Can an Avian Electrocution Risk Model From California Guide Retrofitting Throughout the Western United States?

Extrapolating research results beyond their original scope of inference can be problematic because it is often impossible to know whether the underlying data structure and assumptions match (Tacha et al. 1982; Ellison 1996). However, research on wildlife habitats and behaviors is frequently extrapolated because repeating studies can be prohibitively expensive, result in delayed conservation actions, and be unnecessary when systems are similar. To address whether extrapolation is appropriate, Tacha et al. (1982) recommended defining a study population from which statistical inference is drawn and a target population to which inference will be applied and then carefully considering relationships between the two. Such comparisons can be accomplished via statistical methods in meta-analyses (Normand 1999) or Bayesian statistics (Ellison 1996) when raw data are available. Inference to biological significance can proceed in comparison-based assessments of independent datasets when raw data are unavailable. If patterns are consistent across datasets, the original research may be extrapolated to the target population.

Dwyer et al. (2014) described a predictive model of avian electrocution risk for overhead distribution power lines (hereafter the “2014 model”). To develop the 2014 model, they visited 213 power poles where electrocutions of birds were known (electrocution poles) and 248 poles where no electrocutions were known (comparison poles). They randomly sampled electrocution poles from 440 poles that had electrocuted a bird and were still in service in southern California. Selection of comparison poles occurred by placing random locations throughout the study area and then evaluating the nearest pole to each location. At each pole, Dwyer et al. (2014) collected information on the numbers of jumpers and primary conductors, the presence of grounding, and the type of surrounding habitat. Variables were deliberately simple so the model could be used beyond the study population (Dwyer et al. 2014).

Golden eagles Aquila chrysaetos are affected by electrocution throughout much of their range. For example, the U.S. Fish and Wildlife Service (USFWS) estimates 504 golden eagle electrocutions in the United States annually (95% credible interval: 124–1,494; USFWS 2016a), accounting for 26% of annual anthropogenic mortality. We focus on golden eagles herein because of interest from the USFWS in using the 2014 model to standardize compensatory mitigation requirements under the 1940 Bald and Golden Eagle Protection Act (as amended, USFWS 2021) as an offset for permitted incidental take. We address application of the 2014 model to species other than golden eagles in the discussion. The USFWS defines take of golden eagles as actions to “pursue, shoot, shoot at, poison, wound, kill, capture, trap, collect, or molest or disturb” (USFWS 2009, 2016b) and incidental take as take “associated with but not the purpose of a [lawful] activity” (USFWS 2009). The USFWS may authorize incidental take, such as specific limited impacts during commercial activities, if that incidental take is offset elsewhere (2016b). Preventing electrocutions of golden eagles by retrofitting power poles is a primary mechanism of conservation (USFWS 2016a; Mojica et al. 2018; Bedrosian et al. 2020). For example, the USFWS uses retrofitting high-risk poles as a primary compensatory mitigation tool to offset permitted incidental take of golden eagles (about 17 poles/bird) or nests (about 98 poles/nest; USFWS 2016b). Using the 2014 model to identify high-risk poles is a potential mechanism of transparently prioritizing mitigation for incidental take permits and utility retrofitting programs (Dwyer et al. 2014; Harness and Dwyer 2015). The 2014 model also may be useful as a mechanism for determining equivalencies between pole types (i.e., determining how many low-risk poles yield an equivalent conservation benefit to one high-risk pole). Using the 2014 model to select poles and determine equivalencies will be possible only if it is appropriate to extrapolate the 2014 model beyond the study population.

