Introducing new morphometric parameters to improve urban canopy air flow modeling: A CFD to machine-learning study in real urban environments
Jonas Wehrle, Christopher Jung, Marco G. Giometto, Andreas Christen, Dirk Schindler
Abstract
This study proposes a machine learning (ML) framework generating spatially-distributed mean wind fields at a given height above ground within arbitrary urban canopy geometries. The framework is based on the Random Forest formulation and is trained using building resolving large-eddy simulations of flow over a range of realistic urban environments. The model maps up to 10 morphometric parameters, including three newly developed ones, to the mean wind over a considered horizontal plane. Predictions are computed from an ensemble of models. In independent evaluation areas, the application of the newly developed morphometric parameters increases the prediction accuracy on average by over 34 % with strengths in predicting main flow channels and areas of notably low wind speeds better than previously described morphometric parameters alone. ML-models, such as the one presented herein, are fast and efficient and are therefore suitable for operational use. • Application of Random Forest models for predicting mean wind in urban environments. • Identification of new morphometric parameters for urban canopy layer wind modeling. • The new parameters significantly enhance the predictive accuracy in evaluation areas.