The importance of riparian forests and tree plantations for the occurrence of the European Turtle Dove Streptopelia turtur in an intensively cultivated agroecosystem

Summary Farmland birds represent a large proportion of European avifauna, and the populations of several species have suffered a dramatic decline in recent decades. Among these species, the European Turtle Dove Streptopelia turtur has undergone rapid decline in much of its European range. Therefore, the main aims of this research are to estimate the population density of the Turtle Dove and to investigate its habitat use at home range scale in an intensively cultivated agroecosystem in northern Italy. In the 2015 breeding season we carried out turtle dove counts from 372 point-counts, randomly allocated following a stratified cluster sampling design. The density was estimated by distance sampling, whereas the habitat suitability was assessed by Resource Selection Probability Function. In particular, we followed a presence vs availability approach, using binary logistic regression and the Information-Theoretic approach. During fieldwork, 76 observations of Turtle Dove were collected and a density of 5.0 pairs/km2 was estimated. The Turtle Dove inhabits areas with high tree cover, either semi-natural forests or tree plantations, as well as areas with many shrubs and hedgerows. On the other hand, areas with a high proportion of crops, such as paddyfields, maize, and winter cereals are avoided. For the species’ conservation, it is necessary to maintain a combination of habitat features with suitable nesting and feeding areas, as the degradation of either of these may reduce Turtle Dove populations.


Introduction
Farmland bird species represent a large proportion of European avifauna, and the populations of several species have suffered a dramatic decline in recent decades, especially in Western Europe (Donald et al. 2001(Donald et al. , 2002. The causes of this decline have been identified mostly in the changes of agricultural practices, such as heavy mechanization, increased fertilizer inputs, and a temporal shift in cereal sowing from spring to autumn. In addition, the intensification of agricultural practices and Bird Conservation International (2021) 31:605-619. © The Author(s), 2020. Published by Cambridge University Press on behalf of Bird Conservation International. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited. doi:10.1017/S0959270920000532 European species of global conservation concern (BirdLife International 2017), with an unfavourable conservation status (Gustin et al. 2016). However, to our knowledge there is no other research on this species in Italy, and the few data available are derived from the grey literature or studies not focused on this species (Brichetti and Fracasso 2006). Furthermore, the Turtle Dove being a farmland species, our findings might help also to better understand the ecology of farmland birds in Italy (Brambilla 2019) and, more generally, in Europe, inasmuch they are decreasing sharply as previously stated.

Study area
The present study was carried out in the western Po Plain (north-western Italy), in an area of about 2,900 km 2 (45˚11'N, 9˚05'E) (Figure 1). The Ticino River crosses the study area from north to south, flowing into the Po River that runs from west to east. The Sesia River and the Lambro River flow along the western and the eastern boundaries of the study area, respectively. The landscape is characterized by cultivated areas, especially paddy fields (39.4%), other annual crops (mainly maize, soybean, oil-seed rape, winter wheat, and alfalfa) (29.1%), and tree plantations (6.8%). Broad-leaved forests and built-up areas represent 4.9% and 10.3% of the total area, respectively. The data used to measure the land use cover were obtained from the regional land use map "DUSAF 5.0" (ERSAF 2015) and from the regional forest map "Carta dei tipi forestali reali della Lombardia" (ERSAF 2012), and processed by the software Quantum GIS v.3.2.3 'Bonn'. Continuous forests (composed mainly of Quercus robur, Carpinus betulus, willow Salix sp., poplar Poplar sp., and the invasive Robinia pseudoacacia) are located along the Ticino River and in the southern part of the study area, near the Apennine slopes, whereas remnants of broad-leaved forest fragments are scattered in the cultivated area (95% of which are smaller than 10 ha). Tree plantations mainly consist of monospecific plantations of hybrid poplar species, but also of reforestations and short rotation coppices (SRCs). Due to its particular geographical location, this area is difficult to relate to other European lowlands. Even though the Po Plain has a subcontinental climate, it is separated from the rest of continental Europe by the Alps, which could act as a zoogeographical barrier (Bianco 1990, Hermansen et al. 2011. Furthermore, even if it is attributable to the Mediterranean Basin, the climate, land use, and vegetation in the Po Plain differ from the rest of the Mediterranean area (Mikusiń ski andAngelstam 1997, Capotorti et al. 2012).

