study areas and sound recording

We recorded Dupont's larks throughout Spain in March–June 2004, and in Morocco at the end of February 2005 during the breeding period. Males were recorded in their territories from 1 to 1·5 h before until 1–2 h after dawn. We used a TC-D8 DAT recorder (Sony Corporation, Tokyo, Japan) and a ME67 microphone (Sennheiser, Berlin, Germany) to record birds, and a GPS eTrex(r) Navigator (Garmin, Olathe, KS, USA) to establish male positions during singing bouts.

We recorded 326 singing males, from which we selected a subsample of localities with at least seven birds and where each individual repeated its complete song at least three times (to record the complete song repertoire). Hence, sample size was 204 birds from 20 Spanish localities (196 birds, 7–18 individuals per locality) and one Moroccan site (eight birds) (see Appendix S1). In Spain, the species occurs in patches or groups of nearby patches in each locality (frequently hosting < 10 territories), isolated from other localities by a large matrix of unsuitable habitat. In the Moroccan study area, steppeland cover was more continuous, although locally degraded by overgrazing.

During dawn choruses, males from neighbouring territories approached each other to counter-sing along territory boundaries. Interacting neighbours reduced interindividual distances, often showing a clumped distribution. Singing clusters comprised up to seven birds but birds also sang alone. Males from adjacent territories interacting during bouts of counter-singing were considered as members of the same song cluster (neighbours), for which the mean pairwise distance was 201 m ± 28 SE. Conversely, pairs of males from different singing clusters or birds alone were classified as non-neighbours (mean pairwise distance 1914 ± 363 m).

sound analyses

The Dupont's lark song unit is made up of several discrete sequences (three to 11 per individual); song sequences (types) are made up of one to 13 notes (Fig. 1). The most common sequence of Dupont's lark song is the so-called ‘whee-ur-wheeee’ (Cramp 1988) (Fig. 2). This sequence is composed of a number of notes that varies among localities (from four to 10) but always ends with a characteristic tonal note of increasing pitch. We focused on this sequence to quantify song spectrotemporal matching. In the time domain, we measured total sequence duration and the duration of each note, whereas in the frequency domain we quantified the maximum frequency of each note and its frequency range. In the final tonal element, we also measured the fundamental frequency. We measured 842 sequences (on average four per bird) but all the analyses were carried out using individual means.

To quantify individual repertoire size, we classified song types by inspecting sonograms by eye. Visual inspection of sonograms was carried out by the same investigator, thus avoiding any interobserver bias. To measure song-type sharing, we compared the bird's song repertoire with that of all the recorded males in the same locality. To be considered as shared songs, two sequences had to match at least 3/4 of their component notes and the matching portions had to be similar in note shape, timing and frequency. Classification was facilitated by restricting interindividual comparisons within localities, where borderline cases were rare. Once song type had been classified, we built a binary matrix of song types, scoring the presence (1) or absence (0) of a song type in the pool of an individual.

Sound analyses were carried out with Avisoft SASLab Pro 3·91 (Specht 2003), performing a Fast Fourier Transform (sampling frequency 22 050 Hz, FFT length 512, time resolution 8·9 m s⁻¹, frequency resolution 43 Hz, Window Function: Bartlett).

individual identification

As Beecher (1996) pointed out, the availability of a marked population of birds is a pre-requisite for song-sharing studies because matching would reduce spectrographic recognition of non-marked individuals. In this study, birds were not marked but they were recorded during a single visit only. Each cluster of territories and nearby clusters was visited only once and in the same day, thus avoiding the problems of individual identification on successive days.

The lack of replication and the accuracy of individual characterization of recorded birds was confirmed by: (i) the overlap among singing localizations and individual territories, derived from mapping methods (Tella et al. 2005); (ii) bird fidelity to territory boundaries, determined by radiotelemetry (Garza et al. 2005); and (iii) individual identification, obtained through sonogram inspection. We analysed stereotypy in the spectrotemporal features of songs in two sampling localities (locality n, four groups of neighbours, totalling 14 birds and 100 songs; locality a, 3 groups of neighbours, eight birds and 52 songs) by means of repeatability analyses (Lessells & Boag 1987). Repeatability was significant for 12–18% call variables in neighbours from locality a (r = 0·33–0·42, P < 0·05) and for 47–65% variables in locality n (r = 0·19–0·93, P < 0·05). Hence, despite largely sharing song, neighbours maintained some individual-specific spectrographic features that confirmed field identification. Note that Stoddard (1996) emphasized that song matching does not exclude individual discrimination.

occurrence of song spectrotemporal matching and song-type sharing

We expressed spectrotemporal matching by first quantifying acoustic dissimilarities in terms of Euclidean distances of acoustic variables between individuals of each locality (call dissimilarity d); pairwise spectrotemporal matching was then calculated as 1 − log(d). We measured song-type sharing by means of the sharing index between pairs of individuals, equal to 2n/(R₁ +R₂), where n is the number of shared songs and R₁ and R₂ are the repertoire sizes of the two birds (McGregor & Krebs 1982). To verify whether birds from adjacent territories matched their songs, we tested for differences in the magnitude of spectrotemporal matching and song-type sharing between neighbours and non-neighbours. Differences were tested with Wilcoxon matched pairs test, using localities mean values, i.e. the average spectrotemporal matching between neighbours and non-neighbours and the average sharing index between neighbours and non-neighbours.

