Content uploaded by Gioele Passoni
Author content
All content in this area was uploaded by Gioele Passoni on Oct 07, 2019
Content may be subject to copyright.
Wind farm prioritisation based on potential impacts
on wolf (Canis lupus) habitat in Croatia
Gioele Passoni
September 2015
A thesis submitted for the partial fulfillment of the requirements for the degree of Master of Science
at Imperial College London
Submitted for the MSc in Conservation Science
i
DECLARATION OF OWN WORK
I declare that this thesis,
“Wind farm prioritisation based on potential impacts on wolf (Canis lupus) habitat in
Croatia”,
is entirely my own work and that where material could be construed as the work of others, it is
fully cited and referenced, and/or with appropriate acknowledgement given.
Signature ……………………………………………………………………………………
Name of Student Gioele Passoni
Name of Supervisors Dr. Marcus Rowcliffe; Prof. Josip Kusak
ii
Table of Contents
1 Introduction ..................................................................................................................................... 1
1.1 Problem Statement .................................................................................................................. 1
1.2 Aims and Objectives ............................................................................................................... 3
2 Background ..................................................................................................................................... 5
2.1 The Wolf as a Large Carnivore in Europe ............................................................................... 5
2.2 Wind Farms and Wolves ......................................................................................................... 6
2.2.1 Impact of Wind Farms on Wolves ................................................................................... 6
2.2.2 Avoiding and Minimizing Impacts of Wind Farms on Wolves: an Example ................. 8
2.3 Wolf Homesites ....................................................................................................................... 9
2.4 Habitat Suitability Modelling .................................................................................................. 9
2.4.1 MAXENT ...................................................................................................................... 10
2.4.2 Habitat Suitability Modelling for Wolf Conservation ................................................... 11
2.5 Spatial Planning and Wind Farm Prioritisation ..................................................................... 12
2.5.1 Marxan........................................................................................................................... 13
2.6 Study Site: an Ecological Overview of Croatia ..................................................................... 14
3 Methods ......................................................................................................................................... 17
3.1 Methodological Framework .................................................................................................. 17
3.2 Data Collection and Preparation ............................................................................................ 17
3.2.1 Homesites ...................................................................................................................... 17
3.2.2 Environmental and Anthropic Variables ....................................................................... 18
3.2.3 Wind Farms Data ........................................................................................................... 18
3.3 Habitat Suitability Modelling ................................................................................................ 19
3.4 Wind Farm Prioritisation ....................................................................................................... 20
4 Results ........................................................................................................................................... 22
4.1 Habitat Suitability Modelling ................................................................................................ 22
4.2 Wind Farms Prioritisation ..................................................................................................... 26
5 Discussion ..................................................................................................................................... 30
5.1 Habitat Suitability Model ...................................................................................................... 30
5.2 Wind Farm Prioritisation ....................................................................................................... 33
5.3 Future Implications and Recommendations .......................................................................... 34
6 References ..................................................................................................................................... 37
7 Appendices .................................................................................................................................... 45
7.1 Appendix I ............................................................................................................................. 45
iii
List of Figures
Figure 1.1 Wolf distribution in Europe ................................................................................................. 1
Figure 1.2 Distribution of wolves and proposed wind farms in Croatia ............................................... 2
Figure 2.1 Distribution of the main habitat types in Croatia ..............................................................15
Figure 2.2 Location of the main toponyms mentioned in this thesis ....................................................16
Figure 4.1 The 31 wolf homesite locations collected between 1997 and 2015 ......................................22
Figure 4.2 Response curves for the 6 model predictors. The curves show how the species probability of
occurrence changes with each predictor, maintaining all other predictors at their average sample
value. The red curves represent the mean trends, while the blue shades show the mean +/- the
standard deviation. In each graph, the X axis shows the change in each environmental variable,
while the Y axis shows the species' probability of presence.............................................................24
Figure 4.3 Habitat suitability map obtained with Maxent. Blue indicates low suitability, green
indicates intermediate suitability and red indicates high suitability areas. .......................................25
Figure 4.4 Best solution for the Marxan analysis over 100 repetitions in the 2 km (A) and 4 km (B)
buffer scenarios. The number of each wind farms corresponds to the numbers in table 4.4 ............27
Figure 4.5 Marxan analysis of the hexagonal grid over the area covered by proposed wind farms
with a 4 km buffer. The irreplaceability score is the number of times in which each cell was
selected in the optimal configuration over 100 Marxan repetitions. The figure shows the extent to
which wolf breeding habitat in each cell would be affected by wind farms construction. .................29
List of Tables
Table 4.1 Pearson's correlation coefficients (R) among environmental variables. The threshold to
discriminate correlated variables was R>0.7....................................................................................23
Table 4.2 Main statistical values showing the relative contribution of environmental variables in
Maxent ...........................................................................................................................................23
Table 4.3 Minimum, maximum, mean and standard deviation values for the 6 environmental
variables. For “distance to” variables, nil values indicate that a homesite is located in the same cell
(measuring 250x250 m) of an environmental feature and, thus, do not necessarily pinpoint a distance
of 0 metres. .....................................................................................................................................25
Table 4.4 Marxan values for all wind farms in the 2 km and 4 km buffer scenarios .
IS=Irreplaceability score over 100 Marxan repetitions; MW=Installed capacity in MW; %=percent of
the analogous initial value; % decrease=percent reduction compared to the analogous initial value.
“MW in Best Solution” and “Cost in Best Solution” only show the MW and the cost of selected
wind farms in the best solutions. The numbers in this table correspond to wind farms shown in figure
4.4. Further information about these wind farms can be found in Appendix I ..................................28
iv
List of Acronyms
AUC = Area Under the Curve
EIA = Environmental Impact Assessment
AAPPIEN = Assessment of Acceptability of Plans, Programmes and Interventions for the
Ecological Network
IUCN = International Union for Conservation of Nature
EU = European Union
GPS = Global Positioning System
BACI = Before After Control Impact
HSM = Habitat Suitability Model
SDM = Species Distribution Model
GLM = Generalized Linear Model
GAM = Generalized Additive Model
ROC = Receiver Operating Characteristic
IS = Irreplaceability Score
GIS = Geographic Information System
SGA = State Geodetic Administration
v
Abstract
Wolves (Canis lupus) in Croatia are estimated at nearly 200 individuals and form part of the
Dinaric-Balkan population. As in most of Europe, they are currently expanding in size and
distribution. However, the wolf still faces threats that could hamper its viability. In Croatia,
these threats include the worsening of public attitudes and the construction of wind power
plants in their distribution range. In order to meet the 2020 European targets for renewable
energy production, the Republic of Croatia is planning to build 33 wind farms, with a total
installed capacity of 1,555 MW. However, in order to meet such targets, only 747.25 MW are
necessary.
In this study a suitability model for wolf breeding habitat was carried out using Maxent based
on 6 environmental variables and 31 homesite locations collected between 1997 and 2015.
The prediction of habitat suitability was then used to determine the potential impact of
proposed wind farms on wolves. Lastly, a wind farm prioritisation process was carried out
using the software Marxan. This allowed selecting the wind farms that contributed to the
meeting of the energy targets at the minimum ecological impact on wolf breeding habitat.
The model showed good performance (AUC=0.805) and its prediction was consistent with the
current knowledge and distribution of wolves in Croatia. The main predictors for suitability
were distance to settlements, distance to farmland, distance to roads and distance to forest
edge. Moreover, Marxan allowed the selection of highly cost-efficient wind farms. In fact, in
the best scenario, selected wind farms were 44.5% of the total proposed wind farms and held
only 23.3% of the total initial cost.
In conclusion, this study provides valuable information and useful tools for the conservation
of wolves in Croatia. In particular, the habitat suitability map can be used for the
implementation of the wolf management plan, for the prevention of human-wildlife conflicts
and for future conservation planning. Moreover, the result of the prioritisation will be used to
inform the strategic planning of wind farms in Croatia. Lastly, the framework adopted in this
study can be expanded to multiple infrastructure and multiple large carnivores’ species such
as the Eurasian brown bear and the Eurasian lynx.
WORD COUNT: 14,931
vi
Acknowledgements
This project is fruit of the collaboration of professors, researchers, politicians, students and
friends.
Firstly, I would like to thank my supervisors Prof Josip Kusak and Dr Marcus Rowcliffe for
their constant help and support. In particular, thanks to Josip for teaching me the Art of wolf
research, for engaging me in all kind of field activities, and for the great stories and
discussions in the unforgettable Čorkova uvala. Thanks also to Marcus for his conceptual and
technical advices and for being constantly available, especially when I most needed support.
Special thanks also go to Prof Djuro Huber and Slaven Reljic for involving me in all activities
of the LIFE-Dinalp Bear Project and for their extraordinary kindness. Collaring bears at night
was magnificent!
I would also like to thank Francisco Álvares for kindly giving me useful material about the
impact of wind farms on the wolf. Thanks also to Jasna Jeremić for promptly providing me
with helpful information about land uses distribution in Croatia.
Finally, I want to thank the students of the MSc in Conservation Science and all my friends in
Zagreb. In particular thanks go to Aakash Lamba, Paolo Strampelli, Elliot Newton, Ari
Whiteman, Daniele De Angelis, Peter Haswell, Dario Hipolito and Jakob Janeš.
Data about wolf homesites between 1997 and 2014 were collected by Kosip Kusak. The same
data for 2015 were collected by Josip Kusak and Gioele Passoni. All analyses were entirely
carried out by Gioele Passoni.
1
1 Introduction
1.1 Problem Statement
The grey wolf, Canis lupus, is the second most abundant species of large carnivore in Europe.
With approximately 200 wolves (168-219, Štrbenac 2010), Croatia occupies the western part
of the Dinaric-Balkan population (Kaczensky 2012). As such, it represents a particularly
important area for European wolves, since it may allow the connection of the Dinaric-Balkan,
the Alpine and the Italian Peninsula populations (Figure 1.1) (Fabbri, Caniglia et al. 2014).
Currently, one of the main threats for Croatian wolves is represented by the potential
construction of major infrastructure, notably wind farms, in their core habitat (Kaczensky
2012).
Figure 1.1 Wolf distribution in Europe (Linnell, Salvatori et al. 2008)
2
The wind power capacity installed in Croatia, as of July 2015, is 452.75 MW (MINGO 2015).
However, in order to meet the target of the European Directive 2009/28/EC (EC, 2009), and
according to the Energy Strategy for the Republic of Croatia, wind farms in Croatia have to
reach a total installed capacity of 1,200 MW by 2020 (Croatian Parliament 2009a). To reach
this target, a further installed capacity of 747.25 MW is needed. Notwithstanding, Croatia is
planning to build 33 wind farms, with a total installed capacity of 1,555 MW (MINGO 2015).
Therefore, planned wind farms would provide nearly twice as much installed capacity as
needed to reach the 2020 target.
Although wind is a valuable source of renewable energy, the implementation of wind farms
requires a large amount of land (Kiesecker, Evans et al. 2011). It is thus important that wind
farms are strategically placed in areas where there is the minimum competition with other
land use types, such as agriculture, natural habitats, protected areas and urban areas.
Nonetheless, the vast majority of currently proposed wind farms are located in the wolf
distribution range (Figure 1.2).
Figure 1.2 Distribution of wolves and proposed wind farms in Croatia (Štrbenac 2010, MINGO 2015)
3
Scientific evidence suggests that wind farms have a negative impact on wolves, particularly
on their breeding sites (Àlvares, Rio-Maior et al. 2011, Helldin, Jung et al. 2012, Álvares,
Rio-Maior et al. in press). According to Álvares, Rio-Maior et al. (in press), during wind farm
operation, wolf dens tend to be located further than 4 km from the nearest turbine. This might
be due to several reasons. Firstly, the construction of wind power plants causes substantial
changes in the wolf habitat, including deforestation and fragmentation (Northrup and
Wittemyer 2013). Secondly, a higher density of roads could lead to more collisions with
vehicles, increased disturbance in previously inaccessible areas, and easier access for
poachers (Helldin, Jung et al. 2012). Lastly, the noise produced by rotating turbines could
interfere with wolf howling, which is particularly important during the breeding season
(Harrington, Asa et al. 2003, Helldin, Jung et al. 2012). Moreover, during this season, wolves
are more sedentary and thus may be more sensitive to all these sources of disturbance
(Packard 2003).