Survey design and data collection
In order to obtain a representative sample in this large and heterogeneous study area, a stratified cluster sampling design was performed (Krebs 1999, Sutherland et al. 2004, Sutherland 2006, Barabesi and Fattorini 2013. According to this approach, the area was firstly divided in Landscape Units (LUs), i.e. homogeneous portions of the area in terms of habitat or ecological characteristics likely to induce systematic variations in population density (Sutherland 2006). To this aim, a 2 Â 2 km grid of sampling cells was superimposed on the study area and each cell was assigned to a LU, based on its characteristics. More details on LU definition are reported in Appendix S1 in the online Supplementary Materials. Data were collected in 62 randomly selected sampling cells (approximatively 10% of the study area). The number of sampling cells selected for each LU was proportional to the LU extent, in such a way that all the landscape characteristics of the study area were represented and each stratum was investigated with an effort proportional to its extent (Krebs 1999). Within each selected sampling cell six point-counts were carried out (Bibby et al. 2000, Sutherland 2006) during the breeding season 2015, between May and early June, the peak of the singing and breeding activity (Cramp 1985, Brichetti andFracasso 2006). Data on species occurrence were therefore collected in 372 point-counts. The selection of the six point counts inside the 2-km cell was carried out by multi-level sampling (Sutherland 2006), subdividing the 2-km cells into cells of 250 Â 250 m and randomly selecting six of them ( Figure 1). Using this second grid, it was possible to obtain estimates that are more precise. Each point, placed randomly inside the cells, was surveyed once from dawn to 10:30, and the count lasted for 10 minutes (Chamberlain and Rolando 2014); overall, data collection took place for 32 effective days. During the fieldwork, we measured the exact distance from the observer to the birds with a laser rangefinder (Leica Rangemaster 900; Leica, Solms, Germany). When we did not see a calling bird, we mapped its approximate position on aerial photographs (1:5,000 scale) based on the likely attenuation and direction of its vocalization. We then measured the distance from observer to the position of each calling bird using the software QGIS v.3.2.3 'Bonn'.

Density estimation
The density of the species was estimated through the distance sampling method using the software Distance 7.2 (Buckland et al. 1993, Thomas et al. 2010. For density estimation, we used only the first detection of each located individual or pair, as suggested by Buckland et al. (2001). After a visual inspection of distance distribution, we truncated 10% of the largest observations as suggested by Buckland et al. (1993) and transformed the distance data into equal intervals of 60 m. We chose these intervals because they were the best for assuring a good shape of the distance data, inasmuch as shape criterion is one of the main assumption of distance sampling (Buckland et al. 1993). We tested various combinations of key functions (uniform, half-normal) and series adjustments (cosine, simple polynomial, Hermite polynomial) (Buckland et al. 1993) and we used the Akaike Information Criterion (AIC) (Akaike 1973) to evaluate the fit of each model (Buckland et al. 1993, Thomas et al. 2010. Among the models with the lowest AIC values, the one that appeared most suitable based on a visual examination of histograms and the results of χ 2 goodness-of-fit tests was selected (Buckland et al. 2001, Buckland 2006, Thomas et al. 2010. Furthermore, the average probability of detection was estimated, and the effective detection radius (EDR) was defined. For each estimate the percentage coefficient of variation (CV) and the 95% confidence intervals (CI) were calculated.

