Recognizing temporal variability of lump watershed behaviour and evaluating averaged performance of a hydrologic model

Abstract

Majority of watersheds associated with civil engineering infrastructure projects are ungauged and most commonly used method to determine streamflow in ungauged basins is mathematical modelling with the use of Synthetic Unit Hydrograph (SUH) technique. Mathematical models require watershed characteristics to be spatially and temporally averaged. The SUH is an event based model which estimates direct runoff. Hence model calibration and verification requi: : event based evaluations with a baseflow separation effort or a method incorporating a baseflow model to combine with the SUH model and generate total runoff. In this study, a rainfall runoff model developed using SUH and a linear baseflow concept while selecting the watershed of Attanagalu Oya at Karasnagala as the study area. Other than the SUH parameters to be identified, the conceptual model used for this work consisted of 5 model parameters to be optimised. The main objective of this research was to identify the issues during calibration and verification of this five parameter model. Model calibration was carried out for 30 datasets, selecting the Mean Ration of Absolute Error (MRAE) as the objective function. Optimum model parameters for each event were determined and the most probable range of values for each parameter was computed. Using 30 datasets, model was verification was carried out by assuming that the average of each range would lead to a representative watershed model. A successful calibration produced a good match of observed and calculated streamflows with a MRAE of 0.34. Parameter optimisation revealed the inability to obtain an average initial moisture level for the entire watershed while catering both wet and dry conditions. The runoff coefficients and rainfall thresholds also indicated the need of further investigations. Event based modelling approach in this work provides an insight to the watershed behaviour and to the appropriateness of model parameters, however in order to identify the spatially and temporally averaged parameters it is necessary to carryout optimisations using a lengthy data series together with an appropriate model.