[unav_all] status of IGS orbit products

Jim.Ray jim.ray at noaa.gov
Thu Jan 7 13:06:23 MST 2010

Author: IGS Analysis Center Coordinator

The accuracy of the IGS Final orbits is about 2 cm.  Errors in the along-
and cross-track directions are about 75% larger than in the radial, but
correlations are significant.  The predominant errors are near the semi-
annual (and/or 2nd GPS draconitic harmonic at 175.6 d), the 4th draconitic
harmonic (87.8 d), and fortnightly bands due to modeling deficiencies.
Users should exercise great caution in the interpretation of signals near
these bands in their application results.  The higher-frequency precision
at a few days and shorter is around 7 mm due mostly to rotational scatter
followed by quasi-random variations.  The rotational errors are about equal
in all three components and are probably caused by effects of orbit
mismodeling and reference frame realization.  The performance of the IGS
Rapid orbits is very similar, including sharing the common long-period

The high-frequency precision of the IGS near real-time Ultra-rapid observed
orbits is only about 40% poorer than the later Rapids and is about 3.5 times
worse than the Rapids for the 6-hr predictions.  Rotational scatter also
dominates the Ultra-rapid precision, but much more so for the RZ axial than
the equatorial components.  EOP prediction errors are mainly responsible for
this.  The daily 1D quasi-random WRMS scatter is about 2 cm for the 6-hr
orbit predictions, increasing to nearly 5 cm for predictions over 24 hr.

IGS Product Lines
The main production attributes of each IGS product line are summarized below.

                            IGS Core Product Lines
Product        Acronym    Latency     Update Times    Data Integration Spans
Series                                    UTC                  UTC
Ultra-rapid (GPS only) -- for real-time & near real-time applications:

  predicted       IGU     real-time     03h, 09h,            +- 12 h @
  half                                  15h, 21h        12h, 18h, 00h, 06h
  observed        IGA      3 - 9 h      03h, 09h,            +- 12 h @
  half                                  15h, 21h        12h, 18h, 00h, 06h

Rapid (GPS only) -- for near-definitive but rapid applications:
  observed        IGR     17 - 41 h     17h daily          +- 12 h @ 12h

Final (GPS & GLONASS, separately) -- for definitive applications:
  observed        IGS     11 - 17 d    weekly each         +- 12 h @ 12h

One important feature seen here is the absence of any GLONASS products with
low latency.  It is expected that during 2010-2011 more ACs will begin to
contribute GLONASS products, including some with lower latencies.

Some characteristics of the IGS product inputs are summarized below assuming
all Analysis Center (AC) solutions are available.  In many cases, especially
for the IGUs and IGRs, not all AC submissions are received in time to be
included.  Altogether there are 12 different ACs contributing to the
generation of these products.  One new candidate AC is pending for the Finals
and another AC will soon discontinue their GLONASS processing.

                            IGS Product Inputs
Product          Product           # of Contributing ACs             Output
Series            Type            submit   reject*   used          Intervals
Ultra-rapid:   GPS orbits            7        2        5             15 min
                GPS SV clocks         4        1        3             15 min
                ERPs                  7        2        5              6 h

Rapid:         GPS orbits            8        0        8             15 min
                GPS SV clocks         6        1        5              5 min
                stn clocks            6        1        5              5 min
                ERPs                  8        2        6              daily

Final:@        GPS orbits            8        0        8             15 min
                GPS SV clocks         6        0        6              5 min
                GPS SV clocks         3        0        3             30 sec
                stn clocks            6        0        6              5 min
                GLO orbits            4$       0        4$            15 min
                GLO SV clocks         2        0        0              none
                ERPs                  8        1        7              daily
                TRF                   8        0        8             weekly
  * contributions usually not included in combinations
  $ one GLONASS contribution will end in early 2010
  @ one new GPS contribution for the Finals will start in 2010

Most IGS products have adequate AC redundancy.  However, this is not the
case for IGU GPS clocks nor any GLONASS products.  The IGU products are
particularly fragile because of the strict time requirements so more ACs
would be beneficial.