Habitat suitability
The habitat requirements of the Turtle Dove were evaluated following a presence vs. availability approach (Boyce et al. 2002, Manly et al. 2002. The environmental variables ( Table 1) were measured at the home-range scale (Manly et al. 2002, Morrison et al. 2006) at presence sites and subsequently compared with those of an equal number of availability sites which were randomly selected in the study area without any constraints (Boyce et al. 2002, Manly et al. 2002, Keating and Cherry 2004. By considering the relevance of taking into account the spatial ecology of the species (Brennan et al. 2002, Manly et al. 2002, presence/availability sites corresponded to cells whose extent was defined based on the home range size of the species during breeding, as suggested by the available literature. Specifically, the home range of the species is equal to 1.91-3.08 ha , hence, considering an average of about 2.5 ha, a grid was generated with hexagonal cells with this area (Caprio et al. 2011, Boan et al. 2014). Subsequently, in order to investigate the relationships between the environmental variables and the Turtle Dove, we tested for significant difference between presence and random sites using the non-parametric Mann-Whitney U test (Legendre and Legendre 1998). Then, a Resource Selection Probability Function was formulated (Boyce et al. 2002, Manly et al. 2002 performing a generalized linear model with a binomial error distribution (Keating andCherry 2004, Rushton et al. 2004). In particular, a priori sets of models were first built using all the combination of the environmental variables that seemed to affect species occurrence. For each model the secondorder Akaike Information Criterion (AIC c ) was calculated and the model with the lower AIC c was selected as the best . However, considering that models with Δ AIC c ≤ 2 give substantial support , we showed and discussed them also. For the analysis, all variables considered were standardized by normalization, that is, each variable had a mean of zero and a standard deviation of one   (Quinn andKeough 2002, Zuur et al. 2007). The model's ability to distinguish between occupied and available sites was tested by means of the area under the curve (AUC) of the Receiver Operating Characteristic plot (ROC curve) Ferrier 2000, Fawcett 2006). Moreover, we tested both the residuals' normality by the Kolmogorov-Smirnov test (Legendre and Legendre 1998) and the residuals' spatial autocorrelation by the Moran I test (Zuur et al. 2007, Bivand et al. 2008. We used the Variance Inflation Factor (VIF) with a threshold of 3 to exclude variables' collinearity Monette 1992, Zuur et al. 2010). The explained deviance D 2 was used as a measure of the variance explained by the model (Crawley 1993, Zuur et al. 2007. Data used to measure the environmental variables were obtained from the regional land use map "DUSAF 5.0" (ERSAF 2015) and from the regional forest map "Carta dei tipi forestali reali della Lombardia" (ERSAF 2012) and processed by the software Quantum GIS v.3.2.3 'Bonn'. All the analyses were performed using the statistical software R v.3.3.2 (R Core Team 2019) and the packages MuMIn (Tillé and Matei 2016), car (Fox and Weisberg 2011), and ROCR (Sing et al. 2007).

Density estimation
During fieldwork, 76 observations of Turtle Dove were collected, at an average distance of 113 m (SE = 9.0, min = 0 m, max = 363 m), but six of them were removed from the analysis because of the right truncation. The best detection probability function obtained was the uniform + cosine with two parameters (AIC = 179.57, χ 2 = 0.389, df = 1.0, P = 0.533) ( Table 2). This gave an EDR of 107 m and the average probability of detection was estimated to be 0.20 (CV = 14.0%; 95% CI = 0.15-0.26). Single birds or pairs were usually observed, with an average of 1.04 AE 0.01 (SE) ind/group. There was no correlation between group size and distances of detection (r = 0.181, df = 68, P = 0.934). The density of Turtle Doves estimated was 5.0 pairs/km 2 (SE = 0.01, min = 3.7, max = 7.7, CV = 18.6%) (Figure 2).

Habitat suitability
The analyses showed that there were some differences between presence and availability sites ( Table 3). In particular, in the presence sites there was a greater cover of poplar plantations and other tree plantations, oak forests, riparian willow forests, and shrublands, than in the availability sites. In contrast, there was also a lesser cover of maize and paddy fields. The best model showed that density of poplar plantations, riparian willow forests, shrublands, and hedgerows positively affected the presence of the Turtle Dove ( Figure 3, Table 4). On the contrary, built-up areas, cereal crops, maize, and paddy fields negatively influenced the presence of the species. The set of the best models (ESM Table S1) showed that another three variables positively affected the presence of the Turtle Dove, though they were less important. Specifically, other tree plantations, oak forests, and Robinia pseudoacacia forests. The ability of the best model to distinguish between occupied and unoccupied sites was good, with an area under the ROC curve equal to 0.832 (P < 0.001). The model residuals were normally distributed (D = 0.053, P = 0.726) and did not show any spatial correlation (Moran test, I = 0.484, P = 0.314). The VIF revealed no collinearity among predictors (Table 4) and the explained deviance D 2 was equal to 26.4%. The study area showed an average predicted probability of 0.49 AE 0.008 (SE) (min. 0.05, max 1.0) ( Figure S2).

