Author(s): A. Hassan; I. Ozgen; R. Hinkelmann

Linked Author(s): Reinhard Hinkelmann, Ilhan Özgen-Xian, Elhadi Mohsen Hassan Abdalla

Keywords: Rainfall-runoff modeling; Automatic calibration; Nelder-Mead; Basin-Hopping; Hydrodynamic model

Abstract: In this paper, a simplex-type optimization method (Nelder-Mead) and a global optimization technique (Basin-Hopping) were applied to automatically calibrate the parameters of a hydrodynamic model for rainfall-runoff simulations. Shallow water models have been successfully used for several decades in research and practice of water management, hydraulic engineering, and environmental issues. With their help, water levels, flow fields and the spreading of substances are calculated (e.g. pollutants, nutrients, sediments). Recently, the shallow water models have been applied to simulate rainfall-runoff in natural catchments. These models consist of mathematical equations and parameters, which also describe the physical processes of runoff generation by rainfall and runoff concentration. It is difficult if not impossible to measure the required parameters in the field. Therefore, indirect methods are needed to estimate these parameters. In order to forecast correctly rainfall concentration and flow paths, the model must be calibrated and validated by comparing measurements and model results. As a rule, calibration parameters such as the bottom roughness or the turbulent viscosity must be adjusted by repeatedly changing the values, performing new calculation runs while observing the discrepancies between measurements and calculations. A model is considered calibrated, when measured and calculated results match as good as possible. If this task is carried out by hand, it is referred to as manual calibration. Manual calibration can be carried out following a trial and error method. This means that initial values of the parameters are firstly assumed and, after a computation run, the model’s results are compared with the corresponding measurements. If the results of the simulation are not yet satisfactory, the parameters values are changed and the model is run again, until a desired level of accordance between results and measurements is reached. In most cases, manual calibration requires a lot of time and effort, and depending on the experience of the modeler, the values chosen for the parameters might not be the optimal ones. Instead of manually calibrating the model, it is possible to use numerical optimization methods to steer an automated calibration of the model. This work discusses the application of such automated calibration approach, considering two case studies of rainfall-runoff in natural catchments, showing capabilities and the limitations of the method used. The Hydroinformatics Modeling System (hms) was used as rainfall-runoff model. Hms is a Java-based, object-oriented software framework developed at the Chair of Water Resources Management and Modeling of Hydrosystems, Technische Universität Berlin. Automated calibration of models optimization is carried out by means of: Basin-Hopping method, which is one of the most progressing global optimization techniques primarily based on the Monte Carlo method and it is mandatory to use a searching method for a local minimum (e.g. Nelder-Mead). Nelder-Mead method, which is a simplex-type algorithm that converges to a local minimum, which is ideally the global minimum, of a given objective function. Objective function Root Mean Squared Error, which is chosen to measure the deviations between the model predictions for the candidate parameters (simi) and observed data (obsi). The Basin-Hopping and the Nelder-Mead’s method have been chosen, because of their high success rates to find the global optimum and their higher performance compared with other methods in terms of computing time. The optimization method and technique, which are utilized to implement automatic calibration, were performed by the SciPy optimization package, written in Python. In the first case, rainfall-runoff in a small Alpine catchment was simulated.

DOI: https://doi.org/10.3850/978-981-11-2731-1_130-cd

Year: 2018