A technique for determining the tropospheric ozone (TO) content from the spectra of outgoing thermal infrared (IR) radiation based on the principal component method and neural network approach is proposed. To train the artificial neural networks, TO data calculated from ozone profiles of vertical ozone content derived from ozonesondes are used. The TCO is considered the ozone content in the atmospheric layers from the earth’s surface to pressure levels of 400 and 300 hPa. The error of approximating TO values on training data is 2.7 and 3.6 DU for layers below 400 and 300 hPa, respectively. The methodology is validated on the basis of comparison with ground-based TO measurements at the NDACC international observing network of stations using solar infrared spectra. The mean standard deviations of the differences between the groundbased infrared measurements at 19 stations and the derived TO values from the IKFS-2 data were about 3 DU.
The mean differences depend on the altitude and geographical location of the ground station, varying from +3 to –12 DU. The discrepancies between the ground-based measurements and satellite data correspond to the results of other authors obtained for the IASI satellite instrument, which is close in characteristics. The paper presents examples of the global distribution of mean monthly TO values for different seasons.