Experiments on Improving Temperature and Humidity Profile Retrieval for Ground-based Microwave Radiometer

Ground-based microwave radiometer (MWR) has a crucial role in scientific researches, weather modification service and climate change studies. MWR adopts passive remote sensing technology which has smaller volume, lower power consumption. Ground-based microwave radiometer detects atmospheric temperature and humidity by receiving atmospheric microwave radiation, which can conduct 24-hour unattended, high-resolution observation. It can detect short-time variation of atmospheric elements. Many studies show that different seasons, different weather conditions, quality control algorithms, and changes in environments have certain effects on retrieval results of MWR. In the case of cloudy condition, the uncertainty of cloud absorption coefficient leads to the increase of retrieval error and incorrect data of MWR especially. In order to improve temperature and relative humidity detection capabilities of MWR, the experiment builds BP neural network algorithm with 6 years(from 2011 to 2016) sounding data. The experiment builds two types of retrieval methods because there are some differences of microwave radiation transfer between clear and cloudy samples. The test uses measured brightness temperature data (requiring correction) and cloud data of millimeter-wavelength cloud radar as model inputs and then uses sounding data to evaluate model outputs (temperature and relative humidity profiles) from 2017 to 2018.Results show that correlation coefficients between outputs of 4 models (clear sky sample temperature model, cloudy sample temperature model, clear sky sample relative humidity model and the cloudy sample relative humidity model) and sounding data are 0.99, 0.99, 0.80 and 0.78. Taking sounding profiles as reference, root mean square errors (RMSE) of retrieval results of 4 models are 2.3℃, 2.3℃, 9%, 16%. Comparing with the MWR original profiles, RMSEs of 4 models are reduced by 0.4℃, 0.3℃, 11% and 9%, accuracies are improved by about 30%, 28%, 64% and 45%. In particular, the deviation of temperature model and humidity model within ±2℃ and ±20% account for 68%, 70% and 95%, 78%, which are 7%, 5% and 27%, 23% higher than MWR original profiles. The bias correction of brightness temperature and the training retrieval model of distinguishing weather samples are helpful to improve the retrieval accuracy of MWR temperature and humidity profiles. The network model combined with cloud radar information has obviously better effects on the retrieval result under cloudy samples.Through these experiments, the quality control of brightness temperature, combination of active and passive retrieval algorithms are well improved. The combination of active and passive retrieval effectively improves the performance of MWR, which will lay a foundation for the development of the comprehensive observation system of atmospheric profiles.