Real-Time Rainfall Nowcast Model by Combining CAPE and GNSS Observations

Abstract

Precipitable water vapor (PWV), derived from the Global Navigation Satellite System (GNSS), has contributed significantly to rainfall forecasting. However, another key parameter, convective available potential energy (CAPE), is strongly correlated with increases in extreme rainfall under the background of global warming but has rarely been investigated for rainfall forecasting. Therefore, a real-time rainfall nowcast (RRN) model that combines CAPE and PWV is proposed in this study. In addition, seasonal factors and the time autocorrelation of the predictors were considered. Here, the previous hourly PWV, CAPE, temperature, and rainfall were used to establish the RRN model and simulate/nowcast the next hourly rainfall based on the support vector regression, which was performed in a time span of five years at 23 GNSS stations in Taiwan Province under four designed schemes to validate the performance of the proposed RRN model. The average root mean square (RMS) and correlation coefficients of the proposed RRN model reached 0.34 mm/h and 0.96, respectively. Additionally, the linear relationships between the daily CAPE/PWV and extreme rainfall were investigated, revealing that PWV may contribute more to trigger extreme rainfall than CAPE. Finally, compared to existing quantitative rainfall forecast studies, the RRN model achieved significantly improved rainfall nowcast accuracy and thus has more potential applications in rainfall forecasting.