A multi-point referencing scheme for characterization of uncertainties in near-ground wind speeds for an industrial park using variational inference

Abstract

Accurate characterization of uncertainties in near-ground wind speeds is crucial for robust microclimate research and design, particularly in industrial parks where complex building configurations significantly impact local wind patterns and pollutant dispersion. This study proposes a multi-point referencing scheme (MRS) by integrating wind speed measurements from multiple heights to provide comprehensive uncertainty quantification. The scheme combines exponential profile modeling with variational inference (VI) to estimate probability distributions of profile parameters, offering a more comprehensive approach than traditional deterministic methods. To validate MRS, field measurements are conducted within an industrial park in Shanghai, capturing near-ground wind speeds across five heights (1.55 m - 5.55 m) in a low-rise urban canopy. Results demonstrate that MRS with VI (MRS-VI) achieves robust modeling performance (mean R² of 0.965) while revealing a height-dependent bias that manifests as wind speed overprediction at lower heights. In comparison, MRS using Monte Carlo simulation (MRS-MC) shows notable instability, particularly at lower heights. Although the conventional single-point referencing scheme (SRS) achieves optimal results using topmost height observations, it fails to fully capture wind speed variability across different heights, which is the limitation that MRS-VI successfully addresses. This study provides key references for more reliable characterization of input uncertainties for uncertainty quantification, supporting decision-making for the creation of sustainable built environment.