Atmospheric temperature and humidity profiles are key parameters in weather forecasting and climate studies. This paper addresses the retrieval of atmospheric temperature and humidity profiles from the data acquired by the Microwave Temperature Sounder (MWTS) and the Microwave Humidity Sounder (MWHS) on Fengyun 3E (FY-3E) satellite using deep learning neural networks. The four-layer back-propagation neural network (BPNN) and the back-propagation dendritic neural network (BDNN) are firstly constructed, and then they are trained and validated using the matching samples between FY-3E MWTS/MWHS data and the fifth European Centre for Medium-Range Weather Forecasts(ECMWF) Re-analysis (ERA5) data. The results show that the BDNN method is better than the traditional linear regression method and the BPNN method. Over ocean, the root mean square errors (RMSEs) of the temperature and humidity profiles retrieval are less than 2.0 K and 1.0 g/kg, respectively, whereas they are, respectively, less than3.0Kand 2.0 g/kg over land.
|