杜皓, 郭文飞, 郭迟, 路鹏远, 叶世榕. 针对GNSS-R海面风速反演的自适应CDF匹配方法[J]. 武汉大学学报 ( 信息科学版), 2021, 46(12): 1924-1931. DOI: 10.13203/j.whugis20210253
引用本文: 杜皓, 郭文飞, 郭迟, 路鹏远, 叶世榕. 针对GNSS-R海面风速反演的自适应CDF匹配方法[J]. 武汉大学学报 ( 信息科学版), 2021, 46(12): 1924-1931. DOI: 10.13203/j.whugis20210253
DU Hao, GUO Wenfei, GUO Chi, LU Pengyuan, YE Shirong. Adaptively CDF Matching Method in GNSS-R Wind Speed Retrieval[J]. Geomatics and Information Science of Wuhan University, 2021, 46(12): 1924-1931. DOI: 10.13203/j.whugis20210253
Citation: DU Hao, GUO Wenfei, GUO Chi, LU Pengyuan, YE Shirong. Adaptively CDF Matching Method in GNSS-R Wind Speed Retrieval[J]. Geomatics and Information Science of Wuhan University, 2021, 46(12): 1924-1931. DOI: 10.13203/j.whugis20210253

针对GNSS-R海面风速反演的自适应CDF匹配方法

Adaptively CDF Matching Method in GNSS-R Wind Speed Retrieval

  • 摘要: 利用全球导航卫星系统反射(global navigation satellite system reflectometry, GNSS-R)信号对海面风速进行反演时,连续地球物理模式函数常用于建立时延-多普勒图像(delay/Doppler map, DDM)特征值与风速之间的映射关系。该模型在DDM较少的风速范围内(0~5 m/s及12~20 m/s)存在较大系统偏差。为解决该问题,提出了一种自适应累积分布函数(cumulative distribution function, CDF) 匹配方法。该方法将反演风速序列与参考风速序列进行CDF匹配,并利用最小二乘自适应地寻找最优阶数多项式,对风速偏差序列进行拟合和改正。公开数据实验验证结果表明,在0~5 m/s和12~20 m/s内,改正后的反演风速均方根误差分别减小了6%和15%,系统偏差分别减小了45%和25%,明显提升了少样本情况下的反演精度,反演风速更符合自然界中风速分布。

     

    Abstract:
      Objectives  When the global navigation satellite system reflectometry (GNSS-R) technique is applied for sea surface wind speed retrieval, continuous geophysical model functions are often used to fit the observable extracted from the delay/Doppler map (DDM) to wind speed empirically. A large systematic deviation exists in the above method due to the few samples at 0-5 m/s and 12-20 m/s. To solve the problem, an adaptively cumulative distribution function (CDF) matching method for bias correction is proposed.
      Methods  In this method, CDF matching is performed between the retrieved and reference wind speed sequences, and the least square method is used to adaptively find the optimal polynomial to fit the wind bias sequence and correct it. Public data products are used for validation.
      Results  Test results show that the root mean square errors (RMSEs) after correction are reduced by 6% and 15%, and the biases are reduced by 45% and 25% at 0-5 m/s and 12-20 m/s respectively. And overall bias is improved by 25%.
      Conclusions  Wind retrieval accuracy is obviously improved for wind ranges with few samples, and the probability distribution of wind speed retrievals is more consistent with that in nature. However, the bad performance at high wind speeds shows that the tendencies from DDM observable to wind speed at low and high wind speeds are extremely different because of the low sensitivity of DDM observables at high winds. The piecewise function is a good choice, but it is difficult to determine the piecewise point and keep it smooth.

     

/

返回文章
返回