Winners and losers: How the elevational range of breeding birds on Alps has varied over the past four decades due to climate and habitat changes

Abstract Climate warming and habitat transformation are widely recognized as worrying threatening factors. Understanding the individual contribution of these two factors to the change of species distribution could be very important in order to effectively counteract the species range contraction, especially in mountains, where alpine species are strongly limited in finding new areas to be colonized at higher elevations. We proposed a method to disentangle the effects of the two drivers of range change for breeding birds in Italian Alps, in the case of co‐occurring climate warming and shrub and forest encroachment. For each species, from 1982 to 2017, we related the estimated yearly elevational distribution of birds to the correspondent overall average of the daily minimum temperatures during the breeding season and the estimated amount of shrubs and forest cover. Using a hierarchical partitioning approach, we assessed the net contribution (i.e., without the shared effect) of each driver. Both temperature and shrub and forest cover showed a positive trend along the time series and resulted the most likely causes of the significant elevational displacement for 21 of the 29 investigated birds. While shrub and forest cover was found to be an important driver of the expansion of forest bird range toward higher elevations, the effect of temperature on favouring the colonization of previously climatically unsuitable forests at higher elevations was not negligible. Shrub and forest expansion resulted the main driver of the range contraction for edge and open habitat species, which suffered a distribution shrinkage at their lower elevational boundary. In light of climate warming, these results highlighted how the net range loss for edge and open habitat species, caused by shrub and forest encroachment consequent to land abandonment, should be counteracted by implementing proper conservation management strategies and promoting sustainable economic activities in rangeland areas.


| INTRODUC TI ON
Although habitat transformation is still recognized as one of the major human drivers of species distribution changes at a global scale (Foley et al., 2005), a rising attention has been addressed in the last decades to the effects produced by climate change on biota (Parmesan & Yohe, 2003;Thuiller, 2007;Walther et al., 2002). In fact, the variation in climatic conditions forces species to adjust their spatial distribution according to their ecological niches, provided that resources are available (Thomas et al., 2004;Thuiller, Lavorel, & Araújo, 2005).
Among the changes in climate features, those involving the raise of temperatures (referred as global warming) are certainly the best known and probably the most investigated. The magnitude of temperature change appears to be different in different geographical areas, and their effect can become more severe with increasing latitude or elevation (Loarie et al., 2009). Climate warming may affect species ecology with contrasting, additive or even synergic effects with the other co-occurring processes (e.g., land-use changes, pollution, harvesting, and species interactions; Mantyka-Pringle, Martin, & Rhodes, 2013;Oliver & Morecroft, 2014). This is particularly noticeable in those areas where different relevant environmental changes act simultaneously. Species inhabiting medium latitude mountains, like the Alps, have been documented to suffer not only the effects of rising temperatures, but also those due to changes in agroforestry and pastoral practices, which have played an important role in shaping species' distribution for a long time (Ausden, 2007;Maurer, Weyand, Fischer, & Stöcklin, 2006). Indeed, until a few decades ago, grazing by herds restrained the forest cover in mountain areas, but after the abandonment of pastures, shrubs and forests widely expanded, and even the tree line tended to raise in elevation, probably as a consequence of the synergic action of climate warming and land abandonment (Gehrig-Fasel, Guisan, & Zimmermann, 2007;Leonelli, Pelfini, di Cella, & Garavaglia, 2011;Parolo & Rossi, 2008;Pernollet, Korner-Nievergelt, & Jenni, 2015). In fact, the phenomenon was probably boosted by the increase in temperature, which limiting the period of snow cover, led to the colonization of the higher elevations by shrubs and forests (Gehrig-Fasel et al., 2007).
Overall, in the long-term period, the abandonment of pastures and climate warming may promote the shrub and forest expansion at the expense of open habitat in mountain areas, leading to a general complex pattern that can vary between geographical areas (e.g., Rocchia, Luppi, Dondina, Orioli, & Bani, 2018).
As birds are particularly mobile and sensitive to environmental alteration, they could arguably be one of the first taxa to change its distribution in response to both climatic and habitat changes (Ambrosini et al., 2011;Both et al., 2004;Both & Visser, 2001;Visser, Both, & Lambrechts, 2004).
To quantify objectively the effects of gradual climate and habitat changes on species distribution, long time series of information are essential. Indeed, long time series may help to counteract the intrinsic, often wide, sources of variability and stochasticity, as well as the measurement and sampling noises contained both in biological and environmental data, which could make it difficult to find significant and strong relationships between them (Hilfinger & Paulsson, 2011).
Of course, the data noise due to the stochastic errors of measurement of biological and environmental information can be difficult to manage without a long time series of data. On the other hand, the actual variability of biological data may be due to multiple environmental factors whose interactions are seldom taken into account (Guisan & Zimmermann, 2000).
The main purpose of this study was to investigate the drivers of the elevational range changes of breeding birds in Central Italian Alps, separating the effects produced by climatic (temperature increase) and habitat (shrub and forest recolonization) drivers. We did not take into account precipitations since they did not show a significant long-term trend in the study area. We based our research on the 36-year time series of breeding bird data in Lombardy (from 1982 to 2017), the longest data collection of breeding birds available in Italy, based on point count surveys (see the Methods section for further details). The variation in the elevational distribution of each species along the time was evaluated using a distribution model, which produced a yearly elevational distribution curve, in terms of presence probability, for a bird species along the elevational gradient. Then, we analyzed the contribution of climate and/or habitat changes in shaping the bird elevational range changes.

