Observational evidence of 3‐D cloud effects in OCO‐2 CO2 retrievals

The standard deviations of the distributions of Orbiting Carbon Observatory (OCO‐2) measurements of CO2 (i.e., XCO2) increase in size in the presence of clouds. XCO2 and Moderate Resolution Imaging Spectroradiometer (MODIS) radiance and cloud fields, and OCO‐2 A‐band radiances, are analyzed in order to determine if this behavior is best described as a radiance‐level retrieval artifact or by 3‐D radiative transfer effects. Observations in clear‐sky and fair weather cumulus scenes are analyzed. Scatter diagrams of XCO2 versus MODIS (and OCO‐2) radiances are presented, and averages are calculated for each scene for several radiance bins. The averages vary little in clear skies but decrease markedly for cloudy scenes as radiances increase. These decreases are consistent with an interpretative framework provided by 3‐D SHDOM radiative transfer calculations. Two 3‐D metrics, ΔXCO2 and Have, are calculated and applied. ΔXCO2 is the difference in XCO2 for the smallest and largest radiance bins. Have is a measure of the heterogeneity of the cloud radiance field. Lines of XCO2 and MODIS radiance for four target mode scenes have different slopes for clear and cloudy scenes, contrary to the radiance‐level retrieval artifact interpretation. In contrast, the graph of ΔXCO2 and MODIS Have for the various scenes has a linear correlation coefficient of 0.92, consistent with the 3‐D interpretation. Since the OCO‐2 measurement requirement is 1 ppmv, the cloudy scene XCO2 standard deviations between 1.2 and 2.6 ppmv indicate that 3‐D cloud effects add an important component to the XCO2 error budget. Plain Language Summary The measurement goal of the Orbiting Carbon Observatory (OCO‐2) satellite is to measure CO2 to better to 1% accuracy on a regional scale. OCO‐2 CO2 and Moderate Resolution Imaging Spectroradiometer satellite radiance and cloud fields for a half‐dozen individual scenes are analyzed to demonstrate that three‐dimensional cloud effects contribute to variations in CO2 at local (e.g. 20 km × 20 km) spatial scales. Two three‐dimensional indicators (ΔXCO2 and Have) are calculated and applied. The correlation of ΔXCO2 and Have (0.92) demonstrates that three‐dimensional cloud effects increasingly add to the variations of OCO‐2 CO2 measurements as the cloud field becomes increasingly more complicated. Key Points OCO‐2 XCO2 and MODIS cloud fields are analyzed to provide evidence of 3‐D cloud effects Two 3‐D metrics, ΔXCO2 and Have, are calculated and applied The high co