Performance evaluation and sensitivity analysis of GPU-accelerated QES against wind tunnel experiments for air dispersion modeling under heterogeneous terrain and dense building at an AP1000 NPP site

Accurate air dispersion modeling can effectively support nuclear accident emergency decision-making and emergency planning zone (EPZ) optimization, for improved safety and economic efficiency of advanced nuclear power plants (NPPs). However, fast and detailed air dispersion modeling remains challenging for heterogeneous terrain and dense building environments. This study presents the GPU-accelerated Quick Environmental Simulation (QES) system for simulating high-resolution wind and concentration field in the presence of heterogeneous terrain and dense buildings, and validates it against two wind tunnel experiments of an AP1000 site, evaluating wind direction, speed, and pollutant concentration in both horizontal and vertical profiles. Sensitivity analyses were conducted on solver options, number of released particles per second, non-local mixing coefficients, and grid sizes. The findings demonstrate that GPU-QES successfully reproduces the vortices between buildings and takes only 4 s to simulate a detailed 3 km × 3 km wind field with 35 million grids. For wind fields, more than 50 % of the simulations were within a factor of two of the observations. For concentration fields, the fraction of simulations within a factor of two/five exceeds 51 % and 85 %, respectively, which are 21 % and 15 % higher than the acceptable lower limits, confirming QES's capability for small-scale modeling at this NPP site. Sensitivity analyses suggest the optimal parameters: global memory solver, released particles per second of 200, non-local mixing coefficients of 0.3 m²/s², and grid size of 5 m both horizontally and vertically in similar scenarios. For the site boundary that is important in determining EPZ, the GPU-QES is only 49 % lower than the measurement, whereas the straight-line Gaussian plume model overestimates by 8.23 times, proving GPU-QES's potential in optimizing the EPZ.