一种基于RBF神经网络的大气温度及水汽密度廓线反演方法

An Inversion Method of Atmospheric Temperature and Water Vapor Density Profile Based on RBF Neural Network

利用地基16通道微波辐射计在武汉地区的观测资料(包含16个通道的亮温数据及地面气象信息数据)和对应的探空资料(包含温度廓线及水汽密度廓线)组建神经网络训练样本和测试样本,训练样本中的观测资料和对应的探空资料分别作为训练神经网络的输入和输出数据,测试样本中的微波辐射计观测资料作为输入数据对训练后的神经网络进行测试反演,测试样本中的探空资料作为标准与反演输出结果进行对比。地基16通道微波辐射计所带的BP神经网络反演算法作为该研究构建的RBF网络在不确定条件下的预测准确性和可行性。实验结果表明:RBF神经网络在反演大气温度廓线及水汽密度廓线方面拥有比BP神经网络更快的运算速度,更准确的反演能力和更强的泛化能力。利用RBF神经网络反演大气温度廓线和水汽密度廓线的优势要大于BP神经网络反演方法,将该方法应用在微波辐射计上对于提升其反演技术水平具有重要的意义。

By using the ground-based 16-channel microwave radiometer to analyse the data in Wuhan(including 16 channels of brightness temperature data and surface meteorological information data)and corresponding sounding data(including temperature profile and water vapour density profile)to form the network training samples and testing samples,the observation data in the training samples and the corresponding sounding data are respectively used as input and output data for training the neural networks,and the observation data in the test samples are used as input data to test the trained neural networks.The sounding data of the test sample is compared as standard to the test output.By comparing the training results and the test results of RBF neural network and BP neural network,the prediction accuracy and feasibility of RBF neural network under uncertain conditions are verified.The experimental results show that the RBF neural network has faster calculation speed,more accurate inversion capability and stronger generalization ability than the BP neural network in retrieving the atmospheric temperature profile and the water vapour density profile.The advantages of using RBF neural network to invert atmospheric temperature profile and water vapour density profile are significantly greater than BP neural network inversion method.Applying this method to microwave radiometer is of great significance for improving the level of inversion technology.