Introduction
Habitat selection is a central theme in ecology and conservation biology, since knowledge of the habitat requirements of a species is fundamental to its effective conservation and management. Community interactions and environmental conditions usually vary throughout a species’ range (Guo et al. Reference Guo, Taper, Schoenberger and Brandle2005), and thus it is advisable to assess habitat preferences in the area where the conservation measures are required (Quevedo et al. Reference Quevedo, Bañuelos, Sáez and Obeso2006). More specifically, peripheral populations, considered as those located close to the edge of a species’ distribution, often experience less favourable environmental conditions and are smaller, making them more prone to extinction (Sagarin et al. Reference Sagarin, Gaines and Gaylord2006, Alda et al. Reference Alda, González, Olea, Ena, Godinho and Drovetski2013).
Peripheral populations might exhibit different genetic features, which deserve particular consideration in conservation planning as evolutionarily significant units (Lesica and Allendorf Reference Lesica and Allendorf1995, Dai and Fu Reference Dai and Fu2011), particularly important for maximising adaptive diversity (Thomassen et al. Reference Thomassen, Buermann, Milá, Graham, Cameron, Schneider, Pollinger, Saatchi, Wayne and Smith2010). Moreover, many populations at the edge of a species’ range tend to occur in marginal and atypical habitats, which may lead to local adaptations to particular habitat types (see review in Kawecki Reference Kawecki2008). Indeed, the variability of responses by species throughout their ranges has led to considerable interest in the geographical variation in species distribution models and the assessment of their potential transferability among different areas (Randin et al. Reference Randin, Dirnbock, Dullinger, Zimmerman, Zapper and Guisan2006, Torres et al. Reference Torres, Sutton, Thompson, Delord, Weimerskirch, Sagar, Sommer, Dilley, Ryan and Phillips2015). In this regard, conservation actions based on information obtained from the core area of a species’ range are not always applicable, or may even be counterproductive, at the edges of the range (Hampe and Petit Reference Hampe and Petit2005). In sum, peripheral populations may require specific conservation assessments.
Dupont’s Lark Chersophilus duponti is a habitat specialist that mainly inhabits the shrub-steppes and pseudo-steppes of Europe and North Africa (Suárez Reference Suárez2010). In Europe, it is restricted to flat and open areas in Spain, characterised by sparse small shrubs and bare ground (Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005, Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006). Its populations have been reduced by over 50% in the last two decades to only about 2,200 pairs and several populations were completely extinguished in this period (Tella et al. Reference Tella, Vögeli, Serrano and Carrete2005, Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2014). The species is classified as ‘Near Threatened’ on the IUCN Red List due to its large distribution range (BirdLife International 2015). In Spain it is currently classified as ‘Vulnerable’. The main agents responsible for the decline are habitat loss and fragmentation of natural steppes and a much-reduced grazing pressure (which facilitates shrub and tree encroachment of pseudo-steppes; Suárez Reference Suárez2010). Its populations in eastern and southern Spain are small and far away from the core of the species’ distribution, located in the Ebro Valley and moorlands in the northern plateau (Figure S1 in the online supplementary material). These populations show ecological differentiation associated with their settlement in marginal and atypical habitats, such as alfa grass Stipa tenacissima and rosemary Rosmarinus officinalis steppes (Suárez Reference Suárez2010, Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013), instead of broom Genista spp. or thyme Thymus spp. steppes, which constitute the main habitat in the species’ core area (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006, Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010).
Previous habitat selection studies carried out in the core area of the species in Europe indicate that this lark is a habitat specialist sensitive to patch size and vegetation structure (Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005, Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006, Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010). Such detailed evaluation is lacking for African populations (but see García et al. Reference García, Suárez, Garza, Justribó, Oñate, Hervás, Calero and García de la Morena2008). Thus, research on Spanish peripheral populations of Dupont’s Lark could be a good case study to understand habitat preferences of the species at the limit of its range.
