Ensemble Flood Predictions By Multiple Hydrological Models And Uncertainty Analysis

Human activities and global climate change have increased the frequency of extreme hydrological events,therefore,it has become a trend to find a more accurate and stable hydrological model.This study tries to establish a combined flood forecasting system with different combinations of multiple hydrological models to reduce the impact caused by extreme rainfall events.In this study,historical flood events are collected from Jinhua and Longquan basins and a single black box and a single conceptual model are individually constructed by considering the average lag time of the river calculated by the lag and route method.The set pair analysis and Monte Carlo random sampling methods are then conducted for the uncertainty analysis of various model combinations and the multi-model based combined flood forecasting system can be established after comprehensive evaluation.In this study,3 black box models and 3 conceptual models are selected for the streamflow forecasting through 8 combination methods and 3 strategies.After uncertainty analysis,the results indicate that(1)The lag and route method is introduced to indentify the forecasting capability the conceptual and black box models.The appropriate forecast lead time for the hydrological models built in Jinhua and Longquan basin is 7 and 3hours,respectively;(2)The black box model outperform conceptual models,in which the LSTM(Long Short-Term Memory)and BP(Back Propagation)models provide the best performance in Jinhua and Longquan basins,respectively,and the difference between the best and worst models in the Jinhua and Longquan basins is 370%and 147%,respectively;(3)The unfixed weight combination method outperforms the fixed weight method.Among the 8 methods,the dynamic weight method(DWC)performs the best and the simple arithmetic average method(AM)performs the worst.The smallest difference between the best and the worst combination methods for the Jinhua and Longquan basins is about 14%and 51%,respectively;(4)The combined flood forecasting system conducted by the multiple hydrological models effectively reduce the uncertainty of the forecasts and the largest improvement before and after the combination in Jinhua and Longquan basins is about 659%and 1069%,respectively;(5)By evaluating scatter plot of streamflow,the size of 90%confidence interval,the statistical-index and the uncertainty analysis,the results show that the uncertainty of the forecasts in Jinhua basin is higher than that of Longquan basin,which effectively proves the applicability and efficiency of the set pair analysis method.