Local reference frame bias in the precise daily coordinates of Japanese continuous GNSS array

Accurate and homogeneous coordinate time series of GNSS data over decades is essential for investigation of various geophysical phenomena. In Japan, precise daily coordinates of GEONET, the nationwide GNSS continuous observation network, named as F3 solution, are estimated with a relative positioning method. The F3 solution may be biased because 1) the shape of the network has been changing in response to the increase of sites, 2) an unstable behavior of one site affects the entire network, and 3) the reference frame of the F3 solution is not always consistent with that of input data (e.g. satellite ephemeris). In order to evaluate accuracy and stability of the F3 solution, we reanalyze the daily coordinates of 20 GEONET sites all over Japan from 1996 to 2017 by Precise Point Positioning (PPP) with the Bernese GNSS software ver.5.2 and GFZ repro2 products. Since PPP makes it possible to obtain each site coordinates independently, and we analyze in a fully consistent way, our solution is expected to be less biased. By comparing F3 solution and our PPP solution, we evaluate the bias in F3 solution. We estimate daily Helmert parameters to transform our PPP solution to the F3 solution. After 2004, the scale parameter indicates a systematic bias of +2 3 ppb, which means that the F3 network is defined a little larger than its actual shape. Furthermore, 3 parameters of translation show that F3 network is deviated by 2 7 cm from the true position. All 7 parameters vary as time passes. This means that the bias is not constant but time-dependent. From the point of view of local coordinate, we find that 1) the bias in horizontal component is common to the entire analysis network, 2) in horizontal component, the difference of two solutions is 0.6 cm before 2012 and 1.5cm after 2012, and 3) in vertical component, sites located at lower latitude shows larger bias ( 2.3 cm) than those at higher latitude ( 0.7 cm) after 2012. The bias rapidly changes after 2012 probably because the F3 solution refers to ITRF2005 continuously, although IGS switched its reference frame to ITRF2008. The cause of the bias is unclear, but we find that the difference of tropospheric delay between F3 and our PPP solution is correlated with the coordinates. It is possible that this correlation is responsible for the bias.