Nest box selection and reproduction of European Rollers in Central Italy: a 7-year study

Study area and census methods

About 70 breeding pairs (mostly breeding on man-made artefacts such as farmhouse ruins, barns or in low voltage electricity poles) were estimated to be present in the whole Lazio region (Central Italy) in 2009 and in 2015 (Brunelli et al. 2011; Pezzo and Puglisi 2015). Since 2010 the Italian national electric power company Terna S.p.A. has installed seventy nest boxes specifically designed for rollers on metal pylons of high voltage power lines, across the provinces of Rome and Viterbo in the Lazio region, in order to improve nest site-availability for the species (Fig. 1). The mean distance between each pylon and the nearest one is 445 ± 171.5 m (SD). Birds breeding in nest boxes in these areas were our study population. The area contains both intensive arable farmland and organic cereal crops, fragmented by grasslands with bushes and trees: urban artefacts are present but in low density. Other habitats include fallow fields mainly used for livestock rearing and olive groves.

The nest boxes have been monitored every year, for seven years between 2010 and 2016. In order to assess the nest boxes occupancy, each nest box was periodically checked during the breeding season (one visit per week between the end of April and the end of July). A nest box was considered occupied if at least one egg was laid by a roller (e.g. Finch 2016); unoccupied if remained empty for the whole season or occupied by another species [e.g. Common Starling (Sturnus vulgaris), House Sparrow (Passer domesticus)].

For each nest box, we recorded: (a) clutch size (i.e. number of eggs laid), (b) number of hatched eggs and (c) number of fledged chicks. From these parameters, we calculated the hatching success (young hatched/eggs laid), the fledging success (young fledged/eggs hatched), and the annual breeding success (young fledged/eggs laid). In some cases, nest boxes could not be monitored, resulting in reproductive data being only partially available for that nest boxes (e.g. years 2013 and 2014). These latters were included in the analyses only for the available data.

Environmental data

Statistical analysis and model selection

To investigate the effect of habitat on nest box occupancy, clutch size, hatching, fledging, and breeding success, we performed five series of generalized linear mixed effects models (GLMMs). As response variable, we used nest box occupancy (1 if occupied by rollers, 0 if not), hatching success (calculated as proportion of hatched eggs over the number of laid eggs), fledging success (proportion of fledged young on hatched eggs) and breeding success (proportion of fledged young on laid eggs), under a binomial distribution, whereas for clutch size we used the number of laid eggs under a Poisson distribution. Calendar year was included as a random factor, to account for possible annual differences in meteorological conditions that could affect the parameters of interest. As each nest was sampled for consecutive years, we also included the nest ID as a further random term in each model. The availability of power lines was not used as additional predictor because all nest boxes were installed on power lines, limiting expected variation between sites for this parameter. To select the habitat variables for inclusion in each model we used an information theoretic approach based on the corrected value of Akaike’s information criterion (AIC_c) (Burnham and Anderson 2002) with multi-model inference. To exclude correlation among variables in the same models, we identified all the subsets of uncorrelated (p > 0.05) predictor variables by calculating a matrix of Pearson correlation coefficients and performed the multi-model inference starting from those pre-selected subsets. The model with the lowest AIC_c was elected as the best, with the others ranked according to their AIC_c difference from this (ΔAIC_c). Furthermore, we measured the relative importance of models by their Akaike weights (w_i; Anderson et al. 2000, 2001). Finally, we used model averaging of models with ΔAIC_c < 7 (Burnham et al. 2011) from the top model to obtain the model averaged coefficients of predictor variables and to rank them according to their predictive importance (Σw, sum of Akaike’s weights). Preliminary manipulation of environmental layers was done using the software QGIS (QGIS Development Team 2018); the model fitting and the multi-model inference was carried out in the statistical environment R (R Development Core Team 2018) via the packages lme4 (Bates et al. 2015) and MuMIn (Barton 2016). Other results are reported as mean ± standard deviations.

