Using automatic calibration to improve the physics behind complex numerical models: an example from a 3D lake model using Delft3D (v6.02.10) and DYNO-PODS (v1.0)

Abstract

Models are simplified descriptions of reality and are intrinsically limited by the assumptions that have been introduced in their formulation. With the development of automatic calibration toolboxes, finding optimal parameters that suit the environmental system has become more convenient. Here, we explore how optimization toolboxes can be applied innovatively to uncover flaws in the physical formulations of models. We illustrate this approach by evaluating the effect of simplifications embedded in the formulation of a widely used hydro-thermodynamic model. We calibrate a Delft3D model based on temperature profiles for a case study, Lake Morat (Switzerland), through the DYNO-PODS optimization tool. The results show that higher values of the light extinction coefficient can compensate for neglecting the fraction β of short-wave radiation absorbed at the surface of the water. This leads to unrealistic values of the light extinction coefficient, as the optimization pushes its value toward the limit of no transparency, consistent with the need to reproduce a significant absorption at the surface. Although it is well known that β is significantly larger than zero, its absence from the model was never noticed as critical. Automatic calibration tools provide valuable diagnostic insights into the physical robustness of models, enabling more precise evaluation of their structural integrity and performance.