With this motivation, we aimed to determine the influence of environmental predictors on the breeding spatial distribution of Dupont’s Lark at three different scales (microhabitat, macrohabitat and patch scales) in three isolated and declining populations located in the south of the species’ main distribution area in Spain. Multi-scale approaches to identifying habitat preferences of the species are preferable to single-scale studies because they provide a better overview of the gamut of factors determining the presence of a species, and thus better criteria to define effective conservation measures (Leopold and Hess Reference Leopold and Hess2013, Traba et al. Reference Traba, Acebes, Malo, García, Carriles, Radi and Znari2013).
We built and transferred habitat models among sites and between sampling years, in order to identify those predictors that affect the occurrence of the study species and whether or not their effects are constant or differ among study sites. The evaluation of the transferability of distribution models among populations is important in deciding whether conservation guidelines can be applied over multiple regions (Vernier et al. Reference Vernier, Schmiegelow, Hannon and Cumming2008, Zanini et al. Reference Zanini, Pellet and Schmidt2009, Torres et al. Reference Torres, Sutton, Thompson, Delord, Weimerskirch, Sagar, Sommer, Dilley, Ryan and Phillips2015). Finally, we provide information for the first time about the habitat selection of this lark in two atypical habitats, including rosemary steppe, a habitat type not previously studied and rarely used by the species, and wooded moorland (Suárez Reference Suárez2010). We use our results to identify the conservation requirements of Dupont’s Lark across spatial scales.
Materials and methods
Study area
The study area comprised 13 sites (1,457 ha overall) located in the Rincón de Ademuz region, in the eastern Iberian Peninsula (40°04’N, 1°20’W; Figure 1). These comprised all the potential sites in the region that the species may inhabit, and seven of them are declared as Special Protection Areas (SPA; 1,057 ha, Fig. 1). Potential sites were defined as the areas in which the Dupont’s Lark was found in previous studies or all areas in the Rincón de Ademuz that presented flat relief, low scrub and had an extent larger than 10 ha. Sites were surrounded by an unsuitable matrix of forests and agricultural areas (Pérez-Granados et al. Reference Pérez-Granados, López-Iborra, Serrano-Davies, Noguerales, Garza, Justribó and Suárez2013). Nine sites had been occupied by the species over the last two decades, but currently Dupont’s Lark is present in only five of them (Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013). This area is located at an average of 1,100 m. asl, and has a mean annual temperature of about 10°C and a mean annual precipitation of 550 mm. The area exhibits an undulating topography with large flat areas dedicated to crops, especially almond trees Prunus dulcis, or scrub. Natural vegetation consists mainly of open and sparse shrub communities dominated by thyme, broom Genista scorpius and rosemary, with some interspersed pines Pinus spp. and junipers Juniperus spp.
Figure 1. Location of the study area in Rincón de Ademuz (Valencia). Habitat patches identified by name. The inset shows location of the study area (star) in the Iberian Peninsula. Special Protection Areas (SPAs) are marked with black stripes.
Bird data
We surveyed bird abundance from March to June 2011 and 2012 with the territory mapping method over 53–60 sampling days each year (Bibby et al. Reference Bibby, Burgess, Hill and Mustoe2000). Unoccupied sites were visited three times and occupied ones were visited at least five times each year, since four visits are required to detect 98% of the territories within a site (Pérez-Granados and López-Iborra in press). The walking of plots was scheduled to begin at different places on consecutive visits to reduce biases related to time of day or season, and the routes were designed to cover plots completely (considering a 500-m maximum detection band on each side of the observer). Locations of all birds detected on different days were mapped with ArcGIS 9.3 (ESRI 2008) and a territory was defined when two records of a singing male were made within a radius of 100 m, which corresponds to the radius of a circle comprising the average territory core area (3.7 ha) estimated for the species during the breeding season (104.3 m in Layna moorland; Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005). We determined the centre of each territory after averaging the locations of clumps of registrations located within a radius of 100 m. Censuses were carried out at dawn by the same person (CPG) walking at a constant and slow speed (1–3 km/h) on dry and windless days. The survey was repeated in 2013 and 2014 to test for temporal transferability of models (see below).
