Content uploaded by Roel May
Author content
All content in this area was uploaded by Roel May
Content may be subject to copyright.
Collision risk in white-tailed eagles
Modelling kernel-based collision risk using
satellite telemetry data in Smøla wind-
power plant
Roel May
Torgeir Nygård
Espen Lie Dahl
Ole Reitan
Kjetil Bevanger
692
NINA Publications
NINA Report (NINA Rapport)
This is a electronic series beginning in 2005, which replaces the earlier series NINA commissioned
reports and NINA project reports. This will be NINA’s usual form of reporting completed research,
monitoring or review work to clients. In addition, the series will include much of the institute’s other
reporting, for example from seminars and conferences, results of internal research and review work
and literature studies, etc. NINA report may also be issued in a second language where appropri-
ate.
NINA Special Report (NINA Temahefte)
As the name suggests, special reports deal with special subjects. Special reports are produced as
required and the series ranges widely: from systematic identification keys to information on impor-
tant problem areas in society. NINA special reports are usually given a popular scientific form with
more weight on illustrations than a NINA report.
NINA Factsheet (NINA Fakta)
Factsheets have as their goal to make NINA’s research results quickly and easily accessible to the
general public. The are sent to the press, civil society organisations, nature management at all lev-
els, politicians, and other special interests. Fact sheets give a short presentation of some of our
most important research themes.
Other publishing
In addition to reporting in NINA’s own series, the institute’s employees publish a large proportion of
their scientific results in international journals, popular science books and magazines.
Norwegian Institute for Nature Research
Collision risk in white-tailed eagles
Modelling kernel-based collision risk using
satellite telemetry data in Smøla wind-
power plant
Roel May
Torgeir Nygård
Espen Lie Dahl
Ole Reitan
Kjetil Bevanger
NINA Report 692
2
CONTACT DETAILS
NINA head office
Postboks 5685 Sluppen
NO-7485 Trondheim
Norway
Phone: +47 73 80 14 00
Fax: +47 73 80 14 01
NINA Oslo
Gaustadalléen 21
NO-0349 Oslo
Norway
Phone: +47 73 80 14 00
Fax: +47 22 60 04 24
NINA Tromsø
Framsenteret
NO-9296 Tromsø
Norway
Phone: +47 77 75 04 00
Fax: +47 77 75 04 01
NINA Lillehammer
Fakkelgården
NO-2624 Lillehammer
Norway
Phone: +47 73 80 14 00
Fax: +47 61 22 22 15
www.nina.no
May, R., Nygård, T., Dahl, E.L., Reitan, O. & Bevanger, K. 2011.
Collision risk in white-tailed eagles. Modelling kernel-based colli-
sion risk using satellite telemetry data in Smøla wind-power plant. –
NINA Report 692. 22 pp.
Trondheim, May, 2011
ISSN: 1504-3312
ISBN: 978-82-426-2277-8
COPYRIGHT
© Norwegian Institute for Nature Research
The publication may be freely cited where the source is ac-
knowledged
AVAILABILITY
Open
PUBLICATION TYPE
Digital document (pdf)
EDITION
QUALITY CONTROLLED BY
Jiska van Dijk
SIGNATURE OF RESPONSIBLE PERSON
Research director Signe Nybø (sign.)
CLIENT(S)
NINA / SSE Renewables
CLIENTS’ CONTACT PERSON(S)
Roel May / Chris Marden
COVER PICTURE
© Roel May
KEY WORDS
- Norway, Smøla wind-power plant
- White-tailed eagle
- Satellite telemetry
- Collision risk modelling
NØKKELORD
- Norge, Smøla vindpark
- Havørn
- Satellittelemetri
- Kollisjonsrisikomodellering
NINA Report 692
3
Abstract
May, R., Nygård, T., Dahl, E.L., Reitan, O. & Bevanger, K. 2011. Collision risk in white-tailed
eagles. Modelling kernel-based collision risk using satellite telemetry data in Smøla wind-
power plant. – NINA Report 692. 22 pp.
Large soaring birds of prey, such as the white-tailed eagle, are recognized to be perhaps the
most vulnerable bird group regarding risk of collisions with turbines in wind-power plants. Their
mortalities have called for methods capable of modelling collision risks in connection with the
planning of new wind-power developments. The so-called “Band model” estimates collision risk
based on the number of birds flying through the rotor swept zone and the probability of being
hit by the passing rotor blades. In the calculations for the expected collision mortality a correc-
tion factor for avoidance behaviour is included. The overarching objective of this study was to
use satellite telemetry data and recorded mortality to back-calculate the correction factor for
white-tailed eagles. The Smøla wind-power plant consists of 68 turbines, over an area of ap-
proximately 18 km
2
. Since autumn 2006 the number of collisions has been recorded on a
weekly basis. The analyses were based on satellite telemetry data from 28 white-tailed eagles
equipped with backpack transmitters since 2005. The correction factor (i.e. “avoidance rate”)
including uncertainty levels used within the Band collision risk model for white-tailed eagles
was 99% (94-100%) for spring and 100% for the other seasons. The year-round estimate, irre-
spective of season, was 98% (95-99%). Although the year-round estimate was similar, the cor-
rection factor for spring was higher than the correction factor of 95% derived earlier from van-
tage point data. The satellite telemetry data may provide an alternative way to provide insight
into relative risk among seasons, and help identify periods or areas with increased risk either in
a pre- or post construction situation.
Roel May,
roel.may@nina.no
Torgeir Nygård,
torgeir.nygard@nina.no
Espen Lie Dahl,
Ole Reitan,
espenlie.dahl@nina.no
ole.reitan@nina.no
Kjetil Bevanger,
kjetil.bevanger@nina.no
NINA Report 692
4
Sammendrag
May, R., Nygård, T., Dahl, E.L., Reitan, O. & Bevanger, K. 2011. Collision risk in white-tailed
eagles. Modelling kernel-based collision risk using satellite telemetry data in Smøla wind-
power plant. – NINA Rapport 692. 22 s.
