Range compression of migratory passerines in wintering grounds of the Western Mediterranean: conservation prospects

Summary

Because migrant birds occur in different parts of the world in different seasons, their numbers may be limited by the size of the smallest area they inhabit during the year. In addition, restricted ranges make populations more susceptible to local perturbations such that range size is frequently considered a correlate of species vulnerability. Despite this, little is known about the balance between seasonal ranges in the migrant populations of partially migratory species. These migrants are difficult to segregate from sedentary conspecifics in winter grounds and thus the extent of their ranges is difficult to assess. Here, we studied the extent of potential breeding and wintering ranges of 10 partial migratory passerines moving to winter in the Iberian Peninsula and the Maghreb. After testing migratory connectivity of the individual species, we used niche modelling to calculate the extent of potential breeding and wintering ranges in 1,113 pairs of ring recoveries linking individuals between breeding and wintering localities. The results indicate that most species show migratory connectivity and that all of them show range compression in winter relative to the breeding range, with scores ranging between 19% and 58% (mean 37%) of breeding ranges. We discuss the importance of non-breeding grounds for conserving migratory passerines in the Western Mediterranean Basin, an area under pressure from climate change and agricultural intensification.

Introduction

Range size is considered a main correlate of species vulnerability because restricted ranges make species more vulnerable to local perturbations (Purvis et al. 2000, Lee and Jetz 2009, Di Marco and Santini 2015). However, the study of ranges is usually flawed by methodological difficulties in migrant species, which, as in birds, spend different periods of the year in separate locations (Sherry and Holmes 1995, Klaassen et al. 2014, Runge et al. 2015). As many migratory birds are in decline (Sanderson et al. 2006, Vickery et al. 2014), exploring the distribution and size of seasonal ranges is key to designing conservation measures at appropriate spatial-temporal scales (Marra et al. 2015).

Birds express a variety of migratory movements, with populations spreading or converging in non-breeding grounds (Gilroy et al. 2016). In the case of partially migratory species (comprising both migratory and resident individuals; Chapman et al. 2011), studies on range compression (reduction of non-breeding range size relative to breeding range and vice versa) are lacking. This is concerning because partial migration occurs in many species (e.g. 40% of European birds; Gilroy et al. 2016) and range compression can affect the size and vulnerability of migratory populations (Newton 1995, Rappole et al. 2003, Mills 2006, Norris and Marra 2007, Runge et al. 2015). This lack of information can be explained by the difficulties in separating migratory and local individuals in sympatric non-breeding grounds (de la Hera et al. 2012). This complicates the mapping of non-breeding ranges if compared with long-distance migrants in which range compression can be assessed using current distribution maps available in conventional atlases (Newton 1995, Mills 2006, Gilroy et al. 2016).

Here, we explore the seasonal range compression of some partial migratory passerines (Passeriformes) moving to winter in the Iberian Peninsula and the Maghreb, a region reputed as a primary wintering ground for many European birds (Busse 2001). Widely-dispersed, abundant passerines are often neglected in projects focused on locating and protecting important bird areas for conservation despite the fact that they represent the bulk of the avifauna in most regions and play a significant functional role in some ecosystems (Whelan et al. 2008). Thus, it is interesting to explore whether seasonal changes in the range size of migratory populations reveal the existence of small areas of particular conservation concern. It has been shown that the western Mediterranean Basin is not a homogeneous wintering ground since the best areas are mainly restricted to mild lowlands and coastal areas (Tellería et al. 1999, 2014a, Carrascal et al. 2012). However, we do not know if these geographical features reduce the wintering range of migratory populations compared to their breeding grounds in central Europe.

We use ringing records linking individuals between breeding and non-breeding localities to explore if winter ranges are smaller than summer ranges. Clusters of ringing records have proven useful in detecting the location of ranges of migratory birds (e.g. Clark et al. 2009, Calenge et al. 2010), but do not depict the spatial patterning of areas suitable for populations involved in these movements. However, a set of mathematical algorithms has been developed in the last decade to predict the potential distribution of species by combining occurrence data (latitude and longitude) with environmental variables (Peterson 2001, Elith and Leathwick 2009). We combine these methods with the use of data from ringed birds to estimate potential range sizes and to explore the environmental drivers of bird distribution (Rodríguez et al. 2007, Wisz et al. 2007, Rivera et al. 2011). The potential ranges resulting from these analyses are used to test whether wintering ranges of migratory populations in the Iberian Peninsula and the Maghreb are smaller than their breeding ranges in central Europe.

