The study area is part of the Judean Foothills, approximately 30 km southwest of Jerusalem. The Judean Foothills are among the last remnants of a unique transient ecosystem at the interface of the humid Mediterranean ecosystem to its north and the arid ecosystem to its south (Weizel et al., 1978). The landscape is characterized by a mosaic of different land-use types; mainly natural habitats (scrublands of variable densities and stages of succession mainly lacking trees higher than 5 m, with shrubs and herbs as main plant species), agricultural fields (annual and perennial crops), semi-natural habitats (planted forests comprising mainly pines of the species Pinus halepensis Miller, Pinus pinea L., Pinus brutia Tenore and to a lesser extent pines mixed with planted native broad-leave species), some rural settlements and a few urban and industrial areas (Weizel et al., 1978) (Figure 2).

The sunflower Helianthus annuus L. (Asteraceae) is an annual herb and in Israel grown for seed production but not for oil. Sowing takes place in March and harvest between July and September. Prior to our study, we interviewed local farmers to get information about pest species of the study area. The Rose-Ringed Parakeet was reported as being a severe pest species consuming seeds directly from sunflower heads (see also Nemtzov, 2003; Schäckermann and Weiss, personal observations, August 2010) (Figure 1).

We selected similarly managed sunflower fields, concerning time of sowing, irrigation and pesticide application, of three hybrid seed production varieties (Shelly, Shemesh, Dalet Yod 3). All fields were treated with the fungicide Bayfidan and with the insecticide Endosulfan once during bloom and irrigated with a drip system from April/May until June/July, depending on the time of sowing and harvesting. The sunflower fields were alternated with cotton, chickpea, watermelon, maize or spring wheat in the next season or with winter wheat in a double-cropping system in the same season. The size of fields varied from 1 to 66 ha.

One field with high visible seed predation was chosen in 2010 for field-scale data collection using a grid system. In this field at least half of the sunflower heads at the margins showed signs of seed predation and this was reported by sunflower farmers of the study area as highly predated by birds. Within the grid-collection system, vertical and horizontal rows were selected every 50 m for the entire field. At each of a total of 140 intercept points (when vertical and horizontal rows were crossing), we estimated the percentage of seed predation per each of 50 sunflower heads (summing up to a total of 7000 sunflower heads). The estimation of the percentage of seed predation is described in Estimating Seed Predation.

In 2012, we selected 20 sunflower fields for a landscape-scale data collection to re-assess our findings from 2010 on a larger scale and to integrate landscape variables. We chose 10 fields with visible seed predation at the margins (more than 50% of the heads showed 10–100% seed predation) and 10 fields with little or no visible seed predation at the field margins (more than 50% of the heads showed 0–5% seed predation). Margins were chosen for this assessment because in 2010 we found that they suffer most from seed predation. For each field we selected two sampling points: (1) at field margin and (2) the interior of the field. Minimum distance between margin sampling points of different fields was 1 km. In each field, we used the margin with highest visible seed predation for sampling. In fields without or little seed predation we chose the margin sampling point randomly.

We measured the distances between the intercept points to the closest high tree (tree height > 5 m, single tree or a group of trees with at least one tree taller than 5 m) for all 140 intercept points of the field for the field-scale study in 2010 and for all margin and interior sampling points of all fields in 2012. We recorded the geographic coordinates at all margin sampling points of the fields in 2012 using GIS (ArcGIS, version 9.2., Esri 380 New York Street, Redlands, CA 92373-8100) and calculated in a 1 km radius surrounding the sampling point the (1) percentage of the natural habitat comprising scrubland without soil disturbance in the past 5 years (0–33%), (2) percentage of tree cover comprising high trees (0–20%), and (3) field size in ha. A 1 km radius is known to be an appropriate scale in vertebrate-focused landscape studies (Eilers and Klein, 2009; Guerrero et al., 2012). Land cover data was obtained from the archive of the Hebrew University.

We estimated sunflower seed predation of 50 randomly chosen sunflower heads at each of all 140 intercept points of the field in 2010; we therefore sampled 7000 sunflower heads in this one field. We furthermore sampled 50 randomly chosen sunflower heads each at the margin and interior sampling points in the 20 fields in 2012 (100 heads in each field and 2000 in total for 2012). We estimated seed predation per flower head by estimating the percentage of missing seeds caused by birds (seed predation estimated number of missing seeds eaten by birds/estimated number of all developed seeds in the head^*100). Seeds were absent from sunflower heads through (1) seed predation by birds (Figure 1), (2) touching and rubbing of neighboring sunflower heads (mechanical removal), and (3) no development of seeds. Seeds removed by birds were distinguished from non-developed seeds because seeds removed after full development left “pockets” (palea adhered to the anthodium) (Figure 3) while these “pockets” were missing when unfertilized florets did not develop seeds. Mechanical removal was identified when complete seeds were found under the plant and this was not the case when seeds were eaten by birds.

