Orbiting Carbon Observatory-2 (OCO-2) Cloud Screening; Validation Against Collocated MODIS and Initial Comparison to CALIOP Data

The retrieval of the column-averaged carbon dioxide (CO2) dry air mole fraction (XCO2 ) from satellite measurements of reflected sunlight in the near-infrared can be biased due to contamination by clouds and aerosols within the instrument's field of view (FOV). Therefore, accurate aerosol and cloud screening of soundings is required prior to their use in the computationally expensive XCO2 retrieval algorithm. Robust cloud screening methods have been an important focus of the retrieval algorithm team for the National Aeronautics and Space Administration (NASA) Orbiting Carbon Observatory-2 (OCO-2), which was successfully launched into orbit on July 2, 2014. Two distinct spectrally-based algorithms have been developed for the purpose of cloud clearing OCO-2 soundings. The A-Band Preprocessor (ABP) performs a retrieval of surface pressure using measurements in the 0.76 micron O2 A-band to distinguish changes in the expected photon path length. The Iterative Maximum A-Posteriori (IMAP) Differential Optical Absorption Spectroscopy (DOAS) (IDP) algorithm is a non- scattering routine that operates on the O2 A-band as well as two CO2 absorption bands at 1.6 m (weak CO2 band) and 2.0 m (strong CO2 band) to provide band-dependent estimates of CO2 and H2O. Spectral ratios of retrieved CO2 and H2O identify measurements contaminated with cloud and scattering aerosols. Information from the two preprocessors is feed into a sounding selection tool to strategically down select from the order one million daily soundings collected by OCO-2 to a manageable number (order 10 to 20%) to be processed by the OCO-2 L2 XCO2 retrieval algorithm. Regional biases or errors in the selection of clear-sky soundings will introduce errors in the final retrieved XCO2 values, ultimately yielding errors in the flux inversion models used to determine global sources and sinks of CO2. In this work collocated measurements from NASA's Moderate Resolution Imaging Spectrometer (MODIS), aboard the Aqua platform, and the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), aboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite, are used as a reference to access the accuracy and strengths and weaknesses of the OCO-2 screening algorithms. The combination of the ABP and IDP algorithms is shown to provide very robust and complimentary cloud filtering as compared to the results from MODIS and CALIOP. With idealized algorithm tuning to allow throughputs of 20-25%, correct classification of scenes, i.e., accuracies, are found to be ' 80-90% over several orbit repeat cycles in both the win ter and spring time for the three main viewing configurations of OCO-2; nadir-land, glint-land and glint-water. Investigation unveiled no major spatial or temporal dependencies, although slight differences in the seasonal data sets do exist and classification tends to be more problematic with increasing solar zenith angle and when surfaces are covered in snow and ice. An in depth analysis on both a simulated data set and real OCO-2 measurements against CALIOP highlight the strength of the ABP in identifying high, thin clouds while it often misses clouds near the surface even when the optical thickness is greater than 1. Fortunately, by combining the ABP with the IDP, the number of thick low clouds passing the preprocessors is partially mitigated.