OCO-2, launched in 2014, uses reflected solar spectra and other retrieved geophysical variables to estimate (“retrieve”) the column averaged dry air mole fraction of CO2, termed XCO2. A critical issue in satellite estimates of trace greenhouse gasses and remote sensing at large is the error distribution of an estimated target variable which arises from instrument artifacts as well as the under-determined nature of the retrieval of the quantities of interest. A large portion of the error is often incurred during inference from measurement of retrieved physical variables. These residual errors are typically corrected using ground truth observations of the target variable or some other truth proxy. Previous studies used multilinear regression to model the error distribution with a few covariates from the retrieved state vector, sometimes termed “features.” This presentation will cover the bias correction of XCO2 error attributed to retrieved covariates with a novel approach utilizing explainable Machine Learning methods (XAI) on simulated sounding retrievals from GeoCarb. Utilization of non-linear models (Zhou, Grassotti 2020) or models that can capture non-linearity implicitly (Lorente et al. 2021) have been shown to improve on linear methods in operation. Our approach uses a gradient boosted decision tree ensemble method, XGBoost, that captures non-linear relations between input features and the target variable. XGBoost also incorporates regularization to prevent overfitting, while also remaining resilient to noise and large outliers – a feature missing from other ensemble DT methods. Decision Tree based models provide inherent feature importance that allows for high interpretability. We also approach post training analysis with model agnostic, explainable methods (XAI). XAI methods allow for rigorous insight into the causes of a model’s decision (Gilpin et al. 2018). By applying these techniques, we will demonstrate our approach provides reduced residual errors relative to the operational method as well as yielding an uncertainty estimate in bias corrected XCO2, which is currently not treated separately from the posterior uncertainty estimate derived from the retrieval algorithm.

The Atmospheric Carbon and Transport (ACT) – America NASA Earth Venture Suborbital Mission set out to improve regional atmospheric greenhouse gas (GHG) inversions by exploring the intersection of the strong GHG fluxes and vigorous atmospheric transport that occurs within the midlatitudes. Two research aircraft instrumented with remote and in situ sensors to measure GHG mole fractions, associated trace gases, and atmospheric state variables collected 1140.7 flight hours of research data, distributed across 305 individual aircraft sorties, coordinated within 121 research flight days, and spanning five, six-week seasonal flight campaigns in the central and eastern United States. Flights sampled 31 synoptic sequences, including fair weather and frontal conditions, at altitudes ranging from the atmospheric boundary layer to the upper free troposphere. The observations were complemented with global and regional GHG flux and transport model ensembles. We found that midlatitude weather systems contain large spatial gradients in GHG mole fractions, in patterns that were consistent as a function of season and altitude. We attribute these patterns to a combination of regional terrestrial fluxes and inflow from the continental boundaries. These observations, when segregated according to altitude and air mass, provide a variety of quantitative insights into the realism of regional CO2 and CH4 fluxes and atmospheric GHG transport realizations. The ACT-America data set and ensemble modeling methods provide benchmarks for the development of atmospheric inversion systems. As global and regional atmospheric inversions incorporate ACT-America’s findings and methods, we anticipate these systems will produce increasingly accurate and precise sub-continental GHG flux estimates.