Before being implemented, under the European and Croatian legislation, proposed wind farms
have to undergo an Environmental Impact Assessment (EIA) and an Assessment of
Acceptability of Plans, Programmes and Interventions for the Ecological Network
(AAPPIEN) (Croatian Parliament 2009b, EC 2011). These assessments also need to take into
consideration the impact of wind farms on wolf breeding habitat. However, the spatial
distribution of wolf most suitable breeding areas in Croatia is not fully known and it is
currently based only on expert opinions. Hence, at present, an exhaustive habitat suitability
map that could be used for the assessment and the minimisation of potential impact on wolves
does not exist.
1.2 Aims and Objectives
In a human-dominated region like Europe, the long term viability of large carnivore species
strictly depends on land management decisions and on the coordinated planning of conflicting
land use types (Linnell, Salvatori et al. 2008). Thus, the aim of this study is to provide
scientific material that can influence wind farm implementation and support the long term
viability of the wolf in Croatia.
4
The following objectives will contribute to the achievement of this aim:
1. Gathering of the information and data collection on the location of wolf breeding sites
in Croatia;
2. Creation of a suitability map for wolf breeding habitat through habitat modelling;
3. Systematic prioritisation of proposed wind farms based on installed capacity and
potential impact on wolves;
4. Creation of a map that allows wind farm developers to visualise the most affected
breeding areas within each wind farm;
5. Proposal of a simple and evidence-based framework which can potentially include
multiple infrastructure and other large carnivore species in Europe.
This thesis will start with a background section containing an overview of the wolf
conservation status in Europe, a more detailed explanation of wind farm impacts on wolves, a
review of previous literature, a general explanation of the methods adopted, and an ecological
overview of Croatia. After the background, the thesis will continue with the description of the
methodologies, the presentation of the results, and a final discussion.
5
2 Background
2.1 The Wolf as a Large Carnivore in Europe
The Grey Wolf (Canis lupus) is the second most abundant species of large carnivore in
Europe (Chapron et al 2014). With a total estimated number of 12,000 individuals, the wolf is
expanding across Europe (Chapron et al 2014). However, although some authors consider it a
conservation success, this expansion is not only the results of active conservation actions
(Chapron et al 2014, Boitani 2015). In fact, the social and economic transformations
occurring since the end of WWII have led to an increasing rate of land abandonment in rural
areas (Boitani 2015, Navarro and Pereira 2015). The decrease of human activities in such
areas have led to the regeneration of secondary forests and left available habitat for the
expansion of wolves (Navarro and Pereira 2015).
This expansion may also pose some threats related to the coexistence between people and
wolves (Navarro and Pereira 2015). One of the main causes of human-wolf conflict is
livestock depredation (Navarro and Pereira 2015). Despite the compensation schemes
available in most countries, single farmers can be truly affected by the loss of livestock
caused by wolves (Wilson 2004). Additionally, some people living near wolves are also
concerned for their own and their family’s safety (Røskaft, Händel et al. 2007). Thus, the
coexistence with wolves can lead to negative human attitudes which can be further inflated by
dramatic stories published in the media (Røskaft, Händel et al. 2007, Majić and Bath 2010).
The human-wolf conflict is generally higher in recently recolonised areas, where traditional
livestock-guarding knowledge was lost and people are not used to coexist with large
carnivores (Navarro and Pereira 2015). For all these reasons, in several countries, wolves are
often illegally killed by farmers and poachers (Huber, Kusak et al. 2002, Liberg, Chapron et
al. 2012).
In Croatia, wolves form part of the Dinaric-Balkan population, which spreads across
approximately 10 countries in south-east Europe and include circa 3,900 individuals
(Kaczensky 2012). However, this estimate may not be completely accurate. In particular,
since the population spans across many national borders, the number of individuals may be
inflated. It has been shown that double-counting of bears along national borders between
Norway, Russia, Finland and Sweden led to an inflation of population estimates of up to
119% compared to estimates that took into account “foreign residents” (Bischof, Brøseth et
6
al. 2015). This might also be the case for the Dinaric-Balkan population where many packs
are transboundary (Jeremić 2012). Furthermore, although the population is considered to be
stable, a rigorous assessment of the trend is hindered by the use of a wide range of different
techniques during monitoring (Kaczensky 2012).
In Croatia, wolves are estimated to be around 200 (168-219, Štrbenac 2010). However,
Croatia is located in a strategic area for the long term viability of European wolves, since it
allows the connection between the Dinaric-Balkan, the Alpine and the Italian peninsula
populations (Fabbri, Caniglia et al. 2014). In the eastern Alps, successful reproduction events
between wolves of Dinaric-Balkan and Italian peninsula origins have already been observed
(Fabbri, Caniglia et al. 2014).
Although wolves in Croatia are likely to be increasing, they also face several threats
(Kaczensky 2012). In particular, the main threat is the construction of wind farms in their
distribution range (Skrbinšek and Bath 2010, Kaczensky 2012). Moreover, human attitudes
are worsening as a result of increased livestock depredation in recolonised areas (Kaczensky
2012). In 2011, nearly 1,700 livestock were killed by wolves in Croatia, but compensation
schemes are considered inadequate by farmers (Kaczensky 2012).
Despite and as a result of the overall expansion of wolves in Europe, conservation efforts are
still needed. After having fought to save large carnivores from extinction for decades,
European conservationists have to face the new challenge of peaceful coexistence (Boitani
2015).
2.2 Wind Farms and Wolves
2.2.1 Impact of Wind Farms on Wolves
Wind farms have been shown to have some direct and indirect negative implications on
wildlife (Kuvlesky Jr, Brennan et al. 2007, Northrup and Wittemyer 2013). The most
common impact is caused by direct collision of birds and bats with turbines (Drewitt and
Langston 2006, Kunz, Arnett et al. 2007). However, wind farms also have negative impacts
on non-volant wildlife, although the scientific literature currently lacks the information to
rigorously assess and quantify them (Lovich and Ennen 2013). These impacts on terrestrial
species mainly include habitat modifications and behavioural alterations (Kuvlesky Jr,
Brennan et al. 2007). In particular, besides the installation of turbines, wind farms require the
construction of roads, transformers, substations and transmission lines (Kuvlesky Jr, Brennan
7
et al. 2007). All these infrastructures may cause habitat loss and fragmentation (Northrup and
Wittemyer 2013). Moreover, maintenance facilities may also increase human access to
previously undisturbed areas (Northrup and Wittemyer 2013). This, besides increasing human
disturbance, might also increase the likelihood of collisions of wildlife with vehicles along
roads (Kuvlesky Jr, Brennan et al. 2007). Another indirect impact of wind turbines may be
caused by noise disturbance to those animals that use long distance vocalizations and alarm
calls to communicate (Helldin, Jung et al. 2012).
Some studies have shown that wind farms could potentially affect wolves (Àlvares, Rio-
Maior et al. 2011, Helldin, Jung et al. 2012, Álvares, Rio-Maior et al. in press). This impact
seems to particularly concern breeding success and to cause the displacement of wolf
reproduction sites (Àlvares, Rio-Maior et al. 2011, Álvares, Rio-Maior et al. in press). For
example, Àlvares, Rio-Maior et al. (2011) show that during the construction and operation
phases of one wind power plant, although wolves kept using areas occupied by the wind farm,
they tended to abandon breeding sites and have a decreased reproduction rate in areas closer
than 2 km from the nearest turbine. Moreover, in two case studies in Portugal, wolf breeding
parameters were monitored in a 15-year-long period before, during and after the construction
of wind farms (Álvares, Rio-Maior et al. in press). In this study, the authors showed that,
during the construction phase, wolves kept breeding in the wind farms area with decreased
reproduction rate, while, during the operation phase, wolves started selecting breeding sites
located at least 4 km away from the nearest turbine (Álvares, Rio-Maior et al. in press). GPS-
Telemetry data also showed shifts of home ranges partially away from wind power plants
(Álvares, Rio-Maior et al. in press).
The actual reasons behind the impact of wind farms on wolves have only been proposed
based on current knowledge on the effects of infrastructure on large mammals and are yet to
be thoroughly investigated (Helldin, Jung et al. 2012). The first, most intuitive reason could
be the change and loss of habitat, particularly for reproduction (Àlvares, Rio-Maior et al.
2011). In fact, the construction of wind turbines and other related facilities could cause
significant changes in wolf breeding habitat, including deforestation and fragmentation
(Northrup and Wittemyer 2013). Several studies have shown that, where forests are present,
wolves tend to locate their den in relatively undisturbed and forested areas (Theuerkauf,
Rouys et al. 2003, Person and Russell 2009).
8
Moreover, a higher density of roads and other infrastructures related to wind farms can lead to
increased indirect threats and disturbance. For example, it has been shown that wolves avoid
areas with relatively higher density of roads, houses and human disturbance, particularly
during the breeding period (Theuerkauf, Rouys et al. 2003, Karlsson, Brøseth et al. 2007,
Houle, Fortin et al. 2010). Àlvares, Rio-Maior et al. (2011) found that road traffic in wind
farms increased 20 to 60 fold during construction and 4 to 13 fold during operation, compared
to the pre-construction period. Moreover, Huber, Kusak et al. (2002) found that collisions
with vehicles constituted nearly 20% of wolf mortality cases between 1986 and 2001 in
Croatia. The presence of roads could also facilitate the access of poachers into wolf habitat
and increase mortality due to retaliatory killing (Person and Russell 2008, Helldin, Jung et al.
2012). According to Huber, Kusak et al. (2002), 67.4% of wolf mortality cases in a 15-years
period in Croatia were due to illegal killing.
The last reason that has been hypothesized is related to acoustic disturbance (Àlvares, Rio-
Maior et al. 2011, Helldin, Jung et al. 2012). In particular, the noise produced by operating
turbines could disguise or disturb wolf howling (Helldin, Jung et al. 2012). It has been shown
that howling in wolves has several important functions that tend to peak during the breeding
season, including territorial defence and coordination of movements among separated
packmates (Harrington, Asa et al. 2003, Mech and Boitani 2003).
Although these impacts have been proposed, a rigorous assessment and quantification of their
effects is hampered by the lack of before-after-control-impact (BACI) studies (Lovich and
Ennen 2013). This type of approach requires the assessment of the situation before and after
the construction of wind farms (Lovich and Ennen 2013). However, the time required to
collect adequate data about wolf movements, demography and behaviour is longer than the
time needed for wind farm implementation (Franklin 1989, Management 2005). It is thus very
difficult to have information about the situation before wind farms construction.
2.2.2 Avoiding and Minimizing Impacts of Wind Farms on Wolves: an Example
During the implementation of two wind farms projects in Portugal (wind farm “Alto de
Coutada” - 102 MW, and wind farm “Serra da Nave” – 100 MW), Soares, Duarte et al.
(2011) adopted several measures to avoid or minimise the impacts on wolves. In particular,
during the planning phase, alternative and less impacting positions were identified for the
displacement of some turbines. In the construction phase, all construction activities were
suspended during night time, when wolves are more active. Moreover, all activities carried
9
out during the wolf breeding season were prohibited in some areas with particularly high
habitat suitability. Lastly, during the operation phase, some mitigation measures have also
been adopted like the implementation of barriers along new access roads.
2.3 Wolf Homesites
Wolf homesites, or breeding sites, or reproduction sites, are areas associated with pup rearing,
and may be either dens or rendezvous sites (Harrington and Mech 1978). The former are the
sites where wolf pups are raised during the first 8 weeks from birth (Mech 1970). Dens are
generally located away from the peripheral zones of the territory and are mainly used by the
breeding female and her pups (Packard 2003). There are several types of den which depend
on the type of habitat. In particular, in forested habitats, dens may be formed by a bedding of
leaves or dug under the roots of the trees, while in karstic areas it may be created from
existing burrows between rocks (Packard 2003).
Each home range can have several dens, some of which can be re-used in different years by
the same female (Capitani, Mattioli et al. 2006). Moreover, wolves may move the den site
within a breeding season (Packard 2003). These shifts are usually short (i.e. ca. 250 metres),
especially when the pups are young, although they can also be over several kilometres
(Packard 2003).
After the denning period, between 8 and 20 weeks after birth, pups generally live in
rendezvous sites (Packard 2003). These are areas above ground which include bedding, where
pups huddle while resting, and play areas. Rendezvous sites are generally located in the same
areas as the dens, although, during this period, wolf pups are able to move over longer
distances and can be found far from such areas (Mech 1970).
