Improvement in the retrieval of humidity profiles using hybrid regression technique from infrared sounder data: a simulation study



The profiles retrieved from sounder observations using statistical methods are extensively used as a first guess for physical retrieval. The sounder radiances are nonlinearly related to humidity profiles, making it desirable to use a suitable form of variables in the regression. Traditionally, the natural logarithm of the humidity is used as a predictand. In the present simulation study it is demonstrated that the use of the natural logarithm of humidity as predictand gives better accuracy for a drier atmosphere, whereas direct humidity as a predictand provides better accuracy for a wetter atmosphere. The present study proposes a hybrid regression technique for the improvement in the retrieval of humidity profiles by combining the best features of both the predictands. Compared to the conventional regression technique the accuracy of the humidity retrieval improves by 5% for the lower troposphere in the hybrid regression. Copyright © 2012 Royal Meteorological Society