Store rovfugler, som havørn, er kjent for å være sårbare for kollisjoner med turbiner i vindkraft-
verk. Deres dødelighet er benyttet i modeller for kollisjonsrisiko i forbindelse med planlegging-
en av ny vindkraftutbygging. Den såkalte "Band-modellen" beregner kollisjonsrisiko basert på
antall fugler som flyr gjennom rotorsonen og sannsynligheten for at de blir rammet av de pas-
serende rotorbladene. I beregning av den forventede kollisjonsdødeligheten inngår en korrek-
sjonsfaktor for unnvikelsesatferd. Det overordnede målet for denne studien var å bruke fluktda-
ta og registrert dødelighet til å beregne korreksjonsfaktoren for havørn. Smøla vindkraftverk
består av 68 turbiner, over et område på ca 18 km
2
. Siden høsten 2006 har en søkt etter kolli-
sjonsdrepte fugler ukentlig. Analysene var basert på satellittelemetridata fra 28 havørn merket
med ryggsekksendere siden 2005. Korreksjonsfaktoren for havørn (dvs. "unnvikelsesraten")
inklusive usikkerhetsnivåer som brukes i Band kollisjonsrisikomodellering er beregnet til å
være 99 % (94-100 %) for våren og 100 % for de andre sesonger. Det årlige estimatet, uav-
hengig av årstider, var 98 % (95-99 %). Selv om det årlige estimatet var lik, var korreksjonsfak-
toren for våren høyere enn korreksjonsfaktoren på 95 % basert tidligere på observasjons-
punktdata. Satellittelemetridata kan være en alternativ måte å gi innsikt i den relative risikoen
mellom sesongene, og bidra til å identifisere perioder eller områder med økt risiko, enten før
eller etter utbygging.
Roel May,
roel.may@nina.no
Torgeir Nygård,
torgeir.nygard@nina.no
Espen Lie Dahl,
espenlie.dahl@nina.no
Ole Reitan,
ole.reitan@nina.no
Kjetil Bevanger,
kjetil.bevanger@nina.no
NINA Report 692
5
Contents
Abstract ....................................................................................................................................... 3
Sammendrag ............................................................................................................................... 4
Contents ...................................................................................................................................... 5
Foreword ..................................................................................................................................... 6
1 Introduction ............................................................................................................................ 7
2 Material and methods............................................................................................................ 8
2.1 Study area and study species ......................................................................................... 8
2.2 Searches for collision victims .......................................................................................... 9
2.3 Satellite telemetry data .................................................................................................... 9
2.4 Collision risk modelling .................................................................................................. 10
3 Results .................................................................................................................................. 13
3.1 Background data ............................................................................................................ 13
3.1.1 Collision victim searches .................................................................................... 13
3.1.2 Estimation of population size .............................................................................. 14
3.2 Collision risk modelling .................................................................................................. 14
3.2.1 Seasonal flight activity ........................................................................................ 14
3.2.2 Collision risk including uncertainty levels ........................................................... 17
4 Discussion ............................................................................................................................ 19
5 References ........................................................................................................................... 21
NINA Report 692
6
Foreword
In July 2010 NINA was contacted by Chris Marden from SSE Renewables, Scotland. He asked
whether we could analyze data collected within the Smøla wind-power plant to derive avoid-
ance rates for white-tailed eagles using the so-called ‘Band’ collision risk model. SSE Renew-
ables wished to receive an increased insight into these avoidance rates for use in a pre-
construction collision risk assessment for white-tailed eagles concerning the development of a
wind-power plant in Scotland. The analyses of vantage point data was presented in NINA re-
port 639. This report presents the collision risk modelling results based on satellite telemetry of
white-tailed eagles. This work was financed by SSE Renewables and NINA.
25.05.2011 Roel May
NINA Report 692
7
1 Introduction
The evidence of bird mortality due to large-scale wind energy development is increasing (Hunt
et al. 1998; Johnson et al. 2002; Langston & Pullan 2003; Thelander et al. 2003; Barrios &
Rodriguez 2004; Smallwood & Thelander 2005; Drewitt & Langston 2006, 2008; Madders &
Whitfield 2006; DeLucas et al. 2008, Bevanger et al. 2009), and a particular concern has been
raised regarding raptors. Large soaring birds of prey are recognized to be perhaps the most
vulnerable regarding risk of collisions with turbines in wind-power plants (Barrios & Rodriguez
2004, Hoover & Morrison 2005, Smallwood & Thelander 2008).
These mortalities have called for methods capable of modelling collision risks in connection
with the planning of new wind-power developments both in Norway and in other countries. One
model has been developed that has been widely used, the so-called “Band model” (SNH 2000,
Band et al. 2007). This method is based on 1) estimating collision risk based on the calculated
likelihood of a bird being hit by the rotor blades given that it passes through the rotor-swept
zone (RSZ), multiplied by 2) the estimated number of birds flying through the RSZ throughout
a given time unit (Band et al. 2007). The first step is based on the technical specifications of
the turbines and the morphology (e.g. wing aspect), speed and flight behaviour (flapping or
soaring) of the bird, while the second step involves the use of field observations. The model is
finally adjusted by multiplying its outcome with a correction factor for taking into account,
among others, avoidance.
As part of the BirdWind-project (“Pre- and post-construction studies of conflicts between birds
and wind turbines in coastal Norway”) (cf. Bevanger et al. 2010), juvenile white-tailed eagles
(Haliaeetus albicilla) have been captured and equipped with GPS/satellite transmitter back-
packs at the Smøla wind-power plant. The aim of the satellite telemetry studies was to investi-
gate how the construction of Smøla wind-power plant would affect white-tailed eagle flight be-
haviour (Nygård et al. 2010; Bevanger et al. 2010). In 2008, eagle flight behaviour was ob-
served from 12 vantage points, six inside the wind-power plant area and six in adjacent control
areas (Hoel 2009). This data, together with additional vantage point data collected by Rowena
Langston (RSPB) in May 2009, formed the basis for the ‘Band’ collision risk modelling pre-
sented in NINA report 639 (May et al. 2010).
