Retrieval of water vapor profiles with radio occultation measurements using an artificial neural network

A new method applying an artificial neural network (ANN) to retrieve water vapor profiles in the troposphere is presented. In this paper, a fully-connected, three-layer network based on the backpropagation algorithm is constructed. Month, latitude, altitude and bending angle are chosen as the input vectors and water vapor pressure as the output vector. There are 130 groups of occultation measurements from June to November 2002 in the dataset. Seventy pairs of bending angles and water vapor pressure profiles are used to train the ANN, and the sixty remaining pairs of profiles are applied to the validation of the retrieval. By comparing the retrieved profiles with the corresponding ones from the Information System and Data Center of the Challenging Mini-Satetlite Payload for Geoscientific Research and Application (CHAMP-ISDC), it can be concluded that the ANN is relatively convenient and accurate. Its results can be provided as the first guess for the iterative methods or the non-linear optimal estimation inverse method.