Explicit Physical Knowledge in Machine Learning for Ocean Carbon Flux Reconstruction: The pCO<sub>2</sub>‐Residual Method
Val Bennington, Tomislav Galjanic, Galen A. McKinley
Abstract
Abstract The ocean reduces human impacts on global climate by absorbing and sequestering CO 2 from the atmosphere. To quantify global, time‐resolved air‐sea CO 2 fluxes, surface ocean pCO 2 is needed. A common approach for estimating full‐coverage pCO 2 is to train a machine learning algorithm on sparse in situ pCO 2 data and associated physical and biogeochemical observations. Though these associated variables have understood relationships to pCO 2 , it is often unclear how they drive pCO 2 outputs. Here, we make two advances that enhance connections between physical understanding and reconstructed pCO 2 . First, we apply pre‐processing to the pCO 2 data to remove the direct effect of temperature. This enhances the biogeochemical/physical component of pCO 2 in the target variable and reduces the complexity that the machine learning must disentangle. Second, we demonstrate that the resulting algorithm has physically understandable connections between input data and the output biogeochemical/physical component of pCO 2 . The final pCO 2 reconstruction agrees modestly better with independent data than most other approaches. Uncertainties in the reconstructed pCO 2 and impacts on the estimated CO 2 fluxes are quantified. Uncertainty in piston velocity drives substantial flux uncertainties in some regions, but does not increase globally integrated estimates of uncertainty in CO 2 fluxes from observation‐based products. Our reconstructed CO 2 fluxes show larger interannual variability than smoother neural network approaches, but a lesser trend since 2005. We estimate an air‐sea flux of −1.8 PgC/yr (anthropogenic flux of −2.3 ± 0.5 PgC/yr) for 1990–2019, agreeing with other data products and the Global Carbon Budget 2020 (−2.3 ± 0.4 PgC/yr).