When using a network approach, expressing reliably GNSS position and velocities in a given reference frame (ITRF2014, IGS14, ETRF2000 or ETRF2014) requires the identification of ‘stable’ and ‘reliable’ reference stations. The choice of these reference stations can have a non-negligible impact on the estimated positions and velocities and of course on the derived geodynamic interpretations.

The EPN network has been set up to offer such reference stations: the reference positions and velocities of the EPN stations are kept up to date and are available through the EPN multi-year position and velocity solution. However, not all EPN stations are by definition suitable of reference stations.

To help the identification of the best EPN reference stations, a new station classification was developed. The criteria and thresholds used to define different classes for the EPN stations based on their performances are presented here.

In addition of the classification, we also developed a web tool to help the user to select the the most optimal EPN reference stations in a considered area and for a given observation time. The web tool is accessible here.

The Criteria

Time Series

The residual position time series of the EPN stations are analyzed to quantify:

  • the amplitude of the annual signal A(1Y)N, A(1Y)E, A(1Y)U on North, East and Up components
  • the root mean square of the residual position time series after removing the annual and semi-annual signals RMSN, RMSE, RMSU on North, East and Up components

Annual Signal
Figure 1: Histogram of the amplitude of the annual signal A(1Y)N, A(1Y)E, A(1Y)U on North, East and Up components for the EPN stations
RMS of the residual position time series
Figure 2: Histogram of the RMS of the residual position time series RMSN, RMSE, RMSU on North, East and Up components

Reliability of the Velocity Estimation

The EPN multi-year positions and velocities are estimated using CATREF software (Altamimi et al., 2007). CATREF is based on a least squares adjustment and neglects temporally correlated noise affecting the position time series. Therefore, the estimated velocity errors are too optimistic.
In order to derive more realistic error estimates, the Hector software (Bos et al., 2013) has been used to estimate a linear trend, an annual and semi-annual signals assuming a temporal correlated noise (power-law + white noise). The Hector software (Bos et al., 2013) has been used to derive realistic error estimates. The comparison between Hector and CATREF velocity estimates is used to assess the reliability of the velocity estimation.
The criteria used are:

  • the realistic velocity error estimates from HectorVN, σVEVU) allow to assess the quality of the stations
  • the differences between CATREF and Hector estimations (dVN, dVE, dVU) allow to assess the reliability of the velocity estimation

Hector Sigma
Figure 3: Histogram of the velocity error estimates from Hector (σVN, σVEVU) for the EPN stations
Hector Catref velocity difference
Figure 4: Histogram of the velocity differences between CATREF and Hector (dVN, dVE, dVU) on North, East and Up components for the EPN stations

Velocity Variability

The variability over time of the estimated velocity for a station is obtained by comparing the velocities estimated using data from different moving time windows with the velocities obtained using the full data set of the station as it is included in the multi-year EPN solution.

Project Image
Figure 5: Residual position time series for the station KIRU00SWE (left) and velocity variability plot for the station (right) using a 4-year moving window
Project Image
Figure 6: Residual position time series for the station KIRU00SWE (left) and velocity variability plot for the station (right) using a 8-year moving window


For each station and each component, the velocity variability NdV, EdV or UdV is then quantified as the RMS of the agreement between the station velocities obtained from the moving time windows (> 4 years) and the velocity obtained using the full time series.

Hector Catref velocity difference
Figure 7: Histogram of the velocity variability NdV, EdV or UdV on North, East and Up components for the EPN stations

The Thresholds and the Classification

Based on those criteria, the information has been organized and thresholds have been defined in order to end up with the station classification.

Thresholds

In order to find the stations having the best performances in the EPN network, the thresholds used are based on 3 different percentiles: percentile 75, 85 and 95. If the threshold is fixed at percentile 75, it means that the 25% worst stations for the considered criterion will be excluded. The idea behind this choice is that we want to reject the station that have the worst performances for each criterion.

Figures 8 and 9 show the values for each of the criteria for the stations ACOR00ESP (Figure 8) and BOR100POL (Figure 9).

ACOR00ESP
Figure 8: Plots for the station ACOR00ESP
BOR100POL
Figure 9: Plots for the station BOR100POL

Classification Rules

Table2: the rules applied to define the 8 station classes C0, C1, C2, C3, C4, C5, C6, Short

Name Number Criteria Comment
Velocity
variability
Timeseries
RMS
Amplitude
1Y signal
DVCatref-Hector σHector
C0 64 < Percentile 75 < Percentile 75 Most Stable Stations
C1 38 < Percentile 75 < Percentile 85
C2 47 < Percentile 75 No threshold < Percentile 85 Stable but Noisy or with Seasonal Signals
C3 19 < Percentile 85 < Percentile 85 Less Stable
C4 11 Not Available - Short time series < Percentile 85
C5 72 < Percentile 85 No criteria > Percentile 95 Even Less Stable
C6 146 velocity variability > Percentile 85 and/or 1 or more other criteria > Percentile 95
or Short time series with 1 or more criteria > Percentile 85
Less Reliable
Short 74 < 3yr - not applicable No velocity published