2.4 Habitat Suitability Modelling
Habitat Suitability Models (HSMs), or Species Distribution Models (SDMs), allow ecologists
and conservationists to predict the likelihood of occurrence of species based on their
relationships with environmental variables (Hirzel and Le Lay 2008). HSMs identify the
environmental requirements of a species based on the habitat characteristics in locations
where the species is known to be present (Phillips, Anderson et al. 2006). Once the
environmental requirements are found, they are projected into geographic space and can
provide valuable information about species potential distribution (Phillips, Anderson et al.
2006).
10
Depending on the quality and the quantity of data needed, two main types of SDMs can be
distinguished: presence-only and presence-absence SDMs. In particular, besides presence
localities, presence-absence SDMs require the input of locations where the species in known
to be absent (i.e. absences), in order to generate discriminative rules and statistics to create
habitat suitability maps (Brotons, Thuiller et al. 2004). Examples of presence-absence SDMs
are generalised linear models (GLM) and generalised additive models (GAM) (Brotons,
Thuiller et al. 2004). On the other hand, in order to generate discriminative statistics, most
presence-only models compare the presence localities with a set of random locations where
the species might or might not be present (i.e. pseudo-absences) (Brotons, Thuiller et al.
2004). Examples of presence-only models are Mahalanobis distance, GARP and Maxent
(Wisz, Hijmans et al. 2008).
2.4.1 Maxent
Maxent (Maximum Entropy Modelling) is one of the most used presence-only SDMs. In
recent years, it has been adopted in more than 1,000 publications for a wide range of taxa and
geographic areas (Merow, Smith et al. 2013). According to several studies, it is also one of
the most accurate methods, having very high performances especially at small sample sizes
(Elith, Graham et al. 2006, Hernandez, Graham et al. 2006, Wisz, Hijmans et al. 2008).
Maxent is a machine-learning method that models habitat suitability and species geographic
distribution (Phillips, Anderson et al. 2006). In particular, it relates a set of species occurrence
localities with habitat characteristics by creating simple functions of user-specified
environmental variables, called “features” (Phillips and Dudík 2008). In order to do so,
Maxent identifies the probability distribution of maximum entropy (Phillips, Anderson et al.
2006). This is the most spread out (i.e. most approximated, closest to uniform) probability
distribution that describes an event based on available knowledge (Phillips, Anderson et al.
2006). Hence, Maxent finds the most approximated probability distribution based on the
constraint that the expected value of each feature should match its empirical value (i.e. the
average value at occurrence locations), within an error bound called the “regularization
parameter” (Phillips and Dudík 2008).
The obtained probability distribution is then compared with the probability distribution of the
pseudo-absences, where the probability of occurrence of the species is usually assumed to be
0.5 (Phillips, Anderson et al. 2006). This comparison allows the computation of
discriminative values such as the Receiver Operating Characteristic curve (ROC curve) and
11
the calculation of all related statistical analysis, including the determination of the Area Under
the Curve (AUC) (Phillips, Anderson et al. 2006). Moreover, the projection of the probability
distribution into geographic space enables the creation of a habitat suitability map (Phillips,
Anderson et al. 2006). The model also produces an analysis of the contribution of
environmental variables and response curves showing how the predicted probability of
presence changes based on changes in each environmental variable (Phillips, Anderson et al.
2006).
2.4.2 Habitat Suitability Modelling for Wolf Conservation
Understanding wolf spatial ecology is a crucial step towards its effective conservation (Corsi,
Duprè et al. 1999). For this purpose, habitat suitability models can be very useful tools (Elith
2000). For this reason they have been widely adopted, especially in recent years. The main
methods used were logistic regression-based models (Mladenoff, Sickley et al. 1999, Glenz,
Massolo et al. 2001), Mahalanobis distance (Corsi, Duprè et al. 1999, Cayuela 2004, Ahmadi,
Kaboli et al. 2013) and Maxent (Bassi, Willis et al. 2015).
Previous study found that anthropic-related variables, such as distance to farmland, distance
to roads and distance to settlements, and environmental variables, such as forest cover,
elevation, wild prey availability and distance to water, were the most important predictors for
wolf habitat suitability (Corsi, Duprè et al. 1999, Theuerkauf, Rouys et al. 2003,
Jędrzejewski, Jędrzejewska et al. 2008, Ahmadi, Kaboli et al. 2013, Iliopoulos, Youlatos et
al. 2014, Bassi, Willis et al. 2015). In general, according to these studies, wolves preferred to
locate their homesites in forested areas, near water, and far away from sources of human
disturbance such as villages, roads and farms. Moreover, in some studies, suitable areas were
often found at higher elevations and on rugged or steeper terrains, probably as a consequence
of lower human disturbance (Jędrzejewski, Niedzialkowska et al. 2005, Capitani, Mattioli et
al. 2006).
Despite these general similarities among results, the relative contribution of predictors and the
relationships between variables and habitat suitability show very high variability in the
literature. These differences can be due to the method adopted, ecological differences between
geographic areas, unavailability of some environmental data, unaccounted correlations
between environmental variables and biases in data collection of wolf occurrences (Corsi,
Duprè et al. 1999, Phillips, Dudík et al. 2009, Yackulic, Chandler et al. 2013). For example, it
has been shown that presence locations are often biased towards easily accessible areas (i.e.
12
roads) and that in the big majority of publications this bias is often not mentioned or not taken
into consideration (Phillips, Dudík et al. 2009, Yackulic, Chandler et al. 2013).
From a conservation perspective, wolf HSMs can be used to address several conservation
issues, including conservation planning (Corsi, Duprè et al. 1999, Mladenoff, Sickley et al.
1999, Ahmadi, Kaboli et al. 2013), habitat restoration (Mladenoff, Sickley et al. 1999,
Jędrzejewski, Jędrzejewska et al. 2008), wolf population management (Corsi, Duprè et al.
1999), and human-wildlife conflict prevention (Corsi, Duprè et al. 1999, Mladenoff, Sickley
et al. 1999, Glenz, Massolo et al. 2001, Treves, Naughton‐Treves et al. 2004, Marucco and
McIntire 2010, Ahmadi, Kaboli et al. 2013, Bassi, Willis et al. 2015).
2.5 Spatial Planning and Wind Farm Prioritisation
Strategic spatial planning aims to find the optimal allocation of different land uses across
different spatial scales in order to ensure a sustainable interaction among economic,
environmental, social and political agendas (Albrechts, Healey et al. 2003).
At a landscape scale in Europe, spatial planning is particularly used for nature conservation,
ecosystem services provision and infrastructure development (Albrechts, Healey et al. 2003).
Recently, with the increasing concerns about greenhouse gas emissions and climate change,
many renewable energy sources have been built or are planned to be built as alternatives to
fossil fuels (EC 2009). However, although renewable sources offer “cleaner” energy, they can
also have a significantly negative impact on the environment, especially at a local level
(Kiesecker, Evans et al. 2011). Hence, in order to optimise their environmental benefits and
minimise their socio-economic impact, it is important that they are located strategically within
the landscape.
In recent years, particular attention has been given to wind farm spatial planning and
prioritisation (Baban and Parry 2001, Punt, Groeneveld et al. 2009, Aydin, Kentel et al. 2010,
Tegou, Polatidis et al. 2010, Drechsler, Ohl et al. 2011, Baltas and Dervos 2012, Göke and
Lamp 2012). These studies aimed to identify the optimal allocation of wind turbines by taking
into consideration physical, social and environmental constraints. For example, several
studies analyse these constraints in order to produce suitability maps for future wind energy
implementation (Baban and Parry 2001, Aydin, Kentel et al. 2010, Tegou, Polatidis et al.
2010, Baltas and Dervos 2012). Similarly, other studies identify the areas within a region
where wind farms can meet specific energy production targets at the minimum monetary,
13
social and ecological cost (Punt, Groeneveld et al. 2009, Drechsler, Ohl et al. 2011, Göke and
Lamp 2012). The main constraints considered in these analyses were nature conservation
areas, native habitats and wildlife (e.g. birds, bats, fishes), inhabited areas, wind potential,
archaeological sites and tourist areas (Baban and Parry 2001, Punt, Groeneveld et al. 2009,
Aydin, Kentel et al. 2010, Tegou, Polatidis et al. 2010, Drechsler, Ohl et al. 2011, Baltas and
Dervos 2012, Göke and Lamp 2012). Among the studies mentioned above, only Drechsler,
Ohl et al. (2011) and Punt, Groeneveld et al. (2009) used scientific evidence about species
distribution in the consideration of wildlife outside protected areas.
The main methods that were adopted include multiple-criteria decision analysis (Tegou,
Polatidis et al. 2010), fuzzy-logic-based methods (Aydin, Kentel et al. 2010), numerical
optimisation (Punt, Groeneveld et al. 2009, Drechsler, Ohl et al. 2011), and the use of spatial
planning software such as Marxan (Göke and Lamp 2012).
2.5.1 Marxan
Marxan is originally designed as a conservation planning software. In general, it allows
identifying optimal configurations of complementary areas to be protected in order to meet
specific conservation objectives at the minimum political, social or economic cost (Pressey,
Cabeza et al. 2007, Ardron, Possingham et al. 2008). These areas, called “planning units”, are
generally hexagonal cells that form a grid over a general planning area. Each planning unit
contributes to the meeting of conservation objectives, called “targets” (e.g. protection of a
certain number of species) at a certain monetary, social or ecological cost. Marxan allows the
minimisation of this cost, while ensuring the achievement of the targets (Ardron, Possingham
et al. 2008). Although it is mostly used in protected areas design, some studies have
demonstrated that this software can address the same optimisation problem in other
applications (Rondinini and Boitani 2007, Ban and Vincent 2009, Göke and Lamp 2012).
In order to solve this problem, Marxan uses simulated annealing. With this algorithm, Marxan
repetitively changes current configurations by replacing planning units, thus forming similar,
“nearby” configurations. Each change is accepted if the cost of the new configuration is lower
than the previous one. However, at the beginning of the simulated annealing the algorithm
randomly accepts some configurations with a higher cost. Replacing one planning unit might
lead to a higher cost; however, a further replacement of another planning unit might lead to an
overall better solution compared to the starting situation. The likelihood of accepting a higher
14
cost configuration decreases along the annealing process with a rate that depends on a user-
specified parameter called “temperature” (Ardron, Possingham et al. 2008).
Decreasing temperature along the annealing also reduces the likelihood of the algorithm to
accept big changes (i.e. replacement of more planning units at each time). Therefore, big
changes are more likely to be accepted at the beginning, in order to avoid the resulting
configuration to stand in a “local” minimum cost. As such, replacing one unit may lead to a
lower cost; however, if we change the starting configuration by making bigger changes,
replacing that same planning unit may lead to an even lower cost (Ardron, Possingham et al.
2008).
Marxan applies the simulated annealing over many repeated runs. At the end of the
computation, it produces two main types of output: the best solution among all runs and the
irreplaceability score. The irreplaceability score is the number of times in which each
planning unit was selected among all runs (Ardron, Possingham, and Klein 2008).
2.6 Study Site: an Ecological Overview of Croatia
Due to its location, its shape and its geographic features, Croatia includes four
biogeographical regions: Mediterranean along the Adriatic coast, Alpine along the Dinaric
Mountains, Pannonian in the east bordering with Hungary, and Continental in the remaining
areas (Radovic 2006). For this reason, Croatia hosts an exceptional diversity of habitats and is
one of the most biodiverse countries in Europe (Figure 2.1) (Radovic 2006).
The most common habitat is formed by forests, which occupy 44% of the national territory
(Radovic 2006). Although forests are spread across the whole country, most of them are
found along the Dinaric Mountains, especially in the north-western part. The main species of
trees are beech (Fagus sylvatica), common oak (Quercus robur), silver fir (Abies alba),
Norwegian spruce (Picea abies), durmast oak (Quercus petraea) and common hornbeam
(Carpinus betulus) (Radovic 2006). Other important habitats are grasslands and meadows,
which are mainly found in the Mediterranean ecoregion and in central mountainous areas
(Radovic 2006). A smaller area is occupied by several other habitats, including wetlands,
scrublands, coastal habitat and karstic underground habitats (Radovic 2006). The remaining
area is mainly occupied by agriculture, which holds approximately 23% of the total land share
(EC 2013). Most of agriculture is practiced in the Pannonian region, in the east, and in
Dalmatia, in the south of the country (EC 2013).
