Blackwell Publishing Ltd

Download 1.51 Mb.
Date conversion15.05.2016
Size1.51 Mb.
  1   2   3

Blackwell Publishing Ltd Land-use changes may explain the recent range expansion of the Black-shouldered Kite Elanus

caeruleus in southern Europe


1Universidad de Extremadura, Departamento de Anatomía, Biología Celular y Zoología, Av/de Elvas s/n, 06071 Badajoz, Spain

2Estación Biológica de Doñana, Department of Evolutionary Ecology, C. S. I. C. Pabellón del Perú,

Av/Maria Luisa s/n, 41013 Sevilla, Spain

3Centro para el Estudio y Conservación de las Aves Rapaces en Argentina (CECARA), Facultad de Cs.

Exactas y Naturales UNLPam, Avda. Uruguay 151, 6300 Santa Rosa, La Pampa, Argentina

4Dirección General del Medio Natural. Junta de Extremadura, Av/de Portugal s/n, 06800 rida, Spain

Occasional observations of Black-shouldered Kites Elanus caeruleus in Europe date back to the mid-19th century, but it was only recorded as a breeding species in the early 1960s in Portugal and a few years later in neighbouring Spain. This recent colonization, possibly from Africa where the species is abundant, may be due to climate change, land-use changes in southern Europe, or both. As a first step to understanding this range expansion process we have developed a habitat selection model using data from the current strongholds of its European distribution. Comparing the proportion of different habitat types around 46 breeding sites and 45 randomly chosen plots, we have found that the area of cultivated parklands known as dehesas in Spain is a strong predictor of the current distribution range of breeding pairs of Black-shouldered Kites. Specifically, the percentage of dehesas with planted cereal and a low density of trees (i.e. < 7 trees/ha and thus a savannah-like habitat) within the study plots explained 44.6% of the residual deviance in our model. The minimal adequate model classified 81.3% of breeding sites and random plots correctly. Our results suggest that Black-shouldered Kites may have taken advantage of the gradual increase of cultivated dehesas in the second half of the 20th century to expand its range in Europe. This particular type of dehesa is structurally similar to the African savannahs where the species thrives and may offer a higher density of rodents than traditional dehesas, which primarily contain pastureland for livestock ranching.

Keywords: colonization, dehesa, habitat selection, land-use change, occurrence models, raptors.

The Black-shouldered Kite Elanus caeruleus is a relatively small bird of prey occurring in open grasslands with scattered trees (i.e. savannahs) in Africa, India and southeastern Asia (Brown & Amadon 1968, Del Hoyo et al. 1994). It is classified as vulnerable in the European Union (Tucker et al.

1994) and near-threatened in Spain (Martí & del

Moral 2004). This raptor and its close relatives

*Corresponding author. Email:

(Roulin & Wink 2003) and congeners in the Americas (E. leucurus) and Australia (E. axillaris and E. scriptus) prey on small rodent eruptions at irregular intervals (Mendelsohn 1982, Mendelsohn & Jaksic

1989). Mainly inhabiting the Southern Hemisphere, the Elanus kites of the world have converged both morphologically and ecologically with the nomadic owls of the Northern Hemisphere (Negro et al.

2006). They are nomadic and irruptive themselves (Del Hoyo et al. 1994, Scott 1994), being able to disperse over long distances – up to many hundreds of kilometres between their natal areas and first

breeding sites – and may disperse between successive breeding attempts (Mendelsohn 1983).

During the 19th century the presence of Black- shouldered Kites was recorded in several countries of Western and Central Europe, the majority of records being in France, but also in the Netherlands, Belgium, Italy and Germany. These observations may correspond to vagrant birds and most probably this bird of prey was a very rare visitor before the mid 20th century (Cramp & Simmons 1980). The first records of Black-shouldered Kites in the Iberian Peninsula occurred in 1864 in Portugal (Smiths 1868) and in

1865 in Spain (Lilford 1865). The Spanish individual was an adult bird, shot near the Doñana marshes in southern Andalusia. Those few birds sighted in Europe may have originated from Africa and in particular from Morocco, where the species is abundant (Bergier 1987). The first evidence of breeding was reported in 1963 from the Alentejo, Portugal (England 1963), and in 1975 from Spain in Extremadura and the neighbouring province of Salamanca.

