Measuring domain decomposition effect in estuary model parallelization using high performance computer

The physical process modeling of an estuary is a complex mechanism that comes along with the existence of uncertainty. In contrast, due to significant variability of flows in space and time, an extensive modeling effort must be accommodated with the application of parallelization technology. Therefore, the implication of estuary model parallelization to the model result uncertainty must be examined. In this research, a three dimensional (3D) hydrodynamic model of Columbia Estuary, located in Oregon, United States, was configured in cloud computing environment. The case study was simulated using DFlow Flexible Mesh software from Deltares. The model behaviors were evaluated in terms of water level, velocity, and salinity with the reverence of field observation data. The estuary model was calibrated in respect to physical parameters before parallelization process. Through calibration procedure, it was revealed that the model were sensitive to bed roughness, eddy viscosity coefficient, and eddy diffusivity coefficient. It can be concluded that the model parallelization technique had induced minor uncertainty contribution to the specific estuary model results. The pattern of uncertainties are varies within the range of parallel processes scenarios, as a result of automatic domain decomposition practice that produces additional diffusivity term to model. Meanwhile, there is an optimum parallel process scenario with minimum computational time. Some bargaining alternatives between model uncertainty and computational time are presented for the application of estuary parallel modeling.