人工神经网络在星载散射计海面风场反演建模中的应用  

Application of Artificial Neural Network to Ocean Surface Wind Field Retrieval Modeling for Spaceborne Scatterometer

作　　者：陈克海[1] 解学通[1] 黄舟[2] 方裕[2] 陈晓翔[1] 

机构地区：[1]中山大学遥感与地理信息工程系,广州510275 [2]北京大学遥感与地理信息系统研究所,北京100871

出　　处：《北京大学学报（自然科学版）》2007年第4期460-467,共8页Acta Scientiarum Naturalium Universitatis Pekinensis

基　　金：国家"863"计划项目(2002AA134100);985工程(105203200400006)资助项目

摘　　要：地球物理模型函数(GMF)是散射计风场反演的基础及算法有效运行的前提条件。采用传统的统计方法建立GMF往往需要大量的、多种参数条件下的雷达后向散射测量数据。以圆锥扫描散射计SeaWinds为例,根据其特点,建立了一个两种极化方式下统一的神经网络模型函数(NNGMF),并对风速、相对风向采样间隔和测量值数目对模型精度的影响进行了详细分析。通过与Qscat-1模型进行比较,发现该神经网络模型在采样间隔较大或测量值数目较少的情况下,仍能较好地体现SeaWinds散射计的海面后向散射特性。Geophysical model function（GMF） is the basis of scatterometer wind field retrieval and the prerequisite to effective running of the retrieval algorithm. GMF modeling using traditional statistical method usually needs a great number of backscattering coefficient measurements under various observing parameter conditions. Taking SeaWinds, a type of conically scanning scatterometer, as an example, a unified neural network model （NN GMF） for two polarization modes is designed. Moreover, effects of wind speed sampling interval, wind relative direction sampling interval, and the number of measurements on modeling precision are also analyzed in detail. Compared with Qscat-1 model function, the neural network model is able to reflect the backscattering signature of SeaWinds scatterometer very well, even on the condition that sampling intervals of wind speed and wind relative direction are large or the number of backscattering measurements is small.

关 键 词：散射计 海面风场 地球物理模型函数(GMF) 神经网络 

分 类 号：P714.2[天文地球—海洋科学]