Test Dataset 1

Description of the dataset

Full network

Detail of the network (stations with the open circles are not processed)

Data was processed for the full year of 2000. Three different periods were selected for analysis of the residuals, and to test the residual stacking. The third period, day 297-311, was selected as the main period for which slant delays were computed.

The results of the residual stacking, i.e. improved elevation dependent corrections, improved weighting functions and multipath corrections, have not yet been back-substituted into the analysis to improve the coordinate and ZTD timeseries. However, the multipath corrections were used in the computation of slant delays.

Residual Stacking - Day 297-311 2000

Results of residual stacking in approximately 1 x 1 degree bins over day 297 - 311 of 2000 (15 day period)

Idem, but plotted as function of elevation in 1 degree steps over day 297 - 311 of 2000 (15 day period)

Results of residual stacking in approximately 1 x 1 degree bins over day 311 of 2000 (one day only)

Idem, but plotted as function of elevation in 1 degree steps over day 311 of 2000 (one day only)

Plot of the mean residual, with corrections for multipath from the 15 day maps

Download the results as encapsulated postscript files (eps)

Residual Stacking - Comparison between different periods

Residuals have been stacked over three different periods in 2000. After the first period, there have been some changes in antenna with respect to the second and third, or main, period. This allows us to investigate the effect of antenna changes on the one hand, and sensitivity and repeatability of the residual stacking on the other hand.

The stacked residuals as function of elevation in 1 degree steps are show here

Stacked residual as function of azimuth and elevation, in approximately 1 x 1 degree bins, are shown here

Map of the mean residual (multipath)

The change of antenna had a dramatic effect on the coordinate time series, as can be seen below for Delft, Eijsden and Terschelling. The reference system is ETRS89.

Delft (DELF)

Eijsden (EIJS)

Terschelling (TERS)

The reason for the artifical jumps in the height is the mismodelling of the elevation dependent phase delay due to the antenna and radome that are used. The size of the jumps was

Delft 10.7 mm

Eijsden 17.5 mm

Terschelling 13.3 mm

It can be expected that similar jumps occur in the ZTD, although more difficult to spot because of the higher dynamics of the ZTD. In other words, the bias in the ZTD is changed when antenna's are replaced.

Zenith and Slant Delays

The dataset with slant delays is given here. This is an protected area and you are requested to enter the same userid and password as used on the main tough site for the private area. See also the readme file.

The slant delay is computed from

   SD = hmf(el) * ZHD_0(h) + wmf(el) * (ZWD) +   SD_nonisotropic

      =       SDH          +         SWD     +   SD_nonisotropic

        <---- isotropic slant delay ------->

with hmf(el) Niell's dry mapping function, wmf(el) Niell's wet mapping function, ZHD_0(h) the a-priori zenith dry delay (constant per station), from Saastamoinen, using a standard atmosphere and the height of the station as only parameter (computed from the XYZ coordinates), ZWD zenith wet delay from ZWD = ZTD - ZHD_0, with ZTD estimated total zenith delay (Bernese: EW_00${doy}.TRP) and the SD_noniostropic computed from the multipath corrected residuals.

The Zenith delays are estimated once per hour, whereas the slant delays are given every 30 seconds. This causes problems in the present computation of the slant delays, as illustrated in the two plots below.

Zenith Wet Delay (Delft, day 303)

Slant Total Delay, normalized by hydrostatic mf (Delft, day 303)

To remedy this several options are open

Recompute the ZTD for every epoch. Fortunately, this can be done from the residuals themselves, which actually involves reestimating the ZTD, receiver and satellite clocks on an epoch by epoch basis. Furhtermore, smoothing of ZTD's may be needed
Compute the slant delays using the same interval as the ZTD's. Obviously, this will not work with one ZTD per hour, but the estimation interval for the ZTD's can be changed to e.g. 15 minutes
Treat the non-isotropic slant delay seperately from the ZTD's.
Use double (or single) differences of the slant total delays

At the moment all options are still considered.