Discussion
Our analyses showed that breeding densities of the Turtle Dove in the agroecosystem investigated are between 3.7 and 7.7 pairs/km 2 , with an average of 5.0 pairs/km 2 . These densities are similar or slightly higher than other densities observed in northern Italy, where 1-5 pairs/km 2 were estimated (Brichetti and Fracasso 2006). Unfortunately, due to the small sample size, we could not estimate densities for each land use type, therefore our estimate is related to the whole study area. However, in general the densities are lower in farmlands than in woodlands (Fisher et al. 2018). On the other hand, our estimates are largely lower than those observed in North Africa, where 45 pairs/ha were estimated in citrus groves ; besides, in Moroccan olive groves 16-20 nests/ha were estimated (Hanane and Baamal 2011). The relatively high densities estimated could be partially explained by the average detection probability of 0.20, quite low with respect to the value of 0.56 found in another study (Johnston et al. 2014). However, this low detection could be explained by at least two reasons. First, in our study most of the observations referred to closed habitat types, such as forests, shrublands, and tree plantations, where lower detectability is the rule (Buckland et al. 1993, Gottschalk and Huettmann 2011, Johnston et al. 2014. Second, we observed Turtle Doves also in very heterogeneous landscapes with small patches of forests and it is known that detection probability declined with increasing forest fragmentation (De Wan et al. 2009).  Concerning its habitat requirements, the Turtle Dove selects semi-natural forests, especially riparian willow forests but also oak-dominated forests. Certainly, the species regularly inhabits forests, both broadleaved and needleleaved, as well as mixed (Browne and Aebischer 2003, Browne et al. 2005, Bakaloudis et al. 2009, Dias et al. 2013, Marx and Quillfeldt 2018. This is a rule not only in Central Europe, but also in North Africa, where the subspecies S. t. arenicola is also distributed (Yahiaoui et al. 2014, Hanane and Yassin 2017, Hanane 2018. The selection of riparian forests by the Turtle Dove is already known (Sáenz De Buruaga et al. 2012, Gruychev andMihaylov 2019), although in some studies the distance from open water was not relevant (Gutiérrez-Galán et al. 2018, Hanane 2018. As pointed out by Hanane (2018), this fact could be explained by the ability of the species to travel long distances to reach water points, or by the fact that other factors are more limiting. In our study area, however, we need to consider that a large proportion of wooded areas are placed along the Ticino River, therefore there could be some spatial dependence in relation to forests. In our study area, there is some evidence that Turtle Doves also inhabit sites with high cover of black locust. This fact could be due to the less dense structure of this tree, which might be favorable for the species; indeed, in Morocco there was evidence of the fact that the species does not tolerate a very enclosed microhabitat (Hanane 2018). However, Table 3. Mean values (AE SE) of the environmental variables measured in the presence and in the random sites. We highlighted in bold the variables for which we found significant differences (P < 0.05) by the nonparametric Mann-Whitney U test. considering its statistical weakness, it is reasonable to think that this finding is a random result, especially because black locust is often associated with oaks in our study area. As well as seminatural forests, shrublands and hedgerows have a positive effect on the species' presence. Past research has demonstrated that shrubs are very important for the species because nests are very often concealed inside them (Mason and MacDonald 2000, Browne et al. 2005, Dunn and Morris 2012. In particular, it was estimated that the most suitable should be 4.5 m high and 3 m wide (Sparks et al. 1996. Nonetheless, in the Ebro valley (northern Spain), the abundance of the species is negatively associated with shrublands (Sáenz De Buruaga et al. 2012). Furthermore, hedgerows are undoubtedly very important in other parts of its range , Browne et al. 2005, Dunn and Morris 2012, Dias et al. 2013 and their removal from farmland is one of the reasons for its decline in England . Generally, the most used are composed of thorny species like hawthorn Crataegus monogyna and blackthorn Prunus spinosa (Browne et al. 2005). In addition, the importance of hedgerows could be due to the presence of weeds along these linear features. Indeed, it is well known that one of the main habitat requirements for the species is weed-rich areas (Mason and MacDonald 2000, Browne and Aebischer 2003, Dunn et al. 2015, 2018. This is true because the diet of both adult and juveniles is often composed of seeds of wild species, especially common fumitory Fumaria officinalis and common chickweed Stellaria media (Murton et al. 1984, Gutiérrez-Galán and Alonso 2016, Dunn et al. 2017. The Turtle Dove also selects sites with high cover of tree plantations, such as poplar plantations and reforestations/SRCs. Notably, poplar plantations are selected in spite of the completely absence of shrubby understorey. In fact, in Portugal the species is positively affected by coniferous forests without understorey, probably because they provide a suitable combination of secure nesting sites and food resources in the herbaceous understorey (Dias et al. 2013). In our case, however, poplar plantations are likely not used as nesting sites because their structure is not very dense and therefore cannot provide safe shelter, but it is possible that they are used as feeding sites, as in Portugal. On the other hand, the positive selection for reforestations/SRCs is already known for the species in the study area (Chiatante et al. 2019). This land cover type could be used for nesting because it is often dense and thick, becoming a safe place in which build the nest. As expected, paddy fields and maize have a negative effect on the Turtle Dove because of their unsuitability. Likewise, we found a negative effect of cereals on the Turtle Dove, which concurs with other research Aebischer 2003, 2004). Our research was focused on the first part of the breeding period, so the under-use of cereals is to be expected because Turtle Doves use this habitat only after harvest (Browne and Aebischer 2003). In addition, the species is prone to extinction in arable lands where barley increased (Chamberlain and Fuller 2000). However, cultivated crops, principally wheat, barley, and oil-seed rape, often form the bulk of the diet of both adult and the nestlings Aebischer 2002, Mansouri et al. 2019), and some studies have pointed out the importance of cereal crops for improving the number of chicks hatched and fledged per nest (Kafi et al. 2015, Mansouri et al. 2019. Undoubtedly, the presence of useful crops is important for the Turtle Dove, but probably not so much as the presence of weed-rich areas, especially in intensively cultivated agroecosystems, where the absence of fallows could be a limiting factor Aebischer 2002, 2003). Finally, in our study area the Turtle Dove lives in sites with few built-up areas, which is in accordance with other researches carried out in the Iberian Peninsula and Morocco (Patón et al. 2012, Hanane 2018.