Orbit Performance Metrics
The table below compares each orbit product line to either the IGR or the IGS
orbits as reference.  Shown are the mean and standard deviations over year
2009 for the daily seven Helmert parameters and the RMS, wtd RMS, and median
orbit residuals, averaged over the satellite constellations.  The residuals
are computed from 1D geocentric position differences.  A Total Relative Error
is computed from the RSS of the systematic Helmert differences and the mean
quasi-random WRMS residuals.  This metric is a somewhat pessimistic accuracy
measure since the rotational errors are expressed as equatorial (i.e.,
maximal) displacements and some user applications are more sensitive to
certain error components than others.

                     Orbit Differences wrt Reference Orbits
                                                        1D    1D     1D  TOTAL
          DX    DY    DZ     RX     RY     RZ   SCL    RMS*  WRMS   MEDI ERR $
          mm    mm    mm     mm     mm     mm    mm     mm     mm     mm   mm
                          (equatorial @ GPS altitude)
IGU 6-hr predictions wrt IGR:

  mean   3.5  -0.6   0.3    0.3    0.8    3.1  -0.7   28.9   21.3   15.6  41.7
  SDev   4.7   4.9   3.4   13.8   16.3   27.2   2.6   19.7    8.0    2.6  ====

  + using first 6 hr of predictions from each IGU update
  + 1460 IGUs from 2009-01-01 00:00/1512_4_00 thru 2009-12-31 18:00/1564_4_18

IGU 24-hr predictions wrt IGR:

  mean   1.1   0.3  -0.1   -0.5   -0.6   -0.9  -1.3   64.7   47.3   30.2  80.2
  SDev   1.8   2.0   3.8   21.9   31.2   52.0   1.9   33.3   16.3    6.0  ====

  + using 24 hr of predictions from 00 UTC IGUs only
  + 365 IGUs from 2009-01-01/1512_4_00 thru 2009-12-31/1564_4_00

IGA observations wrt IGR:

  mean   1.2   0.3   0.1   -0.2    0.9    2.6  -1.2    9.0    8.0    7.2  16.3
  SDev   0.8   0.9   1.3    3.4    3.4   12.7   1.5    1.6    1.3    1.2  ====

  + using 24 hr of observations from 00 UTC IGUs only
  + 365 IGUs from 2009-01-01/1512_4_00 thru 2009-12-31/1564_4_00

IGR observations wrt IGS:

  mean  -0.3   0.3   0.2    0.5   -5.3   -4.6   1.2    5.8    5.6    5.1  11.9
  SDev   0.7   0.8   1.2    4.7    3.6    4.6   1.0    0.7    0.7    0.7  ====

  + 360 IGRs from 2009-01-01/1512_4 to 2009-12-26/1563_6

  * unweighted RMS is included for completeness only; users should always
    apply reported SP3 accuracy codes as differential satellite weights
  $ Total Relative Error (wrt Reference) is RSS of all systematic & random
    (WRMS) components

Comparing the IGU 6-hr prediction performance above to that for 2008
reported a year ago (see IGS Mail 5874), we see that the rotational errors,
which dominate, have declined by 34, 33, and 22% for RX, RY, and RZ,
respectively, and that the WRMS residual has dropped by 11%.  The
improvements can be attributed to generally better AC performance and
stricter rejection criteria.  In particular, the two ACs normally excluded
from the IGU combinations have much larger orbit rotations than the other

The largest error component for IGU 6-hr and 24-hr predictions as well as
for the IGA near real-time orbits continues to be RZ rotational scatter,
due to inherently larger errors in predicting UT1 variations.  Quasi-
random WRMS errors are next largest, followed by RX and RY rotational
scatter caused mostly by polar motion prediction errors for the IGUs.

The rotational scatters between IGR and IGS orbits are nearly equal for all
three components, in strong contrast with the IGU/IGA comparisons.  This
suggests that EOP errors do not contribute significantly to the IGR or IGS
orbits and that orbit modeling and/or reference frame rotational effects
are more important.  The mean IGR/IGS rotations are non-zero for RY and RZ,
more so than any other orbit comparisons.  This too points to a non-EOP
source for the rotational errors.  Interestingly, the ratios of the Total
Relative Errors (which are dominated by rotational scatter) to the 1D WRMS
residuals are nearly equal for all four orbit comparisons despite evidently
different sources for the rotational errors: 2.0, 1.7, 2.0, and 2.1,
respectively.  In other words, regardless of the overall orbit accuracy or
the impact of EOP prediction errors, the combined rotational scatter is a
nearly constant factor of ~1.7 greater than the random WRMS scatter for
every product line.

