Artificial neural network for improving the estimation of weighted mean temperature in Egypt

Abstract

Abstract One of the most important parameters in meteorological data is the Precipitable Water Vapor (PWV). It can be measured by radiosonde stations (RS), but the fact is that RS are not available in all times. Therefore, GNSS satellite signals are considered an accurate function to compute it within a conversation factor. The conversation factor depends on the weighted mean temperature ( T m {T_{m}} ) which is non-measurable. In this research, a new idea to estimate T m {T_{m}} is provided, which can potentially contribute to the GNSS meteorology. The T m {T_{m}} was designed, including six RS, over one year in Egypt as input parameters. The machine learning (ML) model has been utilized in the design (IBM SPSS Statistics 25 package). The new model needs to collect the day of year (DOY), site location information and surface temperature to predict the T m {T_{m}} . The results of ML model and four other empirical models (Bevis et al., Wayan and Iskanda, Yao and Elhaty et al. models) are compared. The validation work is carried out, using the radiosonde data, and results indicate that the new T m {T_{m}} model can achieve the best performance with RMS of 1.7 K.