Habitat data
Microhabitat scale (focal points)
The microhabitat structure of the points used by the species during the breeding season across all visits was characterised from May to June in 2011 and 2012 in the three main sites of the study area (Hontanar, Losar 1 and Pinar 1; Figure 1). We selected this subset of the currently occupied sites within the study area because the number of males in the other two occupied sites (Cerrillo 1 and Losar 3) was too low for statistical analyses (mean number of males during the study period was two and five, respectively), despite the large patches they inhabited (125 and 59 ha, respectively). Sites were analysed independently, because they differed in aspects of vegetation that we believed critical for Dupont’s Lark. Hontanar was a scrubland dominated by dwarf thyme shrubs with a few trees interspersed, while Pinar 1 was similar but has a much greater density of trees in some areas than the species tolerates in the rest of its distribution range (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006, Suárez, Reference Suárez2010). In contrast, Losar 1 was a steppe dominated by rosemary rather than thyme shrubs.
In each site, microhabitat features were measured in 25-m radius circles centred on lark locations (“lark points”) and in an equal number of control locations (“random points”) selected at the end of each breeding season and constrained to be at least 50 m from lark points and to not overlap between years. Within each of those circles habitat variables were estimated in nine 10-m diameter sample circles (Fig. 2) and averaged. For this purpose, and given the difficulty of precisely locating this species by ear, we only considered birds detected closer than 25 m to the observer.
Figure 2. Sampling procedure at microhabitat scale. Predictor variables (see details) were measured in nine circles of 5 m radius located in the centre, at 10 and 20 m from the centre of each sampling point in cardinal directions.
In each 10-m diameter sample circle, we recorded the cover of bare rock, pebble, bare soil, herbaceous plants and scrubs, distinguishing thyme, broom and rosemary (following methods in Prodon and Lebreton Reference Prodon and Lebreton1981). We estimated the cover of grasses and each shrub species considering separately three height categories: low (< 20 cm), medium (20–40 cm), and high (> 40 cm), and mean and maximum vegetation height (see Pérez-Granados et al. Reference Pérez-Granados, López-Iborra, Serrano-Davies, Noguerales, Garza, Justribó and Suárez2013). We also recorded the slope of terrain in each 10-m circle (in degrees) with an inclinometer (CJQ-1, G.I.S. Iberica, Cáceres; 1° accuracy). Finally, on the total surface area of the 25-m radius circle we counted the number of trees and visually estimated the percentage covered by crops, after previous training.
Macrohabitat scale (home ranges)
At a larger spatial scale, better suited to describe the individual home ranges (macrohabitat scale, hereafter), we divided the whole area of the three selected sites into a grid of 100 x 100 m squares. Each square was surveyed from May to June 2012 to record the cover of crops, bare soil, grass, open scrub formations of rosemary, thyme or broom (visually estimated percentages) and the number of trees. We distinguished between pines and junipers because of their potentially different effects on Dupont’s Lark according to their morphological structure (pines tend to be taller while junipers have denser foliage), and because pine removal may be authorised as part of habitat management action while junipers are protected under Habitat Directive 92/43/CEE. We further calculated slopes for the 1-ha squares through a Digital Elevation Model obtained from 1:25.000 topographic maps (MDT25) provided by the National Geographic Institute. The mapping of territories carried out in both years of study allowed us to identify occupied and unoccupied squares in each site. A square was considered occupied when at least one territory centre was located within it in any study year.
Patch scale (occupied/unoccupied sites)
We compared habitat variables, from patches occupied or not, in order to identify the main factors that may be related to Dupont’s Lark occurrence (assuming that populations at the worst sites disappeared first and the best sites are the last to remain occupied). A patch was considered occupied when at least one Dupont’s Lark territory was registered within in any study year. Patch scale habitat data were collected from March to June 2011 in the 13 sites in the study area (Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013; Figure 1). We surveyed 15 randomly selected 25-m radius circles in sites larger than 100 ha, and only 10 in smaller sites. In each of these circles we used the same methodology to estimate the same variables as at the microhabitat scale, except for crop cover and slope. Crop cover was estimated using the freely available images from the Spanish National Aerial Orthophotography Plan (PNOA), while slope was calculated through the digital elevation model used for the macrohabitat scale. These data were recorded at random points selected independently from those at the microhabitat scale. PNOA images were also used to calculate site size, considered as the area of natural vegetation potentially adequate for the species, and the distance among patches to the nearest occupied ones.
