Impacts on Noise Analyses of GNSS Position Time Series Caused by Seasonal Signal, Weight Matrix, Offset, and Helmert Transformation Parameters
Impacts on Noise Analyses of GNSS Position Time Series Caused by Seasonal Signal, Weight Matrix, Offset, and Helmert Transformation Parameters
复制标题
季节信号、权重矩阵、偏移和 Helmert 变换参数对 GNSS 位置时间序列噪声分析的影响
DOI：
10.3390/rs10101584
复制
发表时间：
2018
期刊：
影响因子：
5
通讯作者：
Jingnan Liu
中科院分区：
文献类型：
--
作者：
Guo Chen;Qile Zhao;Na Wei;Jingnan Liu
全球导航卫星系统 (GNSS) 位置时间序列的噪声特性可能会受到许多因素的影响，进而影响使用最小二乘法对确定性模型中的参数进行估计。作者评估了季节性信号、权重矩阵、间歇偏移和赫尔默特变换参数对噪声分析的影响。使用 647 个全球站的模拟和真实位置时间序列获得不同的解，并对从叠加解的残差得出的幂律噪声进行比较。由于除了模拟数据之外，无法获得位置时间序列中的真实噪声，因此作者最关注的是由可变因素引起的噪声差异。首先，时间序列中季节信号的参数化可以减少有色噪声并使频谱指数更接近于零（“更白”）。同时，额外的偏移参数还可以将有色噪声变得“更白”，并且确定性模型中更多的偏移参数导致光谱指数更接近于零。其次，无论是真实数据还是模拟数据，从协方差信息导出的权重矩阵都会比单位权重矩阵产生更多的有色噪声，并且协方差信息会导致模拟数据的年度幅度出现较大偏差。第三，模型中考虑的 Helmert 变换参数（三个平移、三个旋转和一个缩放）显示出对幂律噪声的最大影响（幅度和频谱指数的中值分别为 0.4 mm−k/4 和 0.06）。最后，叠加模型中一起使用的变换参数和全权矩阵可以分别导致水平和垂直分量的不同模式，这与不同的主导因素有关。
The noise characteristics of the Global Navigation Satellite System (GNSS) position time series can be biased by many factors, which in turn affect the estimates of parameters in the deterministic model using a least squares method. The authors assess the effects of seasonal signals, weight matrix, intermittent offsets, and Helmert transformation parameters on the noise analyses. Different solutions are obtained using the simulated and real position time series of 647 global stations and power law noise derived from the residuals of stacking solutions are compared. Since the true noise in the position time series is not available except for the simulated data, the authors paid most attention to the noise difference caused by the variable factors. First, parameterization of seasonal signals in the time series can reduce the colored noise and cause the spectral indexes to be closer to zero (much “whiter”). Meanwhile, the additional offset parameters can also change the colored noise to be much “whiter” and more offsets parameters in the deterministic model leading to spectral indexes closer to zero. Second, the weight matrices derived from the covariance information can induce more colored noise than the unit weight matrix for both real and simulated data, and larger biases of annual amplitude of simulated data are attributed to the covariance information. Third, the Helmert transformation parameters (three translation, three rotation, and one scale) considered in the model show the largest impacts on the power law noise (medians of 0.4 mm−k/4 and 0.06 for the amplitude and spectral index, respectively). Finally, the transformation parameters and full-weight matrix used together in the stacking model can induce different patterns for the horizontal and vertical components, respectively, which are related to different dominant factors.