The objective of this study was to back-calculate the correction factor for white-tailed eagles
within the Smøla wind-power plant using satellite telemetry data. These data enables deriving
correction factors for different seasons, and may function as an independent validation of the
correction factors derived for the vantage point data (May et al. 2010). The approach in princi-
ple followed the standard collision risk assessment as promoted by Scottish Natural Heritage
(SNH 2000, 2005, 2010; Band et al. 2007). The overarching approach was to use actual flight
data and actual mortality to back-calculate the correction factor for white-tailed eagles.
NINA Report 692
8
2 Material and methods
2.1 Study area and study species
Smøla is an archipelago located off the coast of Møre & Romsdal County, Central Norway
(63°24´N, 8°00´E) (Figure 1), and consists of a large main island together with approximately
5500 smaller islands, islets and small skerries. The terrain is flat and the highest peak on the
main island is only 64m. The habitats are characterised by heather moors with a mix of small
and large marshes. The Smøla wind-power plant was built in two phases by the Norwegian
energy company Statkraft, the first phase being finished in September 2002, while the second
became operational in August 2005. Since 2005, the wind-power plant has comprised 68 tur-
bines. The wind-power plant covers an area of 17.83 km
2
; represented by the minimum convex
polygon (i.e. envelope) around the outermost turbines including a 200-m buffer. The wind tur-
bines operate in two different gears at 11 and 16 rotations per minute (RPM), depending on
wind speeds: first gear at 11 RPM (≥3
m
/
s
but <6
m
/
s
); second gear at 16 RPM (≥6
m
/
s
but <25
m
/
s
). Below 3
m
/
s
the turbines idle, while at wind speeds ≥25
m
/
s
they stop.
Figure 1. Smøla wind-power plant, central Norway. The green line indicates minimum convex
polygon including a 200-m buffer around the outermost turbines.
The white-tailed eagle is distributed in parts of northern, eastern and central Europe, across
Siberia into China. Its food includes fish, birds, carrion and, occasionally, small mammals.
They generally form monogamous pairs for life, although if one dies, replacement can take
place rather quickly. The nest is a huge edifice of sticks in a tree, on a coastal cliff, or simply
on the flat ground. White-tailed eagles have high territory fidelity. Once they breed, nests are
often reused, sometimes for decades by successive generations of birds (Orta 1994). The ter-
ritory normally covers 30-70km
2
(although smaller on Smøla), usually in sheltered coastal loca-
tions (Gjershaug 1994). Approximately 55-60 white-tailed eagle territories have been reported
in the Smøla archipelago (Bevanger et al. 2010), but recent data from DNA analyses of feath-
ers indicate that this number is too high (own data, unpublished).
NINA Report 692
9
2.2 Searches for collision victims
Searches for dead birds near turbines have been carried out since 1 August 2006 using two
dogs trained to a search image of both feathers and dead birds. A riesenschnauzer was
trained to search for feathers before the start of the project in August 2006. In addition, a briard
was retrained from a human rescue dog to a dog searching for dead birds by reinforcement
when he found dead birds and feathers during the searches. A dog searches mainly by its ol-
factory sense, and therefore covers an area determined by movements of scent in the air. A
dog needs only a few molecules to respond to a scent, and therefore is expected to be more
efficient than what is possible with visual searches alone. By making use of this capacity to-
gether with wind direction and speed we achieved as efficient searches as possible.
Of the 68 turbines in the Smøla wind-power plant area (WPA), 25 were selected as primary
search turbines. These were searched weekly throughout the whole year, i.e. every seven
days (variation mainly 6-8 days). Earlier studies in Altamont Pass, California, have found a
slightly higher collision rate for golden eagles (Aquila chrysaetos) at the end turbines in each
string (Smallwood & Thelander 2005, 2008), and the first nine white-tailed eagle victims at
Smøla were found in the northern part of the WPA. We therefore selected 17 outermost tur-
bines and eight inner turbines as primary search turbines. The other 43 turbines were
searched once each month during periods with expected high activity of birds, mainly March-
June, and less intensively during winter (0-2 times depending on snow conditions). Depending
on the wind direction each turbine was searched upwind, perpendicular to the wind direction,
within a radius of approximately 100 meters from the base of the turbine tower. Objects from
dead white-tailed eagles have been found up to about 120 m from the turbines. In addition to
the search results, dead white-tailed eagles found by Statkraft personnel and the general pub-
lic have been immediately reported and collected. All dead white-tailed eagles have been au-
topsied and X-rayed to verify cause of death.
A possible scavenger removal bias has been investigated. There is an absence of potential
mammalian scavengers on the island of Smøla except for mink (Neovison vison). The main
scavengers on bird carcasses on Smøla seem to be white-tailed eagle, hooded crow (Corvus
cornix) and raven (Corvus corax). Parts of a carcass may be removed, but in general each
carcass seems to be present for many months. The main bias at Smøla WPA may therefore
be the crippling bias (Bevanger 1999), where birds are injured but survive the collision and die
outside the search area.
