Statistical downscaling into the airflows within urban districts based on a numerical simulation datasets

Pedestrian level wind environment is closely related to pedestrian comforts regarding the air quality, thermal comfort, wind load, etc. Now a day, computational fluid dynamic models can successfully simulate the airflows at an urban street scale. However, it still takes much computational costs to run daily similar with the operational weather forecasts.

This study conducts a diagnostic prediction of the microscale wind environment based on the statistical downscaling. This statistical downscaling requires the relationship between the regional meteorological parameters between a coarse grid and local meteorological parameters to be refined. The refined grid is produced by lattice Boltzmann equation model simulations performed on a huge GPU cluster (TSUBAME4.0). Simple linear relationships between the local wind environment and the regional meteorological parameters were observed.

Based on this empirical relationship, the few-meter scale fine velocity distribution within an urban district was reproduced which was downscaled from the regional meteorological observation. The results were compared with the several near-ground observation points in urban districts. A good correlation between the measured and diagnostically estimated velocity distributions was confirmed.

In addition, the regional meteorological data was also replaced by the output of regional weather forecasting model, which the result can also confirm the reproduction of the microscale wind environment