| Study area
The study was carried out in the Alps of Lombardy (Italy), above 600 m asl. The mountain area of Lombardy covers a large (160-km wide) portion of the Italian Alpine chain, and includes the groups of the Lepontine, Rhaetian, and Orobic (or Bergamasque) Alps. The area surface amounts to 7,596 km 2 , about 45% (3,432 km 2 ) of which is currently covered by forests, mainly located (92%) below 1,800 m asl ( Figure 1). The highest peak (Punta Perrucchetti of the Bernina Massif) reaches 4,020 m asl and, overall, about 37% of the study area (2,822 km 2 ) lays above 1,800 m asl. The climate is humid continental, characterized by a high seasonal temperature variation. In winter, the mean temperatures fall far below freezing at 2,000 m asl, while during summer, they can reach values of 10-12°C (RSY, 2015).

| Breeding bird data
Bird data for the period 1982-1988 were obtained from a dataset published by Realini (1988) concerning the mountain-breeding birds of Lombardy, above 600 m asl. This is the first and only big dataset available containing historic data of breeding birds in the Alps of Lombardy. Bird data were collected in the field using the same point count technique adopted for the long-term monitoring project of breeding birds in Lombardy started in 1992 (see below), from the last week of April to the first week of July. Since Realini bird data were published in geographical maps indicating each sampling point, we georeferenced all data. Data were found to be evenly distributed, both geographically and along the elevational gradient. Overall, they consist in 1927 point counts, but we did not considered data col- Bird data for the period 1992-2017 were obtained from the dataset of the long-term monitoring project of breeding birds in Lombardy (Bani, Massimino, Orioli, Bottoni, & Massa, 2009;Fornasari, Bani, De Carli, & Massa, 1998;Massimino, Orioli, Massa, & Bani, 2008). Data were collected using a standardized method based on the 10-min, unlimited-distance point count technique (Blondel, 1981;Fornasari et al., 1998). The technique is considered more effective in detecting bird species belonging to the orders Columbiformes, Cuculiformes, Apodiformes, Coraciiformes, Piciformes, and Passeriformes. Bird surveys were performed every year during the breeding season (10th May to 20th June), to minimize the count of migrants (birds not breeding in the study area) and to survey territorial birds. Censuses were conducted from sunrise to 11.00 a.m., only in good weather conditions, sunny to cloudy, without rain or strong wind (Bani et al., 2009). Overall, above 600 m asl (lower elevation threshold set by Realini), 5,140 point counts were performed, with a yearly mean of 233.6 (range 82-478), but no data were available for 1993, 1994, 1997, and 1998, when the long-term monitoring project of breeding birds in Lombardy was interrupted due to the lack of funding.
The two projects provided a large amount of data collected over wide areas, but since they did not rely on multiple surveys in the same season, it was impossible to account for species detection probability. However, the large dataset used for this research should overcome the potential limit of imperfect detection, reducing the noise produced by stochasticity in species discovery (Dondina et al., 2017). In addition, Fuller and Langslow (1984) highlight that 10-min F I G U R E 1 Study area (northern Lombardy). In black: area between 600 and 1,800 m asl; in gray: area above 1,800 m asl; dashed area: forest and shrub cover (source: DUSAF-2015 cartography; ERSAF, 2010). The white area lays below 600 m asl or pertains to Apennines domain (southwestern corner) and was not investigated in the present study point counts are satisfactory in detecting more than 70% of birds present at the census site and recommend how "counts exceeding 10 min are wasteful of field effort which could be used to improve other aspects of sampling", such as the number of sampling units (see also Matsuoka et al., 2014).
In the long-term monitoring project of breeding birds in Lombardy, data were collected as species abundances (number of individuals of each detected species per point count), while the Realini survey only recorded species presences in each point count. Thus, we merged the two datasets using information on species presences only.
Overall, the two projects provide 28 annual survey in the 36-year time series starting from 1982, making a total of 7,067 point counts, ranging from 600 m asl to 2,700 m asl. The choice of the elevational bounds was due to sampling design constraints, the lower one due to historical data sampling design, the upper one because above this threshold the species' presence drops abruptly in the study area.
Overall, we surveyed 113 bird species breeding in mountain and alpine habitats, for a total of 58,080 bird occurrences.

