微波散射计反演海面风场的神经网络方法研究

发布时间：2018-10-11 12:36

【摘要】：研究利用神经网络方法处理微波散射计数据,反演海面风场。重点研究海洋二号(HY-2)卫星微波散射计数据反演,特别是中高风速条件下的风场反演。其中风速的反演基于后向传播(Back Propagation,BP)神经网络;多解风向的反演基于混合密度(Mixture Density Network,MDN)神经网络,求解过程中的核函数采用高斯分布;网络训练的目标风场采用欧洲中期天气预报中心(European Centre for Medium-range Weather Foresting,ECMWF)模式风场。通过与ECMWF风场的比较,利用神经网络方法反演的风场可以满足HY-2微波散射计风场反演的精度要求。同时通过与国家卫星海洋应用中心发布的HY-2微波散射计L2B级风场产品相比较,表明该方法反演的风场更接近ECMWF模式风场。
[Abstract]:The neural network method is used to deal with the data of microwave scatterometer and to retrieve the wind field of the sea surface. The inversion of microwave scatterometer data of Ocean 2 (HY-2) satellite, especially the inversion of wind field at medium and high wind speed, is studied in this paper. Wind speed inversion is based on backward propagation (Back Propagation,BP) neural network, multi-solution wind direction inversion is based on mixed density (Mixture Density Network,MDN) neural network, and the kernel function is Gao Si distribution. The target wind field for network training is based on the European Center for Medium-Term Weather Prediction (European Centre for Medium-range Weather Foresting,ECMWF) model. Compared with ECMWF wind field, the wind field inversion using neural network method can meet the accuracy requirements of HY-2 microwave scatterometer wind field inversion. By comparing with the HY-2 microwave scatterometer L2B wind field released by the National Satellite Ocean Application Center, it is shown that the wind field obtained by this method is closer to the ECMWF model wind field.
【作者单位】：中国科学院微波遥感技术重点实验室中国科学院国家空间科学中心;中国科学院大学;国家卫星海洋应用中心;
【基金】：中国科学院国家空间科学中心微波遥感技术重点实验室开放课题(MLF2015003)资助
【分类号】：P714.2;TP183

【相似文献】