Line parameter study of ozone at 5 and 10 μm using atmospheric FTIR spectra from the ground: A spectroscopic database and wavelength region comparison

Atmospheric ozone concentration measurements mostly depend on spectroscopic methods that cover different spectral regions. Despite long years of measurement efforts, the uncertainty goal of 1% in absolute line intensities has not yet been reached. Multispectral inter-comparisons using both laboratory and atmospheric studies reveal that important discrepancies exist when ozone columns are retrieved from different spectral regions. Here, we use ground based FTIR to study the sensitivity of ozone columns on different spectroscopic parameters as a function of individual bands for identifying necessary improvements of the spectroscopic databases. In particular, we examine the degree of consistency that can be reached in ozone retrievals using spectral windows in the 5 and 10 mu m bands of ozone. Based on the atmospheric spectra, a detailed database inter-comparison between HITRAN (version 2012), GEISA (version 2011) and S&MPO (as retrieved from the website at the end of 2015) is made. Data from the 10 mu m window are consistent to better than 1%, but there are larger differences when the windows at 5 mu m are included. The 5 mu m results agree with the results from 10 mu m within +/- 2% for all databases. Recent S&MPO data are even more consistent with the desired level of 1%, but spectroscopic data from HITRAN give about 4% higher ozone columns than those from GEISA. If four sub-windows in the 5 mu m band are checked for consistency, retrievals using GEISA or S&MPO parameters show less dispersion than those using HITRAN, where one window in the P-branch of the nu(1) + nu(3) band gives about 2% lower results than the other three. The atmospheric observations are corroborated by a direct comparison of the spectroscopic databases, using a simple statistical analysis based on intensity weighted spectroscopic parameters. The bias introduced by the weighted average approach is investigated and it is negligible if relative differences between databases do not correlate with line intensities. This is the case for the comparison of HITRAN with GEISA in the 10 mu m region and the agreement between the simple analysis and the full retrieval is better than 0.1%. At 5 mu m biases might be as high as 1.4%, and the proposed method is thus limited to the same level of accuracy. Implications of the new data for database improvements and further studies, in particular in the 5 mu m region, are discussed. (C) 2016 Elsevier Inc. All rights reserved.