Methods

Study area

The wintering area covers 1,700,000 km² located between 28° and 44°N at the south-western border of the Palearctic (Figure 1). It receives many migratory birds, most of them from the western side of the migratory divide crossing Europe between 10° and 20°E (Busse 2001, Møller et al. 2011). The most outstanding environmental gradients of this wintering ground are the sharp changes in elevation and the concomitant variation in climatic conditions and vegetation (Figure 1). The area is dominated by cereal fields, vineyards, olive groves and Mediterranean woodlands and scrublands located between the moist broadleaved woodlands and meadows of northern Iberia and the southern expanses of the Sahara.

Figure 1. a) Distribution of the Human Foot Print (HFP) index in Europe. Increasing dark tones show increasingly high values of HFP. The study area is located within the striped square. b) Elevation map of the western Palearctic. Increasing dark tones show increasing elevations, with the darkest tone showing the areas over 1,000 m asl. c) Distribution in breeding and wintering areas of the 1,113 pairs of birds controlled in breeding grounds (black dots) and wintering grounds in the study area (light dots). Vertical lines show the longitudinal range within which are distributed the migratory divides of most migratory birds in Europe, according to Møller et al. (2011). d) Potential distribution of migratory populations as reported by the number of species (0-10) present in breeding and wintering grounds.

Ring recoveries

The data used in this paper refer to all birds recovered/ringed in winter (December to February) in the Iberian Peninsula and the Maghreb that were ringed/recovered in central and northern Europe during the breeding period (April to August; Figure 1). The data, ranging from 1925 to 2011, were provided by the European Union for Bird Ringing (EURING) and the Spanish Office of Migratory Species (Spanish Ministry of Agriculture, Food and Environment). Additional data on the number of ringed birds were recorded from the official page of EURING (http://www.euring.org; Appendix S1 in the online supplementary material). From a previous set of 13 species, we excluded Anthus pratensis, Saxicola torquata and Carduelis carduelis because they had less than 50 ring recoveries, a sample size that we considered too small for a comprehensive view of their seasonal ranges. The final data set included 1,113 pairs of ringed-recovered individuals for 10 species (Figure 1; Table 1). These data provided geo-referenced clusters of ringing records per species, which show the location of the marked individuals in breeding and wintering grounds (Figure S2-1). Finally, we explored the strength of migratory connectivity of populations (most individuals from one breeding population move to the same wintering locations; Webster et al. 2002) by testing if individuals that breed close to each other also winter close to each other. High connectivity will suggest that changes in particular wintering sectors of the Iberian Peninsula and the Maghreb will affect the populations of discrete breeding quarters whilst low connectivity will suggest a widespread effect of changes in breeding grounds (Finch et al. 2017). To test this we conducted a Mantel test (Ambrosini et al. 2009) with the R package ‘vegan’ in which the statistical significance of correlation coefficients was determined by 9,999 random permutations (Oksanen et al. 2015).

Table 1. Number of pairs of ring records used to model the distribution of migratory populations in breeding and wintering grounds. R statistics and P values resulting from Mantel tests. Estimated breeding and wintering ranges of the studied passerines and range compression in wintering grounds in percentage ([wintering area/breeding area]*100).

