Abstract
Background: Kazakhstan is one of the largest oil producers in Central Asia, with a significant share of oil and gas projects associated with the Caspian region (primarily, these include the North Caspian Project at Kashagan, the Tengiz oil field, and the port infrastructure of Aktau). Oil spills in marine waters pose a serious ecological threat to the environment, making it essential to develop effective tools for forecasting their spread. Aim: To develop and test parallel algorithms for the simulation of wind-driven sea currents on graphics processing units (GPUs) for oil spill forecasting. Materials and Methods: Three-dimensional hydrodynamic models with parameterized turbulent viscosity were were used. Numerical schemes were implemented in explicit and implicit forms, parallelized along x and y coordinates using the OpenCL library. Computations were performed on the CPU and GPU for both uniform and non-uniform grids. Results: GPU-based computations demonstrated significant acceleration: up to 23 times faster on uniform grids and up to 13 times faster on non-uniform grids. The accuracy computations on a non-uniform grid with 40 nodes was comparable to that of computations on a uniform grid with 500 nodes. Conclusion: Parallel GPU algorithms substantially reduce computation time while maintaining high accuracy in modeling wind-driven currents. This makes them a promising tool for forecasting oil spill dispersion in the Caspian Sea.