Statistical analyses
We studied microhabitat preferences by comparing lark points with random points (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006). Data from both study years were pooled in this analysis since we consider that the effect of lack of independence in these data should be negligible. Lark points considered for each year to estimate habitat use were spaced at least by 50 m and their number (around 20 per year, see Table 1) is less than the double the number of territories in each site (11–13, Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013). Thus, and given that most, but not all, birds were identified as males, a given individual would have contributed on average with less than two lark points per year. In addition, data were recorded along the whole breeding season, which means that the eventual duplicate records for a bird would be separated by weeks or months, and then these separate records may be considered as independent events of habitat choice. For records obtained in different years of study, it should be taken into account that only about a half of the adult birds may survive between successive years (Laiolo et al. Reference Laiolo, Vögeli, Serrano and Tella2007), which would further decrease the effect of non-independence of data. We tested differences between sampling points using one-way analysis of variance at each site, with P-values adjusted using the False Discovery Rate (FDR, Benjamini and Hochberg Reference Benjamini and Hochberg1995). To identify the main vegetation gradients at this scale we applied a varimax rotated Principal Components Analysis (PCA; Dormann et al. Reference Dormann, Elith, Bacher, Buchmann, Carl, Carré, García Marquéz, Gruber, Lafourcade, Leitao, Münkemüller, McClean, Osborne, Reineking, Schröder, Skidmore, Zurell and Lautenbach2013, Pérez-Granados et al. Reference Pérez-Granados, López-Iborra, Serrano-Davies, Noguerales, Garza, Justribó and Suárez2013) to the lithological and natural vegetation cover variables (arcsine transformed) of the three sites together. We selected the axes of this PCA that had eigenvalues higher than one, and used them as explanatory factors in subsequent analyses. At the macrohabitat scale we tested the differences between environmental variables in occupied and unoccupied squares in each site by one-way analysis of variance corrected by FDR.
Table 1. Mean ± SD of the microhabitat variables measured in lark and random points in the three main populations of the study area. Number of points of each type are shown in parentheses. The p column shows the significance level of the One-way ANOVA tests when difference among points was significant (*P < 0.05, **P < 0.01, ***P < 0.001). Significances were calculated from One-way ANOVA analysis in each site, corrected by False Discovery Rate.
At each site we used Hierarchical Partitioning (HP) analyses to identify the environmental variables with greater influence on habitat use (microhabitat scale) and home ranges (macrohabitat scale) during the breeding season. This method allowed us to identify those predictors with an important independent – as opposed to partial – relationship with the occurrence of Dupont’s Lark in one point or square (Mac Nally Reference Mac Nally2002; and see applications in bird studies in López-Iborra et al. Reference López-Iborra, Limiñana, Pavón and Martínez-Pérez2011, Reis et al. Reference Reis, López-Iborra and Pinheiro2012). We used logistic regression and log-likelihood as the goodness-of-fit measure and tested statistical significances of the independent contribution of each variable by a randomization procedure with 999 bootstraps (Mac Nally Reference Mac Nally2002). We reduced the potential impact of collinearity among variables by removing those with Spearman rank correlation coefficients above a cut-off of |0.7| (Randin et al. Reference Randin, Dirnbock, Dullinger, Zimmerman, Zapper and Guisan2006, Dormann et al. Reference Dormann, Elith, Bacher, Buchmann, Carl, Carré, García Marquéz, Gruber, Lafourcade, Leitao, Münkemüller, McClean, Osborne, Reineking, Schröder, Skidmore, Zurell and Lautenbach2013).