Environmental predictors

The geographical location of ringing records was used to run Maxent by selecting a set of explanatory variables related to the large-scale distribution of birds in breeding and wintering areas. The aim was to obtain useful models and to use the response curves provided by Maxent to explore the relationships between the species occurrence probabilities and the involved variables. We used variance inflation (VIF) analysis to avoid highly correlated variables (Marquardt 1970). From a set of 11 variables (seasonal maximum, minimum and mean temperature, temperature seasonality, seasonal precipitation, precipitation seasonality, sun radiation, NDVI, bare ground, herbaceous cover and tree cover) we selected five environmental predictors per season (VIF scores under 5, range: 1.15–4.56). Mean temperatures in breeding (April to August) and wintering (December to February) ranges were selected because small birds avoid cold areas to limit excessive thermoregulatory costs (Calder and King 1974). We also included mean precipitation in breeding and wintering grounds. Precipitation improves primary productivity in wintering grounds after the Mediterranean summer (Nahal 1981), thus increasing the carrying capacity for wintering birds (Carrascal and Palomino 2012, Tellería et al. 2014b). However, in temperate areas higher precipitation has been related with poor breeding success. Thus, it can be suggested that the moistest sectors will be suboptimal areas for breeding (e.g. Öberg et al. 2015). We also included the annual Normalized Difference Vegetation Index (NDVI) as a surrogate of primary productivity from MODIS satellite-based sensor (30” resolution, or ∼1 km; Carroll et al. 2004, Pettorelli et al. 2011) and, according to the strong effect of vegetation cover on passerine distribution (Wiens 1989), we included woody cover from Vegetation Continuous Fields MOD44B (Di Miceli et al. 2011). Finally, we considered the effect of two different variables related to the seasonal pulses of productivity tracked by migratory birds in breeding and wintering grounds (Wisz et al. 2007, Engler et al. 2014). We used precipitation seasonality (standard deviation of monthly means x 100) as a surrogate for the rain-mediated productive pulses in Mediterranean wintering grounds, and temperature seasonality as a surrogate of the productive output in central and northern European summer grounds. Climate data were recorded from Worldclim 1.4 (Hijmans et al. 2005) and prepared as ASCII raster maps with a resolution of 5x5 km with the raster package in R (Hijmans and Van Etten 2013, R Development Core Team 2015).

Species distribution models

We used Maxent, a machine-learning technique based on the principle of maximum entropy (Phillips et al. 2006), to predict the habitat suitability of marked populations from presence data (ring records) within breeding and wintering ranges. This program has been designed to minimise the relative entropy between two probability densities (one estimated from the presence data and one from automatically generated presences) defined in covariate space (Elith et al. 2011). In this way, it is possible to model the probability of the occurrence of species over a given study area. To select the area with which to calibrate the models in breeding and wintering grounds, we used the minimum convex polygon (convex hull) produced by the full set of ringing records. This is the smallest polygon in which no internal angle exceeds 180° and contains the records of all species. In both seasons and for all species, we ran Maxent (log output; regularisation multiplier b = 1; autofeatures; convergence threshold = 0.00001) in 10 replicates with 70% of the presences as training data and the rest as test data for internal verification. Duplicate records in the same localities (e.g. ringing stations) were removed in each replicate and spatial autocorrelation was minimised by randomly removing occurrences within 5 km of each other. In addition, and since the distribution of ring recoveries can be affected by human presence (e.g. more records will be gathered in the most populated areas; Busse 2001), we used the layer of human footprint (HFP), an index of population density, land transformation and road density (Sanderson et al. 2002), as a bias grid (10,000 background points with a likelihood of presence proportional to HFP; Dudík et al. 2005). We used mean human footprint per country because it is positively related to the density of all ringing recoveries reported per country (r = 0.53, P = 0.036, n = 16), a plain index of the effort applied to control ringed birds (Appendix S1). To explore the importance of each predictor (see below), we carried out jackknife analyses of the regularised gain with training data. The models were evaluated using 10-fold cross-validations measuring the accuracy by means of the area under the receiver operating characteristics curve (AUC; Fielding and Bell 1997) to assess the congruence between observed and detected records in the test data reserved for verification in Maxent (Phillips et al. 2006). Finally, we obtained a map by averaging the suitability models of individual species. With this map, we try to establish a comprehensive view of the main areas used by partial migratory passerines during the breeding and wintering periods. We used the ‘dismo’ package (Hijmans and Van Etten 2013) in R (R Development Core Team 2015) to perform all analyses.

Potential range maps

Geographic ranges can be derived from species distribution models by applying a cut-off threshold value to distinguish suitable from unsuitable areas. The resulting maps from the previous niche modelling process were converted into binary ‘presence–absence’ maps using the threshold selection method based on maximising the sum of sensitivity and specificity (Manel et al. 2001). This threshold is recommended for conservation purposes as omission errors are avoided and sensitivity favoured (Jiménez-Valverde and Lobo 2007, Liu et al. 2013). We considered areas with a habitat suitability above the threshold as ‘presence’ and those below as ‘absence’. The geographic area occupied by species in each season were calculated by converting the ‘presence’ raster to a polygon shape file and calculated polygon areas using the WGS 84 / UTM zone 30N (32630) projection. Finally, we measured winter range compression dividing potential winter range by breeding range (Table 1). We used the raster and ‘dismo’ packages (Hijmans and Van Etten 2013) in R (R Development Core Team, 2015) to calculate the potential range map of the species in both seasons.