2.3 Satellite telemetry data
From the entire dataset of all marked fledgling individuals (>50; Nygård et al. 2010), 28
subadult individuals were included in the analyses. They represented the individuals from their
first winter and onwards whose transmitters rendered at least 20 relocations per season (n =
22,186). In this analysis we assumed that the flight activity of the subadult eagles was repre-
sentative for the entire population. We obtained season-specific bird speed and altitude above
ground level (agl) from 20 white-tailed eagles equipped with GPS satellite transmitters which
rendered 3D location data including information on instantaneous speed (n = 15,716). In order
to be able to estimate the flight activity within the rotor swept zone, we calculated the propor-
tion of time the eagles spent in active flight (instantaneous speed > 0) and within the rotor
swept zone (altitude agl ≥ 29m and ≤ 111m) for each season. The eagles spent between 2.5%
and 5.8% of their time in active flight (0.025, 0.058, 0.026 and 0.047 for winter, spring, summer
and autumn, respectively). When in active flight the eagles spent between 17% and 32% of
their time within the rotor swept zone (0.320, 0.241, 0.231 and 0.170 for winter, spring, sum-
mer and autumn, respectively). Thus, the multiplicative proportion of time spent in active flight
within the rotor swept zone for winter, spring, summer and autumn were 0.008, 0.014, 0.006
and 0.008, respectively.
NINA Report 692
10
2.4 Collision risk modelling
All programming and statistics were performed in the statistical programme R 2.10.1 (R Devel-
opment Core Team 2009).
The modelling of collision risk in white-tailed eagles in principle follows the methodology de-
scribed by the Scottish Natural Heritage (SNH) guidance note (SNH 2000, Band et al. 2007).
For calculation of the number of bird transits (per season) through the rotors within the wind-
power plant area, we followed SNH’ second approach “Birds using the wind farm airspace”.
This approach is most appropriate for birds such as raptors which occupy a recognised terri-
tory, and where observations have led to some understanding of the likely distribution of flights
within this territory. The standard way of estimating the correction factor does not render any
information on the uncertainty involved in the modelling. Here, we have also modelled collision
risk incorporating the (observed) variation in flight activity, day length, and wind and bird speed
to obtain the correction factor (i.e. “avoidance rate”) and associated uncertainty in the Band
modelling. Below follows a stepwise explanation of the approach followed.
Number of birds colliding per season
=
Number of birds flying through the rotor swept zone (Stage 1)
x
Probability of one bird being hit when flying through rotor swept zone (Stage 2)
x
Correction factor for taking into account, among others, avoidance (Stage 3)
Stage 1: Number of birds flying through the rotor swept zone
In order to derive the variation in the number of birds flying through the rotor swept zone
(RSZ), the following calculations were done including variation in flight activity, day length and
bird speed.
1. From the satellite telemetry data, kernel utilization distributions (UD) were calculated
separately for each individual and each season (across years). The UD grids were
bounded within Smøla municipality with a pixel resolution of 200x200 meters. Each UD
was normalized to indicate proportion of time spent within each pixel by dividing each
pixel’s value with the sum over all pixel values for the entire UD. All individual UDs were
thereafter summed for each season separately, divided by the number of individuals and
multiplied by the population size of the Smøla archipelago. This population-UD was then
multiplied by the proportion of time eagles spent in active flight within the rotor swept zone
(see paragraph 2.3) and adjusted for the surface area each pixel represented. This re-
sulted in the estimated total flight activity/hour/km
2
for each season (F).The flight activity
within the wind-power plant area (WPA) was obtained by selecting those pixels overlaying
the WPA-buffer map (see Figure 1). A control subset was established by randomly select-
ing equally many pixels outside the WPA-buffer map, but on the main island.
2. Possible effects of season (winter, spring, summer and autumn) and placement (inside
WPA versus random control outside WPA), including their interaction, on estimated flight
activity were analyzed using linear regression.
3. The wind-power plant area A was defined as the minimum convex polygon (i.e. envelope)
around the outermost turbines including a 200-m buffer (17.83 km
2
; Figure 1).
NINA Report 692
11
4. The period of interest T was calculated by multiplying the number of days for each season
(90-92 days) by the average day length for each season. Day length was defined as the
number of hours between sun rise and sun set for Trondheim, Norway
(
http://www.timeanddate.com/worldclock/astronomy.html?n=288).
5. The bird occupancy n for each season was estimated within the WPA. This is the number
of birds present multiplied by the time spent flying in the WPA for the period of interest for
which the collision estimate is being made: n = F x A x T.
6. The bird occupancy was first calculated for each pixel. From these pixel-based estimates
of bird occupancy, we calculated log-transformed mean and standard deviation from the
pixel-based estimates of bird occupancy n for the entire wind-power plant. These were
used to derive 10,000 randomly created estimates of bird occupancy n assuming a log-
normal distribution.
7. Thereafter, a 'flight risk volume' V
w
was identified, equalling the area of the wind-power
plant multiplied by the rotor diameter (= 82 m).
8. The combined volume swept by the wind-power plant rotors was calculated as V
r
= N x
πR
2
x (d + L) where N is the number of wind turbines (= 68), R equals the rotor length
(=41 m), d is the depth of the rotor back to front (assumed to be 2 m), and L is the length
of the bird (0.8 m; source:
http://blx1.bto.org/birdfacts/results/bob2430.htm).
9. The bird occupancy of the volume swept by the rotor blades is then n x (V
r
/ V
w
) bird-
seconds.
10. The time taken for a bird to make a transit through the rotor disk and completely clear the
rotors was calculated as t = (d + L) / v where v is the speed in
m
/
s
of the bird through the
rotor disk. Using the log-transformed mean and standard deviation in bird speed for each
season, we derived a random dataset of 10,000 estimates of bird speed which was as-
sumed to follow a lognormal distribution.
11. Finally, the number of bird transits through the rotor swept zone, the total occupancy of
the volume swept by the rotors in bird-seconds was divided by the transit time t:
Number
of birds passing through rotor swept zone
= n x (V
r
/ V
w
) / t. Note in this calculation that the
factor (d + L) actually cancels itself out, so only assumed values need be used – it is used
above to help visualise the calculation.
