The atmospheric boundary layer is a major site for human activities, necessitating the establishment of a micrometeorological support system centered on safety and efficiency, particularly regarding the impact of microscale variations in wind speed on flight safety. Coherent Doppler wind LiDAR has the advantages of high accuracy and high temporal and spatial resolution. Computational fluid dynamics (CFD) can simulate the wind field in a refined manner, particularly in the case of complex terrain or building obstructions. In the absence of field observational data, CFD can be used as a complementary method to estimate wind fields. In this paper, a CFD-based multi-source heterogeneous sensing fusion wind field construction method is proposed, which uses wind-speed data obtained from direct observations by multiple wind LiDAR devices and combines data assimilation with a CFD model to realize the real-time construction of a low-altitude three-dimensional wind field in a complex environment. This method provides refined meteorological information for developing low-altitude economies.

First, based on the terrain, vegetation, and building information of the target area, a CFD model was constructed to obtain the flow field data of multiple sectors. After quality control of the LiDAR data, the coordinates were calculated according to the range gate, elevation angle, and azimuth angle of the LiDAR measurement points and then matched with the corresponding grid as the perception input. To ensure the effective transfer of the measured wind speed in the model while retaining an accurate description of the local wind speed and direction by CFD, different fusion strategies were adopted for the wind speed and direction of the nonperception grids. For wind speed, the weight relationship between the LiDAR observation data and CFD was used to transfer the data to non-perception grids. Specifically, the K-Nearest Neighbors (KNN) algorithm was first used to find the k closest perception grids and calculate the deduced wind-speed values of these grids. Then, the inverse distance weighted interpolation method was used to obtain the final wind-speed value. For the wind direction, the KNN algorithm was also used, and the final wind direction was obtained by performing the vector inverse distance weighted interpolation on the k wind directions. Owing to the vector characteristic of wind direction, vector weighting must be used in the calculation. Finally, the accuracy of the fusion model was verified using wind-speed sensors installed on the wind tower.

In the ten-minute wind-speed comparison between the fusion grid and individual sensors (Fig. 7 and Table 3), the wind speed exhibits an increase and subsequent decrease before and after the passage of the typhoon “Babinca”. The sensors are located within a wind-speed range of approximately 2?18 m/s, which clearly reflects the drastic change in wind speed during the passage of the typhoon. Sensor 1 has a significantly higher mean error (ME) than the other sensors owing to the large difference in height from the matching grid. The ME of the other sensors is within 1 m/s of the absolute value, and the mean absolute error (MAE) is approximately 1 m/s. In the ten-minute wind direction comparison with a single sensor (Fig. 8 and Table 4), the wind direction gradually decreases from 360° to 100°, and the rate of change first increases and then decreases with the movement of the typhoon, clearly portraying the trend of wind direction change when the typhoon passes by. Except for sensor 1, which has a large wind direction deviation, the fitting results of the other sensors are excellent, with a coefficient of determination (R²) above 0.9697. The overall evaluation results (Fig. 9 and Table 5) reveal that the ten-minute fit has a higher R² and lower MAE. This is due to the fact that the ten-minute intervals provide more stable time-series data, which helps the model capture trends in wind speed and direction. In addition, the wind LiDAR deployed closer to the wind tower (Fig. 11 and Table 6) is analyzed, and the results indicate that the wind-speed accuracy of the fused grid is improved. The findings reveal that the fusion results depend not only on the neighboring sensing devices but also on the variation of the local wind field to a certain extent.

In complex environments, the traditional meteorological observation means struggle to meet the demand for microscale and high real-time wind field information. To address this challenge, this study adopts high-precision wind measurement equipment, such as wind LiDAR, and combines it with CFD modeling. Through the fusion of high-precision wind measurement data and CFD, the low-altitude 3D wind field is reconstructed. The reliability of our proposed wind field construction method is verified by empirical analysis of wind field data during the impact of typhoon “Bebinca”. This method is capable of generating microscale, high-precision, and real-time wind field information with real-time updates, which is of substantial practical significance for enhancing urban meteorological services and ensuring the safety of low-altitude flights.