Radiometric estimation of tropospheric attenuation: a mixed physically based/machine learning approach

A mixed physically based/machine learning (ML) approach to measure tropospheric attenuation A in all-weather conditions by means of microwave radiometers (MWRs) is proposed. The key idea is to combine the advantages originating from the accurate radiometric A retrievals, provided by the well-established Cosmic background (CB) approach in clear-sky conditions, with the benefits coming from ML techniques. The latter aim at estimating A in rainy situations through a simplified approach able to overcome the issues posed by more complex techniques such as the standard solution of the radiative transfer equation or the Sun tracking (ST) microwave technique. To this aim, an artificial neural network (ANN) is devised to turn the antenna noise temperatures measured by a four-channel MWR (from Ka- to W-band) into tropospheric attenuation at the frequencies of the radiometric channels, namely 23.8, 31.4, 72.5, and 82.5 GHz. The network is properly trained and tested by taking advantage of the concurrent CB and ST measurements collected by the RpG radiometer deployed at Politecnico di Milano, Milan, Italy, under the ESA-funded WRAD project. The proposed approach to retrieve the tropospheric attenuation is intended to overcome the limits associated both with the ST technique (only measurements during the day, link elevation strictly bound to the Sun ecliptic) and to the CB one (unreliable measurements in rainy conditions).