The 2014 model included a relatively small proportion of golden eagle electrocutions (21 [10%] of 213). Data from so few individuals seem intuitively inadequate to extrapolate across the western United States. Extrapolation may be appropriate however, because avian electrocution risk is largely a function of pole configuration and bird size (APLIC 2006, 2018). Regarding pole configuration, General Order 95 specifies regulations for power line configurations in California (State of California 2018), and the National Electrical Safety Code specifies standards of power pole construction elsewhere in the United States (IEEE Standards Association 2017). The two codes are similar in many of the features that influence avian electrocution risk. For example, General Order 95 specifies “11.5 inches” (29.2 cm) of horizontal separation required between distribution conductors carrying up to 7.5 kV. The National Electrical Safety Code specifies “12.0 inches” (30.5 cm) of horizontal separation between distribution conductors carrying up to 8.7 kV. Given the more than 102-cm carpal-to-carpal spread of a golden eagle's wings, the 0.7-cm difference in codes is inconsequential. Regarding bird size, golden eagles are consistently among the largest birds across North American landscapes (Katzner et al. 2020).

Although extrapolation seems appropriate based on pole configuration and bird size, we questioned whether additional supporting data were available. We attempted to answer that by comparing electrocution risk defined by the 2014 model to three independent sources of data: 1) structure configurations (poles, pylons, and towers) associated with golden eagle electrocutions described in global scientific literature, 2) structure configurations associated with other avian electrocutions, and 3) pole configurations associated with previously unpublished golden eagle electrocution data from satellite-tagged golden eagles in the western United States. If, across these datasets, there is an association between more complex structures and greater electrocution rates, as there is in the 2014 model, it is reasonable to conclude that the 2014 model is broadly applicable.

Following Tacha et al. (1982), we defined two study populations: 1) golden eagles and 2) power poles in the service area where the 2014 model originated. We defined golden eagles and power poles in the western United States outside of the study populations as target populations. Following convention within the electric industry, we defined a structure as any construction supporting a power line, including poles, pylons, and towers. Pylons and towers tend to be constructed of conductive concrete or steel. Poles specifically indicate cylindrical structures usually constructed of nonconductive wood in the United States.

We reviewed published studies to identify whether there were patterns in configurations associated with golden eagle electrocutions by searching for the key words Aquila chrysaetos, electrocution, golden eagle, and power line in the Institute for Scientific Information Web of Sciences databases (Mojica et al. 2018). We reviewed publications these searches returned for data relating structure configuration to golden eagle electrocutions, data relating structure configuration to electrocutions of other birds, and citations containing additional data.

Previously published studies describing utility structures associated with electrocutions include potential biases. Detection, scavenging, and search biases can influence the documentation of species and individuals (Cartron et al. 2000; Dwyer and Mannan 2007; Lasch et al. 2010; Harness et al. 2013; Gális et al. 2019). Preconceptions about which structures are dangerous, and assumptions of electrocution as the cause of death, can lead to confirmation bias (Harness et al. 2013; Demeter et al. 2018).

Satellite telemetry can create data unaffected by these biases (Dwyer et al. 2020b). The USFWS is using satellite telemetry of golden eagles to track movements and mortality in the central and southwestern United States (Murphy et al. 2017, 2019; Dwyer et al 2020b). The dataset includes locations for 14 golden eagle electrocutions identified by USFWS personnel or their designees when visiting locations where global positioning system (GPS) tracks terminated in Arizona, Colorado, New Mexico, Oklahoma, Texas, and Utah in the United States and in northern Chihuahua in Mexico. The GPS data are accompanied by photographs and nontechnical descriptions of some of the poles involved. We used these data to assess the 2014 model, but to be useful, we needed configuration data for every pole involved so we could evaluate them consistently. We also needed configuration data for adjacent poles so we could identify how configurations of electrocution poles compared with poles across the landscape. We used Google Earth and Google Street View imagery to generate the necessary data. Google Earth imagery inferred electrocution as a cause of death for osprey Pandion haliaetus (Klaassen et al. 2014) and assessed golden eagle perching on poles (Dwyer et al. 2020b), but to our knowledge has not previously assessed the level of pole-specific detail in the 2014 model. Thus, documenting the effectiveness of using Google Earth was a secondary objective of this study. To evaluate whether Google Earth imagery was effective, we used blind assessments of poles. Specifically, E.K.M. held the USFWS pole descriptions and images, and J.F.D. viewed the poles in Google Earth so we could compare remote sensing assessments to USFWS photos and descriptions. J.F.D. navigated in Google Earth to each of the GPS waypoints provided for the electrocutions, and then following Dwyer et al. (2020b), identified whether a pole was visible within 30 m of the GPS point. If so, we assumed the pole, or the nearest pole if more than one was present, was where the electrocution occurred. J.F.D. examined poles in each Google Earth and Google Street View image available (typically two to three, taken 1–3 y apart) to infer configurations and populate the 2014 model for the electrocution pole and four nearest poles. We followed Dwyer et al. (2020b) in assessing the four nearest poles because those were within approximately 100 m, likely providing a perched golden eagle with a similar view of the landscape. We compared configurations of poles where electrocutions occurred with those of adjacent poles to identify configurations associated with electrocutions given the pole types across the landscape.

