Uncertainty propagation in hydrological modeling

Sellami, Haykel [UCL] Vanclooster, Marnik [UCL] Lajeunesse, Isabelle [UMR CNRS Citeres] Benabdallah, Sihem [Center for Water Research and Technologies]

Uncertainty in hydrological model prediction stems from different sources such as parameter uncertainty, errors of input data, model structure uncertainty, lack of knowledge about the physical characteristics of the watershed and uncertainty in discharge data used to calibrate the model. Portioning and quantifying all these uncertainty sources that affect models and predictions is a difficult and challenging task due to the possible correlation of model parameters, to the propagation and correlation of the errors, to the model complexity, etc. Parameter uncertainty is usually considered as the main source of the model prediction uncertainty while discharge data uncertainty is rarely taken into account when calibrating the model. In this work a new calibration technique of the SWAT model is developed to account explicitly for the uncertainty in discharge data caused by uncertainty in the rating curve. For a given discharge value, a probability distribution function of true discharge was estimated from the constructed uncertain rating curves. This method was applied in three automatic calibration techniques SUFI2, GLUE and ParaSol to calibrate the SWAT model in two Mediterranean watersheds (Vène and Pallas). The uncertainty prediction confidence interval (95PPU) of each of the applied techniques was more successful at bracketing the measured data when uncertainty in the rating curve is accounted for. Discharge uncertainty contributed to total model uncertainty by 8 to 27 % (with an average value of 13 %) for the Vène watershed and by 3 to 20% (with an average value of 9 %) which shows that uncertainty in discharge measurement is not a negligible uncertainty source and has to be considered in hydrological modeling for more accurate model prediction that can be used in decision making for developing sustainable water resource management strategies.