Uncertainty assessment for short-term flood forecasts in Central Vietnam

摘要

Accurate flood forecasts with greater lead-times are very important in development of flood mitigation measures, especially in short response catchments. The flood forecasts based on numerical weather prediction (NWP) and runoff models have demonstrated its breakthrough to extend the forecast lead-time over traditional flood forecast methods, for instance, those are based on rainfall information from rain-gages. However, given the imperfectness either in the specification of initial states or in the formulation of NWP models, rainfall prediction for example, the driving factor for flood forecast, has been recognised as a major source of uncertainty in the generation of river flow. This paper presents the uncertainty assessment for a short-term flood forecast model that is coupled by the short-range global NWP model, 0.5 degree spatial resolution, with the distributed rainfall runoff model, for a large sized basin (Thu Bon River, 3,150km~2) located in Central Vietnam. To reduce uncertainty of runoff forecasts by means of increasing the rainfall prediction skill, first the model output statistic technique has been employed to downstate the large scale prediction forecasts directly derived from the NWP model output to the basin scale by using the artificial neural network with the back-propagation method. Skill scores of the downscaled precipitation are investigated with increasing lead-time and compared to those obtained using the large scale precipitation forecasts. Uncertainties of runoff prediction are assessed by quantifying the relative error of forecasts and estimates of confidence interval for the mean error. Results show that larger uncertainties along with the forecast lead-times are observed; however, the model is able to predict reliable river flows with lead-time of the order of 6-18 hours. This demonstrates great benefits in flood forecasting practices for many developing countries where ground weather observation is scarce and access to high resolution NWP models is limited.