An Adaptive Ensemble Framework for Flood Forecasting and Its Application in a Small Watershed Using Distinct Rainfall Interpolation Methods

Runoff prediction has a pivotal role in the flood warning system. For mountainous small-sized watersheds, establishing a reliable and efficient model to forecast flood is multifarious and disorderly work. The ensemble framework for flash flood forecasting (EF5) provides a new opportunity to model simply and practically. However, the EF5 has not successfully verified its feasibility in mountainous small-sized basins and without satellite rainfall products. This paper used the framework to structure a flood forecast model without any satellite rainfall support for a small watershed in China where flash floods occur frequently. The evaluation indicated that the EF5 model performs well in flood prediction cases, with over 0.9 Pearson's linear correlation coefficient (PCC) values and over 0.85 Nash-Sutcliffe coefficient of efficiency (NSE) values during the validation. In addition, statistical results revealed that the EF5 model can maintain a PCC of more than 0.9, NSE of more than 0.7, and flood peak bias (FPB) of more than -0.2 when the forecast lead time exceeds 3 h. Numerous indicators and plots proved the excellent effect of the model forecasting. Considering the convenience and validity of this framework, the research and verification of the EF5 model in the mountainous small-sized basin are of significance to flood prediction.