The Experiment of Hydrologic Probabilistic Forecast Based on the Precipitation Forecast Calibrated by Bayesian Model Averaging

Based on 24-h accumulated precipitation data of the Huaihe Basin from 1 June to 31 August in 2008 and the corresponding ensemble forecast of 24 h, 48 h, 72 h from T213 model, the method of Bayesian Model Averaging (BMA) is used to calibrate quantitative precipitation forecasts of 15 members from the ensemble forecast based on the training data of 30 days. The calibrated results are verified by continuous ranked probability score (CRPS) and mean absolute error (MAE). Second, the Dapoling—Wangjiaba Catchment in the upper stream of the Huaihe River, which is subdivided into Dapoling—Xixian Catchment and Xixian-Wangjiaba Catchment, is investigated with the hydrological simulation experiment. The Xixian and Wangjiaba hydrologic stations in the upper stream of the Huaihe River are selected as representative stations. The rain process occurring from 23 June to 3 August in 2008 is investigated to simulate the runoff tendency. Then, the 25th and the 75th percentiles of 24-h, 48-h and 72-h precipitation ensemble forecast which are calibrated by BMA are used to force the Variable Infiltration Capacity (VIC) hydrological model respectively to obtain the corresponding runoff, and finally the simulate results are analyzed comparing with daily runoff observations.Results show that the precision of 24-h, 48-h, 72-h precipitation forecast of BMA model is improved after the calibration. The raw ensemble forecast of 24 h is calibrated well by BMA model. As the leading hours increase, the calibration of 48 h and 72 h is as good as that of 24 h. Although BMA calibrates on the raw ensemble forecast, the improvement of calibrated forecast depends on the accuracy of raw ensemble forecast. The valid interval given by BMA model, namely the interval from the 25th percentile to the 75th one of ensemble forecast, is more likely to contain the true value of observations according to the verification analysis of hydrological probabilistic forecast. From this aspect, the performance of BMA forecast outperform deterministic forecast. It can improve the accuracy of forecast and reduce the error by BMA, describing forecast uncertainty in the form of a probability distribution. Known from the analysis of verification index of hydrologic probabilistic forecast, it shows that the hydrological simulation forced by the calibrated precipitation is almost consistent with the runoff tendency of observations. It is effective to grasp the trend of runoff change. It indicates that the precipitation forecast calibrated by BMA can be established coupling with the VIC hydrological model as well as increasing the forecast accuracy significantly. It can meet the more and more objective, quantitative needs of decision-making service, and improve the benefit of weather forecast greatly.