Stage 2: Probability of one bird being hit when flying through the rotor swept zone
This stage computes the probability of a bird being hit when making a transit through the rotor
swept zone. The probability depends on the size of the bird (both length and wingspan), the
breadth and pitch of the turbine blades, the rotation speed of the turbine, and of course the
flight speed of the bird. In order to derive the hit probability including standard deviation, we
obtained the (log-transformed) mean and standard deviation of wind speed and bird speed.
Using these estimates, the hit probability was calculated for each record. Finally, the mean and
standard deviation were calculated from these 10,000 hit probability estimates. The probability
was calculated following the exact formula laid out on an Excel spreadsheet available from the
renewable energy pages of the Scottish Natural Heritage web site:
http://www.snh.gov.uk/docs/C234672.xls (Band et al. 2007; SNH 2000), using the following
input parameters:
• K (3D probability): 1
• Number of rotor blades: 3
• Maximal chord: 3.296 m
• Pitch: 10 degrees
• Bird length: 0.8 m
• Wing span: 2.315 m (average of males and females across age classes; Love 1983)
• Aspect ratio: 0.35 (automatically calculated from the two parameters above)
NINA Report 692
12
• Flight type: (2/π)
F
, with F = 1 (flapping (=0) or gliding (=1))
• Average bird speed: mean and standard deviation derived from GPS satellite transmitters for
each season separately; using these statistics, we derived a random dataset of 10,000 es-
timates of bird speed which were both assumed to follow a lognormal distribution
• Rotor diameter: 82 m
• Rotation period: mean and standard deviation of wind speed at nacelle-height (70 m) were
calculated from the meteorological station within the WPA. Using these statistics, we de-
rived a random dataset of 10,000 estimates of wind speed which was assumed to follow a
lognormal distribution. The estimates were thereafter classified into RPMs as follows: 1 (<3
m
/
s
; idling); 11 (≥3
m
/
s
but <6
m
/
s
; first gear); 16 (≥6
m
/
s
but <25
m
/
s
; second gear); 0.001 (≥25
m
/
s
; stopped).
Stage 3: Number of birds colliding per season – derivation of the correction factor
The number of birds colliding per season was estimated by multiplying the 10,000 estimates of
bird occupancy with the hit probability estimates. The correction factor was derived as follows:
CF = 1 – actual collisions / (number of birds flying through the RSZ x hit probability). From
these 10,000 estimates of bird occupancy and hit probability, the mean and standard deviation
in the correction factor were calculated.
NINA Report 692
13
3 Results
3.1 Background data
3.1.1 Collision victim searches
Altogether, 39 dead or injured white-tailed eagles have been found at Smøla WPA in the pe-
riod from the beginning of August 2005 until 15 November 2010. During these five years, on
average 7.8 dead white-tailed eagles were found per year. This equals on average 0.12 dead
white-tailed eagles per turbine per year.
Of the total 39 dead or injured birds, 28 (72%) were found during a period of 2-2.5 months
each spring. The period with high level of fatalities varied between years due to prevailing
weather conditions. During autumn 6 (15%) dead/injured white-tailed eagles were found (Fig-
ure 2).
Figure 2. Number of white-tailed eagles found dead or injured at the Smøla turbines until 15
November 2010. The first was found in August 2005, and regular searches were initiated in
2006. Winter = December-February; Spring = March-May; Summer = June-August; Autumn =
September-November.
The age distribution of the 39 birds found was 21 (54%) adults, 11 (28%) subadult birds and 7
(18%) juveniles. The adults were mainly found in the spring or autumn, the subadult birds,
mainly in spring, and the juveniles in the autumn and their first spring (Figure 3).
0
1
2
3
4
5
6
7
8
2005
2006
2007
2008
2009
2010
Winter
Spring
Summer
Autumn
NINA Report 692
14
Figure 3. Age distribution of white-tailed eagles found dead or injured at the Smøla turbines
until 15 November 2010. Juveniles represent birds in their first calendar year; subadults calen-
dar year 2-5, and adults calendar year ≥6. See also legend in Figure 2.
3.1.2 Estimation of population size
The number of eagle chicks produced on Smøla during 2006-2010 varied between 21 and 36
(average 27.4, SD = 5.9). At the same time, the average number of breeding pairs was esti-
mated at 45 (90 adults). Based on DNA analyses of feathers found at and around nests, the
number of non-breeding individuals during the same time interval (i.e. “floaters”) was estimated
at ca 30% (circa 27 individuals), resulting in an adult population size of circa 117 adult birds (in
their 6
th
calendar year or older). Based on recorded deaths from dog searches and satellite
tagging, we estimated the yearly mortality rate of juveniles and subadult to be circa 14% per
year, resulting in 76 subadults (calendar year 2-5) present at any one time.
From the extensive satellite telemetry database, most juveniles and subadults have shown to
migrate away from their natal areas on Smøla during summer/autumn. They may then move
large distances, and have mainly moved north (Nygård et al. 2010). This means that the total
population at Smøla shrinks during summer/autumn. We took this into account in the analyses
by calculating the proportion of juveniles/subadults present at Smøla for each season. Based
on the telemetry data, these proportions were 0.64, 0.90, 0.50 and 0.38 for winter, spring,
summer and autumn, respectively.
3.2 Collision risk modelling
3.2.1 Seasonal flight activity
Flight activity (step 1) was calculated for each season. Thereafter we tested for possible ef-
fects of season and placement (inside or outside the wind-power plant) using a linear model
(step 2; Table 3, Figures 4 – 6). The data indicated both significant differences in flight activity
outside versus inside the wind-power plant, an effect that varied over the seasons. Flight activ-
0
2
4
6
8
10
12
Winter
Spring
Summer
Autumn
Juvenile
Subadult
Adult
NINA Report 692
15
ity in white-tailed eagles also showed a significant variation between seasons. Bird occupancy
(step 6) was estimated at 30.30, 148.86, 26.79 and 23.56 hours for winter, spring, summer and
autumn, respectively (median values). For each of these seasons, the median number of bird
transits through the rotor swept zone (step 11) was 304, 1172, 208 and 184.
