This paper addresses the central problem of global positioning system (GPS) time series analysis: the separation of what is considered signal, that is, the systematic variations, from what is considered to be noise, that is, the random variations of usually unknown origin. A functional model is commonly fit to a GPS position time series to represent the signal, consisting of a linear trend, periodic variations, and irregularly timed position offsets as needed. It was realized in the 1990s that weighted least squares fitting for the functional model with formal error propagation yields highly optimistic uncertainties for geodetic parameters if time-correlation is not accounted for (Johnson & Agnew, 1995). Then the discovery that the GPS-based errors are correlated both spatially (Wdowinski et al., 1997) and temporally (Zhang et al., 1997) led to more robust methods to assess and quantify the levels and types of noise in time series of GPS station positions (Mao et al., 1999). The main goal was to obtain more reliable uncertainties for GPS-based velocities, which are needed for many geodynamical studies such as measuring intra-plate tectonic stability and correcting vertical land motion from tide gauge records. A general consensus quickly emerged that GPS time series errors combine mostly white noise (WN) at shorter periods, typically less than a month, with flicker noise (FN) or similar power-law noise (PL) over longer spans, from monthly up to decadal periods (Mao et al., 1999; Zhang et al., 1997), although some level of background random walk (RW) error cannot be excluded. Consequently, velocity uncertainties can be underestimated by as much as an order of magnitude if WN alone is assumed (Mao et al., 1999). Development quickly followed of several mathematical tools to enable users to evaluate models of alternative noise types and amplitudes in observed GPS time series (Amiri