Automatic Calibration of a Shallow Water Model for Rainfall-Runoff Simulations Using Fast Global Optimization

In this paper, an automated calibration tool for shallow water models (ActSwm) was developed utilizing the open-source Python library SciPy. The developed tool links the hms model for rainfall-runoff simulations to global optimization algorithms as a single executable. Three fast global optimization algorithms (Differential Evolution, Dual Annealing, Simplicial Homology Global Optimization) were applied by using two objective functions (Root Mean Square Error, Nash-Sutcliffe efficiency) to automatically calibrate the parameters of two real case studies of rainfall-runoff in natural catchments. In the first case, a small Alpine catchment was simulated and the three optimization algorithms were applied to three parameters, namely the Strickler coefficient, the runoff coefficient and the interflow storage constant. In the second case, an experimental natural catchment was simulated and seven parameters have been optimized using the three optimization algorithms. Three of the parameters belong to a modified Manning friction law, while the other four to the Green-Ampt infiltration approach. The performance of the three algorithms was investigated comparing them with the classical 'manual calibration' considering accuracy of the results and computation time. Results show that the three global optimization algorithms achieve much better results when compared to the classical 'manual calibration' or even enable a calibration which may not be possible by hand; However, using SHGO and Dual Annealing in ActSwm reduce the calibration time by (70-90) % compared with results of Hassan et al. (2021). ActSwm not only provides a framework to also use other objective functions and optimization algorithms, it is also applicable to other shallow water flow models even if there is a large number of parameters to be calibrated and a calibration may not be possible by hand.