Table 3. Analysis-of-variance results from the linear model.
Covariate
df
F-value
P-value
(Intercept)
1,3552
10626.20
<0.001
Inside/Outside
1,3552
101.61
<0.001
Season
3,3552
1932.45
<0.001
Interaction
3,3552
129.73
<0.001
Figure 4. Maps showing the estimated distribution of the total flight activity within the rotor
swept zone per season for the subadult population at Smøla; more intense red indicates in-
creased activity. The placement of the wind-power plant is indicated with the blue broken-line
polygon.
NINA Report 692
16
Figure 5. Map
showing the
estimated dis
tribution of the
total flight activity
within the
rotor swept zone year-round
for the subadult population at
Smøla; more intense red indi-
cates increased activity. The
placement of the wind-power
plant is indicated with the blue
broken-line polygon.
NINA Report 692
17
Figure 6. Box plots showing the estimated total flight activity within the rotor swept zone per
season and placement for the entire white-tailed eagle population at Smøla. The box indicates
the 25
th
and 75
th
percentile; while the whiskers indicate the 5
th
and 95
th
percentile. The thick
line indicates the median (50
th
percentile).
3.2.2 Collision risk including uncertainty levels
The correction factor was estimated including variance in wind speed and bird speed (which
were assumed to follow a lognormal distribution). Median wind speeds at nacelle-height (70 m)
were 6.59
m
/
s
, 5.70
m
/
s
, 4.53
m
/
s
and 6.60
m
/
s
for winter, spring, summer and autumn, respec-
tively (median RPMs: 16, 16, 11, 16). Median bird speeds for these seasons were estimated
from the telemetry data at 12
m
/
s
, 20
m
/
s
, 19
m
/
s
and 23
m
/
s
. Using these estimates, we iterated
the Band-model calculations 10,000 times to produce robust estimates of the correction factor.
Given the variation in wind and bird speed for each season, the median hit probability was
0.193, 0.115, 0.108 and 0.118 for winter, spring, summer and autumn, respectively (stage 2,
Figure 7). The year-round median hit probability was estimated at 0.121. From the number of
recorded collision (paragraph 3.1), the estimated flight activity/occupancy (stage 1) and the hit
probability (stage 2) the median correction factors for the Band-model (i.e. “avoidance rate”,
NINA Report 692
18
stage 3) were estimated at 1.000 for winter, summer and autumn and 0.991 for spring (Table
4, Figure 8). The year-round median correction factor was estimated at 0.975 (interquartile
range: 0.946 – 0.989; 5
th
percentile: 0.864). In this analysis we assumed that the flight activity
of the subadult eagles was representative for the entire population. When only taking into ac-
count the juvenile/subadult collisions, the correction factor could only be derived for spring:
0.989 (interquartile range: 0.933 – 1.000). The correction factors for the other seasons were
estimated at 1.000 (interquartile range: 1.000 – 1.000).
Table 4. Uncertainty statistics for the modelled correction factors using the Band collision risk
model on satellite telemetry data from Smøla.
Season
Median
5
th
percentile
Interquartile range
winter
1.000
1.000
1.000 – 1.000
spring
0.991
0.831
0.942 – 1.000
summer
1.000
0.933
0.977 – 1.000
autumn
1.000
0.800
0.931 – 1.000
Figure 7. Box plot showing the hit probabil-
ity per season. The box indicates the 25
th
and 75
th
percentile; while the whiskers indi-
cate the 5
th
and 95
th
percentile. The thick
line indicates the median (50
th
percentile).
Figure 8.
Box plot showing the correction
factor for the Band collision risk model. The
box indicates the 25
th
and 75
th
percentile;
while the whiskers indicate the 5
th
and 95
th
percentile. The thick line indicates the me-
dian (50
th
percentile).
NINA Report 692
19
4 Discussion
Given our satellite telemetry data the median correction factor (i.e. “avoidance rate”) derived
from the Band collision risk model for white-tailed eagles was estimated at 100% for winter,
summer and autumn and 99.1% for spring. The year-round estimate based on the satellite te-
lemetry data was 97.5%. The SNH guidance note has set the correction factor to 95% based
on flight behaviour and collision monitoring studies. The reason given for this is “because there
is sufficient evidence for their vulnerability to collisions: white-tailed eagle (evidence of a dis-
proportionate number of collisions at Smøla, than might be expected)”. Here they refer to the
annual report from the BirdWind-studies at the Smøla wind-power plant (Bevanger et al. 2008).
The median correction factor based on vantage points was earlier estimated at 95.4% (inter-
quartile range: 0.907 – 0.976) (May et al. 2010). The year-round correction factor derived from
the satellite telemetry data was similar to vantage point-based factors (95-99%; SNH 2010).
However, the correction factor for spring, the season with most collisions, was estimated
higher than the vantage point-based factor. Hopefully, this report can be used to increase our
understanding on how the correction factor may affect the results from collision risk modelling.
It may provide an alternative method for estimating correction factors, unaffected by observer
biases, terrain conditions, and seasonal and daily variation. Also the satellite telemetry data
may provide an alternative way to provide insight into relative risk among seasons, and help
identify periods or areas with increased risk either in a pre- or post construction situation. The
correction factors presented here also illustrate when white-tailed eagles may experience the
greatest collision risk due to their flight activity patterns.
Although the correction factor often is thought to be related to avoidance, we did not find
strong differences in flight activity inside/outside the wind-power plant (Figure 6; May et al.