Results

Mantel test correlations between the geographical location of breeding and wintering sites in most species (all but Robin Erithacus rubecula and Redwing Turdus iliacus; Table 1) support the maintenance of spatial structuring in populations between breeding and wintering ranges. The AUC values resulting from niche modelling suggest a good performance of models according to current standards (AUC >0,7; Elith 2002) and the response curves provide some additional information on the way birds are distributed within breeding and wintering ranges (Table 2). For instance, the habitat suitability of species in breeding grounds was positively related to temperature and strongly affected by temperature seasonality (Table 2). In wintering grounds, most species were related to temperature and woodland cover (Table 2).

Table 2. Estimates of relative contributions of the environmental variables in models predicting habitat suitability of species in breeding and wintering grounds. AUC scores show the fitting of models (max values = 1) and the two following values represent the percent contribution/permutation importance of each variable in the model. Percent contribution indicates the change in regularized gain by adding the corresponding variable. Permutation importance represents, for each environmental variable in turn, the resulting drop in training AUC when the values of that variable on training presence and background data are randomly permuted, normalized to show percentages. Values are averages over 10 replicate runs. Symbols in parentheses show the trend of the response curves for the quantitative variables: +, increase; -, decrease; Ω, hump-shaped = no trend).

Potential range maps suggest that most of the migratory populations wintering in the western Mediterranean Basin extend their breeding grounds in central Europe west to Poland and in southern Scandinavia (Figure S2-2). These maps show that the wintering grounds occupied by migratory populations were smaller (mean 37%) than breeding grounds in all species (t-test for paired samples t = 5.92, P < 0.001, n =10; Table 1). More explicitly, potential winter ranges covered a mean area of 275,600 km² (min-max 134,900–411,190 km², Table 1) accounting for 16% (7.9–24.2%) of the Iberian Peninsula and the Maghreb (Figure 1).

Discussion

Distribution of potential breeding and wintering ranges

The potential breeding ranges of individual species depicted a patchy distribution, with migratory populations reporting the highest occurrence probabilities in different inland or coastal areas of central Europe west of the migratory divide (Figure S2-2). This suggests a diverse geographical origin of populations moving to winter in the western Mediterranean. When the mean potential distribution of these species is represented (Figure 1), the emerging map reveals that the most probable source area of migratory birds is located in a region that extends from the Alps to the North and Baltic seas, and from the French Massif Central to the Slovakian Tatra Mountains, constrained to the east by the migratory divide of European bird populations (Figure 1). This seems to be the core range of the bulk of migratory passerines moving to winter in the Iberian Peninsula and the Maghreb.

The response curves provided by Maxent indicated that in breeding and wintering areas, the potential distributions of migratory populations tend to avoid the colder areas and are sited in most seasonal areas under a given threshold of summer (dryness) and winter (cold) conditions (Table 2). The importance of temperature in shaping migratory bird distribution is widely recognised, and is the result of the high thermoregulation costs of living in areas under bird thermo-neutral ranges (Swanson and Garland 2008, Carrascal et al. 2012, Khaliq et al. 2014). This could explain the avoidance by birds of elevated areas in breeding and wintering ranges (Figure 1). The potential wintering range of migratory passerines depicted a homogeneous pattern, with most birds constrained to similar lowlands and coastal sectors of the Iberian Peninsula and the Maghreb (Figure S2-2, Figure 1), where wood cover was positively (thrushes) or negatively (the rest of species) correlated, supporting the effect of specific habitat preferences on bird distribution (see Tellería et al. 1999, del Moral et al. 2012). These patterns are similar to the wintering maps reported by ring modelling of Meadow Pipit Anthus pratensis, Chiffchaffs Phylloscopus collybita, Blackcaps Sylvia atricapilla and Robin Erithacus rubecula, validated by field sampling of abundances (Tellería et al. 2012, 2014a, 2016).