Effects of habitat variables on nest boxes occupancy

The most important variables affecting nest boxes occupancy were the distance from arable fields and the edge density. Both were present in the top ranked models (totaling a predictive importance Σw = 0.76, with a significant negative coefficient). This indicates that nest boxes close to arable fields and in larger patches (low edge density) had a higher probability to be occupied. In particular, nest boxes at 0 m distance from arable fields had an average predicted probability of being occupied of 65.8%, whereas the furthest nest, which was 1800 m from arable fields, had an 11.0% probability. The effect of arable fields was also shown by the significant and positive coefficient of the percentage of arable fields in the 600 m plot surrounding the nest, although its predictive importance was lower (Σw = 0.24). Solar radiation in the area surrounding the nest box was positively associated with nest occupancy, although with a moderately-low predictive importance (Σw = 0.24). Proximity to orchards positively affected nest occupancy too, with the same predictive importance but the 95% confidence interval of the estimated coefficient overlapped the 0 value, indicating a non-significant effect (Fig. 3; Additional file 1: Table S1 and S2).

Effects of habitat variables on reproductive parameters

Clutch size was not particularly affected by the considered habitat variables. Among all the considered models describing the number of laid eggs, a null model (i.e. including intercept only) was ranked as the best one, in terms of AIC (Additional file 1: Table S3). The top ranked habitat variables were distance from fallow fields and orchards, both with a negative coefficient, suggesting that pairs breeding in proximity to these habitats were likely to lay a higher number of eggs. However, these were not statistically significant and the low sum of Akaike’s weights (respectively 0.20 and 0.16) indicated a low importance of such predictors (Fig. 3; Additional file 1: Table S4).

For hatching success, the total amount of incoming solar radiation in the 600 m circular plot surrounding the nest was the most important variable (Σw = 1.00), being present in all the best models with ΔAIC < 7, and it showed a significant and positive effect on the expected proportion of hatched eggs with an expected increase from 75.7% in nest boxes with the lowest incoming solar radiation to 97.7% in nest boxes with the highest. Percentage of orchards in the 600 m buffer was the second ranked variable for predictive importance (Σw = 0.58), with a negative effect. The negative and significant coefficient of distance from grassland indicates that nest boxes close to grassland had a higher expected proportion of hatched eggs (Fig. 3; Additional file 1: Table S5 and S6).

For fledging success, percentage of fallow fields in the 600 m circular plot around the nest box was the only habitat variable present in almost all the best models, as shown by its sum of Akaike’s weight (Σw = 0.98). The positive and significant coefficient of this variable indicates that pairs breeding in an area surrounded by fallow fields had a higher proportion of fledged young. Altitude was the second ranked variable (Σw = 0.64) and it had a positive effect. Habitat structure seems to play a moderate role, and the positive effect of mean patch size (Σw = 0.52) suggests that habitats with moderate/low complexity are associated with a higher fledging success (Fig. 3; Additional file 1: Table S7 and S8).

Because most losses occurred at the egg stage, results for overall breeding success were similar to those for hatching success. Breeding success, was strongly associated with incoming solar radiation, which showed a high importance Σw = 1 and a significant and positive effect. Percentage of orchards in the 600 m buffer had a high importance (Σw = 0.96) and its coefficient indicates that nest boxes surrounded by orchards have a lower expected breeding success. Other variables with a moderate predictive importance were distance from arable fields (Σw = 0.49), percentage of urban areas (Σw = 0.37) and distance from grassland (Σw = 0.32), all with a positive effect although they failed to reach statistical significance (Fig. 3; Additional file 1: Table S9 and S10).

Effects on nest box occupancy

Habitat variables can play an important role in influencing the choice of nest sites, and thus in shaping the species distribution at a local scale. Factors such as nest location, habitat patch composition and presence of conspecifics and/or predators are known to highly affect nest site selection in European Roller (Parejo and Avilés 2011; Vaclav et al. 2011; Parejo et al. 2013; Bouvier et al. 2014).

