| 应用平滑先验信息方法移除GRACE数据中相关误差 |
| |
| 引用本文： | 詹金刚,王勇,史红岭,柴华,朱传东.应用平滑先验信息方法移除GRACE数据中相关误差[J].地球物理学报,2015,58(4):1135-1144. |
| |
| 作者姓名： | 詹金刚 王勇 史红岭 柴华 朱传东 |
| |
| 作者单位： | 1. 中国科学院测量与地球物理研究所 大地测量与地球动力学国家重点实验室, 武汉 430077;
2. 中国科学院大学, 北京 100049 |
| |
| 基金项目： | 国家自然科学基金项目(41321063,41274084,40974044,41174064,41374086),中国科学院国家外专局国际合作创新研究团队计划(KZZD-EW-TZ-05),大地测量与地球动力学国家重点实验室项目(SKLGED2013-6-1-Z)资助. |
| |
| 摘 要： | 由于GRACE卫星数据解算的时变重力场模型中高阶位系数存在误差,这些误差在重力异常图中表现为南北向的条带噪声,在应用GRACE时变重力场数据时必须进行滤波.本文在空间域引入了一种有效消除GRACE时变重力场条带噪声的平滑先验信息方法,并将其与目前常用的高斯滤波和去相关误差等滤波方法分别应用于合成的质量变化趋势数字模型,检测不同滤波方法消除条带噪声的能力及其对真实信号的影响.滤波结果显示,与目前常用的高斯滤波和去相关误差滤波器相比,本文滤波方法在有效移除条带噪声的同时,具有有效信号幅度衰减小、有效信号形变小以及保存了更多的短波长细节信息等优势;此外,统计结果显示,本文滤波结果在信号最大值、最小值以及残差均方根等方面均与模拟真实信号最为接近.相比300km高斯平滑和组合滤波结果,有效信号振幅的极小值和极大值分别提高了约18%和6%,残差均方根分别降低了25%和33%.说明本文滤波方法移除GRACE相关误差的同时,在保留有效信号方面具有明显的优势.
|
| 关 键 词： | GRACE时变重力场 平滑先验信息方法 相关误差 |
| 收稿时间： | 2014-04-28 |
Removing correlative errors in GRACE data by the smoothness priors method |
| |
| ZHAN Jin-Gang,WANG Yong,SHI Hong-Ling,CHAI Hua,ZHU Chuan-Dong.Removing correlative errors in GRACE data by the smoothness priors method[J].Chinese Journal of Geophysics,2015,58(4):1135-1144. |
| |
| Authors: | ZHAN Jin-Gang WANG Yong SHI Hong-Ling CHAI Hua ZHU Chuan-Dong |
| |
| Institution: | 1. State Key Laboratory of Geodesy and Earth's Dynamics, Institute of Geodesy and Geophysics, Chinese Academy of Sciences, Wuhan 430077, China;
2. University of Chinese Academy of Sciences, Beijing 100049, China |
| |
| Abstract: | The errors, which are present in short wavelength components of time-variable gravity solutions produced by the Gravity Recovery and Climate Experiment (GRACE) satellite, manifest themselves in maps of gravity change as long, linear features generally trending in north-south (i.e., stripes). Thus, filtering is needed to remove them before using the GRACE time-variable gravity solutions. Here, we introduce a smoothness priors method to remove such stripe errors with application to a synthetic mass change model and show significant effects. The method used in this paper is based on the smoothness priors approach and operates like a time-varying finite impulse response high-pass filter. In this paper, the correlative errors in maps of GRACE mass variation are regard as the high frequency signals varying with longitude reserved when the data is filtered, then we subtract them from the original data to get the filtering result. In order to examine the effect of the filter, we generate a model consisting of two parts:'true' geophysical signals and stripe noise. First, following the approaches of Duan et al., we construct a numerical model of mass change trend composed of the NOAH GLDAS model between the latitudes of ±60°, a surface mass change rate corresponding to the gravity variation of a PGR model based on the RF3L20 ice model over Canada and the GRACE mass variation trend over the Arctic and Antarctic areas (90°S—60°S, 60°N—90°N). We use this numerical model as the 'true' geophysical signal of mass change. In addition, we also need a north-south 'stripe' model similar to that in GRACE data to test the effect of the filter and its signal distortion. The stripe model is constructed using the difference between the GRACE monthly time-variable gravity solutions and the filtered field. Its amplitude is reduced to be comparable with that of the numerical mass change trend model. The filtered field is computed using the algorithms of Swenson and Wahr and an isotropic Gaussian filter of radii 300 km. Finally, we apply the smoothness priors method (SPM) to the model to test the effect of the filter. As a comparison, we use the filter of correlated errors (for A=30, K=10), the Gaussian filter of smoothing radius 300 km, the combination of above two methods, as well as the SPM filter to the synthetic model, respectively. To conform with the processing of GRACE data, we transform the mass change of the synthetic model into Stokes coefficients when applying the correlated error filter and Gaussian smoothing, then transform them back to mass change.Compared with other filtered results above mentioned, the result of SPM shows a better signal-to-noise ratio, though a few of slight stripes appearing in the Pacific Northwest. We find that our result has the advantages of less reducing the signal magnitudes, preserving more details of short wavelength components and less signal distortion. For example, the magnitude of the signals at high latitudes such as the positive signals in the Antarctic and southern Chile are not obviously reduced. Some short wavelength components such as the positive signal in middle America and the negative signal in the Jan Mayen Island are also preserved in the filtered result. The positive signal in eastern Brazil is also not distorted compared with the other results.The grid statistic results show that the minimum value and maximum value of the results using the method of Gaussian filtering and the combination method of Gaussian filtering and correlative errors filter are -6.78 cm, 11.06 cm, -6.53 cm and 11.10 cm, respectively. Compared with the 'true' values of -11.45 cm and 13.15 cm, this result shows that the magnitudes of the negative signal are sharply reduced when using the two kinds of filters. Their RMS values are also reduced from 1.448 cm to 1.231 cm and 1.196 cm, respectively. All these statistic data indicate that we have applied a stronger filter on the synthetic field. Compared with the results of the three methods above mentioned, the grid statistics of SPM approach shows a better result. Its minimum value, maximum value and the RMS of the grid statistics are -8.81 cm, 11.88 cm and 1.371 cm, respectively, which are most close to the true values among the filtered results.The smoothness priors method operates like a time-varying finite impulse response high-pass filter, and shows remarkable effect in removing the GRACE stripe noise. The advantage of this filter is that it has less reduction in magnitude of signals and preserves more details of short wavelength components in the filtered field. Another advantage of it is that the filter is flexible in use. |
| |
| Keywords: | GRACE time variable gravity Smoothness priors method Correlated errors |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《地球物理学报》浏览原始摘要信息 |
| 点击此处可从《地球物理学报》下载免费的PDF全文 |
|