| Habitat data
Habitat data came from to the digital cartography available for the study area in order to associate it with the time series of bird data.
We wanted to evaluate the surface covered by shrubs and forests between 1982 and 2017 in the mountain area of Lombardy, above 600 m asl. For the analyses, we pooled together shrub and forest covers, since the distinction between the two land-use classes using remote information is often difficult (Laliberte et al., 2004). (1:50,000), 1999,2007,2009,2012, and 2015 (all 1:10,000; downloadable from http://www.geoportale.regione.lombardia.it/). To avoid possible bias in the evaluation of habitat data due to the different cartographies' resolution, we resampled those with higher resolution to the lower resolution.
For the elevation, we used the Digital Terrain Model of Lombardy Region, with a 20-m ground resolution (downloadable from http:// www.geoportale.regione.lombardia.it/).

| Temperature data
Temperature data were obtained from the E-OBS version 16.0 dataset with a resolution of 0.25 × 0.25 degree (see Haylock et al., 2008 for technical details), which can be downloaded for free at http:// www.ecad.eu. Following the method proposed by Moreno and Hasenauer (2016), we downscaled the E-OBS dataset of daily temperature to a 1-km ground resolution, using the WorldClim 2 map (Fick & Hijmans, 2017). By doing this, we obtained a daily high-resolution map for the annual series of temperature for the whole study area. For each of the 28 available years of breeding bird data, we built two annual temperature maps calculating the overall mean of the daily minimum temperature and the overall mean of the daily maximum temperatures recorded during the main local breeding season, respectively (May and June; see Bani et al., 2009). As the minimum and maximum temperatures were found to be strongly correlated (R = 0.907), all the analyses were performed considering only the overall mean of the daily minimum temperature of May and June.
The rationale of the selection of May and June temperatures is both (a) accounting for the effect of the yearly variability of temperatures during the breeding season on the yearly altitudinal distribution of birds (i.e., inter-annual variability), and (b) accounting for the long-term effect of temperature on the bird species altitudinal distribution (i.e., climate warming).