To take into account spatial autocorrelation in our HP analyses we included an autocovariate in analyses at microhabitat and macrohabitat scales (Lichstein et al. Reference Lichstein, Simons, Shriner and Franzreb2002). At the microhabitat scale, we choose the radius of a circle including the average territory core area reported for the species during the breeding season (about 100 m according to Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005) and we quantified the autocovariate as the total number of lark points within a 100-m radius buffer around a lark or random point. At the macrohabitat scale we estimated the autocovariate as the number of 1-ha squares occupied by the species among the eight neighbouring squares around each grid cell (Segurado et al. Reference Segurado, Araújo and Kunin2006).
To evaluate whether there is regional or temporal variation in habitat preferences, taking into account model uncertainty, we built habitat models for each site at the macrohabitat scale following a model averaging strategy (Cade Reference Cade2015). We addressed the transferability of habitat models in space and time by both a spatial and a temporal cross-validation strategy. We used presence/absence of the species in a 100 x 100 m square as a dependent variable and habitat measurements as potential predictor variables. First, we ranked the possible models according to their Akaike’s information criterion adjusted for small samples (AICc) and selected the subset of models having ΔAICc ≤ 2 (Burnham and Anderson Reference Burnham and Anderson2002). Then, we averaged the suitability estimates for each square weighting the predictions of the models by their AICc (Cade Reference Cade2015). We used these predictions to assess the discrimination ability of the models using AUC (Sing et al. Reference Sing, Sander, Beerenwinkel and Lengauer2005). Large AUC values are associated with higher estimation-observation agreement (Fielding and Bell Reference Fielding and Bell1997). For the spatial cross-validation, we applied the set of models developed for each site to the data sets of the other populations. For the temporal validation, we applied models built with presence-absence data collected during 2011–2012 to a set of data of the same site collected during 2013–2014 by the same observer.
At patch scale we employed Generalized Linear Models (GLM, binomial error distribution; logit-link function) to test the relationship of selected environmental predictors estimated at the site level to species occurrence (presence/absence at each site as a dependent variable). Due to the low sample size (n = 13 potential sites), models were kept simple (just building univariate models). Models were ranked according to their AICc, and we considered a concrete variable to have an effect on Dupont’s Lark occurrence when the model including this variable had an AICc at least two units lower than the null model (Burnham and Anderson Reference Burnham and Anderson2002). Given previous knowledge of the species, we hypothesised that the probability of presence of Dupont’s Lark at a given site may be positively affected by size (larger sites are more likely to be occupied; Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010), and negatively by slope (since the species is largely cursorial; Suárez Reference Suárez2010), population isolation (because of the species’ very low annual dispersal movements; Laiolo et al. Reference Laiolo, Vögeli, Serrano and Tella2007) and tree cover (which the species avoids even if the trees are small; Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006). In addition, habitat structure has an important effect on Dupont’s Lark presence (Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005, Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006), so we developed univariate models for each scrub species, which differ in their height, foliage density and main cover (Dupont’s Lark prefer the smaller and sparser thyme to the other species). Similarly, we hypothesised that herbaceous and bare soil cover could be related to lark occurrence by affecting foraging success (we expected to find the species in patches with lower herbaceous cover and higher bare soil cover).
We conducted the data analysis with R 3.1.1 (R Development Core Team 2009) using specialised packages when needed (hier.part for HP, Walsh and Mac Nally Reference Walsh and Mac Nally2008; MASS for AICc estimates; Venables and Ripley Reference Venables and Ripley2002; MuMIn for model averaging; Barton Reference Barton2011; and ROCR for AUC calculations; Sing et al. Reference Sing, Sander, Beerenwinkel and Lengauer2005). We conducted GIS analysis (random point placement, surface covered by crops, Digital Elevation Model and distance to nearest occupied patch) in ArcGIS 9.3 (ESRI 2008).
Results
Microhabitat scale (focal points)
Dupont’s Lark microhabitat preferences were similar among sites, and lark points differed from random points in a number of predictor variables (Table 1). In the three study sites the slope, vegetation height and number of trees were significantly lower in lark than in random points (Table 1). Larks were only detected in points with slopes lower than 7.7° and with a low number of trees (a maximum of eight trees per sampling point). Likewise, larks were never detected in crops in any of the studied populations (Table 1), and tended to occupy areas with higher cover of small-shrubs (Table 1). This general pattern was independent of the vegetation type of the study sites.
