Uncertainty Assessment of Rainfall-Runoff Modelling Using Quantile Regression Method in Huangchuan Basin

Abstract:

A reliable and accurate rainfall-runoff prediction is very important for water resources planning and management. In this study, based on XinAnJiang (XAJ) model, the hourly streamflow forecasting is carried out in Huangchuan basin in china. The XAJ model is calibrated using 14 flood process and 4 remaining ones are to be predicted. In addition, considering the impact of model input, parameters and structure on streamflow prediction, the quantile regression method is employed to post-process the raw streamflow prediction to obtain probabilistic prediction and assess the predictive uncertainty. The quantile regression establishes the relation between different quantile (i.e. 0.05, 0.5 and 0.95) of prediction error and corresponding forecasting value itself. Finally, based on forecasting value and different quantile of prediction error, the uncertainty of streamflow prediction can be quantified. The results indicates that the quantile regression has a considerable capacity to estimate the predictive uncertainty of streamflow in terms of confidence intervals.