2010). The estimated correction factor therefore may not represent displacement (i.e. not using
the WPA as habitat anymore) or large-scale avoidance (i.e. active behavioural response). It
may however, include fine-scale avoidance, such as flying round the actual physical turbine
structures or last-minute evasion of the rotor blades. It is important to realize that the correction
factor may in fact encompass different sources of error in the model (i.e. stage 1 and 2). The
correction factor likely represents the total effect resulting from many unknown factors:
•
‘Observer’ biases
•
: The kernel utilization distributions (UD) may possibly be affected by the
recording schedule the transmitters were programmed. In the analyses the UDs were nor-
malized by dividing each pixel’s value with the sum over all pixel values for the entire UD, to
take this into account. The transmitters were either battery- or solar-powered; the latter
generally rendered less relocations during the winter half year. For the year-round esti-
mates this means that the UDs overestimate summer activity and underestimate winter ac-
tivity.
Seasonal and daily variation
•
: natural variation in the onset of seasons, daily fluctuations
and weather conditions may affect flight activity differently over the seasons. In the analy-
ses we have pooled the data over the years, and used set seasonal three-month intervals.
This may somewhat affect the true variation in the correction factors.
Species- and site-specific bird density, behaviour and flight activity
•
: the high local density at
the Smøla wind-power plant and resulting high levels of social and/or territorial flight activity
resulting in a disproportionate number of collisions; thus affecting the correction factor.
Model assumptions:
the analyses done here assume that the flight activity of the subadult
eagles is representative for the entire population; which is not necessarily the case. Juve-
nile and subadult eagles migrate northwards during summer/autumn; a pattern which be-
comes less and less pronounced over the years. The effect this may have on the population
size has been taken into account in the analyses. Also, May et al. (2010) noted that the hit
probability (stage 2) never reaches zero; even when the rotor blades are barely moving.
Given the nature of the data, and the possible sources of error involved one rarely has control
over, it is important to visualize the uncertainties to the model outcomes. Although often the
uncertainty connected to this type of modelling is rarely given, we have incorporated the uncer-
NINA Report 692
20
tainty of the calculated estimates in our analyses. This should, ideally, become common prac-
tice. Chamberlain et al. (2006) also point out that relatively small changes in the correction fac-
tor can lead to large proportional changes in mortality rates. The Band model allows for calcu-
lating separate correction values for different seasons and/or geographic regions when such,
and enough, data is available. Splitting the year into seasons may, however, hamper robust
estimates because of lack of enough actual recorded collisions (e.g. the effect of 0 versus 1
collisions is relatively large), and the natural variation in the timing of seasons. Also, the calcu-
lation of the hit probability (stage 2) assumes that all birds approach a turbine up- or downwind
(50-50%). This is not really realistic; birds may also approach the turbine crosswind for exam-
ple. In the original calculations derived by Tucker (1996); he presented two models: one for
up/downwind and one for crosswind. Based on these formulae a stochastic model for estimat-
ing the hit probability more realistically is possible. This model would then include information
not only on average wind and bird speed (as is required now) but also on wind directions and
bird headings, and variations in both speeds and directions. Data on these should be fairly
easy to obtain from weather stations and direct observations, respectively.
NINA Report 692
21
5 References
Band, W., Madders, M. & Whitfield, D.P. 2007. Developing field and analytical methods to as-
sess avian collision risk at wind farms. – In: De Lucas, M., Janss, G.F.E. & Ferrer, M., eds.
Birds and Wind Farms. Risk Assessment and Mitigation. Servicios Informativos Ambien-
tales/Quercus, Madrid. Pp. 259-275.
Barrios, L. & Rodriguez, A. 2004. Behavioural and environmental correlates of soaring-bird
mortality at on-shore wind turbines. - Journal of Applied Ecology 41: 72-81.
Bevanger, K. 1999. Estimating bird mortality caused by collision with power lines and electro-
cution; a review of methodology. – In: Ferrer, M. & Janss, G.F.E. (eds.). Birds and power
lines. Collision, electrocution and breeding. - Querqus, Madrid. Pp. 29-56.
Bevanger, K., Berntsen, F., Clausen, S., Dahl, E.L., Flagstad, Ø., Follestad, A., Halley, D.,
Hanssen, F., Hoel, P.L., Johnsen, L., Kvaløy, P., May, R., Nygård, T., Pedersen, H. C.,
Reitan, O., Steinheim, Y. & Vang, R. 2008. Pre- and post-construction studies of conflicts be-
tween birds and wind turbines in coastal Norway. Progress Report 2008. - NINA Report 409:
55 pp. Norwegian Institute for Nature Research, Trondheim.
Bevanger, K., Berntsen, F., Clausen, S., Dahl, E.L., Flagstad, Ø., Follestad, A., Halley, D.,
Hanssen, F., Johnsen, L., Kvaløy, P., Lund-Hoel, P., May, R., Nygård, T., Pedersen, H. C.,
Reitan, O., Røskaft, E., Steinheim, Y., Stokke, B. & Vang, R. 2010. Pre- and post-
construction studies of conflicts between birds and wind turbines in coastal Norway (Bird-
Wind). Report on findings 2007-2010. - NINA Report 620: 152 pp. Norwegian Institute for Na-
ture Research, Trondheim.
Chamberlain, D.E., Rehfisch, M.R., Fox, A.D., Desholm, M. & Anthony, S. 2006. The effect of
avoidance rates on bird mortality predictions made by wind turbine collision risk models. - Ibis
148: 198-202.
De Lucas, M., Janss, G.F.E., Whitfield, D.P. & Ferrer, M. 2008. Collision fatality of raptors in
wind farms does not depend on raptor abundance. - Journal of Applied Ecology 45: 1695-
1703.
Drewitt, A.L. & Langston, R.H.W. 2006. Assessing the impacts of wind farms on birds. - Ibis
148: 29-42.
Drewitt, A.L. & Langston, R.H.W. 2008. Collision Effects of Wind-power Generators and Other
Obstacles on Birds. - Annals of the New York Academy of Sciences 1134: 233–266.