15
In Croatia there is a great variety of biodiversity with a high quantity of endemic species,
especially for vascular plants (Radovic 2006). Moreover, Croatia hosts some charismatic
mammal species, including Balkan chamois (Rupicapra rupicapra balcanica – 400
individuals) Eurasian brown bears (Ursus arctos arctos – 1,000 individuals), grey wolves
(Canis lupus – 200 individuals) and Eurasian lynx (Lynx lynx – 50 individuals) (Kaczensky
2012, Šprem, Fabijanić et al. 2012). All these species can mainly be found in mountainous
areas along the Dinaric range. Figure 2.2 shows a map with all the toponyms found in this
thesis.
Figure 2.1 Distribution of the main habitat types in Croatia (Ministry of Culture 2005)
16
Figure 2.2 Location of the main toponyms mentioned in this thesis
17
3 Methods
3.1 Methodological Framework
This project aims to inform the strategic prioritisation of planned wind farms in Croatia based
on their potential impact on wolf breeding habitat. A habitat suitability model was utilised to
relate wolf homesites to a set of environmental variables and to produce a habitat suitability
map for wolf breeding habitat. The output of the model was then used to determine the
potential impact of planned wind farms on wolves. Finally, a strategic conservation planning
software was used to prioritise these wind farms so as to minimise their impact while ensuring
the achievement of the targets of the Croatian energy strategy. The methods utilised in this
study were chosen to provide a simple and scientifically-based approach for the prompt
prioritisation of planned wind farms in Croatia based on a relatively limited amount of
available data and information.
3.2 Data Collection and Preparation
3.2.1 Homesites
The locations of homesites were collected in 4 main areas in the wolf distribution range from
April to August between 1997 and 2015.
While dens were located through direct observations, with the help of GPS and VHF-
telemetry data, rendezvous sites were located using the simulated howling survey method as
recommended by Harrington and Mech (1982). All howling surveys were carried out between
July and September. During this time, the packs are still relatively sedentary, the response rate
is high, and young wolf howls are more likely to be distinguishable from adults’ (Harrington
and Mech 1982, Harrington, Asa et al. 2003, Packard 2003). A rendezvous site was
considered as such when howling pups could be easily identified or when only adults were
heard but the presence of pups was confirmed by other signs, such as direct observation,
camera traps photos, dead pups or footprints. Once a rendezvous site was found in the field, in
order to find its location, the direction of the howl was recorded and its distance estimated. In
case some wolves had a collar, GPS location were also used to support the estimate of the
rendezvous site location.
As wolves can use the same rendezvous site through different years (Capitani, Mattioli et al.
2006), locations that were closer than 500 metres were assumed to be part of the same site and
18
were excluded from our sample as suggested by Bassi, Willis et al. (2015). This was a
conservative measure to avoid overestimation by the model of the importance of the variables
associated to those sites.
3.2.2 Environmental and Anthropic Variables
Six environmental variables were chosen as potential predictors for wolf breeding habitat
based on other similar studies (Corsi, Duprè et al. 1999, Theuerkauf, Rouys et al. 2003,
Capitani, Mattioli et al. 2006, Ahmadi, Kaboli et al. 2013): distance to settlements, distance to
farmland, distance to roads, distance to forest edge, altitude, and slope.
In particular, distance to settlements is the distance to the closest village or aggregation of
houses. Distance to farmland is the distance to the closest agricultural land, including arable
land, permanent cropland, livestock farming and permanent pastures. Distance to roads is the
distance to the closest road, including unpaved forest roads. Distance to forest edge is the
distance to the closest forest edge from outside the forest. Thus, a value of “0” means that the
site is located anywhere in the forest. All distances and altitude are expressed in metres, while
slope is expressed in degrees.
The data from which these variables were created were obtained from different sources. In
particular, altitude and slope were obtained from a Digital Elevation Model made available by
the Croatian State Geodetic Administration (SGA). Distance to roads, updated to 2006, was
obtained from a digital topographic map issued by the same institution. Finally, distance to
settlements, distance to farmland and distant to forest edge were obtained from the 2006
Croatian National Habitat Classification (Ministry of Culture 2005). For all these variables a
250x250 m ASCII grid was created for the whole of Croatia using ArcMap 10.2.
Pearson’s correlation coefficients (R) were calculated among all layers before running the
model, in order to avoid collinearity and, thus, the distortion of variables’ relative contribution
in determining habitat suitability (Dormann, Elith et al. 2013). The threshold value to
discriminate correlated variables was set to R>0.7 (Dormann, Elith et al. 2013, Kramer-
Schadt, Niedballa et al. 2013, Syfert, Smith et al. 2013).
3.2.3 Wind Farms Data
For each wind farm the following data were obtained: name of the project holder, name of
project, number of turbines, GPS coordinates for each turbine, and installed capacity (MW).
All data were obtained from the Department of Renewable Resources and Energy Efficiency
19
of the Croatian Ministry of Economy, Labour and Entrepreneurship and are publicly available
(MINGO 2015).
3.3 Habitat Suitability Modelling
The habitat suitability model was performed using Maxent (Version 3.3.3) (Phillips,
Anderson et al. 2006). There are three main reasons why Maxent was chosen in this study.
Firstly, being a presence-only SDM, it does not require absence data, which can be unreliable
and difficult to obtain for elusive and wide-ranging species like wolves (Mech and Boitani
2003, Phillips, Anderson et al. 2006). Secondly, for species that are still expanding in areas
from where they were extirpated, like the wolf in Croatia (Kaczensky 2012, Chapron et al
2014), absence data might be located in unoccupied but suitable habitat (Elith, Phillips et al.
2011). Thus, an absence data might not be indicative of unsuitable habitat and might cause
occupied suitable areas to be considered unsuitable (Elith, Phillips et al. 2011). Thirdly,
among the most commonly used presence–only SDMs, Maxent was shown to have the
highest performance, particularly at small sample sizes (Hernandez, Graham et al. 2006,
Wisz, Hijmans et al. 2008).
After inputting the homesite locations and the environmental variables in Maxent,
“subsampling” was selected as replicated run type, and the model was run for 15 replications.
The “Random seed” setting was activated and the random test percentage was set to 25%,
meaning that, for each replication, 25% of presence localities were randomly set aside and
used as test points to compute the main Maxent outputs.
In order to determine the AUC, Maxent compares the presence localities with a set of pseudo-
absence points randomly selected from a user-specified area (Phillips, Anderson et al. 2006).
However, when the occurrence data are biased (e.g. close to roads for easier access), in order
to avoid the bias to be represented in the whole model, the pseudo-absences can be selected
from an area that shares the same bias as the presence points (Zaniewski, Lehmann et al.
2002, Dudík, Phillips et al. 2005, Phillips, Dudík et al. 2009). In this study, in order to
consider potential biases in the occurrence locations, pseudo-absences were selected from the
sampling distribution of wolf research carried out since 1997 as suggested by Fourcade,
Engler et al. (2014). All the other settings were set as default, as they have been tested and
optimised over a wide and diverse range of studies (Phillips and Dudík 2008).
Lastly, the relative contribution of environmental variables in the model was determined by
three different statistical values: the percent contribution, the permutation importance, and the
20
jackknife on the AUC. In particular, the percent contribution is a relative measure of the
increase in regularized training gain (i.e. the deviance that maximizes the occurrence
probability distribution compared to random) of a variable, compared to the increase in gain
of other variables (Phillips 2005). The permutation importance is a relative measure of how
much the AUC changes when the values of a variable at occurrence and background locations
are randomly permuted (Phillips 2005). Finally, for each variable, the jackknife measures the
AUC value excluding that variable, and including only that variable.
3.4 Wind Farm Prioritisation
The strategic prioritisation of planned wind farms was carried out using Marxan (Version
2.43). Although Marxan is generally used for protected areas design, in this study it was used
with a different approach similar to the one adopted by Göke and Lamp (2012). This method
for prioritising wind farms was chosen for several different reasons. Firstly, it fits the purpose
of this study of meeting specific targets while minimising costs (Ball, Possingham et al.
2009). Secondly, it has been shown that it is relatively easy to handle and flexible to changing
situations and regular data updates (Ardron, Possingham et al. 2008, Göke and Lamp 2012).
Thirdly, unlike other optimisation methods using other types of algorithms, its output
provides several near-optimal alternatives, as opposed to a single best solution (Ardron,
Possingham et al. 2008). In spatial planning, a set of “good” solutions is often preferred to a
single one, since it allows the negotiation among stakeholders and enables the consideration
of other factors that could not be included in the first analysis (Possingham, Ball et al. 2000).
Lastly, Marxan has shown to be suitable for the application to wind farm spatial planning and
it has already been integrated in the planning process of offshore wind farms (Göke and Lamp
2012).
In this study, wind farms were considered as planning units, each of which contributes to the
wind energy production targets at an ecological cost. This cost is represented by the
ecological impact on wolf breeding habitat. Marxan was run in three different scenarios, each
presenting different planning units. In the first and in the second scenarios, each planning unit
was represented by each planned wind farm, with a surrounding buffer of 2 km and 4 km
respectively. The two buffers were chosen according to the information currently available
about the impact of wind farms on wolves (Àlvares, Rio-Maior et al. 2011, Álvares, Rio-
Maior et al. in press). In the third scenario, a grid of hexagons of 1 km per side was created,
and covered the whole area occupied by the planned wind farms with a 4 km buffer. Each
21
hexagonal cell of the grid represented one planning unit. This last scenario was applied to
create a habitat sensitivity map and to allow wind energy producers to visualise the areas
which would have higher impact on wolf breeding habitat.
In all scenarios the cost of each planning unit was determined with the same criteria, using the
output from the habitat suitability model. In particular, each planning unit encompasses
several suitability cells from the habitat suitability map. The ecological cost was calculated by
summing up the suitability values of the cells contained in each planning unit. Hence, the
impact of wind farms on wolves was assumed to be proportional to the habitat suitability in
each cell. The sum of the habitat suitability cells allows taking into consideration both the
average cell value and the area of each planning unit. This approach provides a relative
measure of ecological impact. For example, a wind farm built over a bigger area would have a
higher impact than a smaller wind farm ceteris paribus. Similarly, an area with a higher
average cell value would be relatively more affected than a less overall suitable area. Lastly,
in the cost determination, the presence of operating wind farms was also considered. As such,
in areas where operating and proposed wind farms overlapped, the cost of adding a new wind
farm was considered nil.
On the other hand, each wind farm, with its installed capacity, contributes to the energy
production targets set in the Croatian energy strategy. The installed capacity target for all
planning units in Marxan was set to 747.25 MW and was determined by removing the already
installed capacity (452.75 MW) from the 2020 installed capacity target of 1,200 MW
(Croatian Parliament 2009a).
The analysis was run for 100 repetitions, in each of which Marxan finds a near-optimal
configuration of wind farms to meet the target while minimising the cost. After the analysis,
the best solution over all repetitions is presented together with the so-called irreplaceability
score. The irreplaceability score is the number of times in which a planning unit was selected
in the optimal configuration over all repetitions (Ardron, Possingham et al. 2008). The
penalty cost for not meeting the Marxan target was set sufficiently high for the target to be
met in all repetitions.
Finally, through the setting of the parameter “Boundary Length Modifier”, Marxan allows
taking into account the spatial compactness of the selected configuration. However, since this
component is not relevant in these circumstances, the parameter was not used (i.e. it was set
to 0).
22
4 Results
4.1 Habitat Suitability Modelling
A total of 31 homesites were found between 1997 and 2015 (Figure 4.1). Among these, 24
were rendezvous sites and 7 were actual dens.
Figure 4.1 The 31 wolf homesite locations collected between 1997 and 2015
The correlation among environmental variables, as shown by the Pearson’s correlation
coefficients (Table 4.1), was weak in most cases (R<0.60) and slightly higher for distance to
farmland with altitude (R=0.61), and distance to farmland with distance to settlements
(R=0.64). However, since all values were below 0.7, all variables were accepted in the model.
