Assessment of Uncertainty in Flood Forecasting Using Downscaled Rainfall Data

Flood is one of the most important natural disasters which causes extensive loss of life and properties every year around the world. By forecasting the hydrograph of probable floods in each year, their damages could be reduced by implementing plans, decisions, and actions. Before flood forecasting, it is necessary to forecast rainfalls. By forecasting rainfall, the amount of runoff can be estimated. So it can be determined whether a severe flood would happen or not. The simulated flood hydrograph is affected by uncertainties in rainfall forecasting that must be considered when flood prevention plans are developed. In this study, a long lead flood forecasting model is developed, considering the uncertainties in forecasting. This model uses long lead predicted rainfalls on a daily basis. The predicted rainfalls must be disaggregated to smaller time and space spans (say 6 hours) in order to be used in rainfall-runoff models. So, predicted rainfalls by a GCM (General Circulation Model) have been downscaled using a regression based statistical method. Then HEC-HMS software has been used to develop a rainfall-runoff model of the basin. Using downscaled predicted rainfalls and the rainfall-runoff model of the basin, the predicted flood hydrograph is determined. The effect of uncertainties in rainfall forecasting has been considered in flood forecasting. In daily precipitation, uncertainties have been assessed by comparing means and variances of precipitation, monthly mean dry and wet spells lengths and their confidence intervals and cumulative frequency distributions (CDFs) of predicted floods. The proposed model for flood forecasting has been applied to the Kajoo River located in South Baloochestan region, in the south-eastern Iran.