| Annual elevational distribution curves of species
Since the data came from different projects, we were dealing with different sampling effort along the elevation gradient among years.
This prevented a direct analysis of temporal changes occurred in the species elevational distribution, which would result biased among years according the year sampling efforts. Thus, we were forced to use a modeling approach to first assess the yearly species distribution, and subsequently, analyze its temporal trend. Indeed, for each species, we calculated the annual elevational distribution curve following the "curve response shape" method (Heegaard, 2002;Maggini et al., 2011), which defines a presence probability curve for a bird species along the elevational gradient. The curve was assessed as a smooth function (thin plate regression spline; Wood, 2017) of the elevation, other than northing and easting to account for spatial autocorrelation, using generalized additive models (Maggini et al., 2011;Wood, 2017). We set the maximum degrees of freedom to three for the elevation, in order to avoid overfitting and to obtain unimodal curves, and 15 for the space smooth (geographic coordinates; Maggini et al., 2011;Massimino et al., 2015). The analysis was performed using the mgcv package (Wood, 2003(Wood, , 2011(Wood, , 2017 (Heegaard, 2002). The two central border points were defined by the projections on the elevation axis of the value densities corresponding to OPT*exp(−0.5), while the two outer border points by the projections on the elevation axis of the value densities corresponding to OPT*exp(−2). However, as we were working within a delimited elevational range, we could deal with truncated curves when some reference points fell outside the investigated range. Indeed, for several species, the distribution curve might fall under the 600 m asl or over 2,700 m asl.
For each year, bird occurrence data were bootstrapped (n = 200) and one curve for each bootstrapped sample was fitted in order to assess the estimated variability of the reference points. We considered a reference point estimate as reliable when the fitting procedure succeeded for at least 50% of the bootstrapped samples. Then, for each species and for each of its (up to 5) reliable reference points, we built a time series (reference points' time series). Moreover, we considered the reference points having at least 10 reliable values along the time series as informative for the long-term analysis.

| Reconstruction of the habitat and climate dataset
Since an annual survey of the habitats is not expected within the study area, we built a habitat time series by interpolating the information contained in the available digital cartography. For each species, we calculated the yearly amount of shrub and forest cover within an elevational belt ranging from the minimum and the maximum value of elevation assumed by each of its reference points along the time series. Shrub and forest cover was interpolated using a thin plate regression spline, using a generalized additive model (Wood, 2017), and setting the maximum degrees of freedom to four for the year.
Then, we made a yearly prediction of shrub and forest cover, thus building a habitat time series associated with the reference points' time series (shrub and forest cover time series [km 2 ]). Although annual habitat data were unavailable, we are confident that the information obtained from their interpolation should not deviate very much from the real condition, given the inertia of dynamics involving shrub and forest changes over large areas.
As regards climate data, for each species and for each of its ref-

| Elevational range of breeding birds and its trend over time
First, we assessed the presence of a temporal trend in the species' reference points. To this aim, we fitted a weighted linear regression using the elevation value of each reference point as dependent variable, the year as independent variable, and the number of sampling units as weight. The analysis was performed using the nlme package (Pinheiro, Bates, DebRoy, & Sarkar, 2017) in R. Then, we classified the patterns of elevational range change for all species showing a significant trend in at least one of their reference points as: (a) "contraction" when the changes of the reference points brought to a distance reduction between the extreme reference points; (b) "expansion" when the changes of the reference points brought to an increase of the distance between extreme reference points; (c) "shift" when the changes of the reference points were consistent between each other. Depending on which reference points changed, we referred the observed change to the lower (OBL and CBL) or upper (CBR and OBR) reference points. If the slopes of the year pertaining to the two extreme reference points showed a significant trend with same sign but were significantly different between them, we evaluated if their change caused either a contraction or an expansion of species' range.
We have to acknowledge that reference points' estimation using Moreover, the same kind of error could arise from the impossibility to take into account the imperfect detection that, however, should be quite limited according to Fuller and Langslow (1984).

| Habitat and climate drivers of elevational range changes
In order to assess the environmental drivers of elevational distribution changes, we related the annual value of elevation of each refer- purpose, we used a weighted multiple linear regression, using the annual number of sampling points as weight. In this analysis, we did not account for the possible bias that could be induced by temporal autocorrelation in the dependent variable because this relation may be due to the obvious temporal autocorrelation in the explanatory variables. The temporal autocorrelation was tested with the Durbin-Watson statistic (Durbin & Watson, 1950), performed using the car package (Fox & Weisberg, 2011) in R. However, in very few cases, we found a temporal autocorrelation in the residuals.
In order to assess the contribution of both environmental covariates (habitat and climate), we adopted a hierarchical partitioning approach (Chevan & Sutherland, 1991), using the ecospat package  Sicurella, Orioli, Pinoli, Ambrosini, and Bani (2017) and integrated by the authors where necessary. The association between each pair of categorical variables was assessed performing a log-linear analysis (Agresti, 1996) on the observed bird species' frequencies from a contingency table obtained by crossing the pattern of elevational range change, the breeding habitat, the migration habit, and the drivers of range change.

