OCO-2 v11.1 10-second average data, early release

The file archived here, "OCO2_b11.1_10sec_GOOD_r0.nc4", contains dry air column CO2 mixing ratio (XCO2) data from the Orbiting Carbon Obervatory (OCO-2) averaged over 10-second spans.  It is a preliminary version of files that will be used in the upcoming OCO-2 v11 flux inversion model intercomparison project (MIP), a study designed to quantify various error sources and analysis differences that lead to different surface CO2 flux estimates when the OCO-2 data are used in global CO2 flux inversions.  In particular, this file is the source of the OCO-2 XCO2 10-second average data used in the recently-completed study "An Error Model for Evaluating satellite-based XCO2 products" by Yadav et al.

The Orbiting Carbon Observatory (OCO-2) is a satellite that measures solar radiance reflected from the Earth's surface in two CO2 absorption bands (1.6 and 2.0 um), as well as in the O2 A-band.  By comparing the CO2 / O2 absorption ratio, the dry air mixing ratio of CO2 may be estimated along the observed path, i.e. averaged across the full atmospheric column, though with sensitivity peaking near the Earth's surface (where the impact of surface CO2 fluxes is the greatest).  A radiative transfer model that accounts for the scattering effects of thin clouds and aerosols, water vapor, surface albedo variations, is used, and the vertical profile of CO2 mixing ratio is estimated from the radiance data.  This CO2 profile is then collapsed to a scalar vertical average (XCO2), which is then bias corrected post-hoc against Earth-based Fourier spectrometer data from the Total Carbon Column Observation Network (TCCON), which itself is tied to CO2 measurement standards using in situ aircraft CO2 profile measurements.

OCO-2 takes measurements in a thin swath (up to 10-km wide) underneath the satellite, with a 3 Hz scan rate.  Each cross-scan is divided into 8 individual fields of view (FOVs) of size 2.25km x 1.25km, though the shape of the FOVs is distorted due to the pirouetting of the satellite to keep the sensor slit oriented perpendicular to the Sun-Earth-satellite plane.  These small FOVs increase the chances of seeing through clouds and can reveal details of point-source CO2 emissions, but are generally much finer-scale than can be modeled by the atmospheric transport models used in global CO2 flux inversions, which typically use grid boxes 100s of km on a side.  Rather than assimilating each fine-scale FOV XCO2 value individually and comparing them to modeled XCO2 values that change much more slowly, it is convenient to average the original OCO-2 data to coarser scales beforehand, then assimilating these averaged values in the flux inversions.  This averaging has the beneficial side-effect of reducing the OCO-2 data volume considerably.  In the file presented here, the data have been averaged across 10-second spans, equivalent to an along-track distance of ~67 km on the Earth's surface.

The source of the data averaged in the attached file is the data collection "OCO-2 Level 2 bias-corrected XCO2 and other select fields from the full-physics retrieval aggregated as daily files, Retrospective processing V11.1r (OCO2_L2_Lite_FP)", available at NASA's Goddard Earth Sciences (GES) Data and Information Services Center (DISC):  https://disc.gsfc.nasa.gov/datasets/OCO2_L2_Lite_FP_11.1r/summary?keywords=OCO2_L2_Lite_FP.   The data span for this preliminary version ('Release 0') of the 10-second average file is 20140906-20230430.  The original XCO2 values contained in this data collection, along with other auxiliary variables provided for analysis purposes, have been averaged in the same manner across each 10-second span, with the approach also used with the previous (Version 10) release of the OCO-2 XCO2 data, as described in Section 3.2.1 of Byrne et al. (2023) and Section 3.1.1 of Baker et al. (2022): each value in the span is weighted with the inverse square of its retrieved XCO2 uncertainty value, taken from variable 'xco2_uncertainty' from the v11.1r 'Lite" file.  Data inside each 10-second span are averaged separately based on viewing mode and surface type, as indicated by variable 'data_type'.  Only data that pass the retrieval quality flag (variable 'xco2_quality_flag' in the 'Lite' file equal to zero) are included in the average;  the number of such 'good' data values (1-240) included in each average value is indicated in variable 'N_total_shots'.  It has been found that 10-second spans with fewer 'good' retrievals tend to be affected more by the unwanted effects of aerosols and undetected clouds: these may be mitigated somewhat by not using 10-sec averages for spans with low 'N_total_shots' values.  Variable 'assimilate_flag' separates those data that are not taken in glint viewing mode over either land or water, or in nadir mode over land, from data taken in other modes (target mode, or in transition to/from target mode, or nadir mode over water, or mixed land/water scenes) that are generally not assimilated in flux inversions; it also flags 5% of the assimilable data by orbit for possible use as withheld evaluation data.  The uncertainty on the 10-sec average XCO2 value is given in variable 'xco2_uncertainty'; this accounts for correlations in error between individual scenes (+0.3 over land, +0.6 over ocean) as described in Section 3.2.1 of Baker et al (2022), as well as variability in the averaged XCO2 values not captured by the retrieval uncertainties.  Finally, variable 'model_error' provides an example of errors incurred in attempting to model the computed 10-second average XCO2 value: this could be added in quadrature to the uncertainty from 'xco2_uncertainty' to get the value used in the flux inversions.