Tropospheric delay models for GNSS in Indonesia

Abstract

Tropospheric delay is one of the major error sources in space geodetic observations, including GNSS. It is important to treat these delays accordingly to improve the accuracy of the observations. The tropospheric delay models VMF1 and GMF are widely used in GNSS data analysis. Recently, a new and refined version of VMF1, VMF3 was released. This study focuses on assessing the performance of VMF3 for GNSS observations in Indonesia. Observations from eleven stations of the InaCORS network for the year 2014 were analyzed using PPP technique in Bernese GNSS Software, version 5.2. Currently, the software does not support VMF3 and therefore, some changes in the subroutines had to be made to allow Bernese to apply VMF3 in the analysis. The estimated horizontal coordinates from the three models VMF1, GMF, and VMF3 are generally similar, whereas the mean absolute biases for the height component are larger, typically more than 1 mm. Additionally, while the three models yield similar repeatability for the horizontal components, using either VMF3 and VMF1 provides slightly better results than GMF. The estimated ZWD from all the three models are similar. However, the biases between VMF3-VMF1 are smaller than VMF3-GMF. The biases in ZWD correspond to the biases in the height component. The mean absolute biases of ZWD are roughly half of the biases in height. The slant delays at 5 degrees are around 22-23 meters and 2-3 meters for the hydrostatic part (HD) and the wet part (WD), respectively. The mean differences of WD using VMF3 and GMF range from 1 cm to more than 13 cm, whereas the mean differences between VMF3 and VMF1 range from 0.4-6.5 cm. The estimated gradients from VMF3 scheme tend to be larger than gradients model GRAD. The mean absolute biases for both north and east gradients, Gn and Ge, respectively, are less than 0.1 mm. The standard deviations of GNSS-derived gradients are more than 1 mm, larger than GRAD. The correlation coefficients r between the two gradients are less than 0.5, with Ge being more weakly correlated than Gn. The estimated ZWD can be converted into PWV by using the so-called water vapor weighted mean temperature, Tm. In this study, Tm was first determined using vertical profiles of temperature and humidity from ERA5 data. The results are then compared with Tm from the Bevis model, as determined using ERA5-based Ts. Tm ERA5 has higher variability than Tm Bevis. Generally, Tm Bevis has a cold bias compared to Tm ERA5, typically less than -2 K. The annual range of monthly mean Tm is generally less than 5 K, with the daily variability of less than 2 K. Based on Tm and Ts from ERA5 at 40 grid points, Tm and Ts are not linearly related, with an overall correlation coefficient r of 0.37. Tm-Ts correlation tends to be weaker towards the equator. Site-specific Tm-Ts correlation tends to be stronger than the overall correlation. Additionally, during the summer monsoon, the correlation tends to be weaker than during the winter monsoon. GNSS-derived PWV is then determined at nine stations in 2014 using two types of Tm, namely Tm Bevis and Tm ERA5 for comparison. Tm Bevis was determined using surface pressure from meteorological sensors at the station. The effect of mismodelling Tm to PWV estimation at stations can reach a maximum value of 2-3%, resulting in an error of around 2 mm if PWV is larger than 70 mm. The differences between the two PWV at stations CKEN, CSAU, and CUKE, can reach more than 1 mm at the highest, with a standard deviation of less than 0.5 mm. Based on the PWV as obtained from the estimated ZWD and ERA5-based Tm, PWV in Indonesia is ranging from 20-70 mm. The annual mean of PWV at the stations used in the analysis varies from 44-56 mm, with standard deviations ranging from 4.5-10.7 mm. During the peak of the dry season, between July and September, PWV at most stations tends to decrease. The mean PWV during the dry season ranges from 38-56 mm, and increases to between 49-56 mm during the rainy season. The annual mean PWV tends to increase towards the equator, with the values of 55 mm and 44 mm at the equatorial station CAIR and the southernmost station CREO, respectively. Conversely, the standard deviations are smaller towards the equator, with the values of around 5 mm and 10 mm at CAIR and CREO, respectively.