In this study, we investigate the in‑flight wind speed measurement accuracy of a multi‑rotor unmanned aerial vehicle (UAV drone) equipped with an ultrasonic anemometer. The primary challenge arises from the distorted flow field generated by the UAV drone’s own body and rotating propellers, which introduces systematic errors into the anemometer readings. To address this, we employ computational fluid dynamics (CFD) simulations to quantify the flow field distortion above the UAV drone, and then incorporate the derived correction coefficients into the wind speed calculation formula. Experimental flights over a coastal wetland demonstrate that the proposed correction significantly reduces the standard deviation of the measured wind components, thereby confirming the effectiveness of the method for improving UAV drone‑based wind field monitoring.
1. Introduction
With the rapid advancement of unmanned aerial vehicle (UAV drone) technology, new methods for measuring atmospheric parameters such as temperature, humidity, and air pollution have emerged. Among these, wind speed measurement based on UAV drones has attracted considerable attention due to its low cost, high flexibility, and ability to perform continuous large‑area probing, which overcomes the limitations of traditional fixed‑tower anemometry. Multi‑rotor UAV drones, in particular, offer hovering capability and agile maneuverability, making them suitable for small‑scale, complex environments like urban canopies and mountainous regions.
Existing approaches for multi‑rotor UAV drone wind measurement can be divided into two categories: using onboard sensors (e.g., inertial navigation, GPS, air data system) to estimate wind through flight dynamics, or directly mounting an anemometer (e.g., ultrasonic or multi‑hole pressure probe) on the UAV drone. While direct anemometry provides straightforward wind speed and direction data, it suffers from interference caused by the UAV drone’s own rotor downwash and body‑induced flow distortion. Previous studies have conducted wind‑tunnel tests and field comparisons between UAV drone measurements and tower measurements, showing qualitative agreement but lacking quantitative correction of the systematic errors due to the UAV drone’s perturbed flow field.
Therefore, the objective of this work is to systematically analyze, via CFD simulation, the distortion of the external flow field around a hexacopter UAV drone under various inflow speeds and angles, and to derive a correction factor that can be applied to the raw ultrasonic anemometer data. The corrected wind speed is then validated through a dedicated field experiment. The key contributions include a quantitative relationship between the measurement error and the inflow conditions, a practical correction formula, and experimental evidence of improved accuracy.
2. UAV Drone Wind Speed Measurement System and Method
2.1 Measurement Platform
Our UAV drone wind measurement system consists of a hexacopter platform (model FY‑17), a solid‑state ultrasonic anemometer (FT742‑SM), a small radio transceiver, a data logger, and a laptop computer for real‑time monitoring. The hexacopter has a maximum payload of 15 kg, a flight endurance of 30 min, and a maximum speed of 15 m/s. Its positioning accuracy is ±0.5 m vertically and horizontally, and attitude control is better than ±0.5 m/s in velocity. The ultrasonic anemometer features a measurement range of 0–75 m/s, a resolution of 0.1 m/s, and an accuracy of ±0.3 m/s (see Table 1 for key specifications). The anemometer is powered by the UAV drone’s battery and transmits data at 1 Hz via a serial link to the ground station.
