Performance Evaluation of Single and Multi-Objective Calibration in Rainfall-Runoff Modelling

Abstract

Hydrologic modelling is widely used as a tool for making decisions and predictions in planning and managing water resources. Successful application of a hydrologic model depends upon careful calibration and validation using appropriate performance measures in satisfying corresponding performance evaluation criteria. Performance of MIKE 11 NAM rainfall runoff model was assessed using four calibration objectives and their different combinations, as applied to upper Gin catchment, Sri Lanka. The four calibration objectives measured different aspects of hydrograph: good water balance, good overall agreement of the shape of the hydrograph, good agreement for peak flows, and good agreement for low flows. Their numerical performance was measured using four objective functions from which fifteen calibration schemes were formed (four single objective schemes and eleven multi-objective schemes). Using aggregated distance measure, equal weights were assigned to the four objective functions. Shuffled complex evolution algorithm was used to solve the multiple calibration objective problems. Model performance was evaluated using three criteria: Nash- Sutcliffe coefficient (NSE), percent bias (PBIAS) and coefficient of determination (R2). Results revealed significant trade-offs between the objective functions, highlighting that no single calibration objective was able to depict all the aspects of the hydrograph simultaneously. However, multi-objective calibration yielded more accurate and consistent simulations covering different aspects of the hydrograph, simultaneously with overall best performance shown for combination of the four objective functions satisfying all the performance evaluation criteria (NSE=0.56, R2= 0.56, PBIAS=17%), compared to the single-objective calibration. Instead of using R2 alone, use of the corresponding regression slope as a weighing factor of R2 was recommended following further analysis of simulation results.