Because of the limited level of detail visible in Google Earth, we categorized poles as tangent, intersection, and equipment (Figure 1). Tangent poles supported conductors only, and their construction was in relatively straight lines. Conductors are the long wires connecting poles to one another. Intersection poles supported conductors and jumpers where lines crossed or connected. Jumpers are the short wires connecting conductors to one another or to equipment. Equipment poles supported conductors, jumpers, and equipment. Equipment is any energized pole-mounted device other than conductors and jumpers. Equipment poles also sometimes supported jumpers connecting conductors to equipment. For analysis, we pooled intersection poles and equipment poles as nontangent poles. This addressed small sample sizes of intersection poles and equipment poles. Pooling was reasonable because the presence of jumpers on both pole types made them functionally similar. We used a chi-square test to compare proportions of electrocution poles (tangent vs. nontangent) to proportions of nearby poles (tangent vs. nontangent) to test the null hypothesis that golden eagle electrocutions occurred in proportion to pole types. We considered results significant at α = 0.05.

To further validate the Google Earth assessment, we repeated it with a dataset provided by Powder River Energy Corporation (PRECorp, Gillette, WY; Bedrosian et al. 2020). The PRECorp data included GPS locations for 345 golden eagle electrocutions reported by customers or found by electricians. These locations likely included many of the biases described above, but were useful in evaluating the technique because each datapoint was accompanied by a detailed pole description. We evaluated a random selection of 14 locations to match the sample size from the USFWS, and we analyzed PRECorp data separate from USFWS data.

Excluding Dwyer et al. (2014), we identified 10 publications that quantified relationships between golden eagle electrocutions and some aspect of structure configuration (Table 1). Each publication reported more electrocutions on more complex structures. Simpler structures typically involved few, if any, jumpers, equipment, or grounding. We also identified 23 peer-reviewed scientific publications that quantified relationships between electrocution rates for birds other than golden eagles and some aspect of structure configuration (Table 2). Of these publications, 21 reported more electrocutions on more complex structures than on tangent structures. Many reported electrocutions of various species, often including eagle species, such as bald eagle Haliaeetus leucocephalus (Harness and Wilson 2001), Bonelli's eagle Aquila fasciata (Tintó et al. 2010; Guil et al. 2011), chaco eagle Buteogallus coronatus (Sarasola et al. 2020), eastern imperial eagle Aquila heliacal (Gális et al. 2019), Spanish imperial eagle Aquila adalberti (Janss and Ferrer 2001; Guil et al. 2011), steppe eagle Aquila nipalensis (Matsyna et al. 2011; Dixon et al. 2019), tawny eagle Aquila rapax (Harness et al. 2013), and white-tailed eagle Haliaeetus albicilla (Demeter et al. 2018). Across species, there was an association between electrocutions and more complex structures. One study only evaluated tangent structures and found greater risk with shorter crossarms (Benson 1981). Shorter crossarms placed conductors closer together, similar to how there is a reduction in separations with increased pole complexity. Another study reported electrocutions only on equipment poles. In that study, there was an assumption that tangent poles were lower risk based on prior knowledge (Kemper et al. 2020).

