A Novel Linear Rainfall Forecast Model Based on GNSS Observations and CAPE

In recent years, several models have been extensively investigated for short-term rainfall forecasts based on global navigation satellite system (GNSS)-derived precipitable water vapor (PWV). However, the predictors of these models are mainly developed from the PWV, zenith total delay (ZTD), and their combinations, and new and independent predictors have not been introduced for improving the rainfall forecast accuracy. In addition to PWV, convective available potential energy (CAPE) is an effective index for reflecting the atmospheric stability. We proposed a novel linear rainfall forecast (NLRF) model by combining PWV and CAPE, which includes six independent predictors (PWV value, PWV variation and its derivation, CAPE value, and CAPE variation and its derivation). We selected hourly PWV, CAPE, and rainfall over a period of one year at six GNSS stations in Taiwan to develop the NLRF model. The optimal thresholds for PWV and CAPE predictors were determined based on the percentile theory, which followed the principle of highest true detected rate (TDR) and lowest false forecast rate (FFR), and their optimal percentile thresholds are 5 percentile in most years and seasons at the six stations. The optimal performances of these predictors and their combinations were evaluated using the seven designed schemes. The statistical results indicate that the NLRF model with six predictors can achieve the highest accuracy for rainfall forecasts, with TDR and FFR values of 94.94%–24.09%, respectively. Finally, compared to that of latest studies, the forecast accuracy of the NLRF model used in this study reached the highest level and had more application potential. However, the generalization ability of the NLRF model requires further validation across various research regions (such as extratropical and polar regions), different rainfall types (like non-convective rainfall), and different weather phenomena (like convective gales).