Editorial Type:
Article Category: Research Article
 | 
Online Publication Date: 31 Jan 2025

Unmanned Aerial Vehicles Used to Assess Performance of Automated Detection and Audio Deterrent System for Reducing Wind-Turbine Collision Risk for Golden Eagles

,
,
,
,
, and
Page Range: 1 – 21
DOI: 10.3356/jrr2448
Save
Download PDF

ABSTRACT

DTBird is an automated detection and audio deterrent system designed to discourage birds from approaching spinning wind turbines. With Golden Eagles (Aquila chrysaetos) the focal species of interest, we evaluated DTBird’s performance at two commercial facilities, one situated in a desert landscape in California, USA (Manzana site), and the other in a temperate grassland/scrub landscape on a ridgeline above the Columbia River in Washington, USA (Goodnoe Hills site). To evaluate DTBird’s detection and deterrent-triggering functions, we used fixed-wing unmanned aerial vehicles (UAVs) as eagle surrogates in experimental flight trials, involving planned transect arrays that supported evaluating DTBird responses within a 240-m-radius expected maximum, hemispheric detection envelope. We quantified the probability of detection and used logistic regression to evaluate the influence of several predictors. We also built a general linear mixed-effects model (GLMM) to evaluate the influence of several environmental covariates and flight metrics on DTBird detection and deterrent-triggering response distances. The estimated probability of detection was similar at the two sites (64–66%), increased from morning through afternoon (effects of sun positioning), and was highest when the target flew at moderate distances from the turbine through the midsection of the camera viewsheds. The GLMM analysis confirmed modest variation relating to the five distinct UAV models used, possibly mimicking variation that would apply to eagles of variable size and coloration. The analysis also demonstrated that response distances averaged marginally shorter at the Manzana site and increased (suggesting improved detectability) under uniform cloud cover and as the UAV travel rate and exposure of the UAV profile to the cameras increased. Our experience also emphasized that effective use of foam-bodied, fixed-wing UAVs as eagle surrogates at wind facilities can be strongly limited by complicated topography that restricts centralized flight operations; intervening obstacles such as overhead powerlines that restrict automated flight missions; and excessive wind, inclement weather, and rough landing conditions around many turbines that can easily lead to UAV crashes and damage. Using eagle-like UAVs with immobile wings as eagle surrogates also might have constrained the insights generated from the study. Nevertheless, the indicated relationships can help future system users understand the environmental conditions in which DTBird and similar automated systems are likely to perform best and other factors that can substantially influence the targeting accuracy of such systems.

RESUMEN

VEHÍCULOS AÉREOS NO TRIPULADOS UTILIZADOS PARA EVALUAR EL DESEMPEÑO DE UN SISTEMA DE DETECCIÓN AUTOMÁTICA Y DISUASIÓN SONORA PARA REDUCIR EL RIESGO DE COLISIÓN DE AQUILA CHRYSAETOS CON TURBINAS EÓLICAS

