A proposed neural network model for obtaining precipitable water vapor

Abstract

Abstract The atmospheric Precipitable water vapor (PWV) is a variable key for weather forecasting and climate change. It is a considerable component of the atmosphere, influencing numerous atmospheric processes, and having physical characteristics. It can be measured directly using radiosonde stations (RS), which are not always accessible and difficult to measure with acceptable spatial and time precision. This study uses the artificial neural network (ANN) application to propose a simple model based on RS data to estimate PWV from surface metrological data. Ten RS stations were used to develop the new model for eight and a half years. In addition, two and a half years of data were used to validate the developed model. The study period is based on the data accessible between 2010 and 2020. The new model needs to collect (vapor pressure, temperature, latitude, longitude, height, day of year, and relative humidity) as input parameters in ANN to predict the PWV. The ANN model validations were based on the root mean square (RMS), correlation coefficient (CC), and T-test. According to the results, the proposed ANN can accurately predict the PWV over Egypt. The results of the new ANN model and eight other empirical models (Saastamoinen, Askne and Nordius, Okulov et al., Maghrabi et al., Phokate., Falaiye et al. (A&B), Qian et al. and ERA 5) are compared in addition, the new PWV model can achieve the best performance with RMS of 0.21 mm. The new model can serve as a will be of practical utility with a high degree of precision in PWV estimation.