Wind speed inference from environmental flow–structure interactions

Jennifer L. Cardona, Katherine L. Bouman, John O. DabiriFlow 2021 · DOI

Abstract

This study introduces a physics-based framework for inferring wind speed from the visual deflection of flexible structures, such as trees and cylinders. By modeling the balance between fluid drag and elastic restoring forces, we demonstrate that normalized wind speed can be accurately estimated without prior knowledge of an object's specific material properties. Experimental validation using various cylinders and tree species confirms the model's robustness, offering a non-intrusive alternative to traditional anemometry.

Introduction

Quantifying wind speed is essential for applications ranging from wildfire modeling to weather forecasting. However, traditional anemometers are often impractical to deploy at scale and are susceptible to damage in extreme conditions. This research explores a "physics-based shortcut" to transform common environmental objects into natural flow sensors. By observing how structures like trees sway or bend, we can quantitatively infer wind speed using ubiquitous video data.

Methodology: The Physics of Bending

The core of the proposed method lies in balancing two primary forces: fluid drag and elastic restoration. Fluid drag ($F_D$) is proportional to the square of the wind speed ($U^2$) and the object's frontal area ($A$). Conversely, the elastic restoring force follows Euler–Bernoulli beam theory.

Visual tracking of a flexible cylinder's tip deflection (δ) under wind flow (U).

By equating these forces, the model establishes that wind speed $U$ is proportional to the square root of the structural deflection $\delta$. A critical innovation is the use of normalization: by comparing measurements against a reference state ($U_0, \delta_0$), complex variables like material stiffness and density cancel out, making the model broadly applicable across different structures.

Experimental Validation

To validate the model, researchers conducted experiments in controlled wind tunnel environments. The setup involved subjecting various flexible cylinders and real tree species (Juniper and Bay Laurel) to a range of wind speeds.

Schematic of the wind tunnel setup showing the fan array, flexible cylinder, and camera placement.

High-speed video (240 fps) captured the tip deflections, which were then processed using image tracking algorithms. These visual measurements were compared against ground-truth data from a calibrated air velocity transmitter.

Key Insights and Results

The study revealed that material independence is a significant advantage of the normalized model. For trees, which are "reconfigurable" (they streamline by folding leaves and branches), the model was modified to account for changes in projected height ($h$).

Key findings include:

  • High Accuracy: Inferred wind speeds closely matched anemometer readings, typically within a 10-15% error margin.
  • Robustness: The model performed well for cylinders across normalized speeds of 0.7 to 1.5 and for trees across a wider range (0.3 to 3.5).
  • Linearity Limits: The simple linear model remains accurate until deflection exceeds approximately 20% of the object's length ($\delta/L > 0.2$), at which point nonlinear mechanics must be considered.

Conclusion

This research provides a scalable, non-intrusive method for wind speed estimation. While currently requiring a single calibration point for absolute measurements, the framework demonstrates the potential for using environmental "visual cues" as a reliable data source for meteorological and ecological monitoring. Future work will focus on 3D tracking to handle varying wind directions and testing in complex atmospheric boundary layers.

Summary updated

Abstract

Graphical Abstract
BibTeX
@inproceedings{cardona2021wind,
  author = {Jennifer L. Cardona and Katherine L. Bouman and John O. Dabiri},
  title = {{Wind speed inference from environmental flow–structure interactions}},
  booktitle = {Flow},
  year = {2021},
  doi = {10.1017/flo.2021.3},
}

Loading related content...

Figure