As noted previously, extending the IGU predictions from 6 hr to 24 hr
degrades their net performance by a factor of about two.  This demonstrates
that any delay in delivering IGU products less than ~1 d should have minimal
impact on user applications.  It is also noteworthy that the performance of
the IGA near real-time orbits is only about 40% poorer than the later IGR

Origin and scale variations are minor in all cases.

All of the orbit assessments above rely on comparisons between pairs of IGS
product series, so the individual performance of any one is obscured.  Any
common mode errors are hidden altogether.  To evaluate more objectively the
accuracy of the IGS orbits themselves, the discontinuities at day boundaries
have been computed (e.g., Griffiths and Ray, 2009) for 24 usable satellites
over the period 2005-02-26 to 2007-12-31.  The reprocessed repro1 orbits,
which should be comparable to recent operational Final orbits, have been
used (see acc.igs.org/reprocess.html).  Each daily AC satellite ephemeris
for each pair of consecutive days has been fit to the extended CODE orbit
model, extrapolated to the mid-point epoch between the days, and the
geocentric (and along, cross, and radial) position differences computed to
give time series of orbit repeatabilities. Occasional data gaps have been
filled by linear interpolation.  The mean and standard deviations are given
below for ACR components as well as for the average magnitudes of the 1D
geocentric differences.  Tests indicate that the error introduced by our fit
and extrapolation procedure is less than 4 mm RMS.

                 Day-Boundary Jumps of IGS Reprocessed Orbits $
                   Along        Cross        Radial    1D Geocentric    Number
                   Track        Track                   Differences     of Pts
                     mm           mm           mm           mm
All usable SVs (24)*:
  mean             -0.3         -1.7          0.6          26.1         24572
  SDev             36.8         38.6         21.3          13.0

Block IIA SVs (12)*:
  mean             -4.8          0.0          0.5          24.4         12286
  SDev             34.0         35.9         19.1          11.4

Block IIR/IIR-M SVs (12)*:
  mean              4.1         -3.5          0.7          27.7         12286
  SDev             38.8         41.1         23.2          14.2

  $ results are for the IGS repro1 reprocessed orbits for the period
    2005-02-26 through 2007-12-31
  * the considered satellites are PRNs:
    IIA       -- 01, 03, 04, 05, 06, 08, 09, 10, 24, 26, 27, 30
    IIR/IIR-M -- 02, 11, 13, 14, 16, 18, 19, 20, 21, 22, 23, 28

The mean of the 1D orbit differences is smaller than the average of the ACR
component standard deviations, indicating significant correlations between
error components.  It is surprising to see that all performance metrics for
the older IIA satellites are better than for the modern IIR and IIR-M
satellites, although a similar observation was made previously for a more
recent period (Griffiths and Ray, 2009).  We have no explanation for this
unexpected result.

Using FFT power spectra (see Griffiths et al., 2009) the high-frequency
noise floor is found to be about 10 mm, very similar to the Total IGR/IGS
Relative Error (11.9 mm).  (Note that both these comparisons involve pairs
of orbit differences.)  So the high-frequency precision of the IGR and IGS
orbits must be nearly equal.  The major part of the orbit scatter seen above
for the IGS orbits, about 26 mm, arises from longer period orbit variations
(see Griffiths et al., 2009), mostly semi-annual (probably related to the
twice-yearly eclipse seasons and/or to the 2nd harmonic of the GPS draconitic
year; see Ray et al., 2008), the 4th draconitic harmonic, and a broad
fortnightly band.  These longer-period effects must be mostly common to the
IGR and IGS series since their direct comparison to each other agrees much
better than 26 mm.

Since the day-boundary differences computed above involve orbits for two
consecutive days, the inferred orbit uncertainty for a single day should
be smaller by sqrt(2), or approximately 20 mm.  This estimate is
comparable with recent SLR range residuals (Bar-Sever et al., 2009).

Reprocessed Results
The repro1 reprocessing is being finalized now.  It is expected that results
will be posted by springtime.

--Jim Ray & Jake Griffiths


ACC website:

Status of IGS Ultra-rapid products (IGS Mail 5874):

IGS data reprocessing campaign repro1:

Bar-Sever et al. (2009):

Griffiths and Ray (2009):

Griffiths et al. (2009):

Ray et al. (2008):

More information about the unav_all mailing list