| Parameter category | Parameter | Value |
|---|---|---|
| Wind speed | Range | 0–75 m/s |
| Resolution | 0.1 m/s | |
| Accuracy | ±0.3 m/s | |
| Wind direction | Range | 0–360° |
| Accuracy | ±4° | |
| Operating environment | Temperature range | −40 °C to 85 °C |
| Humidity | 0–100% | |
| Altitude | 0–4000 m |
2.2 Wind Speed Calculation Method
The wind speed measured by the airborne anemometer is the relative speed between the air and the sensor. To obtain the true wind speed in the Earth‑fixed coordinate system, we must account for the motion of the UAV drone. The true wind vector V is given by the vector sum of the measured relative wind Va and the platform’s velocity Vp, with corrections for attitude and rotation:
$$ \mathbf{V} = \mathbf{G}\,(\mathbf{V}_a + \boldsymbol{\omega}_p \times \mathbf{r}) + \mathbf{V}_p $$
where ωp is the angular velocity of the UAV drone (from the inertial navigation system), r is the position vector of the anemometer relative to the UAV drone’s center of rotation, and G is the rotation matrix that transforms the sensor‑fixed coordinates to the Earth frame. The rotation matrix is constructed from the roll (Φ), pitch (θ), and yaw (ψ) angles:
$$
\mathbf{G} = \begin{bmatrix}
\cos\theta\cos\psi – \sin\phi\sin\theta\sin\psi & -\cos\theta\sin\psi – \sin\phi\sin\theta\cos\psi & -\sin\phi\cos\theta \\
\cos\theta\sin\psi & \cos\theta\cos\psi & -\sin\theta \\
\sin\phi\cos\psi + \cos\phi\sin\theta\sin\psi & -\sin\phi\sin\psi + \cos\phi\sin\theta\cos\psi & \cos\phi\cos\theta
\end{bmatrix}
$$
In practice, the anemometer’s raw output is affected by the flow distortion caused by the UAV drone’s body. To account for this, we introduce a scalar correction factor q that multiplies the measured wind speed magnitude. The modified formula becomes:
$$ \mathbf{V} = \mathbf{G}\,(q\,\mathbf{V}_a + \boldsymbol{\omega}_p \times \mathbf{r}) + \mathbf{V}_p $$
The factor q is a function of the free‑stream wind speed and direction, and is determined from the CFD simulations described in the next section.
3. CFD Simulation of the External Flow Field
3.1 Geometry and Computational Domain
We model a hexacopter UAV drone using a simplified geometry that retains the main body and six arms, while omitting the propellers and landing gear. The simplified model is placed inside a hexahedral computational domain of dimensions 30 m × 30 m × 10 m (length, width, height). The UAV drone is located at the center, and the boundaries are placed far enough to avoid interference (see Fig. 2 in the original study). The flow field is meshed using Fluent Meshing, with a global minimum size of 5 mm and maximum size of 500 mm. A boundary layer mesh is applied on the UAV drone surface with a first‑layer height of 10 mm, a growth rate of 1.2, and 10 layers, resulting in a total mesh count of 118,264 cells.
3.2 Simulation Conditions
The nominal flight speed of the UAV drone is 10 m/s, and the ambient wind speed during typical operations ranges from 2 to 8 m/s (Beaufort scale 2–4). Therefore, the relative wind speed experienced by the anemometer lies between 2 and 18 m/s. We select five inflow speeds: 6, 8, 10, 12, and 14 m/s. Additionally, we consider two inflow angles: 20° and 40° relative to the UAV drone’s longitudinal axis, to account for the asymmetric flow pattern. For each case, a steady‑state CFD simulation is performed using the k‑ε turbulence model.
3.3 Flow Field Results
The simulation results reveal that the flow velocity directly above the UAV drone’s center is accelerated due to the obstruction of the body. Figure 3 (described qualitatively) shows that the velocity contours in the YZ‑plane are similar across different inflow speeds, with the highest acceleration occurring near the junction of the rotor arms and the fuselage. We extract velocity profiles along the vertical (Z) axis passing through the center of the UAV drone. Figure 4 (described) shows that the velocity deviation is largest immediately above the fuselage (up to 15% error) and decays with increasing height. For a given height, the absolute error increases with inflow speed.
