Abstract. This paper focuses on the retrieval of temperature and relative humidity profiles through combining ground-based microwave radiometer observations with those of millimeter-wavelength cloud radar. The cloud-base height and cloud thickness from the cloud radar were added into the atmospheric profile retrieval process, and a back propagation neural network method was used as the retrieval tool.

Because substantial data are required to train a neural network, and microwave radiometer data are insufficient for this purpose, eight years of radiosonde data from Beijing were used as a database. The model MonoRTM was used to calculate the brightness temperature of the same channel as the microwave radiometer. Part of the cloud-base height and cloud thickness in the training dataset was also estimated using the radiosonde data.

The accuracy of the results was analyzed by comparing with L-band sounding radar data, and quantified using the mean bias, root-mean-square error and correlation coefficient. The statistical results showed that inversion with cloud information was the optimal method. Compared with the inversion profiles without cloud information, the RMSE values after adding the cloud information were to a varying degree reduced for the vast majority of height layers. These reductions were particularly clear in layers with cloud present. The maximum reduction of RMSE for temperature was 2.2 K, and for the humidity profile was 16 %.