The European population today has been estimated at 1000 – 2000 breeding pairs, occurring mainly in the southwestern portion of the Iberian Peninsula (BirdLife International/European Bird Census Council 2000). In the last 30 years, the Black- shouldered Kite has experienced a range expansion in Europe, with scattered breeding attempts in the majority of the Spanish provinces and also in southwestern France (Ferrero & Onrubia 1998,

2003, Duchateau & Delage 2006).

Predicting habitat suitability has multiple applica- tions in the conservation of species of conservation concern because it allows government administrators to distribute available resources more efficiently among habitats. Identifying the habitat features that favour a species’ survival and reproduction is a fundamental step that should be taken before planning any conservation programme (Manel et al. 2001). The dehesas in Spain appear to be closely associated with Black-shouldered Kite occurrence and consist primarily of scattered acorn and/or cork oak trees (Quercus ilex or Q. suber) in pastureland at a density of about 5 – 50 trees/ha. The dehesas are mainly used for raising livestock (primarily pigs, sheep and cattle; Carrete & Donazar 2005, Martín & Alés 2006). Today, dehesas cover a total area of about 2–2.5 million ha in the Iberian Peninsula, of which about

75% is found in southwestern Spain, and the remaining habitat in Portugal (Costa et al. 2005). This traditional land-use system of dehesas started to change at the

end of the 1950s, with the promotion by the Spanish government of a shift from livestock raising to cereal cultivation in huge extents of the territory.This land-use change involved the introduction of machinery and chemical fertilizers, which led to a large-scale land-use change in the dehesa system. Alvarado (1983) estimated that 1.8 million mature acorn oak trees were destroyed only in Badajoz, one of the two provinces of Extremadura, in just a decade (1967 –

1978). Elena et al. (1987) estimated a loss of about

25% of all oak trees in the whole of Extremadura between 1957 and 1982. The proportion of cultivated dehesas in Extremadura is unknown, but in Andalusia, the neighbouring region to the south, it has recently been estimated at 12% of the total surface area is occupied by dehesas (Costa et al. 2005).

Cultivated dehesas are precisely the habitat where most breeding attempts of the Black-shouldered Kites have been recorded in our study area (Rivera et al. 2006). These cultivated dehesas hold a lower tree density than traditional ones due to tree clearing and the fact that tree regeneration is precluded due to periodic tilling (Costa et al. 2005). Although the association between breeding Black-shouldered Kites and the dehesa habitat has long been recognized (Carbajo & Ferrero 1985, Ferrero & Onrubia 1998), we are not aware of any previous published study considering different dehesa types in terms of tree cover and land use. Therefore, the objectives of this research were two-fold: (1) to test the hypothesis that Black-shouldered Kites positively select cultivated dehesas over other habitat types, and (2) to develop mathematical models using variables describing the habitat at the landscape level that would facilitate identification of suitable breeding habitat for this species outside our study area.

We monitored breeding pairs of Black-shouldered Kites in an area of approximately 4900 km2 located around the town of Badajoz (35°50′N, 6°59′W) in southwestern Spain (Fig. 1). The study area is an agricultural mosaic located on primarily flat relief in the Guadiana river basin. This human-created mosaic is composed of arable land with cereals (mainly wheat, oat and barley), other non-irrigated crops such as beets and chick peas, irrigated fields (mainly corn, tomato, cotton and alfalfa), as well as dehesas. Other common habitats are olive-tree groves, vineyards, fruit-trees, Eucalyptus tree plantations and riparian forests.

Range expansion of the Black-shouldered Kite in southern Europe

Figure 1. Study area showing 46 nest-site locations and 45 random points used to quantify nesting habitat and to build an occurrence model for the Black-shouldered Kite in Extremadura, southwestern Spain.

Breeding territories in this area were searched for and monitored from 2003 to 2007. We restricted our analyses to using data only from the 2004 breeding season. This was because the number of breeding territories have remained constant at about 50 during the study period and the majority of nesting pairs and nests were found in the same territories year on year (usually in the nearest tree). Hence to

avoid pseudoreplication we only analysed data from 2004.

Given that nest-sites are not re-used in this species, from January to July we searched the study area looking for signs of territorial pairs (i.e. territorial displays and delivery of nest material). Territories were considered to be active when the female laid eggs and incubated. In this species, some pairs can


Table 1. Explanatory variables in occurrence models used to predict breeding habitat of Black-shouldered Kites in circular sampling areas of c. 6 km2 located in Extremadura, southwestern Spain.
Code Meaning
TREE % of tree patches, mainly Eucalyptus sp.

