Comparison of GPS strain rate computing methods and their reliability

Abstract

Using modelled and simulated data for comparison of several methods to compute GPS strain rate fields in terms of their precision and robustness reveals that least-squares collocation is superior. Large scale (75°E–135°E and 20°N–50°N) analyses of 1° grid sampling data and decimated 50 per cent data by resampling (then erasing data in two 5°× 10° region) reveal that the Delaunay method has poor performance and that the other three methods show high accuracy. The correlation coefficients between theoretical results and calculated results obtained with different errors in input data show that the order in terms of robustness, from good to bad, is least-squares collocation, spherical harmonics, multisurface function and the Delaunay method. The influence of data sparseness on different methods shows that least-squares collocation is better than spherical harmonics and multisurface function when sample data are distributed from a 2° grid to a 1° grid. Analysis to medium scale (90°E–120°E, 25°N–40°N) in 1°–0.5° grid sampling data reveals that least-squares collocation is superior to other methods in terms of robustness and sensitivity to data sparseness, but their difference is slight. Strain rate results obtained for the Chinese mainland using GPS data from 1999 to 2004 show that the spherical harmonics method has edge effects and that its value and range increase concomitantly with increased sparseness. The multisurface function method shows non-steady-state characteristics; the errors of results increase concomitantly with increased sparseness. The least-squares collocation method shows steady characteristics. The errors of results show no significant increase even though 50 per cent of input data are decimated by resampling. The spherical harmonics and multisurface function methods are affected by the geometric distribution of input data, but the least-squares collocation method is not.