High temporal resolution global PWV dataset of 2005–2016 by using a neural network approach to determine the mean temperature of the atmosphere

Atmospheric water vapour plays an important role in phenomena related to the global hydrologic cycle and climate change. However, the rapid temporal–spatial variation in global tropospheric water vapour has not been well investigated due to a lack of long-term, high-temporal-resolution precipitable water vapour (PWV). Accordingly, this study generates an hourly PWV dataset for 272 ground-based International Global Navigation Satellite System (GNSS) Service (IGS) stations over the period of 2005–2016 using the zenith troposphere delay (ZTD) derived from global-scale GNSS observation. The root mean square (RMS) of the hourly ZTD obtained from the IGS tropospheric product is approximately 4 mm. A fifth-generation reanalysis dataset of the European Centre for Medium-range Weather Forecasting (ECMWF ERA5) is used to obtain hourly surface temperature (T) and pressure (P), which are first validated with GNSS synoptic station data and radiosonde data, respectively. Then, T and P are used to calculate the water vapour-weighted atmospheric mean temperature (Tm) and zenith hydrostatic delay (ZHD), respectively. T and P at the GNSS stations are obtained via an interpolation in the horizontal and vertical directions using the grid-based ERA5 reanalysis dataset. Here, Tm is calculated using a neural network model, whereas ZHD is obtained using an empirical Saastamoinen model. The RMS values of T and P at the collocated 693 radiosonde stations are 1.6 K and 3.1 hPa, respectively. Therefore, the theoretical error of PWV caused by the errors in ZTD, T and P is on the order of approximately 2.1 mm. A practical comparison experiment is performed using 97 collocated radiosonde stations and 23 GNSS stations equipped with meteorological sensors. The RMS and bias of the hourly PWV dataset are 2.87/?0.16 and 2.45/0.55 mm, respectively, when compared with radiosonde and GNSS stations equipped with meteorological sensors. Additionally, preliminary analysis of the hourly PWV dataset during the EI Niño event of 2014–2016 further indicates the capability of monitoring the daily changes in atmospheric water vapour. This finding is interesting and significant for further climate research.