23
Table 4.1 Pearson's correlation coefficients (R) among environmental variables. The threshold to
discriminate correlated variables was R>0.7
Variables
Altitude
Distance to
Farmland
Distance to
Forest Edge
Distance to
Roads
Distance to
Settlements
Slope
Altitude
1.00
0.62
-0.29
0.24
0.56
0.58
Distance to
Farmland
0.62
1.00
-0.25
0.25
0.64
0.39
Distance to
Forest Edge
-0.29
-0.25
1.00
-0.08
-0.21
-0.26
Distance to
Roads
0.24
0.25
-0.08
1.00
0.39
0.15
Distance to
Settlements
0.56
0.64
-0.21
0.39
1.00
0.33
Slope
0.58
0.39
-0.26
0.15
0.33
1.00
Overall, the model showed good performances, indicated by an AUC of 0.805 (SD=0.072).
According to the percent contribution values, the most important predictors for wolf
suitability were distance to settlements, distance to farmlands and distance to roads (Table
4.2), which were all positively correlated with habitat suitability (Figure 4.2). However, based
on the permutation importance values, distance to forest edge seemed also to be very
important, and negatively correlated with probability of occurrence.
Table 4.2 Main statistical values showing the relative contribution of environmental variables in Maxent
Jackknife on AUC
Variable
Percent
Contribution (%)
Permutation
Importance (%)
Without Variable
With Only
Variable
Distance to Settlements
33.46
29.48
0.804
0.763
Distance to Farmland
23.13
14.43
0.801
0.760
Distance to Roads
21.41
11.86
0.751
0.591
Distance to Forest Edge
8.34
33.10
0.796
0.655
Altitude
11.38
8.42
0.802
0.674
Slope
2.27
2.71
0.816
0.590
24
Figure 4.2 Response curves for the 6 model predictors. The curves show how the species probability of
occurrence changes with each predictor, maintaining all other predictors at their average sample value. The red
curves represent the mean trends, while the blue shades show the mean +/- the standard deviation. In each graph,
the X axis shows the change in each environmental variable, while the Y axis shows the species' probability of
presence.
The values of environmental variables at occurrence locations show a very high variability
indicated by high standard deviation values (Table 4.3). Looking at the minimum values, it
can be noticed that some homesites were located very near roads and farmland, while they
tended to be located further from human settlements. Finally, apart in some extreme cases,
homesites were, in average, very close or inside the forest.
25
Table 4.3 Minimum, maximum, mean and standard deviation values for the 6 environmental variables.
For “distance to” variables, nil values indicate that a homesite is located in the same cell (measuring 250x250 m)
of an environmental feature and, thus, do not necessarily pinpoint a distance of 0 metres.
Variable
Min
Max
Mean
STD
Distance to Roads
0.00
2610.08
674.88
625.55
Distance to Farmland
0.00
15337.86
5604.21
3787.99
Altitude
285.47
1496.54
867.94
325.10
Slope
0.17
14.39
6.40
3.69
Distance to Settlements
901.39
10960.16
5106.61
2757.53
Distance to Forest Edge
0.00
2015.56
65.02
356.12
Based on the model output map (Figure 4.3), most suitable areas are found along the Dinaric
Mountains and in smaller, isolated and currently unoccupied areas in the northern and north-
eastern parts of Croatia. In the map, a high breeding habitat fragmentation caused by roads
can also be noticed.
Figure 4.3 Habitat suitability map obtained with Maxent. Blue indicates low suitability, green indicates
intermediate suitability and red indicates high suitability areas.
26
4.2 Wind Farm Prioritisation
Proposed wind farms are mainly located within the current wolf distribution and overlap with
several high quality wolf reproduction areas (Figure 1.2, Figure 4.3, Appendix I). The Marxan
analysis shows that, according to the 2 km and 4 km buffer scenarios (Figure 4.4, Table 4.4),
the 2020 target of the Croatian energy strategy would be met respectively with only 15 and 12
wind farms. These correspond respectively to 44.5% and 36.4% of the 33 total proposed wind
farms (Table 4.4).
In both scenarios, after the selection, the resulting installed capacity would be 748 MW (i.e.
48.1% of the total initial capacity). With respect to the potential impact on wolf breeding
habitat, the optimisation would lead to a decrease of 76.69% in the 2 km buffer scenario, and
of 80.49% in the 4 km buffer scenario. Thus, in the former, 44.5% of the proposed wind farms
would hold only 23.31% of the total initial cost. Similarly, in the latter, 36.4% of the wind
farms would hold only 19.5% of the total initial cost.
This indicates that Marxan allowed selecting highly cost-efficient wind farms. For example,
the wind farm no.1 has the maximum irreplaceability score because it would produce a high
amount of energy for a relatively small impact. On the other hand, the wind farm no.30, with
an installed capacity even bigger than wind farm no.1, has an irreplaceability score of 0, since
its cost is also very high. Finally, the wind farm no.24 has a high irreplaceability score, since,
despite the low installed capacity, it also holds a very small cost. Moreover, the low cost
associated to wind farms no.20, 22, 24 and 27 is due to the fact that these wind farms are
planned around already operating ones. As such, their additional cost is limited.
27
Figure 4.4 Best solution for the Marxan analysis over 100 repetitions in the 2 km (A) and 4 km (B) buffer
scenarios. The number of each wind farms corresponds to the numbers in table 4.4
28
Table 4.4 Marxan values for all wind farms in the 2 km and 4 km buffer scenarios. IS=Irreplaceability score
over 100 Marxan repetitions; MW=Installed capacity in MW; %=percent of the analogous initial value; %
decrease=percent reduction compared to the analogous initial value. “MW in Best Solution” and “Cost in Best
Solution” only show the MW and the cost of selected wind farms in the best solutions. The numbers in this table
correspond to wind farms shown in figure 4.4. Further information about these wind farms can be found in
Appendix I
2 Km Buffer
4 Km Buffer
Wind
Farm
No.
MW
Cost
IS
MW in
Best
Solution
Cost in
Best
Solution
Cost
IS
MW in
Best
Solution
Cost in
Best
Solution
1
117
48.40
100
117
48.40
92.23
100
117
92.23
2
57
149.01
32
57
149.01
224.28
32
3
70
90.62
98
70
90.62
192.12
91
70
192.12
4
22
129.98
0
312.86
0
5
10
45.78
10
173.67
0
6
39
67.03
89
39
67.03
196.63
9
7
48
89.37
80
48
89.37
264.87
9
8
27
21.22
99
27
21.22
3.36
100
27
3.36
9
45
103.07
55
269.53
3
10
33
250.52
0
470.15
0
11
54
117.88
59
54
117.88
169.61
77
54
169.61
12
33
134.64
2
284.25
0
13
45
118.26
20
168.89
40
14
23
58.47
42
23
58.47
65.16
85
23
65.16
15
21
51.99
39
139.26
7
16
64
385.12
0
759.12
0
17
10
47.74
2
188.63
0
18
42
126.77
15
243.79
8
19
80
103.40
99
80
103.40
200.20
99
80
200.20
20
33
7.98
100
33
7.98
16.36
100
33
16.36
21
42
88.42
64
42
88.42
147.49
38
22
48
7.33
100
48
7.33
8.89
100
48
8.89
23
23
133.52
0
81.62
40
24
18
5.96
100
18
5.96
2.37
99
18
2.37
25
45
116.81
28
208.50
14
26
10
21.58
40
80.70
11
27
20
15.36
98
20
15.36
9.31
99
20
9.31
28
45
128.45
19
389.17
0
29
20
133.15
0
237.81
0
30
120
764.02
0
1777.15
0
31
186
516.39
21
708.44
67
186
708.44
32
72
130.20
83
72
130.20
173.44
98
72
173.44
33
33
85.11
23
155.32
19
TOTAL
1555
4293.56
748.00
1000.66
8415.19
748.00
1641.49
%
100
100
48.10
23.31
100
48.10
19.51
%
Decrease
0
0
51.9
76.69
0
51.9
80.49
29
The Marxan analysis for the hexagonal grid shows the cost-efficiency of each hexagonal cell
over the total area covered by the 33 proposed wind farms, considering a 4 km buffer around
each turbine (Figure 4.5). The cells that produce the most with a relatively lower impact are
more likely to be selected by Marxan and, thus, they have a high irreplaceability score. By
comparing this figure with Figure 4.4 it can be seen that, in general, relatively high scoring
cells are located mostly within the areas where wind farms were selected in the 4 km buffer
scenario.
Figure 4.5 Marxan analysis of the hexagonal grid over the area covered by proposed wind farms with a 4
km buffer. The irreplaceability score is the number of times in which each cell was selected in the optimal
configuration over 100 Marxan repetitions. The figure shows the extent to which wolf breeding habitat in each
cell would be affected by wind farms construction.
30
5 Discussion
5.1 Habitat Suitability Model
This study presents the first habitat suitability model for wolf breeding habitat in Croatia. It is
therefore important because it provides valuable information about wolf habitat selection and
potential distribution.
The habitat suitability model obtained an AUC value of 0.805. According to Swets (1988,
Hosmer Jr and Lemeshow 2004) and Elith (2000), such a value indicates good model
performance. Moreover, this value of discriminative power is only slightly lower than the
ones obtained in other similar studies. For example, Iliopoulos, Youlatos et al. (2014)
obtained an AUC value of 0.818, while Bassi, Willis et al. (2015) and Ahmadi, Kaboli et al.
(2013) reached an AUC of 0.876 and 0.894 respectively. However, their sample sizes (i.e. 35,
146 and 35 occurrences) were bigger than in this study, thus increasing the likelihood of
obtaining a higher AUC value.
With respect to model predictors, the Pearson’s correlation analysis showed that there was a
moderate correlation between some of the variables. This correlation never had a coefficient
higher than 0.64 and was therefore considered acceptable as in other studies (Dormann, Elith
et al. 2013, Kramer-Schadt, Niedballa et al. 2013, Syfert, Smith et al. 2013). Nonetheless, the
relative contributions of correlated variables should still be interpreted with caution, as it is
impossible to determine which is the most important in predicting suitability (Baldwin 2009).
For example in this study, distance to settlements and distance to farmland were the two most
important predictors for habitat suitability. However, they were also the most correlated
variables. Thus, the values of the analysis of variable contributions for these two predictors
may not be representative of their independent importance in determining habitat suitability.
Nevertheless, settlements and farmlands are both related to human activities which might
deter wolves from breeding in their proximity. Hence, their relative importance may be
proportional to the type and extent of the disturbance they cause. In fact, it has been widely
shown that wolves tend to avoid humans and to locate breeding sites far away from villages,
farms and roads (Theuerkauf, Rouys et al. 2003, Kusak, Majić-Skrbinšek et al. 2005,
Jędrzejewski, Jędrzejewska et al. 2008, Ahmadi, Kaboli et al. 2013, Bassi, Willis et al. 2015).
In this study, the distance to roads was positively correlated with wolf habitat suitability and
was another important variable. However, looking at the response curve it can be noticed that,
31
with increasing distance, the suitability increases rapidly, reaching a plateau after few hundred
meters. This result is consistent with other studies (Theuerkauf, Rouys et al. 2003, Kaartinen,
Kojola et al. 2005, Ahmadi, Kaboli et al. 2013). Hence, it seems that roads are likely to have
an effect on breeding habitat only for the first few hundred meters. Moreover, it has been
shown that roads may facilitate wolf movements, especially during the denning season, when
adult wolves have to provide food for other pack members (Zimmermann, Nelson et al.
2014).
Among the environmental predictors, the most influential was distance to forest edge, while
altitude and slope only showed minor contributions. However, although similar results are
common for human dominated areas (Theuerkauf, Rouys et al. 2003, Ahmadi, Kaboli et al.
2013), some environmental variables that could potentially have higher contributions, such as
prey availability and water sources, were not considered in this study, since adequate data
were not available. In any case, given the low dependency of wolves on particular habitats, in
human dominated regions like Europe, anthropic variables are more likely to play a major
role in determining habitat suitability (Mech and Boitani 2003, Ahmadi, Kaboli et al. 2013).
The habitat suitability map is consistent with the current knowledge about wolf habitat and
wolf distribution in Croatia (Kaczensky 2012). Most of the predicted suitable areas
correspond to the currently occupied areas, especially in the Dinaric Mountains and in the
region in central Croatia protruding into the north-west of Bosnia and Herzegovina.
