Journal Article

Using Machine Learning to Predict Urban Canopy Flows for Land Surface Modeling

Lu, Yanle, Xiao, Heng, Li, Qi, Zhou, Xu‐Hui

2023

Overview

Developing urban land surface models for modeling cities at high resolutions needs to better account for the city‐specific multi‐scale land surface heterogeneities at a reasonable computational cost. We propose using an encoder‐decoder convolutional neural network to develop a computationally efficient model for predicting the mean velocity field directly from urban geometries. The network is trained using the geometry‐resolving large eddy simulation results. Systematic testing on urban structures with increasing deviations from the training geometries shows the prediction error plateaus at 15%, compared to errors sharply increasing up to 35% in the null models. This is explained by the trained model successfully capturing the effects of pressure drag, especially for tall buildings. The prediction error of the aerodynamic drag coefficient is reduced by 32% compared with the default parameterization implemented in mesoscale modeling. This study highlights the potential of combining computational fluid dynamics modeling and machine learning to develop city‐specific parameterizations. Plain Language Summary Predicting the velocity field in the urban area with fine resolution at the meter scale is computationally expensive. Yet a detailed velocity field is necessary for improving the accuracy of urban land surface representation in weather and climate models. We propose using a convolutional neural network to predict the velocity field from the three‐dimensional (3D) building distribution. The similarity between the predicted velocity fields and LES simulations in the testing geometries illustrates the prediction capability of the trained model. We also investigate the aerodynamic drag coefficient, a key parameter for quantifying the land‐atmosphere momentum exchange. The results indicate that the trained model prediction is much closer to values derived from large‐eddy simulation models than those from the default parameterization scheme, showing the promise of using machine learning to improve urban land surface modeling. Key Points Machine learning (ML) can help develop city‐specific parameterization that fully utilizes urban form data It is a first attempt to develop an ML model for high‐Reynolds number urban canopy flow with multiple bluff‐body obstacles Limitation of the geometry to flow field approach is quantified by accessing the extrapolative capability of the trained model

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: Aerodynamic drag / Aerodynamics / Artificial neural networks / Atmospheric models / Climate models / Coders / Computational efficiency / Computational fluid dynamics / Computer applications / Computing costs / Drag / Drag coefficient / Drag coefficients / Drag reduction / Fluid dynamics / Hydrodynamics / Land surface models / Large eddy simulation / Large eddy simulations / Learning algorithms / Machine learning / Modelling / Momentum / Neural networks / Parameterization / Plateaus / Predictions / Pressure drag / Pressure effects / Simulation / Simulation models / Tall buildings / Testing / Urban areas / urban canopy flow / urban canopy model / Urban structures / Velocity / Velocity distribution / Vortices