GPS Processing Notes

Notes about GPS processing and the creation of GPS time series.

Introduction

Last updated: 2 April 2009

Before making any scientific interpretation of GPS time series data, one should have a full understanding of at least the following items:

• how the data are collected;
• equipment or other changes at a site that may affect the data;
• the data processing;
• any filtering or other adjustment of the time series.

GeoNet provides site metadata, including equipment changes, in the DELTA database. Only part of the database is available to external users, but the information for each site commonly needed for processing GPS data is available as a GeoNet GPS web service.

GeoNet attempts to ensure that the information in DELTA is transcribed accurately to the RINEX headers, and that the same information is used during GPS processing.

GPS daily processing using Bernese software

The GPS processing is done using Bernese v5.0 software developed by the Astronomical Institute of the University of Bern (AIUB), using relatively modern models. However, we do not at present use the latest models and conventions recommended by IERS and IGS. Without going into detail, some of our processing strategies are as follows:

• Solutions calculated each day for data 1, 3, 30 and 90 days behind real time;
• IGS final orbits and earth orientation parameters for data older than 2 weeks;
• IGS rapid orbits and earth orientation parameters for data more recent than 2 weeks;
• Orbits are not solved for;
• IGS relative antenna phase patterns from PHAS_COD.I01 (downloaded from AIUB);
• No satellite antenna phase patterns;
• Ocean loading from ocean tide model TPXO.7.1 calculated by Onsala Space Observatory;
• Global ionospheric model downloaded daily from AIUB;
• Solution done as ~10 clusters, each of < ~20 stations, that are combined using ADDNEQ2;
• Solution aligned to IGb00 realisation of ITRF2000, using a 3-parameter Helmert transformation onto the IGb00 coordinates (propagated to the correct epoch) of a regional network of ~7 sites in central and eastern Australia, the south-west Pacific and the Antarctic.

Three files that are essential for the New Zealand GPS processing are the a priori coordinate and velocity files, and the site information file. Recent versions of these files will soon be available for downloading.

We attempt to keep the processing methodology consistent throughout the time series.

It is our aim to update the processing methodology to use the latest models and conventions, and perhaps to align the solution to a global, rather than a regional, set of stations. We will probably not attempt this until the IGS has completed the reprocessing project that is currently underway.

We also aim eventually to make daily solutions available in loosely-constrained SINEX format so that the solutions can be combined by users who have appropriate software. These users will then be able to generate their own versions of the time series.

Time series

The coordinates and their formal uncertainties are extracted from the daily solution files, and converted to (east, north, up) displacement in millimetres from an initial point.

• For stations in the LINZ PositioNZ network, the initial point is LINZ’s NZGD2000 position of the GPS station.
• For other New Zealand stations, the initial point is either the NZGD2000 position of the station, or is the first point in the time series (a long-term aim is to use the NZGD2000 position for all stations).
• For stations outside New Zealand, which do not have NZGD2000 coordinates, the initial point is the ITRF2000 position (IGb00 realisation) evaluated at 00:00 UT 1 January 2000.

Some “outlier” data points, which are far different to the neighbouring data points, are removed from the time series using an automated procedure. The resulting “raw” time series have no other adjustments made to them, so they may, for example, contain offsets due to earthquakes, offsets due to equipment changes at individual sites, and seasonal signals due to various causes. They also contain a “common-mode” noise signal that is reasonably constant across a wide region. This may arise from the use of slightly erroneous satellite orbits, regional-scale or global-scale mass redistributions (in the ocean, atmosphere or groundwater), the use of non-optimal models in the daily processing, and perhaps other causes. Because this signal is relatively constant across a region it can be estimated by some form of averaging, and subtracted from each time series. This “regional filtering” produces a cleaner-looking time series in which signals of interest may be more clearly visible.

Regionally-filtered time series

To create the regionally filtered time series we in general follow the procedure described by Beavan (2005)¹, treating the east, north and up components at each site separately.

A set of reference stations that are as free as possible of non-linear signals is chosen to estimate the common-mode signal, which is done by simple averaging. We presently use the following reference stations, adopting a different set for the North and South islands.

• North Island: GLDB, MAHO, HAMT, AUCK, WHNG, CHAT
• South Island: GLDB, QUAR, HOKI, WEST, MTJO, OUSD, MQZG

To prepare the reference stations for averaging, we:

• remove any offsets in the time series due to known equipment changes at the station (Table 1), or to known coseismic offsets (Table 2);
• subtract a best fitting linear trend from each time series;
• take the daily mean value point-by-point through the time series, including all reference station solutions that exist on each day;
• remove any very large outliers from the mean time series;
• subtract a best fitting linear trend from the mean time series.

The result is the common-mode time series.

Then, for each raw time series, we:

• remove any offsets in the time series due to known equipment changes at the station (Table 1) (we do not remove coseismic offsets as these represent real ground deformation data²);
• subtract the common-mode time series;
• subtract a best-fitting linear trend and annual sinusoid;
• remove outliers from the result;
• add back the linear trend and the annual sinusoid (but not the offsets due to equipment changes).

We then remove the identified outliers from the raw time series, and repeat these five operations. This is to ensure that all identifiable outliers are removed. The result is the regionally-filtered time series for each station.

Users can see the effect of the outlier removal and regional filtering by comparing the “raw” and “regionally-filtered” time series available on the GeoNet ftp site.

¹ Beavan, R.J. 2005 Noise properties of continuous GPS data from concrete pillar geodetic monuments in New Zealand, and comparison with data from US deep drilled braced monuments. Journal of geophysical research. Solid earth, 110(B8): B08410, doi:10.1029/2005JB003642.
² A better strategy would be to remove an estimate of the coseismic offsets at this stage, and to add these estimates back at the end of the regional filtering process. This would prevent the coseismic offsets from potentially biasing the fitting of the trend and annual cycle. We will change this procedure in due course.

Table 1. Equipment offsets used in regional filtering

Table 2. Coseismic offsets used in constructing common-mode time series