GLOBAL OPTIMIZATION FOR EFFECTIVE CATCHMENT MODEL CALIBRATION

An increasing use of global optimization techniques for the calibration of catchment models has been observed in the 90’s. An evaluation of one technique, an improved version of the traditional genetic algorithm (GA) is presented. A ten-parameter rainfall – runoff model and sixteen years of historical rainfall, potential evapotranspiration and runoff data is used. The improved GA practically locates the global optimum in 9 out of 10 randomly initialized optimization runs with a maximum allowance of 5000 model simulations. Allowing up to 25000 simulations, the global optimum is located in all 10 runs. The quality of model simulations using the global optimum parameters are only marginally better than those using local optimum parameters obtained from the traditional GA. A considerably better parameter consistency and identification of parameter interdependence is however observed with the improved GA. Where time allows, global optimization should be preferred to manual methods of model calibration. The user however needs to specify a suitable range of search and to check that the parameters obtained are not hydrologically unrealistic.