基于主成分分析的GRACE重力场模型等效水高  

Equivalent water height from GRACE gravity model based on principal component analysis

作　　者：穆大鹏[1] 郭金运[1,2] 孙中昶 孔巧丽[1,2] 

机构地区：[1]山东科技大学测绘科学与工程学院,青岛266590 [2]海岛(礁)测绘技术国家测绘地理信息局重点实验室,青岛266590 [3]中科院遥感与数字地球研究所,北京100096

出　　处：《地球物理学进展》2014年第4期1512-1517,共6页Progress in Geophysics

基　　金：国家自然科学基金(40974004;41274018;41374009);国际科技合作计划(2009DFB00130)联合资助

摘　　要：本文使用主成分分析在空间域提取GRACE重力场模型的等效水高,假设信号与噪声之间统计不相关,把由CSR、GFZ和JPL的GRACE重力场模型计算得到的等效水高值作为统计量,通过线性变换来分离信号和噪声.选择2007年3月份GRACE重力场模型作为实验,在直接分离没有取得预期结果后,对GRACE作半径为300km、400km和500km的高斯平滑,然后利用主成分分析得到了3个模式.结果显示,3种平滑半径下,主成分的第1模式的贡献率分别为79%、90%和93%.从主成分3个模式的统计结果以及等效水高图判断,第1模式对应于水文变化信号,其他2个模式为南北条带状噪声.由主成分的第1模式估算了CSR、GFZ和JPL的等效水高,与经验去相关方法进行了对比,结果较为一致.2010年12个月的结果显示,第1模式的贡献率不仅与平滑半径有关,而且与时间有关.Gravity Recovery and Climate Experiment（GRACE）satellite can provide time-variable gravity field every month with spatial resolution 300-400 km.Monthly solutions from GRACE suffered so called north-south stripes noise which needed to be removed by filter approach.Many methods have been proposed and focus on filtering the spherical harmonic coefficients.In this paper,we use principal component analysis（PCA）to extract equivalent water height（EWH）from GRACE in spatial domain,which based on the assumption that signal and noise are uncorrelated. One of the main characteristics of PCA is taking advantage of different GRACE models produced by different groups,i.e.,Center for Space Research（CSR）, Jet Propulsion Laboratory（JPL）,GeoForschungs Zentrum（GFZ）.We view the GRACE gravity model released by CSR,GFZ and JPL as statistical variables,and separate the signal and noise by linear transformation,e.g.,the orthogonal matrix derived by PCA.First of all,we choose the GRACE model of March,2007 as an example to apply PCA.Unfortunately,we didn’t obtain the anticipative result when we decomposed the EWH directly.We did obtain 3 components which are very similar to each other,but none of them can match the signal or noise.The possible reason is that our assumption is too simple for the relation between signal and noise from GRACE,so we choose Gaussian smoothing with300 km,400 km and 500 km to pre-process GRACE solutions,and then decompose the EWH.In this time,3 different components are obtained using principal component analysis.The result showed that the first mode of three radiuses contributes to 79%,90% and 93% of total variation,respectively.According to the statistical results and EWH maps,the first mode can be viewed as signal and others noise.We use the first mode to estimate the EHW of CSR,GFZ and JPL,and they are close to the result of Empirical De-correlation Method.Also,the result of 2010 showed that the contribution of first mode not only depended on smoothing radius,but also depended on time.

关 键 词：主成分分析 GRACE重力场模型 等效水高 

分 类 号：P228[天文地球—大地测量学与测量工程]