Abstract
The Orbiting Carbon Observatory -2 and -3 instruments have been measuring reflected sunlight in the near infrared from low Earth orbit since 2014 and 2019 respectively. A three-channel spectrometer with common entrance optics measures narrow spectral bands centered at 765 nm for oxygen and 1608 and 2065 nm for carbon dioxide. To achieve spectral resolving power exceeding 18000:1, a diffraction grating disperses each spectrum over 1016 columns of the focal plane array. In the spatial dimension, 160 central rows are averaged into eight footprints. For accurate and precise retrievals of carbon dioxide concentration, the calibration process needs to correct for spurious features from the instrument and test sources. The cause and nature of these artifacts varies considerably, including patterns intrinsic to the detector and readout electronics, the signatures of various optical components, and contamination. Most features are accounted for by deriving independent calibration coefficients for each spectral sample, but even a single pixel can disrupt science. A machine learning classifier has been deployed to identify bad pixels in dark and lamp data that are removed in flight software. Additionally, calibrated Earth spectra are checked for outliers that can be ignored by the retrieval before retrospective products are released. These techniques are also relevant to future missions that are planning to incorporate larger format detectors.
Nonuniform Calibration Artifacts in OCO-2 & OCO-3
The Orbiting Carbon Observatory -2 and -3 instruments have been measuring reflected sunlight in the near infrared from low Earth orbit since 2014 and 2019 respectively. A three-channel spectrometer with common entrance optics measures narrow spectral bands centered at 765 nm for oxygen and 1608 and 2065 nm for carbon dioxide. To achieve spectral resolving power exceeding 18000:1, a diffraction grating disperses each spectrum over 1016 columns of the focal plane array. In the spatial dimension, 160 central rows are averaged into eight footprints. For accurate and precise retrievals of carbon dioxide concentration, the calibration process needs to correct for spurious features from the instrument and test sources. The cause and nature of these artifacts varies considerably, including patterns intrinsic to the detector and readout electronics, the signatures of various optical components, and contamination. Most features are accounted for by deriving independent calibration coefficients for each spectral sample, but even a single pixel can disrupt science. A machine learning classifier has been deployed to identify bad pixels in dark and lamp data that are removed in flight software. Additionally, calibrated Earth spectra are checked for outliers that can be ignored by the retrieval before retrospective products are released. These techniques are also relevant to future missions that are planning to incorporate larger format detectors.