefore yielding far more informative results.

Further developing this model, potential applications could include the joint assessment of organic C sequestration and inorganic CO₂ removal. Thus, building on this example, we will discuss how Bayesian approaches could be further developed to support monitoring, reporting and verification (MRV) efforts for enhanced weathering.