Analysis of atmospheric weighted mean temperature models based on radiosonde observations over Iran

The Global Navigation Satellite System (GNSS) has been in existence for several decades, serving the purpose of position determination and finding applications in navigation, military operations, mapping, and earth sciences. Nevertheless, for almost thirty years, this system has also been utilized in meteorology, leading to the establishment of a field known as GNSS metrology. Signals transmitted from GNSS satellites must pass through all layers of the atmosphere to reach receivers located on the ground or at higher altitudes. When passing through these layers, the signals collide with the components of the atmosphere, which changes the speed of propagation of the signals and, as a result, their arrival at the receiver is delayed. This phenomenon, known as total signal delay, is mainly caused by the troposphere and ionosphere layers. Meanwhile, the ionospheric delay, which is caused by the presence of ions and free electrons in this layer, can be eliminated by using observational techniques such as combining observations of the two frequencies L1 and L2.When processing GNSS observations, the tropospheric delay can be estimated along with the coordinate unknowns. This delay is divided into two main parts: the hydrostatic delay and the wet delay, the latter of which is related to the water vapor in the atmosphere. In the process of retrieving precipitable water vapor from the total tropospheric delay using GNSS meteorology, the atmospheric weighted mean temperature (Tm) plays an important role. Bevis et al. (1992) created a mapping model designed to transform wet delay (ZWD) into precipitable water vapor (PWV). This mapping model includes various physical coefficients and Tm to calculate ZWD based on PWV. There are several methods for calculating Tm, among which the use of actual observations of radiosonde atmospheric profiles is considered as the basic method. However, limitations such as very low temporal resolution and poor spatial resolution in most regions of the world, especially in a vast country like Iran, have necessitated the need for an alternative model in the absence of radiosonde observations. Since accurate estimation of Tm requires vertical profiles of atmospheric temperature and humidity, several empirical regional and global models have been developed to estimate it to date. The objective of this research is to assess the precision of several recent global and regional Tm models in Iran, as well as to present a model founded on the SVM algorithm aimed at enhancing the accuracy of Tm estimation within the study area. To develop the model, observations from 11 radiosonde stations collected between 2015 and 2023 were utilized. For evaluation, Tm values derived from data from these stations in 2024 were employed to assess the proposed model alongside other existing models. The results indicated that the proposed model achieved reductions in RMSE of 3.25, 2.43, 1.00, 1.02, 0.58, 0.61, and 0.61 °C compared with the seven selected models: Bevis, hgpt2, gpt2w, gpt3, GTrop, GGNTm, and Rahimi, respectively. Following the 〖IR_SVM〗_Tm model, the GTrop, GGNTm, and Rahimi models demonstrated, on average, better performance than the other models within the study area. Furthermore, the proposed model was evaluated against other models under rainy conditions, and statistical analyses confirmed that it outperformed the alternatives.