In our case, we found that nest boxes in farmland habitats consisting predominantly of arable fields and located in moderately sun-exposed areas had a higher probability of being occupied by rollers. A similar study conducted in southern Hungary showed how such man-made habitats seem to be offering a high-quality breeding habitat for the species (Kiss et al. 2014): in that study, rollers showed higher breeding performance in agricultural areas (where more than 50% of the area was made of arable fields) than in their traditional habitat of natural “steppe” grasslands, because of the higher availability of prey species (different species of insects and small vertebrates).

Effects on reproductive parameters

Landscape composition affected nest box occupancy and also had important effects on the reproductive performance of the breeding pairs. However, these effects were not the same for all the reproductive parameters. More specifically, we found that none of the habitat variables were of importance in predicting the expected numbers of eggs laid per pair. Once a nest box was occupied, the habitat variables had a minor impact on the clutch size. We could hypothesize that, all nest boxes being identical in terms of internal space and size, the expected variation between sites for this parameter has been limited. In future it would be useful to test for differences in clutch sizes between artificial nest boxes of different sizes and natural nest sites which may differ in terms of structure and internal space for laying and incubating eggs (Evans et al. 2002; Lambrechts et al. 2010).

The hatching success was higher in areas where a higher incoming solar radiation was recorded. This is the first study to show such an effect in this species. Relatively high temperatures during the incubation phase may favour chick embryo development and subsequent egg hatching (Parejo et al. 2015). Rollers usually start incubating after that the third egg has been laid (hatching is hence asynchronous), so that parents may leave the eggs unattended for short periods (Cramp 1985; Parejo et al. 2012). In this sense, high temperatures (typical of the climate which characterizes our study area) can maintain eggs in the right conditions during periods when parents are away from the nest, thereby assuring the embryonic survival/growth. In addition, rollers can adjust hatching asynchrony in response to spring climatic conditions and resources availability (Parejo et al. 2015): high spring temperature determines food availability because it favours insect development and outbreaks. It has also been documented that roller nestlings can cope with the constraints imposed by occasional heat waves, being more tolerant/resilient to high environmental temperatures than other sympatric species such as the Lesser Kestrels (Falco naumanni) (Catry et al. 2015). All this means that high temperatures are also important during the first phases of parental cares and food provisioning after the hatching, when chicks are not yet able to properly thermoregulate.

The fledging success was positively associated with fallow fields, indicating that nest boxes closer to these patches had a higher proportion of fledged birds. At the same time, nest boxes located in areas with lower landscape complexity also showed a higher fledging success. This is counterintuitive, as in other studies (e.g. Avilés and Parejo 2004; Kiss et al. 2016), habitat heterogeneity was a positive factor for fledging success. However, this unexpected result may be explained by the fact that prey species richness and abundance in fallow and arable fields was already very high in our area, and likely sufficient to sustain chick rearing. Furthermore, areas with higher habitat structure included larger areas of woodland and pastures, which possibly had lower prey abundance than freshly threshed arable fields and mowed fallow fields. In our study area, pastures are usually much overgrazed, and therefore show very poor soil structure and consequently low amounts of large invertebrates (van Klink et al. 2015), while woodland is not a primary foraging habitat for rollers (Cramp 1985). Finally, the breeding success was higher when nest boxes were sited in areas with higher incoming solar radiation and when relatively distant from orchards and urban areas. This is in accordance with other studies who found that nest boxes located near urban areas or where human disturbance is high, are not selected for nesting and/or yield low breeding success (Tiefenbach 2009; Rodriguez et al. 2011).

Overall, these results show that both the composition and the configuration of the landscape are fundamental to the occurrence of breeding rollers as well as the subsequent reproductive performance, and thus the long term sustainability and resilience of local populations. We also found that specific factors differently influence the two mechanisms, such as the nest site selection and reproductive success, during the breeding season. In particular, a good solar irradiation, arable fields and a moderate/low habitat matrix complexity appear to facilitate rollers’ occurrence and provide suitable foraging resources for chick rearing, and thus to ensure the success of reproduction.