Conclusion
In summary, in an intensively cultivated agro-ecosystem in northern Italy, the Turtle Dove has densities of 5.0 pairs/km 2 ; however, considering the low detectability (equal to 0.20) this estimate could be biased (MacKenzie et al. 2006). This species inhabits areas with higher tree cover, either semi-natural forests or tree plantations, as well as areas with many shrubs and hedgerows, that are likely used for nesting. In addition, it avoids areas with a high proportion of crops, such as paddy fields, maize, and cereals. Furthermore, we point out the importance of maintaining weed-rich areas, such as hedgerows, which provide food both for adults and for juveniles. We have to keep in mind that these landscape features should also be maintained in areas in which there are no trees, shrubs, or hedgerows suitable for nesting, because the Turtle Dove can fly for many kilometres to reach feeding areas Aebischer 2003, Gutiérrez-Galán et al. 2018). For conservation purposes, these should be encouraged also within or adjacent to cropped fields. They should be produced by sowing seed mixes that contain species eaten by Turtle Doves (Browne and Aebischer 2003, Dunn et al. 2015, Rocha and Quillfeldt 2015, Fisher et al. 2018. Besides, these landscape features should not be sprayed with herbicides or other pesticides because of the sensitivity of the species to biocides (Ruiz-Suárez et al. 2015, Fisher et al. 2018. Hence, Turtle Doves require a combination of habitat features with suitable nesting and feeding areas, both in close proximity and in the surrounding context of the home range, and the degradation of either of these may reduce Turtle Dove populations (Gillings and Fuller 1998, Dias et al. 2013, Fisher et al. 2018).

Supplementary Materials
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/ S0959270920000532.