| Descriptive statistics of habitat and climate changes
In the Alps of Lombardy (above 600 m asl), the shrub and forest cover

| Changes in the elevational range of breeding birds
Overall, we identified 125 reliable reference points for the analysis of the changes in the elevational distribution range for 29 mountainand alpine-breeding birds (Table 1). Forty-nine reference points, pertaining to 21 species, showed a significant (p < 0.05) positive trend, which means that their elevational distribution significantly moved upward along the time series. The remaining 75 reference points did not show any significant variation along the time series (none showed a significant downward shift).
Among the 21 species that changed their elevational distribution, eight showed an elevational range shrink due to a lower boundary upward contraction, four displayed an elevational upward shift, while nine showed an elevational range enlargement due to an upper boundary upward expansion.
We found a consistent upward displacement of the extreme reference points (OBL and OBR), for two edge species, the Dunnock (Prunella modularis) and the Willow Tit (Poecile montana), which should be classified as an "upward shift". However, in both cases, we classified the elevational range change as a "lower boundary contraction" since the regression coefficients (β y ; Table 1) of the OBL reference point was significantly higher than the regression coefficients of the OBR reference point, which indicates a net shrinkage of the elevational range for the two species. In fact, these two species seemed to partially compensate the loss of part of their lower range due to the disappearance of ecotones after forest and shrub encroachment by colonizing areas at higher elevations.

| Shrub and forest cover and temperature as drivers of species elevational range changes
The contributed deviance of temperature in affecting the trend of reference point varied between zero and 0.722, while that of shrub F I G U R E 2 Descriptive statistics between 1980 and 2017 of (a) the shrub and forest cover [km 2 ] and their trend (between 600 m and 1,200 and above 1,200 m asl), and (b) the average of daily minimum temperatures [°C] in the months of May and June and their trend in three elevational belts, in the Alpine areas of the Lombardy Region. For the shrub and forest cover, data were interpolated using a generalized additive model on available survey data (filled circles). For temperatures, data were derived from the E-OBS dataset. See the text for details on data sources and forest cover varied between zero and 0.526. The unexplained deviance was often small, higher than 0.3 in five cases only, and always lower than 0.567. The shared deviance (i.e., the common contributed deviance) of temperature and shrub and forest cover varied between 0.099 and 0.993, and was often high, being higher than 0.5 in most cases (35 out of 49; Table 1).
Among the 49 reference points that showed a significant positive trend along the time series, 29 were found to be affected by shrub and forest cover, 10 by the temperature, and one by both shrub and forest cover and temperature. Nine reference points were affected by a not distinguishable effect of shrub and forest cover and temperature (Table 1).
In five cases in which we found a statistical significance of the habitat driver, and in three cases in which the driver was not distinguishable, shrub and forest cover showed an opposite effect with respect to the known ecology of the species. Indeed, the upper reference points were significantly higher in elevation, the lower was the forest and shrub cover. Nevertheless, since the shrub and forest cover decreases with the elevation, the effect of the driver should be obviously spurious. Therefore, the upward displacement in elevation of the reference points could be due to other causes (temperature also showed a nonsignificant effect), which will be described in the Discussion section.
In the case of the Coal Tit (Periparus ater), the driver of the upward shift of the OBR reference point should be not distinguishable (nd) according to our hierarchical partitioning criteria. However, we identified "temperature" as a driver since the temperature effect was consistent with those of the lower reference points and the only significant covariate for the OBR reference point, while habitat cover showed an opposite effect on species ecology and was not significant in the weighted multiple linear regression (Table 1).
The log-linear analysis highlighted a significant relationship between the pattern of elevational range change of birds and (a) bird's habitat; (b) bird's migration habit; and (c) range change driver. The statistical significance of these relationships is shown in Table 2, and the proportion of species for each association is indicated in Figure 3.