An alternative method to verify the occurrence of song matching within localities involves spatial autocorrelation analysis (Sokal & Neal 1978), which considers acoustic dissimilarity as a function of geographical distance. The profile of the distograms, i.e. the plots of acoustic dissimilarity vs. geographical distance, serves to interpret the spatial structure of song variation. If distograms possess a stabilizing profile (significant coefficients at the first distance classes and non-significant coefficients at the last distance classes; Fig. 3), discrete patches of acoustic variability are found, with song matching between close birds and random variation among birds farther from each other. If there is a continuous increase in acoustic dissimilarities with increase of geographical distance, we expect a clinal variation through space (Diniz-Filho & Telles 2002). In the latter conditions song transmission does occur, but without clear-cut divergence between neighbours and non-neighbours. By using discrete distance classes (500 m), we applied a permutation procedure using Monte Carlo simulations to test for significant deviation from spatial random distribution at each distance class (1000 permutations; Manly 1997). A 95% confidence interval was used to infer statistical differences between observed and random values. Pairwise acoustic dissimilarity was expressed in terms of city-block distances (Deichsel & Trampisch 1985) of spectrotemporal variables and song repertoire. SGS software was used for this analysis (Degen, Petit & Kremer 2001).

factors affecting spectrotemporal matching and type sharing

We divided the study localities into two groups, the first with singing clusters physically separated by habitat other than steppe (fragmented groups, n= 12) and the second with non-neighbours occurring in the same continuous steppe (unfragmented groups, n= 9). In the first case, groups were separated by main roads, cereal fields, orchards, pine plantations, young woodlots, railways or (in 64% of cases) a combination; in all cases, the width of unsuitable habitat exceeded 100 m. We carried out a two-way ancova, with song-type sharing and spectotemporal matching as dependent variables, the level of fragmentation (fragmented vs. unfragmented groups) and the relationships between birds (neighbours vs. non-neighbours) as fixed effect, and latitude, longitude, mean interindividual distance and recording period as covariate. Recording period was included to account for temporal variation in the intensity of song matching (Krebs, Ashcroft & Van Orsdol 1981; Beecher et al. 2000). We divided the breeding season into seven periods (period 0, 26 February−15 March; 1, 16–31 March; 2, 1–15 April; 3, 16–30 April; 4, 1–15 May; 5, 16–31 May; 6, 1–15 June). By applying Bonferroni correction at P < 0·05, P-values were accepted as significant if < 0·017.

As an alternative method, we used generalized linear models to investigate the relationships between song-matching variables and a series of ecological, ethological and geographical predictors.

We considered the following song-matching variables (locality mean values):

As predictors, we considered the following.

Before carrying out analyses, data were log-transformed, and a normal distribution and an identity link function were used in models. By applying Bonferroni correction, a model was considered significant for P-values < 0·008. Analyses were carried out with the package Statistica 6·1 (StatSoft 2003).

spatial autocorrelation analyses

The song repertoire within sites included 11–34 song types. Overall song-type sharing within localities varied between 0·16 and 0·54, whereas spectrotemporal matching comprised between 0·38 and 0·96 (Table 1). Differences among localities were significant (song-type sharing: F_19,914 = 61·9, P < 0·001; spectrotemporal matching: F_19,914 = 24·6, P < 0·001). A significant correlation resulted between the two measures of acoustic similarity (Rs = 0·54, n= 21, P < 0·05).

Spatial autocorrelation analyses showed that only four localities out of the 21 sites had no clear patterns of acoustic differentiation with spatial distance among individuals, either in spectrotemporal matching or in song-type sharing. In 12 localities we found song-type dissimilarities, and in 10 localities we found spectrotemporal dissimilarities that were significantly lower than expected by chance for distances shorter than 500 or 1000 m (Table 1). Divergence was never significantly greater than expected by chance for longer distances, implying that a stabilizing profile in dissimilarity patterns was achieved, with song matching occurring within a limited spatial radius (Fig. 3).