The main area predicted suitable outside the wolf current distribution is the region around the
Papuk Mountain. This area may potentially accommodate a future expansion from northern
Bosnia. However, this expansion is very unlikely in the near future, since the area is
completely surrounded by farmland and is isolated by a fenced highway without crossing
structure. This area is also rather far from currently occupied sites. All other areas which were
predicted to be suitable mainly correspond to confined forest patches in mountainous areas
and are too small and isolated to represent potentially meaningful expansion areas.
In unsuitable areas, especially in the currently occupied range, wolves might still be regularly
present. This study only models dens and rendezvous sites, and does not consider the winter
time, nor wolf movements in the breeding season. In fact, it was shown that, during the
breeding season, adult wolves in North America may walk up to 48 km from the den to obtain
food (Mech 1988). Despite this distance being smaller in Europe (Kusak, Majić-Skrbinšek et
al. 2005), it is still likely that wolves spend a large part of their time in unsuitable breeding
32
habitat. Moreover, although wolves avoid human disturbance for locating dens and
rendezvous sites (Theuerkauf, Rouys et al. 2003, Kusak, Majić-Skrbinšek et al. 2005,
Jędrzejewski, Jędrzejewska et al. 2008, Ahmadi, Kaboli et al. 2013, Bassi, Willis et al. 2015),
they still visit and feed from highly humanized places, including villages, roads, farms and
garbage dumps (Ciucci, Boitani et al. 1997, Kusak, Majić-Skrbinšek et al. 2005).
Nonetheless, they always tend to minimize their direct contact with people, mainly by
segregating their activity pattern during night time (Kusak, Majić-Skrbinšek et al. 2005).
In Croatia, this situation is particularly common in Dalmatia, where human-wolf conflict is
more intense (Kusak, Majić-Skrbinšek et al. 2005). In this area, livestock depredation by
wolves is the main cause of conflict (Kusak, Majić-Skrbinšek et al. 2005, Majić and Bath
2010). This often leads to retaliatory killing, which represents one of the main causes of wolf
mortality in Croatia (Huber, Kusak et al. 2002, Kusak, Majić-Skrbinšek et al. 2005). This
conflict, resulting from the coexistence of humans alongside wolves, might be one of the
reasons why habitat suitability is predicted to be lower in Dalmatia than in other parts of the
current wolf range. In fact, as it can be noticed in the habitat distribution map (Figure 2.1),
forests in Dalmatia are more fragmented by a relatively large amount of farmland, compared
to other areas in the wolf range.
In spite of the good performance, the model also presents some limitations mainly related to
the difficulties in wolf data collection, notably in Karstic and highly rugged terrains. In
particular, the main limitation was the large time interval over which the homesites were
spread. In the model, the occurrence localities collected from 1997 to 2015 were related and
projected to environmental variables fixed in 2006. Hence, the model was carried out
assuming that general habitat conditions did not change substantially between 1997 and 2015.
Among the environmental predictors, apart from altitude and slope, which are obviously
invariable across time, forest cover showed an increase of only 2.79% from 1997 to 2012
(FAO 2015). With respect to the predictors related to human disturbance, the changes were
slightly higher. From 1996 to 2014, the total population decreased by 5.73%, with most of
this decline occurring in rural areas, while arable land decreased by 8.11% between 1997 and
2012 (Croatian Bureau of Statistics 2011, FAO 2015). Although these changes were higher,
they were still considered acceptable. This decision was consistent with that of Jędrzejewski,
Jędrzejewska et al. (2008), who accepted a population increase of circa 5% over the time
interval in which data were collected. Moreover, several other studies that share similar time
33
discrepancies between occurrence localities and environmental variables have overlooked this
type of limitation (Corsi, Duprè et al. 1999, Treves, Naughton-Treves et al. 2004, Iliopoulos,
Youlatos et al. 2014, Bassi, Willis et al. 2015).
5.2 Wind Farm Prioritisation
This study presents important results to support the spatial planning of wind energy in Croatia
and it will contribute to the assessment of environmental impacts of proposed wind farms.
The prioritisation carried out in this study would potentially lead to a reduction of wind farm
impacts on wolf breeding habitat of up to 80.5% with a decrease of only 52% in potential
installed capacity. This reduction was similar in both buffer scenarios and was due to the
selection of wind farms with a high capacity and a low cost. This low cost is not necessarily
associated to unsuitable wolf habitat. For example it can be due to the presence of already
existing wind farms. In fact, locating proposed wind farms near others already in operation
would likely reduce their additional impact on the wolf, by avoiding disturbance in new
“undisturbed” areas. The Marxan analysis also produced a sensitivity map across the whole
planning area. This output will allow wind farm planners to identify the areas where wind
turbines are more likely to have higher impacts and require modifications (e.g. displacement
to other areas).
Although several studies have been published on the spatial planning of wind farms (Baban
and Parry 2001, Punt, Groeneveld et al. 2009, Aydin, Kentel et al. 2010, Tegou, Polatidis et
al. 2010, Drechsler, Ohl et al. 2011, Baltas and Dervos 2012, Göke and Lamp 2012), the
obtained results are highly specific to the area, the type of environment, the nature of the
costs, the planning units considered, and the method adopted. It is therefore difficult to
compare these outputs and their effectiveness with the ones of other studies. However,
Marxan is considered ideal in optimisation problem solving, also in different applications
from strategic conservation planning (Ardron, Possingham et al. 2008, Göke and Lamp 2012).
For example, Marxan was used in the context of wind farm spatial planning in at least one
occasion. In particular, in an offshore wind energy implementation project in the Baltic Sea, it
was integrated as a support tool in the spatial planning process (Göke and Lamp 2012). In this
case study, Marxan has shown to be an adequate and successful method also in addressing
wind farm prioritisation (Göke and Lamp 2012). For these reasons, and considering the type
of problem addressed and the nature of the outputs required in this study, Marxan was
preferred to the other conventional methods.
34
Despite the choice of the most suitable method, this study presented some limitations. In
particular, the main weakness was related to the determination of the ecological cost in the
areas where two or more proposed wind farms overlapped. The Marxan analysis was carried
out by assuming that each wind farm would be built independently from other farms.
However, if two or more proposed wind farms share the area over which they may have a
potential effect (i.e. the 2 km or 4 km buffer), they would also share the ecological cost.
Hence, if considered together, they would have the same installed capacity than if both were
considered singularly, but they would have a lower cumulative cost. Unfortunately, this
shortfall could not be prevented, since Marxan cannot handle overlapping planning units
(Ardron, Possingham et al. 2008). Nonetheless, for the 2 and 4 km buffer scenarios the total
overlapping area was only around 8% and 11% respectively, and was distributed equally
across many wind farms. Hence, it is likely that no particular areas would benefit from wind
farms being built together in clusters. Moreover, this limitation would have existed also in the
other conventional methods used in previous similar studies.
This approach to Marxan can also be extended to other infrastructure and to the other two
species of large carnivores in Croatia. However, some complications could arise when
considering multiple and incommensurate costs in optimisation processes (Göke and Lamp
2012). In particular, in order to be minimised, the different types of costs have to be merged
in a single overall cost (Punt, Groeneveld et al. 2009, Drechsler, Ohl et al. 2011, Göke and
Lamp 2012). As such, each single cost has to be given a subjective weight that reflects its
importance in the calculation of the total cost. For large carnivores, it may be difficult to
determine these weights, since detailed information about the extent of wind farms’ impacts
on each species are not available. Moreover, once the total cost is minimised, it should be
verified that the minimisation occur equally for each single cost and that all cost are
satisfactorily minimised.
5.3 Future Implications and Recommendations
This study provides valuable tools for the future conservation of wolves in Croatia and
Europe. In particular, the habitat suitability model offers a better understanding of breeding
wolf environmental requirements and provides a useful map showing the potential
distribution of wolves in Croatia. This information can be used for the improvement of the
management plan for the wolf in Croatia, for future conservation planning, for the prevention
of human-wolf conflicts, for environmental impact assessments and for awareness raising
35
campaigns (see section 2.4.2). Moreover, the results of the prioritisation show the optimal
configuration of wind farms to meet the Croatian target at the lowest impact on wolf habitat,
and will contribute to a large EIA for wind farms in Croatia. Hence, this study presents a
scientific and evidence-based framework to support the sustainable implementation of
proposed infrastructure, and contributes in ensuring the long-term viability of wolves and
other charismatic and wide-ranging species in Croatia and Europe.
However, despite the usefulness and practicality of these outputs, more work is required to
improve the accuracy of scientific findings and increase the effectiveness of science on policy
and decision making. Notably, more effort should be put into the identification of more wolf
homesites in a more restricted time interval, in order to bypass the limitations highlighted in
this study and produce a yet more accurate map of breeding habitat suitability.
Moreover, the qualitative and quantitative impact of wind farms on wolves and other non-
volant animals should be clarified (Lovich and Ennen 2013). For this purpose, constant wolf
monitoring should be carried out in and around areas where wind farms are proposed or built.
Energy consumption in Croatia is projected to increase further by 2030 and wind energy has
been identified as the main source of renewable energy (Ministry of Economy and UNDP
2008). Therefore, wolf monitoring should also be realized in areas where wind energy may
potentially be implemented in the longer term. This would provide data over longer periods to
carry out more accurate BACI analysis in the future. From this perspective, a more thorough
and regular communication of intents and objections among scientists, politicians and wind
power developers would be beneficial and is, thus, urgently required. In an environmental
context, conservationists, politicians and wind farms developers should share at least part of
their values and objectives. However, a regular communication and cooperation among these
stakeholders in Croatia is currently lacking (Švarc 2006). It is therefore important that these
parties find a common thread, while still acknowledging their differences in short-term stakes.
Lastly, a more direct communication is essential to enable the adoption of an adaptive
management approach for wolf monitoring and wind-energy-related decision making.
Furthermore, although prioritisation is a useful way to reduce wind farms’ potential impacts
on wolves during planning processes, other measures for the minimisation of these impacts
should also be taken into consideration during the construction and operation phases. For
example, in the construction phase of a wind power plant in Portugal, critical areas within the
future wind farms were identified and all construction activities were prohibited during the
36
denning period (Soares, Duarte et al. 2011). All activities in all areas were also forbidden
from sunset to sunrise (i.e. the period when wolves are most active). Furthermore, since roads
seem to represent one of the main factor impacting wolves around wind farms during the
operation phase (Huber, Kusak et al. 2002, Àlvares, Rio-Maior et al. 2011, Helldin, Jung et
al. 2012), in Portugal it has been proposed to close access roads in order to reduce traffic and
direct human disturbance (Àlvares, Rio-Maior et al. 2011).
From a wider perspective, wind is a source of renewable energy that could curb our
dependence on fossil fuels and significantly decrease greenhouse gasses emissions (Sims,
Rogner et al. 2003). Thus, wind energy has several environmental advantages and represents
an outstanding opportunity towards anthropogenic climate change mitigation (Edenhofer,
Pichs-Madruga et al. 2011). On the other hand, the generation of energy through the use of
wind turbines could have a negative environmental impact related to the large amount of land
required for wind energy implementation (Kiesecker, Evans et al. 2011). In anthropic-
dominated regions like Europe, where the available surface of land is limited, the landscape is
the result of competition among agriculture, urbanisation, conservation, energy production
and other land use types. This competition is the main factor that determines the state of our
environment and economies (Rounsevell, Reginster et al. 2006). It is, therefore, crucial that
land management decisions take into consideration the environmental, social and economic
opportunities and implications of each land use activity.
In conclusion, this study presents a systematic and repeatable framework for infrastructure
prioritisation based on its ecological impact on wide-ranging carnivore species. In particular,
it offered scientific evidence of the spatial distribution of wolf breeding habitat and adopted it
in the strategic prioritisation of planned wind farms in Croatia. As such, it provides
fundamental information for wolf conservation and represents a small step towards a more
equal and sustainable land management in Europe; environmentally, socially and
economically.
37
6 References
Ahmadi, M., M. Kaboli, E. Nourani, A. A. Shabani and S. Ashrafi (2013). "A predictive spatial model
for gray wolf (Canis lupus) denning sites in a human-dominated landscape in western Iran."
Ecological research 28(3): 513-521.
Albrechts, L., P. Healey and K. R. Kunzmann (2003). "Strategic spatial planning and regional
governance in Europe." Journal of the American Planning Association 69(2): 113-129.
