Wake reconstruction of a building using sparse data of virtual Doppler wind lidar

Understanding how wind flows in urban environments is vital for the low-altitude aviation economy, aviation safety, and building monitoring. However, even Doppler lidar, currently the most effective remote sensing tool for wind field measurement, faces challenges in urban environments due to insufficient spatial resolution and signal blockage by buildings. To solve these problems, we propose UrbanFlow-Dif, a novel artificial intelligence framework designed to reconstruct fine-scale wind information from sparse lidar data. The model learns from high-resolution simulations of airflow around buildings generated by computational fluid dynamics and uses a diffusion-based generative model, which can remove noise and recover detailed flow structures. Subsequently, the sensitivity of model performance with different scanning configurations is studied to reveal the influence mechanism of lidar position angle, spatial resolution, and scanning step angle. These results demonstrate that UrbanFlow-Dif successfully reconstructs the complex wake structures, achieving a root mean square error below 1.284 m/s even under sparsity conditions (finer than 30 m/1°). Ensemble averaging can further improve the reconstruction performance. This work provides a new foundation for fine-scale urban wind field detection and reconstruction.