Accelerating Urban Aerodynamic Drag Estimation: Evaluating AI, POD, and Linear Interpolation Approaches

Traditionally, acquiring drag forces and their complex distribution relies heavily on computationally expensive Computational Fluid Dynamics (CFD) simulations or wind tunnel measurements. This study proposes and evaluates simplified, accelerated methods to estimate building drag forces in diverse urban blocks.

Based on a comprehensive CFD dataset of 32 generic urban neighborhoods encompassing uniform and varying building heights across distinct plan area densities (sparse to dense), we compared three simplified estimation techniques against the baseline CFD results. The methods evaluated include linear interpolation, Proper Orthogonal Decomposition (POD) interpolation, and Artificial Intelligence (AI)-based interpolation utilizing an Artificial Neural Network (ANN). The comparative analysis focused on the accuracy and computational efficiency in estimating building drag force, neighborhood sectional, and overall drag.

Preliminary findings indicate that while linear interpolation provides rapid estimations, it struggles to capture non-linear local shelter effects, particularly in varying-height and highly dense neighborhoods. POD interpolation successfully captures the dominant vertical drag profiles with a moderate computational cost. The AI-based interpolation demonstrates the highest predictive accuracy for both bulk and distributed drag forces, effectively mapping complex morphological parameters (e.g., plan area density and height variation) to localized aerodynamic responses, albeit requiring initial training time.

AI- and POD-based interpolations offer robust, computationally efficient alternatives to traditional CFD for estimating urban canopy drag. These simplified methods can significantly accelerate urban parametrization processes, providing rapid insights for urban planners in designing aerodynamically efficient neighborhoods.