Soil and vegetation characteristics of the study sites were summarised in four PCA axes (69.1% of total variance of data; see Table S1). The first axis (PCA I; 31.0 % of the variance) contrasted rosemary-dominated scrub with broom or thyme communities. The second (PCA II; 14.1% of the variance) describes a gradient of grass cover, while the third axis (PCA III; 14.4 % of data variability) defined a gradient of broom cover. The fourth axis (PCA IV, 9.6% of the variance) distinguished areas dominated by bare rock and vegetated areas.
PCA II and crops showed a high correlation (Rs = 0.72). Thus, we excluded crops in HP analyses due to their low presence in the study area (Table 1). Slope, PCA I and PCAII Axis, number of trees and the spatial term had a significant and similar effect on Dupont’s Lark territory selection among study sites (Table 2). However, the independent contribution of the remaining predictors varied greatly among populations (Table 2).
Table 2. Habitat variables related to species preference according to hierarchical partitioning (HP) analyses at the microhabitat scale. PCA axes come from a varimax rotated Principal Components Analysis performed with the lithological and vegetation variables. Individual contribution of each variable is shown as a percentage (I%) of the total deviance explained by the variables. The sign of the effects are obtained from univariate regression models. The z-test column shows the significance level of the randomization tests for the independent contributions (*P < 0.05, **P < 0.01, ***P < 0.001). % Dev is the percentage of deviance accounted by a logistic regression model including all variables.
Macrohabitat scale (home ranges)
In general, the species inhabited non-cultivated squares with gentler slopes, more small chamaephytes (thyme) and fewer trees, independent of the vegetation type of the study sites (Table 3). However, HP analyses showed less consistency among study sites at this scale than at a microhabitat scale, since only the number of pines and the spatial term was significant in all sites (Table 4).
Table 3. Mean ± SD of the habitat features measured in occupied and unoccupied 1-ha squares (macrohabitat scale) and habitat patches (landscape scale) by Dupont’s Lark in the three main populations of Valencia. At the macrohabitat scale the p column shows the significance level of the One-way ANOVA tests, corrected by False Discovery Rate, when difference among type 1-ha square was significant, whereas at the landscape scale P shows the significance of the logistic regression models. In both cases (*P < 0.05, **P < 0.01, ***P < 0.001).
Table 4. Habitat variables related to species distribution according to hierarchical partitioning (HP) analyses at a macrohabitat scale. The individual contribution of each variable is shown as a percentage (%I) of the total deviance explained by the variables. The signs of the effects are obtained from univariate regression models. In analyses that included more than 9 variables (Hontanar and Losar), with results sensitive to the ordering of variables (see Olea et al. Reference Olea, Mateo-Tomás and de Frutos2010), the table shows the mean of %I and mean of variable ranking calculated from 10 HP runs with variables randomly ordered, while in Pinar only 1 HP was run since it was not sensitive to the ordering of variables. Numbers in parentheses show the range found in the 10 runs. The z-test column shows the significance level of the randomization tests for the independent contributions (*P < 0.05, **P < 0.01, ***P < 0.001). Rosemary % was not used in the Pinar analysis since this shrub was lacking this area. % Dev is the percentage of deviance accounted by a logistic regression model including all variables.
A large number of similarly plausible models was found for each site. The number of candidate models and the number of times each predictor was selected in those models for each site can be found in Table S2. Predictions agreed with the observed occurrence according to AUC (Fig. 3), and their transferabilities (as measured by AUC) were very high both in space and time (Fig. 3).
Figure 3. Summary of AUC values of the plausible models for each study site. Arrows indicate the direction of the spatial cross-validation of the models to the other sites, while brackets indicate the temporal transferability of the models calculated with an independent set of data collected during 2013–2014 at the same sites.