| D ISCUSS I ON
The results of this research showed how the environmental transformations undergoing in mountain areas are leading to a shift toward higher elevations of a large part of the species living there (see also Rocchia et al., 2018;Ferrarini, Alatalo, & Gustin, 2017; but see Scridel et al., 2018). However, the upward shift was not only driven by temperature increase. Local processes, such as land-use changes, also appeared to play an incontrovertible and likely synergistic role in exacerbating the displacement of species toward higher elevations. Although these evidences were largely derived from information on relatively common species, it is likely that even rare species pertaining to several different taxa, which are usually more demanding in terms of climatic and habitat factors, may be moving upward to a similar extent (e.g., Pernollet et al., 2015;Maggini et al., 2011).
Most of the bird species threatened by environmental changes in mountain areas were found to be open habitat species, which are mainly pushed upward by the loss of their physical space due to shrub and forest encroachment (Chamberlain, Negro, Caprio, & Rolando, 2013). This phenomenon is chiefly linked to the land abandonment caused by the widespread loss of economic performance of mountain pastoral activities (Schermer et al., 2016). Indeed, in a previous study conducted between 2006, Rocchia, Luppi, Dondina, Orioli and Bani (2018 found that most of the breeding birds showed a significant change of their elevational range and that a large part of these birds were forest species. This pattern, as stated by the authors, is probably due to a significant upward shift of forests (i.e., forest expansion) during the period considered, since they did not find a significant positive trend for temperature.
In this research, which covered a long time period (1982-2017), we found a significant positive trend for temperatures in the Central Italian Alps, a trend that was virtually consistent at all elevation. At the same time, we also found a significant increase of shrub and forest cover, which was more conspicuous at lower elevations. Thus, the trends of the two covariates overlapped to some extent, making it difficult to disentangle the net effect of each driver on boosting the elevational changes in species distribution. However, the methodological approach adopted (i.e., hierarchical partitioning) allowed us to evaluate at least the partial (i.e., pure) contribution of each of the two drivers (i.e., the individual contribution without the joined contribution due to covariance). In addition, the availability of a long, almost continuous, time series of breeding bird data strengthened the power of statistical tests adopted in detecting the effect of the investigated drivers, even in the presence of possible stochastic noises, sources of uncertainty in bird data, and the possible effect of many other general or local factors affecting the elevational distribution of birds.
The physical habitat of an animal species is certainly a more important limiting factor compared with the local temperature, which can, however, be critical for the habitat to be colonized (Huey, 1991).
For example, the local temperature may be considered a secondary limiting factor that could affect the colonization of a new area by a forest-dwelling species, provided that there is an available forest habitat (primary limiting factor). Thus, it is not surprising that shrub and forest cover plays a primary role as a physical habitat in driving the elevational change of reference points for a large portion of the studied species. This point was widely confirmed by the upward shift of most of the lower reference points of bird species living in open or edge habitats, for which the encroachment of shrubs and forests (which is greater at lower elevations) led to a decrease of the available habitat, probably regardless of temperature increase.
Moreover, for these species, we did not observe a corresponding upward shift of the upper reference points, which has consequently led to a net shrinkage of their elevational range, with a lower boundary contraction.
Conversely, forest species showed an upward shift with a displacement of one or more of their central reference points toward higher elevations. In other cases, they even showed an upper TA B L E 1 Temporal trend of elevation of species' reference points assessed by weighted linear regressions and corresponding pattern of range change for each species; β y and p y : slope and level of significance for the year, respectively (left side of the table), and effect of forest and shrub cover and temperature as drivers of the elevation of species' reference points (right side of the Notes. Effect of forest and shrub cover and temperature as drivers of the elevation of species' reference points assessed by a weighted multiple linear regression, and relative estimation of covariates' contributed deviance assessed by hierarchical partitioning approach; β tn and p tn : slope and level of significance for the temperature, respectively; β SF and p SF : slope and level of significance for the shrub and forest cover, respectively; R adj : adjusted R-squared of the weighted multiple linear regression. Range change driver: tn: temperature; SF: shrub and forest cover; nd: driver not distinguishable; dde: density-dependent effect (in brackets in the "range change driver" formerly deduced by the hierarchical partition approach). a The tn was