Nowadays, the Global Navigation Satellite Systems (GNSS) have become an effective atmospheric observing technique to remotely sense precipitable water vapor (PWV) mainly due to their high spatiotemporal resolutions. In this study, from an investigation for the relationship between GNSS-derived PWV (GNSS-PWV) and heavy precipitation, it was found that from several hours before heavy precipitation, PWV was probably to start with a noticeable increase followed by a steep drop. Based on this finding, a new model including five predictors for heavy precipitation prediction is proposed. Compared with the existing 3-factor model that uses three predictors derived from the ascending trend of PWV time series (i.e., PWV value, PWV increment and rate of the PWV increment), the new model also includes two new predictors derived from the descending trend: PWV decrement and rate of PWV decrement. The use of the two new predictors for reducing the number of misdiagnosis predictions is proposed for the first time. The optimal set of monthly thresholds for the new five-predictor model in each summer month were determined based on hourly GNSS-PWV time series and precipitation records at three co-located GNSS/weather stations during the 8-year period 2010–2017 in the Hong Kong region. The new model was tested using hourly GNSS-PWV and precipitation records obtained at the above three co-located stations during the summer months in 2018 and 2019. Results showed that 189 of the 198 heavy precipitation events were correctly predicted with a lead time of 5.15 h, and the probability of detection reached 95.5%. Compared with the 3-factor method, the new model reduced the FAR score by 32.9%. The improvements made by the new model have great significance for early detection and predictions of heavy precipitation in near real-time.

Global navigation satellite systems (GNSSs) have become an important tool for remotely sensing water vapor in the atmosphere. In GNSS data processing, mapping functions and gradient models are needed to map the zenith tropospheric delay (ZTD) to the slant total tropospheric delay (STD) along a signal path. Therefore, it is essential to investigate the spatial–temporal performance of various mapping functions and gradient models in the determination of STD. In this study, the STDs at nine elevations were first calculated by applying the ray-tracing method to the atmospheric European Reanalysis-Interim (ERA—Interim) dataset. These STDs were then used as the reference to study the accuracy of the STDs that determined the ZTD together with mapping functions and gradient models. The performance of three mapping functions (i.e., Niell mapping function (NMF), global mapping function (GMF), and Vienna mapping function (VMF1)) and three gradient models (i.e., Chen, MacMillan, and Meindl) in six regions (the temperate zone, Qinghai–Tibet Plateau, Equator, Sahara Desert, Amazon Rainforest, and North Pole) in determining slant tropospheric delay was investigated in this study. The results indicate that the three mapping functions have relatively similar performance above a 15° elevation, but below a 15° elevation, VMF1 clearly performed better than the GMF and NMF. The results also show that, if no gradient model is included, the root-mean-square (RMS) of the STD is smaller than 2 mm above the 30° elevation and smaller than 9 mm above the 15° elevation but shows a significant increase below the 15° elevation. For example, in the temperate zone, the RMS increases from approximately 35 mm at the 10° elevation to approximately 160 mm at the 3° elevation. The inclusion of gradient models can significantly improve the accuracy of STDs by 50%. All three gradient models performed similarly at all elevations and in all regions. The bending effect was also investigated, and the results indicate that the tropospheric delay caused by the bending effect is normally below 13 mm above a 15° elevation, but this delay increases dramatically from approximately 40 mm at a 10° elevation to approximately 200 mm at a 5° elevation, and even reaches 500–700 mm at a 3° elevation in most studied regions.