We did not find a single study indicating tangent structures were equally dangerous or more dangerous to golden eagles specifically, or to birds generally, than nontangent structures. Patterns illustrating the risk relationship between equipment and grounding were particularly apparent in India, Hungary, and Mongolia (Harness et al. 2013; Demeter et al. 2018; Dixon et al. 2019), where nearly all configurations, including tangents, involved grounded structures. In those locations, there was an association of tangent structures with increased numbers of electrocutions, but even then, intersection and equipment structures had greater numbers.

We evaluated 28 electrocution poles including 14 from the USFWS (hereafter USFWS poles; Table 3) and 14 from PRECorp (hereafter PRECorp poles; Table 4). Each pole and some wires and equipment were visible in at least one Google Earth image (Figure 1). Grounding was not consistently identifiable. When we could confirm grounding present or absent, we report that assessment only. When we could not confirm grounding, we report scores with grounding present and with grounding absent so readers can see the role that grounding plays in electrocution risk.

In assessing USFWS poles, at seven poles (50%) it was not possible to compare risk scores from remote sensing to known pole configurations because known configurations were not given (n = 5), the GPS location was midway between two poles with different configurations (n = 1), or the structure was a transmission pole not suitable for the 2014 model (n = 1). Of the remaining poles, our identification matched the reported configuration in six cases (86%). For one case, the USFWS reported three transformers, where it was clear in Google Earth there was only one transformer. We concluded the USFWS assessment was incorrect.

Limiting our assessment to these seven poles, three (43%) electrocutions involved three-phase poles, five (71%) involved intersection or equipment poles (nontangent poles), and two (29%) involved poles with grounding above the lowest energized component. The total count exceeds seven because some poles included combinations of factors. Overall, six poles (86%) where satellite-tagged golden eagles were electrocuted had one or more risk factors predicted by the 2014 model. Considering electrocution and comparison poles together, five electrocutions occurred across nine nontangent poles and two electrocutions occurred across 26 tangent poles (data from seven electrocution poles and four adjacent poles at each location). Electrocutions were significantly (χ² = 9.57; df = 1; P = 0.002) more likely on nontangent poles than on tangent poles.

Stated another way, proactively retrofitting the nine nontangent poles in the dataset would have prevented 71% of electrocutions. Retrofitting all 15 nontangent poles and tangent poles with grounding above the lowest conductor would have prevented 87% of electrocutions. By contrast, proactively retrofitting all 20 tangent poles without grounding above the lowest conductor would have prevented only one (13%) electrocution. Accounting for the distribution of pole types on the landscape, electrocutions occurred at a rate 8.0 times greater on intersection and equipment poles, and on tangent poles with grounding above the lowest energized conductor than on tangent poles without grounding above the lowest energized conductor.

In assessing PRECorp poles, we correctly identified the electrocution pole in all 14 cases, but did not correctly identify configurations for 3 cases. In two of those cases, crossarm-mounted lightning arresters were not visible in Google Earth, and in one case, switches were not visible, leading to underestimates of the number of jumpers in each case. After correcting those errors, 12 (86%) of 14 electrocutions occurred on three-phase poles and 10 (71%) of 14 occurred on equipment or intersection poles (one also included grounding above the lowest conductor). Only four (29%) electrocutions occurred on tangent poles, and only one (7%) electrocution occurred on a single-phase tangent pole. Six (11%) of 56 nearby poles were equipment or intersection poles; 50 (89%) were tangent poles. We did not detect grounding above the lowest energized conductor on any adjacent pole. Overall, 10 electrocutions occurred across 16 nontangent poles and 4 electrocutions occurred across 54 tangent poles (data from 14 electrocution poles and 4 adjacent poles at each electrocution pole). Electrocutions were significantly (χ² = 23.41; df = 1; P < 0.001) more likely to occur on nontangent poles than on tangent poles. Proactively retrofitting all 16 nontangent PRECorp poles would have prevented 71% of electrocutions. By contrast, proactively retrofitting all 54 tangent poles would have prevented only four (29%) electrocutions. Accounting for the distribution of pole types on the landscape, electrocutions occurred at a rate 8.4 times greater on intersection poles, equipment poles, and poles with grounding above the lowest energized conductor than on tangent poles without grounding above the lowest energized conductor.

