The impact of observation losses on IVS-R1/R4 VLBI sessions

TL;DR

This work tackles the problem of observation losses in IVS-R1/R4 VLBI sessions and their impact on geodetic parameter estimation. It applies the VieSched++ simulator to 1030 sessions from 2014–2023, comparing the ideal, scheduled observations with the subset usable in analysis to quantify degradation and derive a practical scale factor for simulations. The findings show a median 25.3% loss of observations, translating to median degradations of about 18.8% for UT1-UTC, 19.2% for 3D station coordinates, 11–12% for X/Y nutation, and 22–29% for x/y polar motion; X/Y nutation is relatively robust while polar motion is highly sensitive. The results support applying a scale factor when using simulations to forecast real-world performance and highlight the need to reduce data loss through scheduling improvements, automation, and better network communication to safeguard geodetic accuracy.

Abstract

Global VLBI observations, to measure Earth orientation and station positions, are organised into 24-hour sessions. Each session has a bespoke schedule created, optimised for the particular time period and the station network that is available during it. Due to various factors, whether it be station outages, sensitivity issues or source effects, not all scheduled observations are available, or of sufficient quality, to be included in the final geodetic analysis. In this paper we derive statistics about the number of missing observations, as well as their effect on the expected precision of geodetic parameters such as station positions and Earth Orientation Parameters. We investigate the impact of observation loss on the weekly rapid turnaround IVS-R1 and IVS-R4 geodetic VLBI sessions over a decade period from 2014 - 2023. Across our 1030 sessions we find on average 25.3\% of observations scheduled do not make it to analysis. This results in median performance losses, when compared to the scheduled versions, of 18.8%, 19.2%, 12.1/11.3% and 28.7/22.9% for UT1-UTC, 3D station position, X/Y nutation and x/y polar motion respectively. We find that the estimation of X/Y nutation is particularly robust to typical observation loss seen from these 24-hour sessions. Conversely, we see high-rates of critical degradation in performance (a doubling of the scheduled repeatability) for other geodetic parameters at observations losses of between 15 - 19%, which is less than the median loss of 25.3% that we find across this 10-year period.

Figures (5)

• Figure 1: Top panel: Distribution (number of sessions) for the number of observations contained within the vgosDB of a session versus scheduled. Middle panel: Same as above, with the additional filter of the observations requiring a quality code $\ge5$. Lower panel: Distribution (number of sessions) of number of observations used versus scheduled in R1 and R4 IVS analysis reports over the same 10 year period. The vertical dashed lines represent the median values for each distribution.
• Figure 2: Histograms (number of sessions) of parameter performance between the session as observed versus the theoretical perfect session. For example, a value of 0.5 represents the observed session having twice the repeatability of the ideal scheduled case case. Vertical dashed line represents the median value of each distribution.
• Figure 3: Annual median values of relative performance for all considered parameters. The blue columns represent the annual median number of observations scheduled per session.
• Figure 4: Scatter plot for each parameter of relative performance versus fraction of observations that make it to analysis. Red lines represent the square root relationship ($f(x) = \sqrt{x}$) that is expected from a sample of independent, equally weighted data points. The blue smoothing spline displays the broad trend of the data for ease of comparison. Shaded areas represent the minimum levels of observation loss where 10% or more sessions have double the scheduled repeatability (relative performance $<0.5$).
• Figure 11: Histograms (number of sessions) of parameter performance comparing R1 and R4 sessions over the 10 year period. Vertical dashed line represents the median value of each distribution. Only parameters with distributions deemed significantly different (based on the Kolmogorov-Smirnov tests discussed in Section \ref{['sec:r1_vs_r4']}) have been included here for comparison.