DEHECEREAL L1 % of dehesas with dispersed density of trees < 7 trees/ha (L1 = Level 1) over non-irrigated crops DEHECEREAL L2 % of dehesas with medium density of trees 7–20 trees/ha (L2 = Level 2) over non-irrigated crops DEHELIVESTOCK % of dehesas used for livestock raising with a tree density of 20 – 40 tree/ha

SCRUB % of scrub

NIRRIGA % of non-irrigated crops IRRIGA % of irrigated crops VINEYARD % of vineyard

OLIVE % of olive groves

PASTU % of pasture

DRIVER distance (m) to nearest river DROAD distance (m) to nearest road DVILLA distance (m) to nearest village

lay two clutches in different nests within the same area during a single breeding season (Ferrero et al.

2003). If we suspected the latter had occurred, only the first active nest was considered for analysis. We measured 13 variables inside circular sampling plots to describe landscape features in 46 breeding territories and in 45 areas generated at random (Table 1). Analysis of the landscape features was based on circular plots centred on the focal spot (nest-site or random point) of each territory or of each randomly generated area. Plots had a 1425-m radius (half the nearest neighbour distance (NND) in the study area) in order to get a circular area of 637.9 ha (approxi- mately 6 km2).

The spatial scale selected to study habitat prefer- ence may influence our results (Sánchez-Zapata & Calvo 1999, Martínez et al. 2003, Román-Muñoz et al. 2005). The scale we selected encompassed an area that was used by individuals not only for breeding but also for foraging, according to visual monitoring of hunting birds coupled to our radio- tracking data on 11 breeding individuals (authors’ unpubl. data). The Black-shouldered Kite hunts its staple prey (rodents) in open habitat, field margins and irrigated fields after the harvest of cereals (authors’ unpubl. data). Therefore, we feel that our chosen spatial scale does well to represent the home range of the Black-shouldered Kite, which usually results in better performance of the predictive models (Ferrer & Harte 1997, Suárez et al. 2000, Fernández et al. 2003, Martínez et al. 2003, Balbontín

2005). Random points were chosen inside the habitat available for adult Black-shouldered Kites. Available habitats were considered to be all habitats inside our study area having minimum requirements

for nesting. Therefore, generated random points located in areas lacking at least a suitable tree for nesting, on water courses or reservoirs, points closer than 19 m to the nearest road or less than 812 m away from the nearest village (i.e. the minimum distance recorded in our sample from a nest-site to a road or a village, respectively) were rejected. Because the study area was intensively searched, we are confident that we found the majority of nesting sites and hence those random points lacking a nest- site were considered as absences in the occurrence (presence/absence) models.

Landscape characteristics were analysed by means of a Geographic Information System. We used colour digital orthophotos taken from an airplane during summer 2002 with 0.5-m resolution (1 : 5000). The orthophotos were obtained from the Extrema- dura Regional Government (Junta de Extremadura). Percentages of habitat variables inside circular plots were measured after digitizing them using the Habitat Digitizer Extension for Arcview 3.2 (available at: Other recorded variables included the distance from the focal point (nest-site or random point) to the nearest source of human disturbance (road or village) or the nearest stream/river.

Statistical analyses
Mean values for the different variables collected at nest-sites and random plots were compared using Wilcoxon rank sum normal statistics with correction tests for the differences between means. All tests were two-tailed and statistical significance was set at P < 0.05; means are given ± sd.

We built occurrence (presence–absence) models using logistic regression via a Generalized Linear Model (using the GLM procedure of S-Plus 2000, Mathsoft 1999) to identify the set of variables that best separated breeding territories from random areas. We first built an environmental model using the explanatory variables that described habitat at the landscape level (Table 1). We used a binomial error distribution and a logit-link function. Investi- gation of dispersion plots suggested incorporation of some independent variables as main and quadratic effects in a maximal model to test if the inclusion of second-order effects would improve the final model. The explanatory variables included as second-order polynomials were: irrigated crops, different types of dehesas, scrub, pasture and trees. The statistical significance of each variable was tested in turn in the model using a backward stepwise procedure and models were fitted using a maximum-likelihood method (McCullagh & Nelder 1989). Because neighbouring circular plots could have similar environmental conditions, residuals from a fitted model might exhibit spatial autocorrelation (Bustamante & Seoane 2004). To control for possible pseudoreplication due to geographical proximity of sampled plots we added a term called autocov that averages the number of occupied squares among a set of ki neighbours of squares i (Augustin et al.

