Improving water quantity simulation & forecasting to solve the energy-water-food nexus issue by using heterogeneous computing accelerated global optimization method

Highlights: • We proposed a novel parallel SCE-UA method. • We implemented the parallel SCE-UA on multi-core CPU and many-core GPU. • The parallel SCE-UA can obtain global optimum much faster than original serial version. - Abstract: With continuous population increase and economic growth, challenges on securing sufficient energy, water, and food supplies are amplifying. Water plays the most important role in the energy-water-food (E-W-F) nexus issue such as energy supply (clean hydropower energy generation), water supply (drinking water), and food supply (agricultural irrigation water). Therefore, water quantity simulation & forecasting become an important issue in E-W-F nexus problem. Water quantity simulation & forecasting model, such as rainfall-runoff (RR) hydrological model has become a useful tool which can significantly improve efficiency of the hydropower energy generation, water supply management, and agricultural irrigation water utilization. The accuracy and reliability of the water quantity simulation & forecasting model are significantly affected by the model parameters. Therefore, demand of effective and fast model parameter optimization tool for solving the E-W-F nexus problem increases significantly. The shuffled complex evolution developed at University of Arizona (SCE-UA) has been recognized as an effective global model parameter optimization method for more than 20 years and is highly suited to solve the E-W-F nexus problem. However, the computational efficiency of the SCE-UA dramatically deteriorates when applied to complex E-W-F nexus problem. For the purpose of solving this conundrum, a fast parallel SCE-UA was proposed in this paper. The parallel SCE-UA was implemented on the novel heterogeneous computing hardware and software systems which were constituted by the Intel multi-core CPU, NVIDIA many-core GPU, and PGI Accelerator Visual Fortran (with OpenMP and CUDA). Performance comparisons between the parallel and serial SCE-UA were carried out based on two case studies, the Griewank benchmark function optimization and a real world IHACRES RR hydrological model parameter optimization. Comparison results indicated that the parallel SCE-UA outperformed the serial one and has good application prospects for solving the water quantity simulation & forecasting model parameter calibration in the E-W-F nexus problem.