We analyzed the field-scale count data from 2010 with generalized linear mixed models using a quasipoisson error distribution to account for over dispersion and penalized quasi likelihood (GLMM; packages = “nlme,” “nlme4,” “multcomp,” “vegan”), (Script S2) (Pinheiro et al., 2007; Bates et al., 2013). We used generalized linear mixed models because of our non-normal distributed data (we did a visual test to find out if our data was normally distributed) and to fit in fixed factors as well as random factors. This approach is recommended in the literature (Bolker et al., 2009). Seed predation was used as response variable and distance from the sampling point to the closest high tree in meters as explanatory variable. For the landscape scale data comprising 20 fields in 2012, we also used generalized linear mixed models (Script S2). Models were analyzed using a poisson error distribution (to model the count data) or quasipoisson error distribution in case of over dispersion. Models with seed predation as response variable included the following explanatory variables as fixed factors: (1) distance from the sampling point to the closest high tree (in meters), (2) sampling point location (margin, interior), (3) percentage of tree cover in a 1 km radius surrounding the sampling point, (4) percentage of natural habitat in a 1 km radius surrounding the sampling point, (5) field size in ha. We used the additional explanatory variable (6) presence of high trees within 50 m for the landscape scale (tree or trees present (yes), no trees (no), because we detected similar patterns and a threshold of around 50 m at both the field-scale and landscape-scale analyses (Figure 4), when testing the effect of distance to the closest high tree on seed predation. Field and location within the fields were included as random factors if they were not included as explanatory variable (e.g., Nelder and Wedderburn, 1972). Correlations between explanatory variables were tested with a correlation matrix based on Spearman (Table 1). Because some of the explanatory variables were correlated, all variables were tested in separate models. In the following we present the standard results given for mixed model summaries (Zuur et al., 2009). Amongst others we give p-values (P) of our models. The p-value can be defined as the probability (therefore has a value between zero and one) how likely it is to obtain such a sample result or a more extreme, if the null hypothesis is true. The threshold used was 5% (0.05). Furthermore, we give the t-value (T) which measures how many standard errors the coefficient is away from zero. Generally, any t-value greater than +2 or less than −2 is acceptable. This is also the threshold we used for all our analyses. The higher the t-value, the greater the confidence we have in the coefficient as a predictor. Low t-values are indications of low reliability of the predictive power of that coefficient. All statistical analyses were conducted using R (R Development Core Team, 2010, version 2.12.0).

Seed predation increased strongly with decreasing distance to the closest high tree or tree group within 50 m in our study. Because birds were the main sunflower seed predators in our study area, our results agree with the findings of Hanspach et al. (2011) who found that scattered trees were key habitat structures for birds in semi-natural open areas. We observed flocks of a few hundreds of birds of the Rose-Ringed Parakeet in our study area, (Personal Observation and reports by local farmers) (Figure 1), one of the main bird pests to agriculture in Israel (Nemtzov, 2003). However, in other agricultural areas in Israel, flocks with up to a few thousands of birds were reported by local farmers (Personal Communication, Yoav Motro, Ministry of Agriculture and Environment).

In areas with higher bird abundance, the pressure on individual birds to locate food should be stronger, and they may therefore exceed the main foraging distance of 50 m we found in our study. When comparing seed predation at the margin sampling points to the interior sampling points of the field- and at the landscape scale, we found that the margins were more affected, showing that birds live outside the sunflower fields and gradually forage toward the interior, starting at the margins. Many species tend to feed on field margins and may only move to the field interiors when food resources become limited (Kollmann and Buschor, 2002). This was also supported by our results for seed predation at the field scale throughout an entire field. Here, we found the highest seed predation at the margin intercept points of the field adjacent to a high tree or small group of trees. Hence, a field that seems highly predated at the margin does not necessarily show predation in the interior. High trees or tree groups are usually located outside the fields and therefore sunflower plants in the interior are generally further away than plants at the margin.

