Improving offset detection algorithm of GNSS position time-series using spline function theory

Abstract

SUMMARYIt is well known that unmodelled offsets in Global Navigation Satellite System (GNSS) position time-series can introduce biases into the station velocities. Although large offsets are usually reported or can be visually detected, automated offset detection algorithms require further investigation. This problem is still challenging as (small) geophysical offsets are usually covered by coloured noise and remain undetected. An offset detection algorithm has recently been proposed, which can detect and estimate offsets in both univariate and multivariate analyses. Although efficient in truly detecting offsets, this method still suffers from a high rate of detected fake offsets. To improve the offset detection performance, we attempt to stabilize the offset power spectrum to reduce the number of false detections. The spline function theory is adopted in the smoothness process of the power spectrum. The algorithm modified using the spline functions, referred to as As-mode, is compared with its original counterpart, called A-mode. The GNSS position time-series consisting of a linear trend, seasonal signals, offsets, and white plus coloured noise are simulated for the numerical comparison. The overall performance of the algorithm is significantly improved using the As-mode algorithm. The multivariate analysis shows that the truly detected offsets' percentage (true positive) increases from 52.9 per cent for A-mode to 61.1 per cent for As-mode. Further, the falsely detected offsets' percentage (false positive) is reduced from 40.6 per cent to 29.8 per cent. The algorithm was also tested on the DOGEx data set. The results indicate that the proposed method outperforms the existing solutions, with TP, FP and FN being 33.3 per cent, 32.3 per cent and 34.4 per cent, respectively. Also, in 90 per cent of the station, velocities are estimated at a 0.8 mm yr−1 distance from the simulated values.