A Probabilistic Forecast Method for the Runoff of Snowmelt Rivers

Uncertainties are inevitable when using hydrological models for runoff forecasting,especially the runoff forecasting results of snow-melt rechargeable rivers have greater uncertainty.The Xin’anjiang model considering snowmelt and therecession curve method were established respectively for the runoff forecast in the flood season (including the snowmelt season) and in the dry season.The error autoregressive correction model was integrated to improve the real-time forecast accuracy.A probabilistic forecast model of daily runoff was established by combining the deterministic forecast results and the hydrological uncertainty processor,and the probabilistic forecast accuracy was assessed by two types of indicators:precision evaluation and reliability evaluation.In order to verify the validity of the probabilistic forecasting framework,we carried out a case study with the watershed above the Houziyan Reservoir of the Dadu River as the study area.The multi-year average Nash-Sutcliffe efficiency coefficient of the Xin’anjiang model was 0.85 in both calibration and validation periods,and the overall accuracy was high.With the error correction,the forecast accuracy was improved,and the forecast accuracy could be guaranteed on the basis of extending the forecast period.The forecast accuracy of the median of the probability distribution function was better than the original deterministic forecast to a certain extent.The coverage rate and dispersion degree of the 90% confidence interval were about 90% and below 0.40,respectively,indicating that it can cover most of the measured discharge values in a relatively narrow interval and has high reliability.The findings can provide technical support for flood control power generation scheduling and risk management in the study area.