摘要
环境负荷通常会引起GNSS垂向坐标时间序列发生非线性变化,对其影响进行精细改正是GNSS坐标时间序列研究中的一项重要内容.传统的物理模型环境负荷改正方法在模型建立与参数求解等过程中需引入部分简化与近似,导致改正不够精细.本文引入数据驱动的广义回归神经网络(Generalized Regression Neural Network,GRNN)方法改善环境负荷修正效果.以川滇地区GNSS测站的垂向坐标时间序列为研究对象,首先基于变分贝叶斯独立分量分析(Variational Bayesian Independent Component Analysis,vbICA)技术分离坐标序列,分析得到周期性分量,发现大气及陆地储水负荷是引起测站坐标发生季节性变化的重要原因.然后通过GRNN建立与大气及陆地储水相关的环境因素数据和坐标时间序列数据之间的关联,进而消除坐标时间序列中两种环境负荷的影响.经数据驱动的GRNN建模修正大气及陆地储水负荷影响后,各测站坐标残差序列的RMS值平均降低了21.56%,而采用传统的物理模型方法修正后平均降低幅度仅为9.29%,可认为基于GRNN方法的改正效果更好.另外顾及地下温度、冰浓度、比湿、降雨率四种气候因素的影响建立GRNN模型,结果表明地下温度因素对川滇地区GNSS测站垂向坐标影响稍大.
Environmental loading typically contributes to the non-linear change of GNSS vertical coordinate time series,and addressing its influence is a crucial component of GNSS coordinate time series research.The model construction and parameter solution processes used in traditional method of environmental loading correction based on physical model need to involve various simplifications and approximations,resulting in the correction is not precise enough.In this paper,generalized regression neural network(GRNN),a data-driven method,is introduced to improve the effect of environmental loading correction.Taking the GNSS vertical coordinate time series of the Sichuan-Yunnan regional stations as the research object,we first separate the coordinate time series based on the Variational Bayesian Independent Component Analysis(vbICA)technique,analyze the obtained periodic components,and find that the atmospheric and land water storage loadings are the important causes of the seasonal changes of the station coordinates.The influence of two environmental loadings in the coordinate time series is then eliminated by using GRNN to build a connection between the data of environmental parameters linked to the atmosphere and land water storage and the coordinate time series data.After correcting the effects of atmospheric and land water storage loadings by GRNN modeling,the RMS values of the stations coordinate residual series are reduced by 21.56%on average,while the average reduction is only 9.29%after correcting by the traditional physical model method.It can be considered that the data-driven approach based on GRNN is more effective.Additionally,the GRNN models are established taking into account the effects of four climatic factors:temperature(below ground),ice concentration,specific humidity,and rainfall rate.The results show that the temperature(below ground)factor has a slightly greater impact on the vertical coordinates of the stations in the Sichuan-Yunnan region than the other three.
作者
高菡
匡翠林
楚彬
GAO Han;KUANG CuiLin;CHU Bin(School of Geoscience and Info-physics,Central South University,Changsha 410083,China;Hunan Institute of Geomatics Sciences and Technology,Changsha 410007,China)
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2024年第9期3357-3366,共10页
Chinese Journal of Geophysics
基金
国家自然科学基金(42388102,41774040)
中南大学研究生自主探索创新项目(2022ZZTS0591)资助。
关键词
GNSS坐标时间序列
环境负荷
广义回归神经网络
数据驱动
GNSS coordinate time series
Environmental loading
Generalized regression neural network
Data-driven
