A fast and automated approach for urban CFD simulations: integration with meteorological predictions and its application to drone flights

TL;DR

The paper addresses the challenge of rapid, accurate urban wind simulations by automating geometry reconstruction from LiDAR and cadastral data and integrating meteorological predictions into CFD. It combines an automated workflow with a wind-tunnel–style drone evaluation to reduce computation time while preserving fidelity, and validates the approach against ground-truth meteorological data. Key findings include very high concordance with observations using corrected OpenMeteo inputs (e.g., $ρ_c$ for wind direction ≈ 0.985 and wind speed ≈ 0.853) and strong agreement across multiple weather services, with OpenMeteo typically performing best. The decoupled wind-tunnel method yields comparable drone-load results to full-city simulations but with orders of magnitude lower computation time, enabling practical UAV planning and urban management insights.

Abstract

In past years, several studies have proposed new methods and applications for urban wind simulations. In this article, we present a fast and automatic methodology for reconstructing airflows within urban environments using LiDAR and cadastral data coupled with Computational Fluid Dynamics (CFD) simulations. Our approach integrates meteorological predictions with computational techniques to simulate the complex interactions between wind currents, buildings, vegetation, water zones and terrain morphology within urban environments. Accurate boundary conditions based on meteorological predictions are introduced into a coupled methodology that directly creates the terrain shape inside the simulation environment, simplifying the geometry creation process, which is one of the most prevalent problems in CFD urban simulations. The simulation results are confronted against ground-truth real data obtained from a meteorological station, showing strong agreement with the outcomes generated by the proposed CFD model, with a concordance correlation coefficient up to $ρ_c = 0.985$ for the wind direction and $ρ_c = 0.853$ for the wind speed. The results from these simulations are then used for validating a wind tunnel approach that mimics the interaction between a moving drone and the extracted wind currents, demonstrating a great improvement in computation times when compared to the most straightforward approach that consists in embedding the drone within the full urban landscape. This research contributes to the advancement of urban CFD modeling, and it has significant implications for various applications, providing valuable insights for urban development.

Figures (8)

• Figure 1: (a) Geometry used for the calibration process representing the Campus Sur in the University of Santiago de Compostela (Spain). The red dot represents the meteorological station from the official Galician weather service (MeteoGalicia), which we will use as a ground-truth reference value to assess the validity of our methodology. (b) Mesh used for the simulations. (c) Example of the resulting triangulation extracted from the Ear Clipping algorithm.
• Figure 2: (a) Example of a boundary condition generated for the calibration geometry in the University of Santiago de Compostela. Note that the ABL equation for velocity is preserved at every point. (b) Vegetation zones inside the domain. The $z_0$ values are assigned at every point inside the simulation, mimicking the effect of vegetation on the velocity profile.
• Figure 3: Overview of the simulation process. We start by selecting an appropriate mesh for both (a) the case of the complete domain over the city reconstruction and (b) the wind tunnel, with the same Base Size for the spherical region around the drone, extracted from a mesh independency test. The dimensions of the tunnel are described relative to the drone's diagonal wingspan $C = 0.4$ m, using the reference values from prabhakararao2014cfdhassan2014numerical. Then, (c) the whole city domain is initialized by running a steady simulation, and the wind data is extracted following a given trajectory (brown line). A transient simulation with a moving drone in the whole domain is performed and compared with (d) the wind tunnel approach with lower computational times.
• Figure 4: Visualization of the velocity and vorticity fields in a plane 5 m above ground level for an example simulation in our calibration domain at the Campus Sur of Santiago de Compostela, Spain.
• Figure 5: Boxplot of wind speed and direction errors at 10 m for each model considering the averaged real-life values over an hour around 9:00 AM.