Item

Abstract

<!--!introduction!--><b></b><div id="haloword-lookup"><p>In the past three decades, ground-based Global Navigation Satellite System (GNSS) has been used to retrieve atmospheric Integrated Water Vapor (IWV). It shows unique advantages in severe weather event monitoring such as, e.g., its all-weather availability. <p>Traditionally, real-time GNSS IWV sensing using the analytical physical model needs to obtain meteorological data at the location of the observation station, such as surface pressure and atmospheric weighted mean temperature. However, real-time acquisition of the collocated meteorological observations is a very challenging task for most GNSS stations. Although empirical models such as <em>Global Pressure and Temperature 3</em> (GPT3) can provide meteorological estimates, their accuracies are limited. In particular, it is found that the GPT3 prediction errors can be time-series correlated in specific regions (e.g., Central and Northern Europe). <p>In view of the above problems, this study implements a PWV inversion model based on deep learning Long Short-Term Memory (LSTM) network, which realizes real-time GNSS IWV sensing without the actual need for meteorological data. Results show that the Root Mean Square Errors (RMSEs) of the prediction residuals of the developed model are significantly lower than those from GPT3, especially in Central and Northern Europe. The seasonal patterns of the prediction residuals are also mitigated. The developed model provides a broad application prospect for real-time GNSS IWV sensing without meteorological data.</div>