Àlvares, F., H. Rio-Maior, S. Roque, M. Nakamura, D. Cadete, S. Pinto and F. Petrucci-Fonseca
(2011). Assessing ecological responses of wolves to wind power plants in Portugal: methodological
constraints and conservation implications. Proceedings, Conference on Wind Energy and Wildlife
Impacts, 2–5 May 2011. K. B. Roel May. Trondheim, Norway.
Álvares, F., H. Rio-Maior, S. Roque, M. Nakamura and F. Petrucci-Fonseca (in press). Ecological
response of breeding wolves to wind farms: insights from two case studies in Portugal. Wildlife and
Wind Farms: Conflicts and Solutions. Volume 1: Onshore. M. R. Perrow, Pelagic Publishing. Volume
1: Onshore: 432.
Ardron, J. A., H. P. Possingham and C. J. Klein (2008). "Marxan good practices handbook." Pacific
Marine Analysis and Research Association, Vancouver.
Aydin, N. Y., E. Kentel and S. Duzgun (2010). "GIS-based environmental assessment of wind energy
systems for spatial planning: A case study from Western Turkey." Renewable and Sustainable Energy
Reviews 14(1): 364-373.
Baban, S. M. and T. Parry (2001). "Developing and applying a GIS-assisted approach to locating wind
farms in the UK." Renewable energy 24(1): 59-71.
Baldwin, R. A. (2009). "Use of maximum entropy modeling in wildlife research." Entropy 11(4): 854-
866.
Ball, I. R., H. P. Possingham and M. Watts (2009). "Marxan and relatives: software for spatial
conservation prioritisation." Spatial conservation prioritisation: quantitative methods and
computational tools. Oxford University Press, Oxford: 185-195.
Baltas, A. and A. Dervos (2012). "Special framework for the spatial planning & the sustainable
development of renewable energy sources." Renewable Energy 48: 358-363.
Ban, N. C. and A. C. Vincent (2009). "Beyond marine reserves: exploring the approach of selecting
areas where fishing is permitted, rather than prohibited." PloS one 4(7): e6258.
Bassi, E., S. G. Willis, D. Passilongo, L. Mattioli and M. Apollonio (2015). "Predicting the Spatial
Distribution of Wolf (Canis lupus) Breeding Areas in a Mountainous Region of Central Italy." PloS
one 10(6): e0124698.
Bischof, R., H. Brøseth and O. Gimenez (2015). "Wildlife in a politically divided world: insularism
inflates estimates of brown bear abundance." Conservation Letters.
38
Boitani, L. (2015). The Return of Large Carnivores to Europe. ICCB - International Congress for
Conservation Biology. S. f. C. Biology. Montpellier, Society for Conservation Biology.
Brotons, L., W. Thuiller, M. B. Araújo and A. H. Hirzel (2004). "Presence‐absence versus presence‐
only modelling methods for predicting bird habitat suitability." Ecography 27(4): 437-448.
Capitani, C., L. Mattioli, E. Avanzinelli, A. Gazzola, P. Lamberti, L. Mauri, M. Scandura, A. Viviani
and M. Apollonio (2006). "Selection of rendezvous sites and reuse of pup raising areas among
wolvesCanis lupus of north-eastern Apennines, Italy." Acta theriologica 51(4): 395-404.
Cayuela, L. (2004). "Habitat evaluation for the Iberian wolf Canis lupus in Picos de Europa National
Park, Spain." Applied Geography 24(3): 199-215.
Chapron et al. (2014). "Recovery of large carnivores in Europe's modern human-dominated
landscapes." Science 346(6216): 1517--1519.
Ciucci, P., L. Boitani, F. Francisci and G. Andreoli (1997). "Home range, activity and movements of a
wolf pack in central Italy." Journal of Zoology 243(4): 803-819.
Corsi, F., E. Duprè and L. Boitani (1999). "A large‐scale model of wolf distribution in Italy for
conservation planning." Conservation Biology 13(1): 150-159.
Croatian Parliament (2009a). Energy Strategy of the Republic of Croatia. Official Gazette 130/09. P.
o. t. R. o. Croatia. Zagreb.
Croatian Parliament (2009b). Ordinance on the assessment of acceptability of plans, programs and
interventions for the ecological network (OG 118/09). Ministry of Environment and Nature Protection
& Parliament of the Republic of Croatia. Zagreb. OG 118/09.
Croatian Bureau of Statistics (2011). Croatian Population Census, The World Bank- World
Development Indicators.
Dormann, C. F., J. Elith, S. Bacher, C. Buchmann, G. Carl, G. Carré, J. R. G. Marquéz, B. Gruber, B.
Lafourcade and P. J. Leitão (2013). "Collinearity: a review of methods to deal with it and a simulation
study evaluating their performance." Ecography 36(1): 27-46.
Drechsler, M., C. Ohl, J. Meyerhoff, M. Eichhorn and J. Monsees (2011). "Combining spatial
modeling and choice experiments for the optimal spatial allocation of wind turbines." Energy Policy
39(6): 3845-3854.
Drewitt, A. L. and R. H. Langston (2006). "Assessing the impacts of wind farms on birds." Ibis
148(s1): 29-42.
Dudík, M., S. J. Phillips and R. E. Schapire (2005). Correcting sample selection bias in maximum
entropy density estimation. Advances in neural information processing systems.
39
EC, European Commission (2009). Directive 2009/28/EC of the European Parliament and of the
Council of 23 April 2009 on the promotion of the use of energy from renewable sources and amending
and subsequently repealing Directives 2001/77/EC and 2003/30/EC. Official Journal of the European
Union. 140: 16-62.
EC, European Commission (2011). "Environmental Impact Assessment (EIA) Directive 85/337/EEC,
97/11/EC, 2003/35/EC, 2009/31/EC codified in 2011/92/EU."
EC, European Commission (2013). Agriculture, forestry and fisheries statistics - 2013 edition.
Agriculture and fisheries. P. O. o. t. E. Union. Luxembourg, European Commission: 249.
Edenhofer, O., R. Pichs-Madruga, Y. Sokona, K. Seyboth, S. Kadner, T. Zwickel, P. Eickemeier, G.
Hansen, S. Schlömer and C. von Stechow (2011). Renewable energy sources and climate change
mitigation: Special report of the intergovernmental panel on climate change, Cambridge University
Press.
Elith, J. (2000). Quantitative methods for modeling species habitat: comparative performance and an
application to Australian plants. Quantitative methods for conservation biology, Springer: 39-58.
Elith, J., C. H. Graham, R. P. Anderson, M. Dudík, S. Ferrier, A. Guisan, R. J. Hijmans, F. Huettmann,
J. R. Leathwick and A. Lehmann (2006). "Novel methods improve prediction of species' distributions
from occurrence data." Ecography 29: 129-151.
Elith, J., S. J. Phillips, T. Hastie, M. Dudík, Y. E. Chee and C. J. Yates (2011). "A statistical
explanation of MaxEnt for ecologists." Diversity and Distributions 17(1): 43-57.
Fabbri, E., R. Caniglia, J. Kusak, A. Galov, T. Gomerčić, H. Arbanasić, D. Huber and E. Randi
(2014). "Genetic structure of expanding wolf (Canis lupus) populations in Italy and Croatia, and the
early steps of the recolonization of the Eastern Alps." Mammalian Biology-Zeitschrift für
Säugetierkunde 79(2): 138-148.
FAO (2015). Land Use Statistics, The World Bank - World Development Indicators.
Fourcade, Y., J. O. Engler, D. Rödder and J. Secondi (2014). "Mapping species distributions with
MAXENT using a geographically biased sample of presence data: a performance assessment of
methods for correcting sampling bias."
Franklin, J. F. (1989). Importance and justification of long-term studies in ecology. Long-term studies
in ecology, Springer: 3-19.
Glenz, C., A. Massolo, D. Kuonen and R. Schlaepfer (2001). "A wolf habitat suitability prediction
study in Valais (Switzerland)." Landscape and Urban Planning 55(1): 55-65.
Göke, C. and J. Lamp (2012). Case Study: Systematic Site Selection for Offshore Wind Power with
Marxan in the Pilot Area Pomeranian Bight. BaltSeaPlan Reports. B. Publications. Rostock, Germany,
BaltSeaPlan. Report 29.
Harrington, F. H., C. S. Asa, L. Mech and L. Boitani (2003). "Wolf communication." Wolves:
Behavior, ecology, and conservation: 66-103.
40
Harrington, F. H. and L. D. Mech (1978). "Howling at two Minnesota wolf pack summer homesites."
Canadian Journal of Zoology 56(9): 2024-2028.
Harrington, F. H. and L. D. Mech (1982). "An analysis of howling response parameters useful for wolf
pack censusing." The Journal of Wildlife Management: 686-693.
Helldin, J. O., J. Jung, W. Neumann, M. Olsson, A. Skarin and F. Widemo (2012). "The impacts of
wind power on terrestrial mammals." Naturvardsverket, Swedish Environmental Protection Agency,
Report 6510: 1-51.
Hernandez, P. A., C. H. Graham, L. L. Master and D. L. Albert (2006). "The effect of sample size and
species characteristics on performance of different species distribution modeling methods." Ecography
29(5): 773-785.
Hirzel, A. H. and G. Le Lay (2008). "Habitat suitability modelling and niche theory." Journal of
Applied Ecology 45(5): 1372-1381.
Hosmer Jr, D. W. and S. Lemeshow (2004). Applied logistic regression, John Wiley & Sons.
Houle, M., D. Fortin, C. Dussault, R. Courtois and J.-P. Ouellet (2010). "Cumulative effects of
forestry on habitat use by gray wolf (Canis lupus) in the boreal forest." Landscape ecology 25(3): 419-
433.
Huber, Đ., J. Kusak, G. Gužvica and T. Gomerčić (2002). "Causes of wolf mortality in Croatia during
1986-2001." Veterinarski arhiv 72(3): 131-139.
Iliopoulos, Y., D. Youlatos and S. Sgardelis (2014). "Wolf pack rendezvous site selection in Greece is
mainly affected by anthropogenic landscape features." European journal of wildlife research 60(1): 23-
34.
Jeremić, J. K., J.Huber, Đ.Štrbenac, A. (2012). Report on Wolf Population in Croatia. Zagreb, State
Institute for Nature Protection: 37.
Jędrzejewski, W., B. Jędrzejewska, B. Zawadzka, T. Borowik, S. Nowak and R. Mysłajek (2008).
"Habitat suitability model for Polish wolves based on long‐term national census." Animal
Conservation 11(5): 377-390.
Jędrzejewski, W., M. Niedzialkowska, R. W. Mysłajek, S. Nowak and B. Jędrzejewska (2005).
"Habitat selection by wolvesCanis lupus in the uplands and mountains of southern Poland." Acta
Theriologica 50(3): 417-428.
Kaartinen, S., I. Kojola and A. Colpaert (2005). Finnish wolves avoid roads and settlements. Annales
Zoologici Fennici, JSTOR.
Kaczensky et al. (2012). Status, management and distribution of large carnivores – bear, lynx, wolf \&
wolverine – in Europe.
Karlsson, J., H. Brøseth, H. Sand and H. Andrén (2007). "Predicting occurrence of wolf territories in
Scandinavia." Journal of Zoology 272(3): 276-283.
41
Kiesecker, J. M., J. S. Evans, J. Fargione, K. Doherty, K. R. Foresman, T. H. Kunz, D. Naugle, N. P.
Nibbelink and N. D. Niemuth (2011). "Win-win for wind and wildlife: a vision to facilitate sustainable
development." PLoS One 6(4): e17566.
Kramer‐Schadt, S., J. Niedballa, J. D. Pilgrim, B. Schröder, J. Lindenborn, V. Reinfelder, M.
Stillfried, I. Heckmann, A. K. Scharf and D. M. Augeri (2013). "The importance of correcting for
sampling bias in MaxEnt species distribution models." Diversity and Distributions 19(11): 1366-1379.
Kunz, T. H., E. B. Arnett, W. P. Erickson, A. R. Hoar, G. D. Johnson, R. P. Larkin, M. D. Strickland,
R. W. Thresher and M. D. Tuttle (2007). "Ecological impacts of wind energy development on bats:
questions, research needs, and hypotheses." Frontiers in Ecology and the Environment 5(6): 315-324.