Gjershaug, J.O. 1994. Hekkefuglenes utbredelse og bestandsstatus i Norge. – In: Folkestad,
A.O. (ed.). Norsk Fugleatlas. Norsk Ornitologisk Forening, Klæbu, Norway.
Hoel, P.L. 2009. Do wind power developments affect the behaviour of White-Tailed Sea Ea-
gles on Smøla? – Master thesis, NTNU, Trondheim. 45 pp.
Hoover, S.L. & Morrison, M.L. 2005. Behavior of red-tailed hawks in a wind turbine develop-
ment. - Journal of Wildlife Management 69: 150-159.
Hunt, W.G., Jackman, R.E., Hunt, T.L., Driscoll, D.E. & Culp, L. 1998. A population study of
golden eagles in the Altamont Pass Wind Resource Area: population trend analysis 1997. -
Report to National Renewable Energy laboratory, Subcontract XAT-6-16459-01. Predatory
Bird Research Group, University of California, Santa Cruz.
Johnson, G.D., Erickson, W.P., Strickland, M.D., Shepherd, M.F., Shepherd, D.A. & Sarappo,
S.A. 2002. Collision mortality of local and migrant birds at a large-scale wind-power devel-
opment on Buffalo Ridge, Minnesota. - Wildlife Society Bulletin 30: 879-887.
Langston, R.H.W. & Pullan, J.D. 2003. Wind farm and birds: an analysis of the effects of wind
farm on birds, and guidance on environmental assessment criteria and site selection issues. -
Report T-PVS/inf (2003) 12, by BirdLife International to the Council of Europe, Bern Conven-
tion. RSPB/BirdLife in UK.
Lehner, P.N. 1979. Handbook of Ethological Methods. - Garland STPM Press, New
York/London.
Love, J.A. 1983. The return of the Sea Eagle. - Cambridge University Press, Cambridge.
Madders, M. & Whitfield, D.P. 2006. Upland raptors and the assessment of wind farm impacts.
- Ibis 148:43-56.
NINA Report 692
22
May, R., Hoel, P.L., Langston, R., Dahl, E.L., Bevanger, K., Reitan, O., Nygård, T., Pedersen,
H.C., Røskaft, E. & Stokke, B.G. 2010. Collision risk in white-tailed eagles. Modelling collision
risk using vantage point observations in Smøla wind-power plant. – NINA Report 639. 25 pp.
Nygård, T., Bevanger, K., Dahl, E.L., Flagstad, Ø., Follestad, A., Hoel, P.H., May, R. & Reitan,
O. 2010. A study of White-tailed Eagle movements and mortality at a wind farm in Norway.
BOU Proceedings – Climate Change and Birds. http://www.bou.org.uk/bouproc-
net/ccb/nygard-etal.pdf
Orta, J. 1994. White-tailed Sea-eagle. – In: del Hoyo, J., Elliott, A. & Sargatal, J. (eds). Hand-
book of birds of the world. Vol. 2. New World vultures to guineafowl. - Lynx Edicions, Barce-
lona, Spain. Pp. 122.
R Development Core Team. 2009. R: A language and environment for statistical computing. -
R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0, URL
http://www.R-project.org.
Smallwood, K.S. & Thelander, C.G. 2005. Bird mortality at the Altamont Pass Wind Resource
Area March 1998-September 2001. - BioResource Consultants, Ojai, California.
Smallwood, K.S. & Thelander, C. 2008. Bird mortality in the Altamont Pass Wind Resource
Area, California. - Journal of Wildlife Management 72: 215-223.
Scottish Natural Heritage (SNH). 2000. Windfarms and birds: Calculating a theoretical collision
risk assuming no avoiding action. – SNH: 10pp.
Scottish Natural Heritage (SNH). 2005. Survey methods for use in assessing the impacts of
onshore wind farms on bird communities. – SNH: 50 pp.
Scottish Natural Heritage (SNH). 2010. SNH Avoidance Rate Information & Guidance Note.
Use of Avoidance Rates in the SNH Wind Farm Collision Risk Model. – SNH, 10pp.
Thelander, C.G., Smallwood, K.S. & Rugge, L. 2003. Bird risk behaviors and fatalities at the
Altamont Pass Wind Resource Area. Period of performance: March 1998 to December 2000.
- Report to National Renewable Energy Laboratory, Golden, Colorado. Work performed by
BioResource Consultants, Ojai, California: 86 pp.
Tucker, V.A. 1996. A mathematical model of bird collisions with wind turbine rotors. - Journal of
Solar Energy Engineering 118: 253-269.
Whitfield, D.P. 2009. Collision Avoidance of Golden Eagles at Wind Farms under the ‘Band’
Collision Risk Model. – Natural Research Information Note 1 (revised). Natural Research Ltd,
Banchory, UK: 35 pp.
Whitfield, D.P. & Madders, M. 2006a. Deriving collision avoidance rates for red kites Milvus
milvus. Natural Research Information Note 3. – Natural Research Ltd, Banchory, UK: 14 pp.
Whitfield, D.P. & Madders, M. 2006b. A review of the impacts of wind farms on hen harriers
Circus cyaneus and an estimation of collision avoidance rates. – Natural Research Informa-
tion Note 1 (revised). Natural Research Ltd, Banchory, UK: 32 pp.
Norwegian Institute for Nature Research
NINA head office
Postal address: P.O.Box 5685, NO-7485 Trondheim, NORWAY
Visiting address: Tungasletta 2, NO-7047 Trondheim, NORWAY
Phone: + 47 73 80 14 00
Fax: + 47 73 80 14 01
Organisation number: NO 950 037 687 MVA
www.nina.no
Graphic design: NINA
Cover photos: Per Jordhøy, Børre Dervo,
Knut Kringstad, Tycho Anker-Nilssen
ISSN: 1504-3312
ISBN: 978-82-426-2277-8
692