We evaluated whether the 2014 model might be too biased by geography and sample size to be useful in determining retrofitting priorities for golden eagles outside the study population. We could not test that quantitatively, so we considered similarities in pole construction practices and golden eagle sizes, consistency between the 2014 model's variables and structure-specific factors identified as influential in other studies, and new data from satellite-tagged electrocuted golden eagles. We found high levels of consistency, suggesting the 2014 model can be extrapolated. Although our focus was on golden eagles, consistency with studies of other electrocuted birds suggests that our conclusions also may apply to other species.

The 2014 model has potential for widespread use, but also has limitations. One practical limitation is that correctly identifying grounding on poles is essential to generating accurate results. Google Earth imagery will likely be insufficient for identifying grounding, so users may need to physically visit poles to correctly determine whether a path to ground is present. Another limitation is that the 2014 model's input variables are deliberately simple to make the model accessible to a wide range of users (Dwyer and Harness 2012). Although the choice for simplicity was deliberate, it nevertheless results in a weakness wherein the model considers the number of conductors, but not their arrangement.

We used Google Earth imagery to develop risk index scores. The approach was mostly successful, but the resolution of the imagery was insufficient to see pole ground wires. The resolution was often sufficient to see where neutral wires were attached and thus to infer whether grounding was likely. In an actual retrofitting program, questions regarding grounding would be resolved through visiting and evaluating poles being assessed. We also failed to detect lightning arresters on two poles and switches on one pole where electrocution of golden eagles occurred, leading to incorrectly low risk index scores. Again, in a mitigation program resolution of this issue could be achieved through visiting and evaluating poles in person. Despite these errors, there was a correct association between most electrocution locations and equipment poles. Given these patterns, we conclude that using Google Earth to characterize poles was effective but imperfect.

In the United States, determining how to prioritize poles for retrofitting is typically left to the discretion of electric utilities. In some cases, this is changing when the USFWS becomes involved through compensatory mitigation permitting. In such cases, the USFWS helps guide incidental take permittees and their utility partners to which poles can be retrofitted to fulfill permit conditions. To create consistency and transparency in the process, the USFWS and electric industry need a mechanism of scaling the mitigation benefits of retrofitting various pole types. We describe what might be the least conservation-conservative equivalency, a highly conservation-conservative equivalency, and a middle ground for this scaling. The least conservation-conservative equivalency is to assume that all poles are equal so retrofitting any pole is equivalent. Clearly, not all pole configurations pose equal electrocution risk, and a 1:1 equivalency is inappropriate. A highly conservation-conservative equivalency might be to count pieces of equipment needed to retrofit a pole configuration that electrocutes a relatively large number of golden eagles. Typically, there is an association between three-phase transformer banks and high electrocution risk (see citations in Tables 1 and 2). Retrofitting this configuration frequently requires 1 conductor cover or dead-end cover, 9 jumper covers, 3 lightning arrester covers, 3 cutout covers, and 3 bushing covers, for a total of 19 pieces of retrofitting equipment. If this configuration is used as a baseline, installation of 19 pieces of retrofitting equipment could be determined as the equivalent of retrofitting one high-risk pole regardless of how many poles the equipment is actually distributed across. Many intersection and equipment poles do not require as many pieces of equipment to retrofit however; so, a 19:1 equivalency is probably also inappropriate. Equivalency between tangent poles and nontangent poles appears to be somewhere between 1:1 and 19:1.

