Marina Fennell Dissertation Defense
Title: Development and Application of QUODcarb: A Bayesian solver for over-determined datasets of seawater carbon dioxide system chemistry
Abstract: Accurately quantifying the ocean's uptake of anthropogenic carbon requires a comprehensive understanding of seawater carbon dioxide (CO2) system chemistry. Over-determined datasets (containing three or more measurements) often reveal persistent inconsistencies between thermodynamic model-calculated and measured values (e.g., pH, AT, CT, pCO2). This dissertation introduces QUODcarb, a novel Bayesian CO2-system solver designed to resolve such inconsistencies.
QUODcarb produces a probability distribution for the true CO2-system state of a water parcel by combining prior thermodynamic information describing the acid-base chemistry of CO2 in seawater with measured parameters and their uncertainties. A key finding is that QUODcarb achieves the GOA-ON 1% uncertainty target for carbonate ion concentrations using the GOMECC-3 dataset–unattainable with exactly-determined solvers. Furthermore, over 98% of samples yield state estimates consistent with measurement uncertainties, a significant improvement over prior methods.
We highlight the novel nuance that both measurement and dissociation constant uncertainties influence posterior parameter and uncertainty values. Notably, reducing dissociation constant uncertainties had a greater impact on posterior accuracy than reducing measurement uncertainties. The posterior uncertainties, combined in quadrature with the measurement uncertainties, are representative of a normal distribution for pH and pCO2 only, whereas the combined uncertainties for CT, AT, and [CO32-] are too small to sufficiently represent the measured-minus-calculated residuals. QUODcarb improves consistency when all parameters are input into the calculation, but the over-determined combinations that exclude the parameter of interest are not found to calculate the parameter with a residual that achieves the GOA-ON climate targets.
QUODcarb identifies systematic biases by leveraging the redundancy in over-determined calculations to make within-uncertainty adjustments to the calculated (posterior) values. These adjustments to the posterior values offer critical new insights to the inconsistencies observed in the marine CO2 system. QUODcarb represents a pivotal advancement for marine CO2-system studies and has far-reaching implications for global carbon cycle monitoring, model validation, and reconciling discrepancies in surface ocean pCO2 inventories.