Improving OCO-2 XCO2 Retrieval at Critical Albedo with Neural Network A-priori Constraints of Aerosol

Sihe Chen,  California Institute of Technology,  sihechen@caltech.edu (Presenter)
Zhao-Cheng Zeng,  University of California, Los Angeles,  zcz@gps.caltech.edu
Vijay Natraj,  Jet Propulsion Laboratory,  vijay.natraj@jpl.nasa.gov
Yuk-Ling yung,  California Institute of Technology,  yly@gps.caltech.edu

For any particular wavelength and aerosol type, the critical albedo is defined as the surface albedo when the top of the atmosphere continuum radiance loses sensitivity to the aerosol loading. Accurately retrieving aerosol information from passive remote sensing measurements over surfaces with reflectance near the critical albedo has remained a challenge. The Orbiting Carbon Observatory-2 (OCO-2) measures the column-averaged dry-air mole fraction of CO2 (XCO2) from space using three near-infrared bands. However, the retrieved aerosol optical depth (AOD) from the OCO-2 full physics retrieval algorithm correlates poorly with data from ground-based AERONET stations. Failing to correctly characterize atmospheric aerosols is known to be a major source of error in OCO-2 retrievals (Zhang et al., 2016). In addition, the globally fitted bias correction approach used by OCO-2 L2 LITE products may not be able to fix the particular issue of critical albedo.
Dust aerosol over desert constitutes a commonly encountered example of the critical albedo issue. The OCO-2 retrieval of XCO2 over deserts tends to wrongly interpret spikes in AOD as XCO2 plumes. In this study, we use AOD and aerosol layer height (ALH) information derived from Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) as the truth. We use a spectral sorting approach (Zeng et al., 2020) to train a two-step neural network that retrieves both AOD and ALH and their uncertainties. The neural network prediction shows a high correlation with CALIPSO measurements (R2 for AOD and ALH are 0.47 and 0.56). This information is used as a priori for retrieving XCO2, in order to correct for the bias in the Level-2 standard retrieval results. In addition, we use synthetic simulations to show that the correction with an improved a priori is physically realistic. Our results show that this correction generates significantly different results than the globally fitted bias correction in OCO-2 L2 LITE products. Specifically, the LITE products have a difference of up to 2 ppm from the retrieval with improved aerosol characterization for the scenarios studied over Riyadh. The proposed technique could have important implications for flux inversions.

Poster: Poster_Chen__90_25.pdf 

Presentation Type: Poster

Session: 2.5a Retrieval algorithms and methods for inter-instrument and product Cal/Val

Session Date: Tuesday (6/15) 12:00 PM