Uncertainty Considerations for the Comparison of Water Vapour Derived from Radiosondes and GNSS

Abstract

The integrated water vapour (IWV) can be estimated from the tropospheric delays of GNSS signals. These estimations are usually validated by radiosonde observations. However, very limited information is available on the precision of the IWV determined by radiosondes. In this paper the methodology of the computation of IWV retrieved from radiosonde data is revised using the atmospheric profiles of pressure, relative humidity and temperature. The formulae to calculate the uncertainty of the estimated values are derived, where the correlation of the neighbouring atmospheric layers is also taken into account. The results show that the mean uncertainty of the IWV from radiosonde observations reaches the level of ±0.26 kg/m² in case of the Vaisala RS-92 radiosondes in Central and Eastern Europe. However, it increases to ±0.7 to 0.8 kg/m² in summertime.Since the zenith hydrostatic delay (ZHD) must be modeled accurately to estimate the IWV from GNSS observations, the Saastamoinen, the Hopfield and the Black tropospheric delay models have been validated with ZHD values computed from radiosonde observations in Central Europe. Moreover some local models have also been derived in order to minimize the bias in IWV caused by the existing tropospheric models. In order to take the effect of the masses above the topmost level of the radiosonde profile in consideration, the International Standard Atmosphere has been used. Since the radiosonde observations terminate at different altitudes and pressure levels, which certainly affect the accuracy of the computed ZHD values, the omission error has been modeled with a simple exponential function. The results showed that the best ZHD model fitted to the radiosonde observations with the bias and standard deviation of +0.8 and ±1.2 mm, respectively. This means that the GNSS derived IWV is biased by −0.1 kg/m². This value is approximately 50 % lower than the bias caused by the Saastamoinen model.Finally, the calculation of the scale factor between the zenith wet delay (ZWD) and the IWV is studied. Various models exist to determine this scale factor. There are models that derive the scale factor as a direct function of the surface temperature, while other models use a linear regression model of the surface temperature to compute the mean temperature of water vapour in the troposphere and derive the scale parameter from physical equations. Radiosonde profiles were used to test the two approached in Central and Eastern Europe. The results showed that the prior model showed no bias, while the latter one showed a relative bias of approximately 0.3 %.

Keywords