Wilcoxon matched pairs tests confirmed that song-type sharing and spectrotemporal matching were greater between neighbours than between non-neighbours (spectrotemporal features: T = 13, n = 21, P < 0·001; song type: T = 6, n = 21, P < 0·001; Fig. 4).

effects of steppe fragmentation and other predictors on song matching

Two-way ancova showed significant differences in song sharing between neighbours and non-neighbours (F_1,34= 27·4, P < 0·001) and highlighted the effects of fragmentation on neighbour–non-neighbour sharing patterns (neighbour–non-neighbour fragmentation: F_1,34 = 10·5, P= 0·0027). This implied that the existing differences in sharing between neighbours and non-neighbours were significantly affected by the presence of elements of fragmentation in the steppe (Fig. 5). In the case of spectrotemporal matching, differences were no more significant.

The proportion of steppeland cover varied from 31% to 100% (average 83·2%) in the study localities. Fragmentation at the local scale was the result of agriculture (cereal fields, orchards), forestry (pine plantations) and, to a lesser extent, urbanization (highways, buildings, railways, etc.) and woodland encroachment. The index of male competition (relative proportion of singing males) comprised between 1 (all birds singing) and 0·4 (40% of birds singing).

Latitude, longitude, interindividual distances, index of male competition, steppe cover and recording period were uncorrelated (all Rs < 0·35, n= 21, NS), thus they were entered in generalized linear models as independent predictors. Dupont's lark males shared significantly more song types with non-neighbours and less with neighbours in areas with a larger proportion of continuous steppeland (Table 2). Spectrotemporal similarities between non-neighbours were also significantly correlated with steppeland cover and increased with the level of competition in the site (Table 2). Overall spectrotemporal matching significantly increased with the intensity of male competition and with steppeland cover (Table 2). Recording period, distance and geographical location had no effect on the degree of song-type sharing and spectrotemporal matching, either between neighbours or non-neighbours (range of P-values 0·054–0·99). Distances between counter-singing neighbours were significantly shorter in fragmented than in continuous steppes (171 ± 33 m vs. 221 ± 47 m, Mann–Whitney U= 23, n= 21, P= 0·027), whereas distances between clusters of individuals were not affected by fragmentation.

song-matching patterns

Male Dupont's larks from nearby territories shared more song types and matched the physical features of songs better than non-adjacent birds. Spatial autocorrelation analyses confirmed that in most localities song type or spectrotemporal similarities were highest within clusters of adjacent territories, and declined with distance. While short-distance dissimilarities tended to be significantly lower than expected by chance, long-distance dissimilarities were never greater. This suggests that vocal units within localities tended to show a discrete (or patchy) spatial distribution, and that acoustic sequences only diffused over very short distances.

The resulting patterns of song variation in Dupont's lark might derive either from the combination of innate behaviour and extreme philopatry or from learning processes occurring after dispersal. In most animal taxa vocal signals develop without significant environmental input, but in songbirds they are know to be learned (Beecher & Brenowitz 2005). Many species learn their song from their neighbours after dispersal, during their first breeding season (Kroodsma 1982). These learning mechanisms would determine the static nature of song variants, which may remain spatially localized even when singing individuals are actually moving. In the light of this, a higher degree of song spectrotemporal matching and type sharing between neighbours than between non-neighbours can be interpreted as the outcome of song-learning processes (Beecher 1996). According to the acoustic mimicry hypothesis, young males or floaters may take advantage of song learning and matching from territorial adult males (Craig & Jenkins 1982). Floaters of song sparrow have been shown to match perfectly the songs of adult males, and obtain their first territories by inserting themselves between the territories of their tutors (Beecher, Campbell & Stoddart 1994). These new birds might be mistaken initially by each neighbour for the other, and this may reduce aggression. Stoddard (1996) pointed out that this mimicry could function particularly well when adult males are otherwise occupied in caring the young. Dispersing young males of Nuttall's white-crowned sparrow Zonotrichia leucophrys nuttalli Forster better matched the frequency and temporal characteristics of their neighbours’ rather than their fathers’ songs, supporting the idea that song matching of neighbours is an integral component of territory settlement by juveniles (Bell, Trail & Baptista 1998). Song sharing may also provide advantages to young birds when co-operative interactions with neighbours occur, because sharing songs might increase familiarity between group members, which in turn could bring fitness advantages (Beletsky & Orians 1989). Sharing songs to increase familiarity can also be beneficial when neighbours do not co-operate, and might serve as a signal of respect of territory boundaries among old neighbours (Dear Enemy Hypothesis; Beecher & Brenowitz 2005).

