CO Total Column Retrieval from SCIAMACHY: Full-Mission Validation with NDACC and TCCON

Abstract

Verification and validation are critical elements of any code development, and mandatory for the assessment of spaceborne remote sensing products. The objective was to perform intercomparisons of CO columns estimated from SCIAMACHY with temporal coincidented and co-located retrievals provided by ground-based Fourier transform infrared spectroscopy. More specifically, we used column averaged dry air mole fractions from several NDACC (Network for the Detection of Atmospheric Composition Change) and TCCON (Total Carbon Column Observing Network) stations. Like SCIAMACHY's channel 8 the TCCON instruments utilize the 2.3 mue band of carbon monoxide, whereas NDACC observes the CO mid infrared absorption. In most cases, satellite validation is based on statistical comparison (specifically true for SCIAMACHY with its large signal to noise ratio per observation (Gimeno Garcia et al. 2011)) with reference data. However, satellite and reference measurements do neither exactly match in time and space (mistime and misdistance) nor address the same volume of air (misintegration). Hence, the natural atmospheric variability leads to differences between both data sets and these differences must not be interpreted in terms of a satellites instrument malfunction (Verhoelst et al. 2015). The validation strategy presented here accounts for spatial (and in selected cases temporal) induced mismatches. The outcome of the study demonstrates that spatial (and temporal) averaging is required in order to minimize representation errors. Increased deviations of the spaceborne and ground-based columns in the later years of the mission clearly demonstrated the impact of the degrading channel 8 detector. Therefore, in order to perform a comprehensive full-mission (2003 - 2012) validation of the SCIAMACHY dataset, an approach providing more observations within a given time interval and sampling area had been utilized. The results reveal that the combination of these methods lead to acceptable agreement of SCIAMACHY CO data with most g-b reference sites (within the standard deviation). In addition the effect of whether using the mean or median of the datasets for comparison was investigated. The outcome demonstrates that the differences between the SCIAMACHY CO mean and median values are within 1-10 percent in most cases and within the standard deviation of the reference observations.