To determine the optimal mounting height for the anemometer, we analyze the vertical distance at which the flow perturbation becomes negligible. Table 2 lists the wind speed errors at a height of 750 mm above the UAV drone center, which we selected as the installation point because it offers a good balance between minimal flow distortion and tolerable vibration.
| Inflow speed (m/s) | Speed error (m/s) | Relative error (%) |
|---|---|---|
| 6 | 0.173 | 2.883 |
| 8 | 0.210 | 2.625 |
| 10 | 0.195 | 1.950 |
| 12 | 0.226 | 1.883 |
| 14 | 0.236 | 1.686 |
Similarly, we evaluate the effect of inflow angle at a constant inflow speed of 10 m/s. The results show that the error increases slightly when the inflow comes from an oblique direction due to the non‑symmetric shape of the UAV drone. Figure 5 (qualitatively described) shows that at a 40° inflow angle, the velocity contour becomes denser on the windward side, indicating stronger acceleration. Nevertheless, the overall error pattern remains consistent, and the correction factor can be expressed as a function of the measured wind direction.
3.4 Derivation of Correction Factors
Based on the simulation data, we divide the inflow speed range into five intervals: 5–7, 7–9, 9–11, 11–13, and 13–15 m/s. For each interval, we compute the average relative error and express the correction factor q as the inverse of the relative error factor. Since the ultrasonic anemometer provides both wind speed and direction, the correction must be applied to the two orthogonal components (north and east) separately. The correction factors for the roll‑axis direction (u‑component) and pitch‑axis direction (v‑component) are given by:
$$
q_u = \begin{cases}
\frac{1}{2.883}\,\cos\gamma & \text{for } 5 \le V_a < 7\ \text{m/s} \\[4pt]
\frac{1}{2.625}\,\cos\gamma & \text{for } 7 \le V_a < 9\ \text{m/s} \\[4pt]
\frac{1}{1.950}\,\cos\gamma & \text{for } 9 \le V_a < 11\ \text{m/s} \\[4pt]
\frac{1}{1.883}\,\cos\gamma & \text{for } 11 \le V_a < 13\ \text{m/s} \\[4pt]
\frac{1}{1.686}\,\cos\gamma & \text{for } 13 \le V_a < 15\ \text{m/s}
\end{cases}
$$
$$
q_v = \begin{cases}
\frac{1}{2.883}\,\sin\gamma & \text{for } 5 \le V_a < 7\ \text{m/s} \\[4pt]
\frac{1}{2.625}\,\sin\gamma & \text{for } 7 \le V_a < 9\ \text{m/s} \\[4pt]
\frac{1}{1.950}\,\sin\gamma & \text{for } 9 \le V_a < 11\ \text{m/s} \\[4pt]
\frac{1}{1.883}\,\sin\gamma & \text{for } 11 \le V_a < 13\ \text{m/s} \\[4pt]
\frac{1}{1.686}\,\sin\gamma & \text{for } 13 \le V_a < 15\ \text{m/s}
\end{cases}
$$
where γ is the wind direction measured by the anemometer relative to the UAV drone’s forward direction. For wind speeds outside these intervals, a linear interpolation is applied.
4. Experimental Verification
4.1 Test Site and Flight Plan
The field experiment was conducted on November 7, 2024, at a coastal wetland park in Guangdong Province, China. The test area consists of multiple 40 m × 40 m heterogeneous vegetation patches with flat terrain. The UAV drone flew a quadrilateral route at a constant altitude of 50 m above ground level. The flight speed was set to 10 m/s (ground speed) using a constant‑velocity mode. To ensure data quality, the UAV drone’s pitch and roll angles were kept within ±10°, the angular rate during turns was limited to 3 °/s, and the altitude fluctuation was controlled within ±5 m. Each leg of the quadrilateral lasted about 1–2 minutes.
4.2 Data Preprocessing
The raw data from the inertial navigation system and the ultrasonic anemometer were synchronized using timestamps. Outliers were identified using a sliding window variance method: any data point deviating more than 4 standard deviations from the local window mean was considered an outlier and replaced by linear interpolation. Figure 6 (qualitative description) shows the raw wind components and the processed data after outlier removal. The preprocessing effectively eliminates spikes while preserving the overall trend.
