LOESS用于GNSS垂向坐标时间序列噪声识别与提取

针对GNSS垂向坐标时间序列噪声复杂、精度较差等特点，采用局部加权回归模型(locally weighted regression, LOESS)对中国大陆构造环境监测网络中289个GNSS站的垂向坐标时间序列进行降噪分析。首先利用LOESS方法对预处理后的时间序列进行降噪处理，得到降噪后的时间序列及噪声序列；然后采用Durbin-Watson(DW)检验对降噪后的噪声序列进行自相关性检验，同时采用Wilcoxon秩和检验方法对降噪前后序列的标准差、噪声项、速度不确定度等指标进行显著性检验；最后采用降噪前后序列的信噪比及前3个指标来定量评价降噪效果。结果表明，各测站降噪后的噪声序列不存在自相关性，采用LOESS方法降噪处理后各评价指标均有显著改正，表明LOESS方法能够有效减少GNSS垂向坐标时间序列中的噪声，进一步提高GNSS垂向坐标时间序列的精度。

Noise Recognition and Extraction of GNSS Vertical Coordinate Time Series Based on LOESS

Aiming at the characteristics of complex noise and poor precision of GNSS vertical coordinate time series, we use the locally weighted regression(LOESS) method to denoise the vertical coordinate time series of 289 GNSS stations in the crustal movement observation network of China. Firstly, we use the LOESS method to denoise the preprocessed time series, and obtain the denoised time series and noise series. Then, we use the Durbin-Watson test for autocorrelation of denoised series, and the Wilcoxon rank sum test method for significance test of the standard deviation, noise term and velocity uncertainty of the series before and after denoising. Finally, we use the signal-noise ratio and the above three indicators to quantitatively evaluate the denoising effects. The results show there is no autocorrelation in the noise series of each station after denoising, and each evaluation index has significant correction after noise reduction by using LOESS method. LOESS method can effectively reduce the noise of GNSS vertical coordinate time series, and improve the accuracy of the GNSS vertical coordinate time series.