Author(s): Sang-Bo Sim; Hyung-Jun Kim
Linked Author(s):
Keywords: Urban flood forecasting; GPU acceleration; Heterogeneous computing; 1D-2D coupled model
Abstract: High computational costs of physically-based 1D-2D coupled models hinder real-time urban flood forecasting. To address this, we developed KICT-UF-HPC, a high-speed model utilizing heterogeneous computing. It employs a hybrid parallelization strategy: CUDA OpenACC (GPU) for 2D surface flows and OpenMP (CPU) for 1D sewer networks to maximize efficiency while maintaining the numerical accuracy of the original KICT-UF model. Validation confirmed numerical precision against experimental flume data [2] and evaluated speed using the 2024 flood event in the Sangpyeong district, South Korea. Results demonstrate that the model maintains physical accuracy while achieving a 35x speedup, proving its viability as a core engine for real-time forecasting systems.
Year: 2026