Combining Numerical Rainfall Forecasts and Realtime Observations to Improve Early Inundation Warnings

Abstract

Numerical weather predictions can extend the forecast lead time of an early flood warning system and provide decision makers with extra time to take necessary measures and prevent the loss of lives and property. However, the complexity of a climate system increases the spatiotemporal uncertainty of numerical rainfall forecasting, which complicates effective decision making by local authorities during emergency response situations. This study developed an early flood warning system that evaluates the flood risk at the township level by comparing rainfall forecasts and five rainfall thresholds (1-, 3-, 6-, 12-, and 24-h rainfall durations). The system provides a rapid method of evaluating the flood risk, but the spatiotemporal accuracy of the rainfall forecasts directly affects the forecasting accuracy. Therefore, this study combined real time rainfall observations and numerical rainfall forecasts to calculate cumulative rainfall. The flood risk was estimated according to the cumulative rainfall and rainfall thresholds. Additionally, this study linked multiple rainfall values to a township to obtain cumulative rainfall values. These approaches were developed to decrease the spatiotemporal uncertainty of the rainfall forecasts. Five typhoons that made landfall in Taiwan in 2016 were used to investigate the performance of the system. The results showed that the forecast accuracy of the 1-d ahead forecast was significantly improved compared to that of previous methods. The combination of real time observations and numerical rainfall forecasts has demonstrated promising potential in both decreasing uncertainty when using numerical rainfall forecasts and improving the performance of flood forecasting.