Detecting the probable offsets is a crucial step in the preprocessing of the Global Positioning System (GPS) coordinate time series. Undetected offsets lead to the biased estimation of time series parameters and their uncertainties resulting in data misinterpretation. In the current research, a DIA (detection, identification, and adaptation)‐based procedure in maximal overlap discrete wavelet transform (MODWT) rough space has been introduced to address the location of offsets in long GPS time series without a priori information of the functional or stochastic models. A remarkable property of a wavelet rough (WR) at lower‐scale (j ≤ 5) details is to reflect the local regularity of the time series, being large where the signal is irregular and small where it is smooth. Performance and effectiveness of the proposed approach have been shown with DOGEx (Detection of Offsets in GPS Experiment) data set, which was a simulated time series that mimicked realistic GPS data consisting of a velocity component, seasonal cycle, offsets, and white and flicker noises composed in an additive model. The results showed that the fifth percentile range (5% to 95%) in velocity biases was equal to 1.24 mm/yr, which was smaller than all tested automatic solutions. Furthermore, the offsets detected by this method were split into 34.3% of true positive (TP), 36.5% of false positive (FP), and 29.2% of the false negative (FN), offering the proposed method as the best automatic solution.

中文翻译：

---

基于WR的统计实验新方法在GPS实验中的偏移检测。

在全球定位系统（GPS）坐标时间序列的预处理中，检测可能的偏移量是至关重要的一步。未检测到的偏移量会导致时间序列参数及其不确定性的估计偏差，从而导致数据误解。在当前的研究中，已经引入了在最大重叠离散小波变换（MODWT）粗糙空间中基于DIA（检测，识别和自适应）的程序，以解决长GPS时间序列中的偏移位置，而无需功能的先验信息。或随机模型。在比例较低（小波粗糙（WR）的一个显着特性Ĵ  ≤5）细节是为了反映时间序列的局部规律性，在信号不规则的情况下较大，而在信号平滑的情况下较小。已通过DOGEx（GPS实验中的偏移量检测）数据集展示了该方法的性能和有效性，该数据集是模拟的时间序列，模拟了真实的GPS数据，包括速度分量，季节周期，偏移量以及白噪声和闪烁噪声由加法模型组成。结果表明，速度偏差的第五个百分位数范围（5％到95％）等于1.24 mm / yr，小于所有测试的自动解决方案。此外，通过该方法检测到的偏移被分为34.3％的真阳性（TP），36.5％的假阳性（FP）和29.2％的假阴性（FN），从而提供了建议的方法作为最佳的自动解决方案。