Patch scale (occupied/unoccupied sites)
Only one univariate model for the probability of occurrence across sites was clearly better than the null model (AICc = 19.3). This model describes a positive relationship between site size (AICc = 17.1) and Dupont’s Lark presence. Dupont’s Larks were detected on average in patches two times larger than unoccupied ones, with a minimum site size of 59 ha (Table 3).
Discussion
We found that Dupont’s Lark show similar habitat preferences between the core area and peripheral populations at all scales analysed. Slope degree, density of trees and crop cover seem to be good environmental predictors in both population types, filtering the distribution of the species throughout its range (Garza et al. Reference Garza, Suárez, Herranz, Traba, García de la Morena, Morales, González and Castañeda2005, Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006). It is revealing that there were similar fine-grained preferences for focal points, even among sites whose habitat configuration differed notably. In our view, this indicates that birds are able to find their preferred niche within arguably atypical habitats (e.g. rosemary shrubs and sparse woodlands), which contrasts with the classical view of Dupont’s Lark as a strict low-shrub steppe specialist.
Along these lines, an enlightening result of our study is the first description of the microhabitat preferences of Dupont’s Lark in rosemary steppe. Rosemary is a well-distributed shrub species in Spain (López Reference López2007). However, there are only five known populations of Dupont’s Lark in this habitat type, most of them located in peripheral areas where the species cannot find more typical habitats (Suárez Reference Suárez2010). Our results suggest that the species successfully occupies alternative habitats, similar to those preferred in its core areas, by exploiting favourable microhabitat features. Specifically, the species had a strong preference for thyme-dominated areas even at sites where rosemary was the dominant vegetation.
Dupont’s Lark is reported to be intolerant of trees in its core areas in the Iberian Peninsula (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006, Suárez Reference Suárez2010). Thus, we were surprised to find that the larks completely avoided trees at focal points, but they tolerated a higher density of trees within their home ranges than previously described. However, we attribute this tolerance to the clumped spatial distribution of trees, which allows the birds to exploit the open areas found within home ranges. The natural encroachment of the study sites –which were formerly heavily grazed – is likely to be squeezing the regional distribution of Dupont’s Larks into the ever-diminishing natural and semi-natural open areas. Indeed, and contrary to our assumptions, the abundance of juniper trees had a positive effect on the species’ distribution in the wooded moorland. We think that this is due to the fact that junipers segregate spatially from pines, which are taller and strictly avoided by larks, and thus male territories end up in areas with a relatively high density of junipers.
The high habitat model performance and transferability among sites provide evidence against local habitat specialisation in the studied peripheral populations. Similarly, transferability across time suggests inter-annual stability of habitat preferences. Thus, we think our models would likely be useful for conservation planning and management over a larger geographic area, including reserve design and assessment (Guisan and Thuiller Reference Guisan and Thuiller2005, Tuanmu et al. Reference Tuanmu, Viña, Roloff, Liu, Ouyang, Zhang and Liu2011). Such a high transferability could be favoured by the aggregated distribution of the species, since clumped occurrence patterns make it easier to predict spatial distributions than sparse ones (Vallecillo et al. Reference Vallecillo, Brotons and Osborne2010). If so, we suspect that our model performance would decrease if transferred to areas inhabited by Dupont’s Larks in other climatic regions, with less aggregated distribution or different land uses (Pearson et al. Reference Pearson, Dawson and Liu2004, Luoto et al. Reference Luoto, Pöyry, Heikkinen and Saarinen2005).
At the larger scale, the size of habitat patches was the only relevant variable explaining current Dupont’s Lark distribution, as has been found elsewhere (e.g. Nogués-Bravo and Aguirre Reference Nogués-Bravo and Aguirre2006, Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010). This is not surprising given that our studied populations are declining and occupy a highly fragmented and isolated landscape (Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013). Indeed, all of the studied plots might fulfil the fundamental habitat requirements of the species, and therefore those currently unoccupied sites could be colonised in future if the total population grows (Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010).
