An Improved Model for Detecting Heavy Precipitation Using GNSS-Derived Zenith Total Delay Measurements

In recent years, precipitable water vapor has been widely used in heavy precipitation prediction, which is obtained from a conversion of the zenith total delay (ZTD) of the GNSS signal. Since the parameter directly estimated for the tropospheric delay from Global Navigation Satellite Systems (GNSS) data processing is the ZTD, this study investigated the feasibility of directly using ZTD to predict heavy precipitation. Based on the finding that, prior to a heavy precipitation events, ZTD was likely to start with a duration of continuous rise followed by a sharp drop, a new heavy precipitation detection model containing seven predictors derived from ZTD was established. The seven predictors reflect not only the ascending and descending trends but also long-term and short-term variations in the ZTD time series. Three criteria, representing different situations for the formation of heavy precipitation, were also constructed using the predictors to detect heavy precipitation. The optimal set of thresholds for the seven predictors for each summer month were determined based on hourly ZTD and precipitation records at a pair of co-located GNSS/weather station-HKSC-KP in Hong Kong over the period 2010-2017. The model was evaluated using the predictions in 2018 and 2019 to compare against the corresponding precipitation records. Results showed that the new model correctly predicted 98.8% of the heavy precipitation events, with a mean lead time of 4.37 h. Compared with the existing models, the new model also reduced the false alarms by 32.3%. Similar results were also obtained from the other three pairs of co-located stations. These results suggest that it is rational and effective to use the new ZTD-based model for improving the performance of heavy precipitation detection.