GPS Time Series Analysis Using BFAST Algorithm

With the development of GPS technology and the improvement of positioning precision, GPS technology has played an important role in more and more fields. At present, the Continuous Operational Reference System has accumulated up to several years or even decades of observational data. GPS time series analysis can obtain the plate motion, ground surface settlement, earth oscillation cycle information, and other applications. However, the accuracy of GPS time series is degraded by offsets, seriously deteriorating the accuracy of tectonics and non-tectonics signals derived from GPS time series analysis. Especially the unknown-reason offsets, usually too small to be reliably distinguished from noise, have no effective detection methods. Offsets detection method in GPS time series can be divided into two categories:manual methods and automated or semi-automated methods. At present, manual methods almost give better results than automated or semi-automated methods. With increasing of the amount of GPS stations, the length of time series data increases rapidly, and it is inefficient for manual detection method to handle the offsets in massive data. Therefore it is very necessary and urgent to find the automated detection method in GPS time series analysis.BFAST(stands for Breaks For Additive Seasonal and Trend) integrates the decomposition of time series into trend, seasonal, and remainder components with methods for detecting change within time series. BFAST iteratively estimates the time and number of changes and characterizes change by its magnitude and direction. According to the characteristics of the GPS time series, we can apply the BFAST algorithm to GPS time series analysis with two restrictions:1.the velocity before and after the offset remains unchanged in determine secular rate for long station position time series, while the original BFAST algorithm using piecewise linear fitting to estimate the slope with the different velocities in each sections; 2. Offsets in the season component are inexistent in GPS time series.We design simulative GPS time series based on the noise level of the GPS time series data from Crustal Movement Observation Network of China. Then, we combine the time series together with the artificial offsets with different magnitudes, and use the improved BFAST algorithm to detect the offsets. Monte Carlo method is employed to test the feasibility of BFAST algorithm. Results show:BFAST has good capacity in detecting offsets in simulated GPS time series, and exports linear trend and seasonal component of high quality simultaneously. The accuracy of detecting offsets increases with the increasing of the offsets amplitude, while the accuracy of the fitting trend and the seasonal components decreases with the increase of the noise amplitude.Additionally, BFAST was applied to the GPS time series analysis from Crustal Movement Observation Network of China, achieving the comparative results of the offsets, trend, and seasonal components released by Crustal Movement Observation Network of China. The accuracy of the estimation of seasonal component is very good, with discrepancy of less than 2mm in the seasonal annual and semi-annual amplitudes between the BFAST-estimated values and the released values. The fifth percentile ranges in velocity bias for BFAST in north, east and up components are1.45mm/yr,2.38mm/yr, and 3.63mm/yr, respectively. The estimated velocities are in good agreement with the released values for most stations, while the inconformity of detected offsets, to a large extent, generates the difference of velocity.Finally the improved BFAST algorithm is compared with other semi-automated methods for offset analysis, suggesting that the improved BFAST algorithm has high ability in estimation of the offsets, trends and seasonal components. The improved BFAST algorithm would be used as an automatic method in GPS time series analysis for the advantages of high precision, strong robustness, automation and so on.