The dependence of seed-predating birds on high trees may also explain the high spatial variance in seed predation between different fields. If there were no high trees adjacent to the fields, we only observed very little seed predation, but there was a dramatic increase in crop seed predation by birds in crops adjacent to high trees. While birds seemed to depend on the presence of high trees to scope parts of the landscape as potential feeding areas, a higher percentage of tree cover did not cause higher seed predation in crop fields. Therefore, the high spatial variance in seed predation seems to be related to the appearance of high trees used for perching by birds. This supports findings of Fischer et al. (2010) where the number of trees was rather negligible and only the general presence of a single tree drove the overall pattern of bird richness in an Australian livestock grazing landscape. Therefore, individual trees should be taken into account in land-management planning, since at least in Israel it is not an easy option to cut trees in agricultural areas, because permissions for doing so are needed for every tree. We furthermore observed Hooded Crows picking single sunflower seeds and consume them outside of the crop fields, while the Rose-Ringed Parakeet consumed many seeds without moving at top of the sunflower heads in the fields (Schäckermann, personal observations and see also Figure 1). Even though the crow and the parakeet are both granivorous birds and crows have a higher body mass, their actual contribution to seed predation may vary considerably mediated by their predation activity. Therefore, different granivorous bird taxa should be considered (Notman et al., 1996). Hence, feeding behavior as well as local abundances of main bird species found in agricultural areas, that can influence the high spatial variance in seed predation between different fields, should be taken into account when predicting the impact of birds on seed predation. Even though high trees were the main perch (high structures for birds to sit on and perch) in our study, we also observed high seed predation by birds in one field with no trees in its surroundings (outlier in Figure 5). However, we did find electric pylons and power lines next to the field and small buildings in the area. Therefore, the availability of perches may influence seed predation in sunflowers. Future research should therefore investigate the impact of these man-made landscape structures (alone and in combination with high trees) and different habits of different seed predators. This may lead to more detailed management recommendations regarding the interaction effects of different natural and man-made landscape structures on biodiversity and ecosystem services or dis-services relationships.

Natural habitats can cause services (benefits) or dis-services (negative effects) or a combination of both to agriculture (Zhang et al., 2007; Bommarco et al., 2013; Martin et al., 2013; Ango et al., 2014). In our study, natural habitats were found to be unsuitable for sunflower seed-predating birds because they did not contain trees that could be used as perching structures. Also other studies found that natural and semi-natural field margins are not suitable as breeding habitats for birds feeding on crops, but offered shelter to a broad range of bird species (for example Song Sparrow Melospiza melodia Wilson, Vesper Sparrow Pooecetes gramineus Gmelin, Savannah Sparrow Passerculus sandwichensis Gmelin, Red-winged Blackbird Agelaius phoeniceus L. and American Goldfinch Carduelis tristis L.) potentially useful for biological pest control (Jobin et al., 2001). High natural trees rarely occur in our study area in Israel, but introduced high-grown Eucalyptus trees can be frequently found. In our study area, high Eucalyptus trees are often located close to settlements, probably grown to spend shade in summer (Schäckermann, personal observations). Because it is known that many bird species prefer high landscape structures for perching (Bull et al., 1992; Sonerud, 1992; Holl, 1998; Miller and Cale, 2000), the presence of trees taller than 5 m improves the habitat quality, especially for birds originated in forest areas, like the Rose-Ringed Parakeet, one of the main seed predators on sunflower in Israel. Furthermore, Rose-Ringed Parakeets depend on cavities as nesting sites (Strubbe and Matthysen, 2007). In our study area, Rose-Ringed Parakeets used holes in buildings as nesting places (Weiss, personal observation) and crops they used as food resources are often planted close to settlements. Knowing that the Rose-Ringed Parakeet, which is an invasive bird in Israel and an agricultural pest, uses an introduced tree species for perching indicates, that humans can unintentionally create suitable habitats for pest bird species. Ornithologists and nature conservationists in Israel are concerned that the high numbers of Parakeets negatively influence the population density of native cavity-nesting birds, by occupying their nesting sites and reducing their breeding success (Weiss, communication with local conservationists and ornithologists). Birds of concern are the Hoopoe (Upupa epops L.), Syrian Woodpecker (Dendrocopos syriacus Hemprich and Ehrenberg) and European Scops Owl (Otus scops L.). Research about the effects of the introduced Rose-Ringed Parakeet on native cavity-nesting bird species and other invasive birds showed, that there are complex interactions between invasive birds, these interactions had an effect on native cavity-nesting birds (Orchan et al., 2013) showing, that a complex interaction network between native and invasive bird species exists. Globally invasive species have enormous ecological and economic costs and are an important threat to biodiversity (Wilson, 1992; Pimentel et al., 2005; Shine et al., 2009). There are many hypotheses on why birds established and/or are becoming invasive (e.g., Case, 1996; Blackburn and Duncan, 2001; Duncan et al., 2003; Cassey et al., 2004; Blackburn et al., 2009). Nevertheless, more knowledge is needed about the impact of invasive bird species on the environment and economy in the invaded range.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.