In Dupont's lark, a high number of floater males occurs because of the limited availability of suitable habitat (Garza, Suárez & Tella 2003), strongly male-biased sex ratio among adults (Tella et al. 2004) and territory saturation (at least in some localities; Garza et al. 2005). Hence a shortage of females and habitat most probably shaped song-matching behaviour in Dupont's lark males, although the causative factors promoting it still remain open to debate (e.g. harsh antagonism for territories and mates or co-operative interactions or respect of territory boundaries).

Differences among localities in song matching and in song differentiation over distance might be a result of variation in competition levels, as at least the magnitude of spectrotemporal matching was correlated with the proportion of singing males. In song sparrows, marked geographical variation in the amount of song sharing possibly resulted from life-history traits such as mortality or variation in philopatry (Hill et al. 1999). Dupont's lark was studied in fragments that were isolated from each other by a variety of factors. The availability of space and habitat corridors might have affected the number of floaters and, in turn, the intensity of both competition and song matching.

effects of fragmentation on song matching

We found that habitat loss in the steppe matrix markedly affected song-type sharing mechanisms in Dupont's lark. The occurrence of anthropogenic habitat barriers seems to hinder cultural transmission of song types over distances, resulting in an intensification of the differences between non-neighbours and increasing mimicry between neighbours. This suggests that males from fragmented habitats perceived as rivals only the close neighbours with which they engaged in counter-singing. During interactions, they reduced interindividual distances and sharpened the precision of matching, possibly as a result of the fierce competition for the remaining patches of suitable habitat. Conversely, individuals located on the opposite side of barriers might be viewed as a milder threat, taking into account bird propensity to confine daily home ranges to pure steppe habitat (Seoane et al. 2004; Garza et al. 2005). The increase of song sharing and spectrotemporal matching between non-neighbours in continuous steppeland might result from a tendency to patrol further from the strict boundaries of close territories when barriers are lacking. Alternatively, the presence of continuous habitat might facilitate turnover of territories among clusters of individuals, thus promoting bird movements and the partial circulation of their songs.

In this study, we found that Dupont's lark perceives the spatial geometry of its habitat and behaves accordingly. Local habitat fragmentation seems to alter behavioural processes by hampering the local transmission of song cultures, in particular the spatial range of individual acoustic niche contracts at increasing fragmentation levels. Notably, patch connectivity and fragmentation at a broader scale affected variation in another vocalization of Dupont's lark, the short territorial call. Although the patterns of variation of songs and territorial calls differed (the latter being less variable than the former), we found that the differentiation of territorial calls was affected both by isolation by distance and isolation by habitat unsuitability mechanisms, and call dissimilarities continued to increase at long geographical distances (P. Laiolo & J. L. Tella, unpublished data). The pattern of complex song variation differs from this linear isolation by distance model, and song-matching mechanisms might be responsible for the observed deviation of the curve.

Few studies to date have documented an alteration in communication systems in response to habitat loss (Holland, McGregor & Rowe 1996; Seibt et al. 2002) or have directly analysed the relationships between landscape geometry, song variation and anthropogenic disturbance. We believe that communication systems might become reliable indicators of human disturbance (behavioural indicators; sensuBlumstein & Fernandez-Juricic 2004), which could assist our understanding of how and why species are vulnerable to human impact.

Vocalizations are rapidly evolving traits because of their cultural nature, in which transmission is both vertical (learning from fathers), horizontal (learning from members of the same generation) and oblique (learning from unrelated birds of different generations) (Cavalli-Sforza et al. 1982; Baptista & Gaunt 1997). The effects of fragmentation on natural populations can thus be detected over short time scales by analysing variation in communication systems. The study of song-matching patterns may therefore be used as an early warning system to identify anthropogenic disturbances, when other potential indicators, such as genetic markers, may not show any effect, either because the genetic composition of the population has not changed yet or because of the difficulty of separating recent events from historical ones (Pearse & Crandall 2004). We predict that habitat fragmentation will affect the communication systems of habitat-specialist species more than those of generalist ones, and that this phenomenon would be easier to detect when the boundaries between suitable and unsuitable habitats are straightforward and clearly identifiable.

Attention to the behavioural detail could help in developing effective habitat management and population recovery programmes, especially if the fitness consequences of behavioural mechanism are properly understood (Alonso, Switzer & Mangel 2003; Béchet, Giroux & Gauthier 2004; Norris 2004; Tarlow & Blumstein, in press). In the case of Dupont's Lark, our analysis of acoustic variability suggests that habitat conservation should be coupled with the maintenance of steppe corridors, to allow bird movements and prevent the isolation of separate nuclei of individuals. It has been argued that populations may split into smaller units if breeding only occurs within dialect groups, and this in turn could have detrimental effects on individual fitness, eventually leading to extinction (McGregor, Holland & Shepherd 1997; Vos & Chardon 1998).