Optimization of a Conceptual Rainfall-Runoff Model using Evolutionary Computing methods

Application of artificial intelligence (AI) in hydrologic modeling is receiving increasing attention to explore new optimization approaches to the complex nonlinear high-dimensional models. This paper presents the application of two evolutionary computing (EC) methods to calibrate the parameter of a conceptual rainfall runoff model (i.e., the Tank model). Tank model has been selected as for several reasons including its simplicity in presenting the runoff processes, its flexible adaption and its capability to produce real catchment hydrograph if suitable parameters' calibration is attained. To this end, two global optimization methods (GOM), the Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), are used to optimize the Tank model parameters, applied to actual rainfall runoff hydrograph. The performance of two methods have been compared under different scenarios and environments based on two objective functions namely, the root mean square error (RMSE) and Nash-Sutcliffe model efficiency coefficient (NSE). The results proved the capability of GOM to calibrate the 4-stage Tank model with 16 parameters, even under insufficient data availability (as in the case of UAE watersheds). Both techniques performed satisfactorily with slight superiority of GA over the PSO. Also, NSE proved that it can be considered a superior criterion, to use as objective function, compared to RMSE. The research concludes by recommending the best optimized 16 parameters of the tank model for the selected watershed and provides promising results for UAE watersheds.