A machine learning approach for calibrating ABL profiles in large-eddy simulations

Atmospheric boundary layer (ABL) inhomogeneity is a common problem encountered in computational wind engineering (CWE) where inflow profiles experience unintended flow adaptation while travelling from the inlet boundary to the location of interest inside the domain. This causes the region of interest to experience flow conditions different from those intended by the modeller thereby introducing error in the simulation. Solutions to ABL inhomogeneity have been proposed for RANS models but this issue remains problematic in scale-resolving simulations. In this study, a machine learning (ML) approach is proposed for calibrating ABL profiles to achieve target flow properties in large-eddy simulations (LES). The proposed method is demonstrated for ABL flow over a suburban terrain based on AS 1170.2, leading to a considerable reduction of inhomogeneity error from 49.6% to 4.6%. Sensitivity studies are also presented to investigate the influence of numerical parameters on profile calibration. While the proposed approach does not resolve underlying theoretical limitations resulting in ABL inhomogeneity, it provides a practical solution for achieving target ABL profiles which can help improve confidence in the reliability of LES for wind engineering applications.