Evaluating the performance of the rainfall forecast model based on GNSS ZTD and ERA5 CAPE for Hong Kong

Abstract

The Earth's troposphere delay provides valuable information for rainfall forecasting. Tropospheric delay products (e.g., Zenith Total Delay (ZTD)) derived from the Global Navigation Satellite System (GNSS) have become an indispensable approach for accessing tropospheric information. Several studies have established rainfall forecast models based on the GNSS-derived ZTD. However, the forecast predictors used in these models were all developed from the ZTD or its highly correlated parameters, resulting in high virtual forecast results. Therefore, a novel predictor, Convective Available Potential Energy (CAPE), was introduced to enhance the existing rainfall forecast model and is composed of the ZTD and CAPE rainfall (ZCR) model. The ZCR model includes six predictors: the ZTD value, ZTD variation, its derivation, CAPE value, CAPE variation, and its derivation. Hourly ZTD, CAPE, and rainfall data from three stations in Hong Kong with a time span of one year were collected to perform the experiment. The forecast accuracy of the ZCR model was evaluated using the corrected forecasted rate (CFR) and false alarm rate (FAR). The optimal thresholds of the predictors were determined using the percentile method, following the highest CFR and lowest FAR. The forecast accuracies of the ZTD and CAPE models, composed of their own three predictors, were comparable at the seasonal and monthly scales. Finally, we combined the ZTD and CAPE predictors into a ZCR model. Compared with the single ZTD or CAPE models, the combined ZCR model can significantly improve the accuracy of the FAR, with the CFR and FAR values of 93.53 % and 26.82 %, respectively. And the ZCR model also illustrates that the forecast accuracy on the monthly scale is higher than that on the seasonal scale.