Development of a generic automated instrument for the calibration of morphodynamic Delft3D model applications

More Info
expand_more

Abstract

The numerical flow model Delft3D simulates the flow of water and movement of submerged sediment by means of solving simplified equations that describe these processes. Because of these simplifications, parameterizations are introduced. These parameters represent amongst others the processes that take place on scales smaller than the spacing of the numerical grid. Furthermore, parameters exist that represent measurable attributes of the system, such as the grain diameter. Lastly, scaling parameters are found in the model, with which the relative importance of certain processes in the model can be indicated. All these parameters have to be calibrated using measured data; a model application has to be tuned to produce the best fit with a set of measurements. Although calibration is an essential part in the development of any numerical flow model, no automated or standardized calibration approaches exist within Deltares. As many parameters are available within Delft3D, calibration can be a complex, time consuming process. Automated calibration is objective and can save time, whilst in the process improving the model performance; i.e. minimizing the error between the model results and the measurements. Objective of this Thesis therefore was to develop an efficient method to improve performance of and insights in the Delft3D model throughout complex morphodynamic applications. A careful reading showed that a generic open source platform for calibration of hydrodynamic Delft3D model applications already existed in the form of the software package OpenDA. This software systematically alters parameter values and compares the corresponding model results with the measurements provided. The results are judged by means of a cost function; a performance indicator which represents the goodness of fit between the model results and the measurements. OpenDA was however not yet applicable on morphodynamic model applications. Therefore, the software was adjusted, resulting in an upgraded version of OpenDA; OpenDA MOR. Apart from adding morphological parameter and -result readers, two additional performance indicators were implemented in the code; the Brier skill score and Kirchhofer scores. OpenDA MOR is suited for both calibration and sensitivity analysis of morphodynamic model applications. The OpenDA MOR calibration instrument has been tested by means of TWIN experiments. From these tests, rules of thumb have been deduced on how to apply the tool. It was shown that a calibration was most likely to succeed when no more than 2 parameters were calibrated at once, these parameters are non-inter-related, the parameter(s) are sensitive and that the initial values of the parameters are within a 75% range of their optimum values. An extensive sensitivity analysis was performed on 17 parameters using four model applications. A large overlap in sensitivity throughout the model applications was found; i.e. the same parameters were sensitive in all model applications. Furthermore, from the parameter sensitivity, conclusions could be drawn on the relative importance of the different processes in the various model applications. Lastly, the sensitivity analysis has resulted in insights in the inter-relationships that exist between the various model parameters. After successfully testing the calibration instrument, different calibration cases have been set up to investigate two questions; can OpenDA MOR pinpoint sensitive parameters automatically? Do optimum parameter values differ for different model applications and transport formulations? It was found that automatic pinpointing was not possible, implying that a sensitivity analysis has to be performed separately before starting a calibration. From the second calibration case it became clear that different optimum parameter values were found depending on the model application and transport formula used. This shows the importance of calibration and that there is no such thing as a universal best suited calibration strategy. What parameters should be calibrated depends completely on the goal of the model; perfectly calibrated hydrodynamic parameters do not necessarily produce the best morphodynamic results and vice versa. The test results and calibration cases have shown that the calibration instrument OpenDA MOR has successfully been developed and is applicable on any morphodynamic model application. However, the applicability on complex models was shown to be difficult from a practical point of view, because of the very long runtimes of these models. Therefore, further research is needed on how to decrease the runtimes of complex model applications. Moreover, a new cost function should be developed which enables a more accurate judgment of the morphodynamic model results, as it was found that the Brier skill score and Kirchhofer scores are not suited to replace the standard cost function used during this Thesis.