Stratified rainfall forecast method using GNSS observations

Abstract

Different rainfall intensity have varying effects on human activity. Previous studies on rainfall forecasting using observations from global navigation satellite system (GNSS) mainly focused on whether rainfall occurred. However, forecasting rainfall with different grades is rarely investigated before, which becomes the focus of this paper. A stratified rainfall forecast method (SRFM) was proposed in this study, where GNSS observations were used to forecast rainfall with different grades. The SRFM considers three independent predictors: precipitable water vapor (PWV) value, PWV variation, and its first derivation. The relationship between PWV and zenith total delay (ZTD) was investigated in detail. The optimal thresholds of the three predictors were determined by innovatively stratifying the rainfall with different grades and introducing the percentile method. In addition, a novel weight ratio method (WRM) was proposed to determine the weight distribution for each grade of rainfall, which overcomes the shortcomings of unreasonable weight distribution under empirical equal-weights or a single accuracy index when evaluating the total forecast accuracy of the SRFM. The proposed SRFM was validated, and experiments were performed using seven GNSS stations over 2015–2016 from the Continuously Operating Reference System (CORS) network in Taiwan Province and five GNSS stations during September 1, 2014 to August 31, 2015 from the CORS network in Zhejiang Province, China. The statistical results show that the accuracies of the total true detected rate (TDR) and false forecasted rate (FFR) of SRFM for forecasting different grades of rainfall were 94.3% and 37.9%, respectively. Compared to existing rainfall forecast models, both with and without forecasting rainfall intensity, the proposed SRFM can well forecasted arbitrary grades of rainfall in each season while guaranteeing the highest TDR and lowest FFR.