Estimation of tropospheric profiles using COSMIC GPS radio occultation data with neural networks

In this work we have proposed a method based on neural networks to directly retrieve dry and wet refractivity and dry pressure profiles in troposphere (which in turn can be used to obtain temperature and humidity profiles) by using FORMOSAT-3/COSMIC GPS radio occultation data. To overcome the constraint of an independent knowledge of one atmospheric parameter at each GPS occultation, we trained different neural networks with refractivity profiles as input computed from the geometrical occultation parameters, while the targets were the dry and wet refractivity profiles and the dry pressure profiles obtained from the contemporary European Centre for Medium-Range Weather Forecast data. We have considered a data set of 445 available satellite radio occultations covering the entire land spanning within the Tropics during the summer of 2006, by splitting it into a desert and a vegetation data set that are referred to desert zone and vegetation zone respectively. For each subset we have used 90% profiles for training the neural networks and the remaining 10% ones for the independent test. Finally, selecting two cases, we have compared the estimated pressure, temperature and water vapour partial pressure, for each neural network approach, with the corresponding radiosounding profiles.