Conservation implications
Multi-scale habitat studies offer important tools for identifying requirements of rare and elusive species at different scales (Razgour et al. Reference Razgour, Hanmer and Jones2011), which are essential to guide conservation management practices (Gonthier et al. Reference Gonthier, Ennis, Farinas, Hsieh, Iverson, Batáry, Rudolphi, Tscharntke, Cardinale and Perfecto2014). Different processes may drive the population trends of endangered species at the different scales, and thus specific strategies should be adopted at each scale (Marty et al. Reference Marty, Daeden, Moutet, Vogiatzakis, Mathevet, Potts, Tzanopoulos, Henle., Potts., Kunin., Matsinos., Similä., Pantis., Grobelnik., Penev and Settele2014). Our patch scale results show that less than half of the potential habitat patches were occupied by Dupont’s Lark, and thus a first objective must be to increase the number of patches occupied. Site size is related to occupation probability, but crop development and shrub and tree encroachment are ongoing processes in the study area that are reducing patch sizes to critical thresholds. Indeed, those phenomena seem to have resulted in species extinction in some previously occupied patches in the study area (Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2013). Thus, our main recommendation at this scale is to avoid further reduction in the size of the remnant patches and to study the possibility of returning some of them to their previous extension. Interestingly, the distance to occupied patches has not been found to have a significant effect on the species’ occurrence, despite the fact that Dupont’s Lark is considered to be a poor disperser (Laiolo et al. 2008). Therefore, facilitating connectivity among patches, which has been commonly proposed to ensure the viability of the species (Tella et al. Reference Tella, Vögeli, Serrano and Carrete2005, Suárez Reference Suárez2010, Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010), does not appears to be a priority in this region, and the emphasis should be placed on increasing the extent of the patches.
Population viability is related to its size (Mills Reference Mills2007), and thus habitat should be managed to favour the settlement of new territories, especially within the larger sites. Slope is an important environmental predictor at both macrohabitat and microhabitat scales. Slope is usually out of the reach of habitat management actions, but should be taken into account in proposed management measures only in areas with the appropriate slope. The vegetation feature most important at the macrohabitat scale is the density of trees, especially pines. Thus, we expect that reducing their number, at least to a density similar to that found in occupied squares in our study patch with more pines (five pines/ha), should increase the habitat suitability for the species, and thus the number of potential territories within a patch. Likewise, reforestation should be prohibited within all potential sites, since it would be detrimental for the species (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006).
At the microhabitat scale, vegetation management may act to increase the availability and quality of preferred places for feeding, nesting, etc. At this scale our results have shown that although the species may tolerate rosemary, it favours areas dominated by small and medium-sized thymes, and to a lesser extent small brooms. Therefore, these features could be promoted by clearing specific scrubs or promoting traditional low-level extensive grazing. In broad terms, our recommendations coincide with those proposed for core populations at specific scales elsewhere (Seoane et al. Reference Seoane, Justribó, García, Retamar, Rabadán and Atienza2006, Vögeli et al. Reference Vögeli, Serrano, Pacios and Tella2010). Unfortunately, to our knowledge, these measures have never or rarely been put into practice (own data) and the decline of the Dupont’s Lark populations continues in Spain (Pérez-Granados and López-Iborra Reference Pérez-Granados and López-Iborra2014). Our multi-scale analysis of habitat selection has provided a multi-scale strategy for planning the conservation of this species that should be tested as soon as possible in the field. Likewise, this approach may also be useful for designing conservation strategies in other species.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0959270916000356
Acknowledgements
This study was supported by a pre-doctoral fellowship, financed by Levantina y Asociados Minerales, S.A. within the project “Estudios aplicados a la conservación de las poblaciones de alondra ricotí (Chersophilus duponti) en el entorno del municipio de Vallanca”. We are grateful to Eva Serrano-Davies for her collaboration during field-work. We also thank two anonymous referees and Paul Donald whose helpful comments contributed to improving the paper. We wish to thank the Servicio de Vida Silvestre of Conselleria D´infraestructures, Territori i Medi Ambient (Generalitat Valenciana) for supporting our work and to Sarah Young for help with the English manuscript. This research is part of the doctoral thesis of C.P.G. at the University of Alicante.
