A NN-based algorithm for estimation of water vapor content using AVHRR data over ocean

Abstract

A neural network-based algorithm for retrieving precipitable water vapor (PWV) using the Advanced Very High Resolution Radiometer (AVHRR) data is proposed. The neural network (NN) model is combined with the radiative transfer calculations using MODTRAN 4.0 with the latest global assimilated data. The selected NN is a multilayer feed-forward neural network. The input variables are the top-of-atmosphere (TOA) brightness temperatures in AVHRR channels 4 and 5 and the sea surface temperature (SST), and the output variable is the PWV. In order to evaluate the performance of the trained neural network, simulations are carried out for the mid-latitude summer (MS) model atmosphere, with a RMSE of 0.33 g/cm2. Furthermore, the new approach is validated with the AVHRR data of the Pacific Ocean. The water vapor contents derived from AVHRR image are compared with that derived by radiosonde data, with a difference of 0.12 g/cm2. The advantages of the proposed algorithm are discussed briefly. The preliminary results show that the new algorithm is able to provide an accurate estimation of PWV from AVHRR data.