DTBird® es un sistema de detección automática y disuasión sonora diseñado para desalentar a las aves de acercarse a las turbinas eólicas en movimiento. Considerando a Aquila chrysaetos como la especie de interés, evaluamos el desempeño de DTBird en dos instalaciones comerciales localizadas en un paisaje desértico de California y en un paisaje de pastizal/arbustos templados en la cima de un cordón montañoso que bordea el río Columbia en Washington. Para evaluar las funciones de detección y activación de disuasión de DTBird, usamos vehículos aéreos no tripulados (VANT) de ala fija como sustitutos de las águilas en ensayos de vuelo experimentales, utilizando esquemas de transectos previamente planificados que permitieran la evaluación de las respuestas de DTBird dentro de un radio máximo esperado de 240 metros en un área de detección hemisférica. Cuantificamos la probabilidad de detección y usamos regresión logística para evaluar la influencia de varios predictores. También construimos un modelo de efectos mixtos lineales generales (MEMLG) para evaluar la influencia de varias covariables ambientales y métricas de vuelo en las distancias de detección y activación de disuasión de DTBird. La probabilidad de detección estimada fue similar en los dos sitios (64-66%), aumentó de la mañana a la tarde (efectos de la posición del sol) y fue mayor cuando el objetivo volaba a distancias moderadas de la turbina a través de la sección media de los campos de visión de la cámara. El análisis MEMLG confirmó una variación moderada relacionada con los cinco modelos distintos de VANT utilizados, posiblemente imitando una variación que podría aplicarse a águilas de tamaño y coloración variables. El análisis también demostró que las distancias de respuesta fueron ligeramente más cortas en el sitio de Manzana y aumentaron (sugiriendo una mejor detectabilidad) bajo una cobertura de nubes uniforme y a medida que la velocidad de desplazamiento del VANT y la exposición de su perfil a las cámaras aumentaban. Nuestra experiencia también destacó que el uso efectivo de VANTs de cuerpo de espuma y ala fija como sustitutos de A. chrysaetos en instalaciones eólicas puede verse fuertemente limitado por una topografía complicada que restringe las operaciones de vuelo centralizadas; por obstáculos intermedios como líneas eléctricas que restringen las misiones de vuelo automatizadas; y por viento excesivo, condiciones climáticas adversas y condiciones de aterrizaje difíciles alrededor de muchas turbinas, que pueden llevar fácilmente a accidentes y daños en los VANT. El uso de VANTs con alas inmóviles como sustitutos de las águilas también podría haber limitado las conclusiones generadas en el estudio. No obstante, las relaciones indicadas pueden ayudar a los futuros usuarios del sistema a comprender las condiciones ambientales en las que DTBird y sistemas automáticos similares son más efectivos y otros factores que pueden influir sustancialmente en la precisión del sistema.

Copyright: © 2025 The Raptor Research Foundation, Inc. 2025
pdf
Figure 1.
Figure 1.

Layout of Manzana Wind Power Project in southern California (upper panel) and Goodnoe Hills Wind Farm in Washington (lower panel) showing the locations of DTBird installations and where UAV flight trials were flown.


Figure 2.
Figure 2.

Vertical cross-section illustrating theoretical DTBird detection envelope and audio deterrent triggering zones calibrated for Golden Eagles, with the approximate rotor swept zone of the turbine shown as a central gray circle and cameras positioned on the turbine tower 4–5 m above ground where the base of the inverted-cone detection envelope begins.


Figure 3.
Figure 3.

Images portraying 4 of 6 UAVs deployed during flight trials conducted during this study in California (images A-AES Custom, and B-AUV Custom) and Washington (images C-Clouds, and D-Ranger). A fifth aircraft not depicted and a sixth aircraft similar to the one shown in image C were also flown in Washington.


Figure 4.
Figure 4.

Example array of 100 randomly selected linear transects prepared to evaluate DTBird’s detection performance at an individual turbine during a half-day series of automated flight missions with an eagle-surrogate UAV. The darker gray hemisphere portrays the projected 240-m-radius maximum detection envelope around the focal turbine, with delineated transects beginning and ending 50 m beyond that range.


Figure 5.
Figure 5.

Example array of tracks recorded during a half-hour UAV flight-trial mission at an individual DTBird-equipped turbine, illustrating use of external loiter points for 30 sec between preplanned flight segments to produce independent sampling flights.


Figure 6.
Figure 6.

GLMM predicted probabilities of DTBird detection for individual eagle-surrogate UAV flight transects and modeled relationships with influential independent variables at wind facilities in California and Washington.


Figure 7.
Figure 7.

Deviations from the estimated global average DTBird response distance associated with different turbine-specific installations (upper panel) and UAV models (lower panel), estimated as random effects nested within study sites in the multi-site GLMM developed for the study.


Figure 8.
Figure 8.

GLMM predicted relationships between DTBird detection and deterrent-triggering response distances and influential categorical predictors. Within panels, shared letters indicate pairwise differences that are not significant (P > 0.05).


Figure 9.
Figure 9.

GLMM predicted relationships between DTBird detection and deterrent-triggering response distances and UAV ground speed and wind speed, as measured by UAV avionics during sampling flights.


Figure 10.
Figure 10.

GLMM predicted relationships between DTBird detection and deterrent-triggering response distances and the interactive influences of UAV roll angles (left and right wing dips) and pitch angles (nose up or down), as measured by UAV avionics during sampling flights.


Contributor Notes

 Present address: BOW Renewables, 1813 Main Street, Blakeley, PA 18447 USA.

 Corresponding author: jsmith@harveyecology.com
Received: 18 Jun 2024
Accepted: 18 Oct 2024
  • Download PDF