Precipitable Water Vapor Retrieval Using Neural Network from Infrared Hyperspectral Soundings

Abstract

A neural network (NN) based algorithm for retrieval of precipitable water vapor (PWV) from the Atmospheric Infrared Sounder (AIRS) observations is proposed. An exact radial basis function (RBF) network is selected, in which the at-sensor brightness temperatures are the input variables, and PWV is the output variable. The training data sets for the RBF network are mainly simulated from the fast radiative transfer model (Community Radiative Transfer Model, CRTM) and the latest global assimilation data. The algorithm is validated by retrieving the PWV over west area in China using AIRS data. Compared with the AIRS PWV products, the RMSE of the PWV retrieved by our algorithm is 0.67 g/cm2, and a comparison between the retrieved PWV and radiosonde data is carried out. The result suggests that the RBF neural network based algorithm is applicable and feasible in actual conditions. Furthermore, spatial resolution of water vapor derived by RBF neural network is superior as compared to that of AIRS-L 2 standard product. Finally a PCA scheme is used for the preliminary investigation of the compression of AIRS high dimension observations.