Auto calibration of complex watershed models using simulationoptimization framework

A physics based hydrologic model (either lumped or distributed) is generally characterized by a multitude of parameters. The calibration of such models is always a complex task. Though, manual calibration is by far the most widely used approach for parameter estimation of such models, it is highly time consuming, and uncertainty of obtaining the best fit exists. Attempts to develop automatic calibration procedures that use optimization algorithms for distributed watershed models have resulted in long computational time and effort. This was mainly because the model has to perform simulations a number of times before it can evaluate the objective function of the algorithm during the parameter search. This study presents a novel framework for calibration of watershed models under simulation-optimization environment. The framework identifies the parameters of the model through an optimization algorithm that uses the objective function from a pseudo simulator, which reduces the computational time significantly. The pseudo simulator will map the objective function in terms of the parameters thus avoiding complex simulation every time. The methodology is demonstrated by calibrating SWAT model for Illinois watershed in Arkansas, USA. The calibrated model is found to simulate the flows reasonably good. The model performance indices when computed on a daily basis, the R2 value between the simulated and measured flows ranged from 0.36 to 0.60 during calibration period and are 0.77 for the validation period. The model resulted in an efficiency of 0.75 during the validation period. Overall, the study reports an auto-calibration procedure for watershed models that can be used for calibration of models irrespective of their complexity.