Objectives Latest release of international terrestrial reference frame has been generated with an enhanced modeling of nonlinear station motions, while more investigations on the strategies applied for the terrestrial reference frame modeling nonlinear station motions still needed.

Methods A global GPS (global positioning system) network has been reprocessed, using the latest GNSS (global navigation satellite system) data processing models and an integrated strategy, and then accurate and consistent input data are obtained; the post-seismic deformation models of 13 CMONOC (crustal movement observation network of China) stations are established and used as the input data, and FREQ2016 based on the function model, and GFZ2016 and EOST2016 considering the environmental loading correction are realized; these GPS terrestrial reference frames with different models are compared in terms of the goodness for fit, velocities and seasonal signals, and then the optimal model is explored.

Results The results show that the weighted root mean square averages of the postfit residuals of FREQ2016 are smaller than those of GFZ2016 and of EOST2016; FREQ2016 shows more significant seasonal signals, especially for semi-annual signals; the horizontal velocity field differences of these three are almost negligible (< 0.08 mm/a), and the large vertical velocity differences, for some stations, reach 1.3 mm/a; the reduction of FREQ2016 horizontal velocity formal errors is about 10%.

Conclusions This paper suggests that the strategy based on the function model is better than that of considering the environmental loading correction.