identified as the driver of the OBR reference point for the Coal Tit although should be set to nd following the criteria of the hierarchical partitioning approach (see the text for the explanation). * Significant temporal autocorrelation at Durbin-Watson test. As expected, the habitat played the role of main driver. However, even the contribution of temperature was not negligible, especially for forest species, whose shift or expansion toward higher elevations was statistically significant. Temperature probably led to an improvement of climatic conditions in preexistent forest areas, which were previously not suitable. Indeed, the temperature driver was found to be more active in the upper part of the elevational range than the shrub and forest cover driver, which can be expected for species that usually inhabit lower and medium-elevation forests, Our research showed that the elevation increase of the upper reference points was in contrast with the ecology of forest species in several cases. In fact, it is important to note how, in those cases, the elevation of the upper reference points was inversely related to shrub and forest cover, without a positive temperature effect on the colonization of potentially suitable habitats at higher elevations.
This pattern was recurrent, especially for the most common forest species, and seems to indicate that a specific process is in act. An upward shift in the reference points at lower elevations, induced either by an increase in habitat availability or by an improvement in the climatic condition due to temperature increase (which corresponds to a likely increase in population density), seemed to determine a sort of density-dependent effect that may cause a colonization of even Species living in open habitats could also colonize preexisting suitable vacant areas at higher elevations, provided that these fall within their climate niche. However, many of these species had a truncated elevational distribution (i.e., lack of higher reference points), which suggested that they were already at the elevational limit reached by their habitats. Indeed, habitat formation is a slow process, especially at higher elevations, since the previous soil formation is likely to be slower than the speed at which the birds' upward displacements could theoretically progress under environmental pressures (Edwards, Scalenghe, & Freppaz, 2007;Freppaz, Filippa, Caimi, Buffa, & Zanini, 2010;Körner, 2003).
Each of the observed patterns of range change was strictly associated with a particular bird ecological group and depended on a specific driver. Forest birds were found to colonize higher elevations shifting their range or expanding upwards, while birds living along edges or in open habitats suffered a contraction of their elevational range due to the shrinkage of the lower boundary. When effective, temperature was found to be the main driver affecting range expansion toward higher elevations for forest species. For many species suffering a range contraction, it was not possible to ascertain a pure contribution of one of the two investigated drivers, although a large portion of these species was found to be affected by shrub and forest encroachment. Finally, long-distance migrants were all affected by a range contraction. Such a particularly unfavorable situation for these migratory birds can be emphasized by the "trophic mismatch", as highlighted by different researches (Bairlein & Hüppop, 2004;Both et al., 2009;Pearce-Higgins, Eglington, Martay, & Chamberlain, 2015).
In sum, forest species were clearly the winners in a context of shrub and forest encroachment in mountain areas, besides being fa- Alps over a 29-year period (Pernollet et al., 2015). However, these patterns are not universal and are probably more difficult to observe in shorter time periods (e.g., Archaux, 2004, Popy, Bordignon, & Prodon, 2010. Other taxa, usually represented by slow-reacting organisms, such as tree species and bryophytes, also moved upward Both habitat transformation and climate warming are harsh threatening factors to biodiversity across the world, but while the first can be thwarted effectively from a local to a medium scale, the second requires shared and harmonized policies at a global scale (Goodstein, 2007).
In conclusion, how can we promptly and effectively counteract the negative trend showed by edge and open habitat birds in mountain areas?
The conservation of open mountain habitats (grassland and prairies) is strictly linked to local ad hoc management practices that should also be implemented through several economic activities (e.g., marketing of local typical products, recreational outdoor activities, and food services) that depend on open habitats, even within the wide system of protected areas of the Central Italian Alps (Sicurella et al., 2017). Of course, these for-profit activities should be addressed toward sustainable forms by adequate governance policies, in order to maintain a viable mountain economy in the longterm, which should also be perceived as an attractive employment opportunity and produce a positive and broader perception in terms of ecosystem services (Schermer et al., 2016).

ACK N OWLED G M ENTS
We thank the General Directorate for Agriculture of the Lombardy Region (D.G. Agricoltura della Regione Lombardia) and the Regional Professor Renato Massa, who first conceived and started the monitoring project of breeding birds in Lombardy.

CO N FLI C T O F I NTE R E S T
None declared.