Linear and nonlinear GNSS PWV features for heavy rainfall forecasting

Abstract

Heavy rainfall events usually require sufficient water vapor and intense upward movement of airflow. However, current heavy rainfall prediction models using Global Navigation Satellite System (GNSS)-derived precipitable water vapor (PWV) usually use linear features of PWV (PWV increase and decrease, PWV derivative, etc.), which ignore the process of airflow motion. In this study, an hourly heavy rainfall prediction model using the linear and nonlinear features of GNSS-derived PWV is proposed. It uses five linear and three nonlinear features of PWV, together with the corresponding meteorological data, and uses the Random Forest algorithm to predict heavy rainfall or not. The dataset uses the HKSC station observations in Shamshuipo, Hong Kong and the co-located meteorological station data each summer between 2013 and 2020. The results show that the true detection rate and false alarm rate of the proposed model are 97.4% and 19.6% respectively, which are significant improvements compared with other models.