Abstract
A finite-volume hydrodynamic simulator based on centroidal Voronoi tessellation (CVT) is proposed to realize smoothly varying local mesh refinement. The fundamental properties and generation techniques of CVT, weighted CVT and boundary-fitted CVT are explained with a suitable scalar advection scheme. Hybrid parallelization of the shared and distributed memory models has been implemented to achieve efficient long-term high resolution predictions. The simulator has been applied to the brackish lake Abashiri and the salinity fluctuations in the halocline were compared with the measured data. The measured salinity fluctuations due to the internal Kelvin wave are well simulated both in phase and magnitude.
