Comparison of different techniques in atmospheric temperature-humidity sensing from space

Abstract

Numerical closed-loop experiments on retrieving atmospheric temperature and humidity profiles by high-resolution measurements of the outgoing thermal infrared (IR) radiation using a Russian Fourier spectrometer (IRFS-2) were performed. Three techniques were used: multiple linear regression (MLR), the iterative physical-mathematical approach (IPMA), and artificial neural networks (ANNs). The MLR technique gives significant root mean square (RMS) errors in the retrieval of the temperature profile, especially in the troposphere region; these errors may be as great as 2–3 K. The ANN and IPMA techniques are considerably more accurate, giving approximately equal RMS errors of 1.0–1.5 K at altitudes of 2–30 km. For all interpretation techniques, a growth of errors of retrieval of temperature in the lower troposphere is observed and is especially substantial (up to 3 K for the near-surface temperature) in thermal sensing over land. The systematic errors of temperature retrieval for the ANN technique are practically zero, and for the other two techniques, they do not exceed 0.4 K. The differences in thermal sensing of the atmosphere over water and land manifest themselves in the appearance of an additional five determined coefficients of expansion of the spectral dependence of the IR emissivity of land in principal components. This leads to increased errors on thermal sensing in the lower troposphere, up to ~0.5 K for all interpretation techniques. The information content of the IRFS-2 device measurements with regard to the atmospheric humidity profile is relatively small because of the values of the errors of measurements of the outgoing radiation in the shortwave range, and in particular, in the water vapour absorption band 6.3 µm. The ANN technique makes it possible to determine relative humidity in the troposphere with RMS errors of 10–15%. In the case of observations over water, the mean errors of the ANN technique are practically equal to zero, and for the MLR and IPMA techniques, they are of an approximately equal order of magnitude, namely 2–4% of relative humidity. The IPMA and MLR techniques give RMS humidity errors of 15–20% and up to 40%, respectively.