Previous studies done for the TROPOMI mission (e.g. Meister et al (2017), Hu et al (2016), Landgraf et al (2016)) has shown that inhomogeneous illumination within the field of view causes distortions in the instrument spectral response function (ISRF). The differences in the distorted ISRF and the ISRF measured in the pre-flight calibration lead to retrieval biases. It is known through laboratory trials that these distortions increase with wavelength and various methods have been employed to mitigate these issues, including continuous within-slit scanning and temporal oversampling. Recently, a hardware solution, called a slit homogenizer, has shown promise in reducing distortions due to inhomogeneous illumination. In this work, we used an instrument model of the Geostationary Carbon Cycle Observatory, GeoCarb, to simulate observations in the 0.76um, 1.6um, 2.05um, and 2.3um spectral bands with varying illuminations. We found that the posterior scatter and bias of Greenhouse Gas retrievals grow linearly with the coefficient of variation of illumination within a footprint. We also found that the ISRF distortion effects of inhomogeneous illumination are significantly reduced through the use of a slit homogenizer, though not completely resolved.

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.

Enhanced CO2 mole fraction bands were often observed immediately ahead of cold front during the Atmospheric Carbon and Transport (ACT)-America mission and their formation mechanism is undetermined. Improved understanding and correct simulation of these CO2 bands are needed for unbiased inverse CO2 flux estimation. Such CO2 bands are hypothesized to be related to nighttime CO2 respiration and investigated in this study using WRF-VPRM, a weather-biosphere-online-coupled model, in which the biogenic fluxes are handled by the Vegetation Photosynthesis and Respiration Model (VPRM). While the default VPRM satisfactorily parameterizes gross ecosystem exchange, its treatment of terrestrial respiration as a linear function of temperature was inadequate as respiration is a nonlinear function of temperature and also depends on the amount of biomass and soil wetness. An improved ecosystem respiration parameterization including enhanced vegetation index, a water stress factor, and a quadratic temperature dependence is incorporated into WRF-VPRM and evaluated in a year-long simulation before applied to the investigation of the frontal CO2 band on 4 August 2016. The evaluation shows that the modified WRF-VPRM increases ecosystem respiration during the growing season, and improves model skill in reproducing nighttime near-surface CO2 peaks. A nested-domain WRF-VPRM simulation is able to capture the main characteristics of the 4 August CO2 band and informs its formation mechanism. Nighttime terrestrial respiration leads to accumulation of near-surface CO2 in the region. As the cold front carrying low-CO2 air moves southeastward, and strong photosynthesis depletes CO2 further southeast of the front, a CO2 band develops immediately ahead of the front.