Range compression in wintering grounds

Compression of winter ranges reported by common distribution maps of species seems to occur in half of the Western Palaearctic migratory birds (Gilroy et al. 2016). This reveals the importance of a phenomenon that varies between species (and populations) according to the idiosyncratic features of the species and regions involved in this process. Because of this potential variability, range contraction is a context-dependent process whose effect on the conservation of migratory populations must be addressed in particular migratory populations within their geographical ranges (Gilroy et al. 2016).

This study, which has focused on migratory passerines wintering in the Iberian Peninsula and the Maghreb, supports a sharp reduction of potentially suitable areas in winter ranges affecting all the study species, with compression scores ranging from 19% to 58% (mean 37%) of breeding ranges (Table 1). This effect of range compression affecting all species is higher than that reported in Palearctic land birds moving to winter south of the Sahara, where 69% of species (n = 62) experienced a reduction of ranges in non-breeding grounds (Newton 1995). In fact, our results are more similar to those reported in North American land birds moving to winter in Mesoamerica, where most birds (94%, n = 89) showed a reduction of non-breeding range (Mills 2006). In this case, range contraction was explained by the sharp reduction of emerged lands available to birds in the Mesoamerican corridor (Mills 2006). We suspect a similar effect in the south-western border of the Palaearctic, where migratory populations are restricted to a minor part of emerged lands (mean range covers 16% of the Iberian Peninsula and the Maghreb) because suitable areas are constrained by coastlines, cold highlands and the dry expanses of the Sahara (Tellería et al. 2014a,b).

Conservation prospects

The results in this study suggest that, because of small range size, the wintering grounds in the Iberian Peninsula and the Maghreb may act as ‘bottlenecks’ of passerine populations involved in these migratory movements (Weber et al. 1999, Gilroy et al. 2016). This will likely occur in species tied to scarcer habitats, or limited by food availability. However, evidence available in the study area does not clearly show such limitations in the case of common passerines. Most migratory passerines arrive at this area in October, when autumn rains begin a period of primary productivity, invertebrate activity and fruit abundance (Nahal 1981, Guo et al. 2012). Common passerines rely on seeds, invertebrates and fruit available in extensive human-modified landscapes (e.g. cereal fields, olive groves, etc.) and Mediterranean scrublands and woodlands (Rey 2011, Del Moral et al. 2012, Tellería et al. 2014b) where field studies do not support food restriction (Díaz y Tellería 1994, Hampe 2008, Tellería et al. 2013). We acknowledge that this balance will be disrupted in winters of food scarcity, a situation that many birds cope with by food-tracking (Rey 2011, Tellería et al. 2014c). However, movement among suitable areas will not be enough if changes affect simultaneously the carrying capacity of the whole range. The south-western border of the Palearctic is expected to suffer the strong effects of climate change, with increasing temperature and decreasing rainfall (Giorgi and Lionello 2008, Schilling et al. 2012). In this context, the depletion or altitudinal shifts of lowland habitats suitable for wintering birds have been predicted (Ruiz-Labourdette et al. 2012, Ponti et al. 2014, Tellería et al. 2016).

The response of birds to these changes in wintering areas is difficult to predict (La Sorte et al. 2009) but the potential carry-over effects of these environmental changes (Norris et al. 2004) will probably affect populations in larger breeding areas of Europe. In this context, the study of migratory connectivity is essential to understanding the effects of these changes on breeding populations (Creswell 2014). Migratory connectivity may vary along a continuum, from strong to weak (Finch et al. 2017), and our results support this variability. Most passerines wintering in the Iberian Peninsula and the Maghreb show significant but different correlation coefficient scores in Mantel tests, and two species report lack of migratory connectivity (Table 1). This lack of connectivity in Redwing and Robin could result from within-range movements typical of some birds (Tellería et al. 2014c, Stutchbury et al. 2016) or the relatively small scale of approach in this study (Creswell 2014; e.g. Ambrosini et al. 2016 report migratory connectivity of Robins at continental scales). But, as a whole, the results in this paper suggest that connectivity is expected to underpin the response of breeding populations in central Europe to local changes in non-breeding grounds of the Western Mediterranean. We agree that this may affect just a part of populations on breeding grounds if some individuals spread out to alternative wintering grounds or are able to shift their ranges in response to changes (Finch et al. 2017).

Supplementary Material

To view supplementary material for this article, please visit https://doi.org/10.1017/S0959270917000120