4.3 Correction and Analysis
We applied the correction factors from Eqs. (4) and (5) to the measured wind speed components. The northbound and eastbound wind speeds before and after correction are compared in Table 3. For the entire flight, the mean northbound wind speed changed from 5.012 m/s (uncorrected) to 4.984 m/s (corrected), while the standard deviation decreased from 0.8446 to 0.3093. The eastbound wind speed mean changed from 3.612 m/s to 3.581 m/s, and the standard deviation dropped from 0.6649 to 0.4702. These reductions in standard deviation confirm that the correction effectively mitigates the flow‑induced errors.
| Component | Statistic | Uncorrected | Corrected |
|---|---|---|---|
| Northbound | Mean (m/s) | 5.012 | 4.984 |
| Std dev (m/s) | 0.8446 | 0.3093 | |
| Eastbound | Mean (m/s) | 3.612 | 3.581 |
| Std dev (m/s) | 0.6649 | 0.4702 |
Furthermore, we analyzed the quadrilateral legs individually. Table 4 lists the mean wind speed and variance for each leg after correction. The northbound component shows a maximum average error of 2.95% across the four legs, indicating good consistency. The eastbound component has a larger spread, with legs 3 and 4 showing higher deviations, likely due to changes in the ambient wind field during those legs rather than systematic UAV drone effects.
| Leg | Northbound mean (m/s) | Northbound variance | Eastbound mean (m/s) | Eastbound variance |
|---|---|---|---|---|
| 1 | 5.01 | 0.08 | 3.65 | 0.21 |
| 2 | 4.98 | 0.07 | 3.58 | 0.19 |
| 3 | 4.96 | 0.09 | 3.52 | 0.26 |
| 4 | 5.00 | 0.08 | 3.60 | 0.23 |
5. Discussion
The CFD simulation results clearly demonstrate that the flow above the UAV drone is accelerated, leading to an overestimation of wind speed by the ultrasonic anemometer. The relative error ranges from about 1.7% to 2.9%, depending on the free‑stream speed. By placing the anemometer at a height of 750 mm, we achieve a balance between reducing flow disturbance and maintaining structural stability. The correction factor derived from the simulation successfully reduces the variance of the measured wind components, with the northbound standard deviation decreasing by 63% and the eastbound standard deviation by 29%.
One limitation of our approach is that the CFD model neglects the influence of the rotating propellers. In reality, the rotor downwash can interact with the ambient flow, potentially altering the flow field above the UAV drone. However, previous studies (e.g., Li et al., 2023) have shown that the region directly above the center of a hexacopter is less affected by rotor wakes compared to the vicinity of the rotor disks. Our choice of 750 mm is also consistent with the mounting height used in several field experiments. Future work could include a full rotor‑resolved CFD simulation to refine the correction factor.
Another source of uncertainty is the wind direction measurement. The ultrasonic anemometer has a directional accuracy of ±4°, which, when multiplied by the wind speed, gives an error of up to 0.5 m/s in the cross‑wind component. The correction formula uses the measured direction γ, so any bias in γ propagates into qu and qv. Nevertheless, the overall reduction in variance suggests that the dominant error is captured by the magnitude correction.
6. Conclusion
In this study, we have developed and validated a method to improve the accuracy of wind speed measurement using a multi‑rotor UAV drone equipped with an ultrasonic anemometer. Through CFD simulation, we quantified the acceleration of the flow above the UAV drone body and derived a piecewise correction factor that depends on the measured wind speed and direction. Field experiments over a coastal wetland demonstrated that the corrected wind components exhibit significantly lower standard deviations compared to the raw data, with the northbound component’s standard deviation dropping from 0.8446 to 0.3093 and the eastbound component from 0.6649 to 0.4702. These results confirm that our systematic approach to compensating for the UAV drone’s perturbed flow field can substantially enhance the reliability of UAV drone‑based wind field monitoring. The methodology can be extended to other UAV drone platforms and anemometer types by repeating the CFD analysis for the specific geometry. Future work will focus on incorporating rotor effects and validating the method under turbulent and gusty conditions.