With an increase in the intensity and frequency of extreme rainfall events, there is a pressing need for accurate rainfall nowcasting applications. In recent years, precipitable water vapor (PWV) data obtained from GNSS observations have been widely used in rainfall prediction. Unlike previous studies mainly focusing on rainfall occurrences, this study proposes a transformer-based model for hourly rainfall prediction, integrating the GNSS PWV and ERA5 meteorological data. The proposed model employs the ProbSparse self-attention to efficiently capture long-range dependencies in time series data, crucial for correlating historical PWV variations with rainfall events. Additionally, the adoption of the DILATE loss function better captures the structural and timing aspects of rainfall prediction. Furthermore, traditional rainfall prediction models are typically trained on datasets specific to one region, which limits their generalization ability due to regional meteorological differences and the scarcity of data in certain areas. Therefore, we adopt a pre-training and fine-tuning strategy using global datasets to mitigate data scarcity in newly deployed GNSS stations, enhancing model adaptability to local conditions. The evaluation results demonstrate satisfactory performance over other methods, with the fine-tuned model achieving an MSE = 3.954, DTW = 0.232, and TDI = 0.101. This approach shows great potential for real-time rainfall nowcasting in a local area, especially with limited data.

Extreme rainfall events pose escalating threats to urban infrastructure, agricultural productivity, and public safety. Water vapor is essential for both precipitation formation and heavy rainfall monitoring and forecasting. This study comprehensively analyzes the spatiotemporal characteristics of water vapor variation throughout an extreme rainstorm in Hong Kong from October 7 to 9, 2021, utilizing Global Navigation Satellite System (GNSS) tomography, radiosonde observations, rain gauges and ERA5 reanalysis data. Results show that precipitable water vapor (PWV) increases steadily before the storm, briefly declines due to cloud droplet condensation, and returns to average levels after the rainfall. Prior to the rainfall, significant vertical water vapor disturbances are observed, with a root mean square deviation (RMSD) of 13.53 mm/km, about three times higher than that in stable conditions. This marks atmospheric instability and strong convection contributing to the storm. Key findings reveal that during the storm, PWV fluctuates around 70 mm for 2.5 days, reaching the 96.77th percentile of 2021 PWV records—a statistically extreme anomaly. Analysis indicates that synergistic forcing from typhoon systems and midlatitude weather patterns drove sustained water vapor convergence, enhancing moisture transport and accumulation at high PWV levels. By delineating the spatiotemporal interplay between water vapor and rainfall, this study offers a new perspective on monitoring extreme rainfall precursor. The identified links between persistently high PWV anomalies and the dynamic rainfall system evolutions offer insights for refining rainfall prediction models. Thus, this study hopefully helps improve extreme rainfall early warning and mitigate associated hydrological risks.

Precipitable water vapor (PWV) retrieved from Global Navigation Satellite System (GNSS) has been successfully applied in rainfall forecast. This paper shifts to a new focus, aiming at nowcasting lightning, for its relatively scant GNSS-PWV investigation. Unlike previous studies that mainly explored statistical correlations between PWV and lightning, this approach integrates GNSS-PWV and other meteorological parameters with advanced automated machine-learning algorithms to accurately predict lightning occurrences with a lead time up to 30 minutes. In this paper, the relationship between lightning occurrences and PWV variations is firstly examined through comprehensive statistical analysis. Next, a machine-learning-based lightning nowcasting model is established in this study, with the input of GNSS-PWV and common meteorological parameters. The training and test datasets are sampled every 10 minutes from 7 collocated GNSS stations and automatic weather stations (AWS) along with the lightning location information during the period from 2018 to 2022 in Hong Kong. A comprehensive evaluation is conducted on the performance of the lightning nowcasting model. Results indicate that the proposed model possesses convincingly more excellent performances over four evaluation metrics: Probability of Detection (POD, 89%), False Alarm Ratio (FAR, 30%), Threat Score (TS, 0.64), and Heidke Skill Score (HSS, 0.77). It is also revealed that this model achieves impressive innovativeness, performance and competitive advantages, compared with other existing methods and previous lightning forecast models.