大气角动量函数在日长变化非线性预报中的重要作用  

Importance of Atmospheric Angular Momentum Function in Non-linear Prediction of Length of Day Variation

作　　者：王琪洁[1] 廖德春[1] 周永宏[1] 廖新浩[1] 

机构地区：[1]中国科学院上海天文台

出　　处：《天文学报》2008年第1期93-100,共8页Acta Astronomica Sinica

基　　金：国家自然科学基金(10633030、10373017、10673025)资助项目

摘　　要：日长变化具有复杂的时变特性,传统的线性时间序列分析方法往往难以取得良好的预报效果.采用非线性人工神经网络技术对日长变化进行预报,网络模型的拓扑结构由最小均方误差法来确定.考虑到日长变化与大气环流运动间的密切关系,在神经网络预报模型中引入轴向大气角动量序列.结果表明,联合日长和大气角动量序列,比起单独采用日长资料,预报精度得到显著的提高.Prediction of the variations of the length of day （LOD） is of great importance in both scientific issues and practical applications. However, due to the complex time-variable characteristics of the LOD variation, it＇s usually difficult to obtain satisfied prediction results by conventional linear time series analysis methods. The artificial neural networks （ANN） is a non-linear information processing system. This study employs the ANN to predict the LOD change. The topology of the ANN model is determined based on the criterion of minimization of the root mean square error （RMSE）. For most of the studies that use ANN to predict the LOD, the influence of global atmospheric movements on the variations of the LOD hasn＇t been considered. Considering the close connection between the LOD variation and the atmospheric circulation movement, and the capability of simulating and forecasting the axial atmospheric angular momentum （AAM） function with global atmospheric circulation pattern, the axial AAM is added into the ANN model as an additional input parameter to predict the LOD variation. The daily LOD series in this study are from the C04 series of the International Earth rotation and reference systems service （IERS）, spanning from 1962 to 2005. We first removed the contributions of the 62 zonal Earth tides from the LOD changes with periods from 5 days to 18.6 years according to IERS Convention 2003, and the effects that can be described by functional models, e.g. the annual and semi-annual oscillations, the terms whose periods are 1, 1/2, 1/3 of the length of the whole data set. Only the residuals between the modeled and the observed LODRs, are used for training. Likewise, the axial AAM series are also de-trended. The residuals of LODR and axial AAM series are used to train the networks. The trained networks are applied to predict the LODR variation for time interval of 1 to 40 days. For comparisons, we also use the LODR only to construct the ANN model and to predict the LODR variations. The res

关 键 词：天体测量 地球自转 方法 数值 

分 类 号：P127[天文地球—天体测量]