Our results may be useful in identifying a middle ground. In this study, electrocutions occurred 8.0–8.4 times more frequently on intersection poles, equipment poles, and poles with grounding above the lowest energized conductor (nontangent poles) than on tangent poles without grounding above the lowest energized conductor. This ratio exceeds the ratio of 5.25:1 calculated by Mojica et al. (2022) in their assessment of electrocution data in Lehman et al. (2010). Because the Lehman et al. (2010) data were drawn exclusively from an oil and gas field where the density of nontangent poles was unusually high, we suggest the ratio of 5.25:1 is informative in that case, but not applicable across the entire western United States. Instead, we suggest the ratio of poles in Lehman et al. (2010) may indicate that the range in our findings be refined to its low end of 8.0:1.

Using the 8.0:1 ratio, if the developer of a wind energy facility is required to retrofit 10 high-risk poles (in the vernacular of the USFWS; Mojica et al. 2022), 10 nontangent poles could be retrofitted, 80 tangent poles could be retrofitted, or the ratio could be applied to facilitate a mix of nontangent and tangent poles. This would enable electric utilities to focus either on nontangent poles or on circuits. Importantly, this does not imply that electric utilities cannot retrofit additional tangent poles if data exist indicating electrocutions are occurring on them in their service areas. If an electric utility knows that golden eagle electrocutions are occurring, poles should be retrofitted regardless of whether funding from an external compensatory mitigation plan is available.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author.

Reference S1. Harness RE. 2000. Effectively retrofitting powerlines to reduce raptor mortality. Louisville, Kentucky: Institute of Electrical and Electronics Engineers, Rural Electric Power Conference.

Available: https://doi.org/10.3996/JFWM-21-046.S1 (720 KB PDF)

Reference S2. Harness RE, Dwyer JF. 2015. Avian electrocution risk assessment predictive model. Final report submitted to the USFWS Western Golden Eagle Conservation Team. Fort Collins, Colorado: EDM International.

Available: https://doi.org/10.3996/JFWM-21-046.S2 (3.513 MB PDF) and https://ecos.fws.gov/ServCat/Reference/Profile/83429

Reference S3.[USFWS] U.S. Fish and Wildlife Service. 2013. Eagle conservation plan guidance: module land-based wind energy, version 2. Washington, D.C.: U.S. Fish and Wildlife Service.

Available: https://doi.org/10.3996/JFWM-21-046.S3 (2.063 MB PDF) and https://www.fws.gov/migratorybirds/pdf/management/eagleconservationplanguidance.pdf

Reference S4.[USFWS] U.S. Fish and Wildlife Service. 2016a. Bald and golden eagles: population demographics and estimation of sustainable take in the United States, 2016 update. Washington, D.C.: U.S. Fish and Wildlife Service.

Available: https://doi.org/10.3996/JFWM-21-046.S4 (3.724 MB PDF) and https://www.fws.gov/migratorybirds/pdf/management/EagleRuleRevisions-StatusReport.pdf

Tracy Jones of Powder River Energy Corporation (Gillette, WY) and Robert Murphy of Eagle Environmental (Santa Fe, NM) provided location data for the golden eagle electrocutions evaluated in this study. Their studies were funded by the United States Army and the USFWS. EDM International (EDM; Fort Collins, CO) provided financial support for this study. None of the contributing or funding parties had any role in determining power poles assessed. Conflict of Interest statement: EDM provides consulting support to electric utilities implementing avian retrofitting throughout North America and internationally. We thank Duncan Eccleston, Rick Harness, Matthew Stuber, Gary Williams, the USFWS Western Golden Eagle Team, USFWS National Raptor Program, and the Journal's Associate Editor and anonymous reviewers for comments that improved this work.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.