1996). For this purpose, we built a grid covering our

study area of 70 × 70 km formed by squares of similar size as the breeding territories (grid size

2850 m (2 × 1425 m)). Each nest-site was assigned to the centre of the square i where it belongs. Those squares lacking a nest-site were considered as absence squares and those with a nest as presence squares. We also checked for autocorrelation by adding the geographical coordinates (longitude and latitude) as two new independent variables into our models.

We used two different methods to compare the performance of the model. First, we used a cross- validation leave-one-out re-sampling procedure to compare the performance of our model. We checked the stability of our final minimal adequate model by comparing the percentage of agreement between presence/absence predictions of the original model with those predictions obtained when using the cross-validation re-sampling procedure. This measures the average effect of a single observation on the model predictions. The presence or absence of predicted values was accepted at a threshold probability at which the sum of sensitivity and specificity was maximized (Albert & Harris 1987,

Zweig & Campbell 1993). For this purpose, we defined an objective cut-point above which to consider the species as present, classifying prediction values as present at all cut-points between 0.0 and

1.0 with an interval of 0.1 (Suárez-Seoane et al.

2002). The percentage of cases (random and nest-sites) classified correctly (CC) at selected cut-points was calculated and an optimum set at that cut-point where presence and absence sampling plots are equally predicted by the GLM model. Finally, we constructed a confusion matrix (Fielding & Bell

1997) and Cohen’s Kappa was calculated (Cohen

1960). This statistic objectively computed the chance-corrected percentage of agreement between observed and predicted group memberships. Values of 0.0 – 0.4 indicate slight-to-fair, values of 0.4 – 0.6 moderate, 0.6 – 0.8 substantial and 0.8 –1.0 almost perfect model performance (after Landis & Koch

1977). Secondly, we used a data-splitting strategy, developing the models with a random selection of

75% (n = 69) of the sample (the training set) and then using the remaining 25% of the data (n = 22) to evaluate the models (the test set). The cross-validated model using a data splitting strategy (i.e. prediction success) was assessed using the area under the receiver- operating characteristic (ROC) (Beck & Shultz

1986, Fielding & Bell 1997, Osborne et al. 2001).

Black-shouldered Kites tended to breed in cultivated dehesas with a low tree density (< 7 tree/ha). The average percentage of this type of habitat at breeding sites within circular plots was 22.5% (143.5 ha), differing significantly from 1.53% (9.75 ha) found in circular plots generated at random (P < 0.001, Table 2). Breeding territories also had a greater area of cultivated dehesas with a medium density of trees (7 – 20 tree/ha) than random circular sampling areas. Breeding areas were also characterized by having a significantly greater area of non-irrigated crops (P = 0.009) and a lower area of pastureland (P < 0.001) than random areas. Variables related to human disturbance showed that breeding territories were located significantly further away from villages than random areas (P = 0.02). Also, we detected a trend (although not statistically significant) for breeding territories being located further away from the nearest road than random areas, indicating a tendency to avoid areas used by humans.

The occurrence GLM model was highly signifi- cant (P < 0.0001), explaining a total of 60.4% of the

Table 2. Difference between breeding sites and random sites in mean surface occupied by 11 habitat types measured within circular sampling areas of habitat deemed available to Black-shouldered Kites. Mean differences in the distance to the nearest road, village and water courses (e.g. rivers, reservoirs and streams) are also reported.


Nesting sites (mean ± sd)

Random sites (mean ± sd)



TREE (%)

0.14 ± 0.53

0.90 ± 2.36




22.5 ± 19.5

1.53 ± 3.24


< 0.0001


2.63 ± 6.49

0.46 ± 1.54




11.0 ± 16.2

22.7 ± 32.3




0.61 ± 1.93

1.03 ± 2.89




20.6 ± 15.2

15.7 ± 20.9




27.7 ± 21.8

28.3 ± 34.8




6.02 ± 8.60

8.36 ± 13.5




3.53 ± 4.84

8.60 ± 13.8




2.21 ± 4.74

6.83 ± 11.3




1483.0 ± 1185.7

1568.1 ± 1585.0




1650.6 ± 1317.0

1317.1 ± 1396.5




5772.3 ± 2775.3

4530.7 ± 2765.9



  1   2   3

The database is protected by copyright © 2016
send message

    Main page