Kusak, J., A. M. Majić-Skrbinšek and D. Huber (2005). "Home ranges, movements, and activity of
wolves (Canis lupus) in the Dalmatian part of Dinarids, Croatia." European Journal of Wildlife
Research 51(4): 254-262.
Kuvlesky Jr, W. P., L. A. Brennan, M. L. Morrison, K. K. Boydston, B. M. Ballard and F. C. Bryant
(2007). "Wind energy development and wildlife conservation: challenges and opportunities." Journal
of Wildlife Management 71(8): 2487-2498.
Liberg, O., G. Chapron, P. Wabakken, H. C. Pedersen, N. T. Hobbs and H. Sand (2012). "Shoot,
shovel and shut up: cryptic poaching slows restoration of a large carnivore in Europe." Proceedings of
the Royal Society of London B: Biological Sciences 279(1730): 910-915.
Linnell, J., V. Salvatori and L. Boitani (2008). "Guidelines for population level management plans for
large carnivores in Europe." A Large Carnivore Initiative for Europe report prepared for the European
Commission (contract 070501/2005/424162/MAR/B2) 83.
Lovich, J. E. and J. R. Ennen (2013). "Assessing the state of knowledge of utility-scale wind energy
development and operation on non-volant terrestrial and marine wildlife." Applied Energy 103: 52-60.
Majić, A. and A. J. Bath (2010). "Changes in attitudes toward wolves in Croatia." Biological
conservation 143(1): 255-260.
Management, B. o. L. (2005). Record of decision: Implementation of a wind energy development
program and associated land use plan amendments. B. o. L. M. U.S. Department of the Interior.
Washington , DC.
Marucco, F. and E. McIntire (2010). "Predicting spatio‐temporal recolonization of large carnivore
populations and livestock depredation risk: wolves in the Italian Alps." Journal of Applied Ecology
47(4): 789-798.
Mech, L. (1970). "The wolf: the ecology and behavior of an endangered species,[1st] edition.
Published for the American Museum of Natural History by the Natural History Press." Garden City,
NY.
Mech, L. D. (1988). The arctic wolf: living with the pack, Voyageur Press.
42
Mech, L. D. and L. Boitani (2003). "Wolf social ecology." Wolves: behavior, ecology and
conservation: 1-34.
Merow, C., M. J. Smith and J. A. Silander (2013). "A practical guide to MaxEnt for modeling species’
distributions: what it does, and why inputs and settings matter." Ecography 36(10): 1058-1069.
MINGO, M. o. E. (2015). "Interactive Map of Energy Sources in Croatia." Retrieved July 2015,
2015, from http://oie-aplikacije.mingo.hr/InteraktivnaKarta/.
Ministry of Culture (2005). Ordinance on habitat types, habitat maps, endangered and rare habitat
types and on measures for conservation of habitat types. Ministry of Culture of the Republic of
Croatia. Zagreb, Official Gazette. 532-10-01 / 5.4.05.
Ministry of Economy and UNDP (2008). Update/Upgrade of the Energy Strategy and of the
Implementation Programme of the Republic of Croatia. Zagreb, Ministry of Economy, Labour and
Entrepreneurship and UNDP.
Mladenoff, D. J., T. A. Sickley and A. P. Wydeven (1999). "Predicting gray wolf landscape
recolonization: logistic regression models vs. new field data." Ecological Applications 9(1): 37-44.
Navarro, L. M. and H. M. Pereira (2015). Rewilding abandoned landscapes in Europe. Rewilding
European Landscapes, Springer: 3-23.
Northrup, J. M. and G. Wittemyer (2013). "Characterising the impacts of emerging energy
development on wildlife, with an eye towards mitigation." Ecology letters 16(1): 112-125.
Packard, J. M. (2003). Wolf behavior: reproductive, social and intelligent. Wolves: Behavior, ecology
and conservation. L. D. M. a. L. Boitani, The University of Chicago Press: 35-65.
Person, D. K. and A. L. Russell (2008). "Correlates of mortality in an exploited wolf population." The
Journal of Wildlife Management 72(7): 1540-1549.
Person, D. K. and A. L. Russell (2009). "Reproduction and den site selection by wolves in a disturbed
landscape." Northwest Science 83(3): 211-224.
Phillips, S. (2005). "A brief tutorial on Maxent." AT&T Research.
Phillips, S. J., R. P. Anderson and R. E. Schapire (2006). "Maximum entropy modeling of species
geographic distributions." Ecological modelling 190(3): 231-259.
Phillips, S. J. and M. Dudík (2008). "Modeling of species distributions with Maxent: new extensions
and a comprehensive evaluation." Ecography 31(2): 161-175.
Phillips, S. J., M. Dudík, J. Elith, C. H. Graham, A. Lehmann, J. Leathwick and S. Ferrier (2009).
"Sample selection bias and presence-only distribution models: implications for background and
pseudo-absence data." Ecological Applications 19(1): 181-197.
Possingham, H., I. Ball and S. Andelman (2000). Mathematical methods for identifying representative
reserve networks. Quantitative methods for conservation biology, Springer: 291-306.
43
Pressey, R. L., M. Cabeza, M. E. Watts, R. M. Cowling and K. A. Wilson (2007). "Conservation
planning in a changing world." Trends in ecology & evolution 22(11): 583-592.
Punt, M. J., R. A. Groeneveld, E. C. Van Ierland and J. H. Stel (2009). "Spatial planning of offshore
wind farms: A windfall to marine environmental protection?" Ecological Economics 69(1): 93-103.
Radovic, J., KristijanTopic, Ramona (2006). Biodiversity of Croatia. Zagreb, Croatia, State Institute
for Nature Protection, Ministry of Culture - Republic of Croatia.
Rondinini, C. and L. Boitani (2007). "Systematic conservation planning and the cost of tackling
conservation conflicts with large carnivores in Italy." Conservation biology 21(6): 1455-1462.
Rounsevell, M., I. Reginster, M. B. Araújo, T. Carter, N. Dendoncker, F. Ewert, J. House, S.
Kankaanpää, R. Leemans and M. Metzger (2006). "A coherent set of future land use change scenarios
for Europe." Agriculture, Ecosystems & Environment 114(1): 57-68.
Røskaft, E., B. Händel, T. Bjerke and B. P. Kaltenborn (2007). "Human attitudes towards large
carnivores in Norway." Wildlife biology 13(2): 172-185.
Sims, R. E., H.-H. Rogner and K. Gregory (2003). "Carbon emission and mitigation cost comparisons
between fossil fuel, nuclear and renewable energy resources for electricity generation." Energy policy
31(13): 1315-1326.
Soares, V., S. Duarte and T. Monteiro (2011). Avoiding and minimizing impacts in Iberian Wolf
populations - EDP Renováveis Portugal experience. Workshop Business and Biodiversity, Lisbon,
EDP Renováveis Portugal.
Šprem, N., N. fabijanić, K. protrka, Z. popović, A. bulić and B. šabić (2012). "The applicability of
camera trapping to estimate population density of chamois in Biokovo Nature Park." Journal of
Central European Agriculture 12(4).
Swets, J. A. (1988). "Measuring the accuracy of diagnostic systems." Science 240(4857): 1285-1293.
Syfert, M. M., M. J. Smith and D. A. Coomes (2013). "The effects of sampling bias and model
complexity on the predictive performance of MaxEnt species distribution models." PloS one 8(2):
e55158.
Štrbenac, A. K., J.Huber, Đ.Jeremić, J.Oković, P.Majić-Skrbinšek, A.Vukšić, I.Katušić, L.Desnica,
S.Gomerčić, T.Bišćan, A.Zec, D.Grubešić, M. (2010). Wolf Management Plan for Croatia. Zagreb,
State Institute for Nature Protection.
Švarc, J. (2006). "Socio-political factors and the failure of innovation policy in Croatia as a country in
transition." Research Policy 35(1): 144-159.
Tegou, L.-I., H. Polatidis and D. A. Haralambopoulos (2010). "Environmental management
framework for wind farm siting: Methodology and case study." Journal of environmental management
91(11): 2134-2147.
Theuerkauf, J., S. Rouys and W. Jedrzejewski (2003). "Selection of den, rendezvous, and resting sites
by wolves in the Bialowieza Forest, Poland." Canadian Journal of Zoology 81(1): 163-167.
44
Treves, A., L. Naughton‐Treves, E. K. Harper, D. J. Mladenoff, R. A. Rose, T. A. Sickley and A. P.
Wydeven (2004). "Predicting human‐carnivore conflict: A spatial model derived from 25 years of data
on wolf predation on livestock." Conservation Biology 18(1): 114-125.
Wilson, C. J. (2004). "Could we live with reintroduced large carnivores in the UK?" Mammal Review
34(3): 211-232.
Wisz, M. S., R. Hijmans, J. Li, A. T. Peterson, C. Graham and A. Guisan (2008). "Effects of sample
size on the performance of species distribution models." Diversity and Distributions 14(5): 763-773.
Yackulic, C. B., R. Chandler, E. F. Zipkin, J. A. Royle, J. D. Nichols, E. H. Campbell Grant and S.
Veran (2013). "Presence‐only modelling using MAXENT: when can we trust the inferences?"
Methods in Ecology and Evolution 4(3): 236-243.
Zaniewski, A. E., A. Lehmann and J. M. Overton (2002). "Predicting species spatial distributions
using presence-only data: a case study of native New Zealand ferns." Ecological modelling 157(2):
261-280.
Zimmermann, B., L. Nelson, P. Wabakken, H. Sand and O. Liberg (2014). "Behavioral responses of
wolves to roads: scale-dependent ambivalence." Behavioral Ecology 25(6): 1353-1364.
45
7 Appendices
7.1 Appendix I
Further information about the proposed wind farms In Croatia.
Wind
Farm
Number
Capacity
(MW)
N. of
Turbines
Project Holder
Project Name
1
117
37
Cannon Libertas Co.
VE Konavodska brda
2
57
20
VE MRAVINJAC d.o.o. za
proizvodnju enerjie
VE Mravinjac
3
70
27
Vjetroelektrana Rudine d.o.o.
VE Rudine
4
22
10
Hyperborea d.o.o.
VE Rujnica
5
10
5
Vjetroelektrana Orjak d.o.o.
VE Kom-Orjak-Greda
6
39
12
Vjetroelektrana Katuni d.o.o.
VE Katuni
7
48
16
Vjetroelektrana lukovac d.o.o.
VE Lukovac
8
27
9
Dalekovod Professio
VE Voštane
9
45
30
Vjetroelektrana Jelinak d.o.o.
VE Čemernica
10
33
11
Zelovo d.o.o.
VE ST3-1/2
11
54
18
Aiolos Projekt
VE Ogorje
12
33
22
Vjetroelektrana Opor
VE Opor
13
45
30
Acciona Enerjia
VE Boraja II
14
23
10
Vjetroelektrana Glunča
VE Glunča
15
21
8
IVICOM Consulting GmbH. -
PODRUŽNICA ZAGREB
VE Mideno brdo
16
64
36
Jura Energija
VE Svilaja
17
10
5
Vjetroelektrana Ljubač
VE Ljubač - faza 1
18
42
31
C.E.M.P.
VE Krš Padene-Proširenje
19
80
38
C.E.M.P.
VE Krš Padene (KPA) 1. faza
20
33
11
Ventus Flatus
VE ZD3P
21
42
11
Vjetroelektrana Orljak
VE Orljak
22
48
15
Kunovac
VE ZD2P
23
23
10
Venti
VE Krug - Bikina Glava
24
18
7
EKO Zadar DVA
VE ZD4P
25
45
15
Poštak
Proširenje ZD6 (dio) snage oko
45 MW
46
Wind
Farm
Number
Capacity
(MW)
N. of
Turbines
Project Holder
Project Name
26
10
10
C.E.N.S.U.R. - Zrmanja
Kompleks male vjetroelektrana
Jasenice
27
20
6
IN POSTERUM d.o.o.
VE ZD5
28
45
15
Vjetroelektrana Bruvno
VE Bruvno
29
20
6
Dalekovod Professio
VE Mazin 2
30
120
61
LIKA-FENIKS
Kompleks vjetroelektrana
Udbina 120MW
31
186
63
Energija Projekt
VE Senj
32
72
49
EURUS d.o.o. Za projektiranje i
nadzor
VE Goli
33
33
13
WPD ENERSYS d.o.o.
VE Bila Ploča
