We studied whiskered terns in two of the major breeding populations in France, about 200 km apart from each other: the Brenne region (studied in 2001–2004), and the Lake of Grand-Lieu (studied in 2004 only). The Brenne (46°46′N, 01°10′E) covers about 80 000 ha, and consists of a mixture of natural grasslands, forests, and about 2300 artificial fish-ponds that cover approximately 8000 ha [35]. It currently supports 27–40% of the French breeding population of whiskered terns (687–1083 pairs, J. Trotignon, personal communication). In the Brenne, whiskered terns breed on a wide variety of floating and emergent plants [36], though in a restricted number of small- and medium-size ponds each year (on average 15 to 23 ha depending on the years, range = 1–117 ha). Most ponds are owned privately, hence their management depends solely, or mainly on economic needs, i.e. hunting and fish farming [36]. In Brenne, 11 colonies (i.e. 11 ponds that support breeders) were studied: one colony was visited for three years and one colony for two years (see details in Table 1). All colonies were visited only once within the incubation period. Nests were reached by wading, since ponds were shallow enough.

Lake Grand-Lieu is a very large (4000 ha in summer and 6300 ha in winter) shallow, turbid, eutrophic natural freshwater ecosystem in western France (47°05′N, 1°39′W), surrounded by wet grasslands. It is covered with extensive beds of floating plants (770 to 978 ha, depending on the years), comprising mainly waterlily (Nymphaea alba) beds. Currently, Lake Grand-Lieu supports a whiskered tern breeding population of ca. 700 pairs at 10 to 11 colonies (S. Reeber, personal communication) that were well spaced by several hundred metres. Nesting platforms are always built on waterlily beds and consist mainly of waterlily leaves and stems and common clubrush (Scirpus lacustris) stems. Six colonies were visited once a week after clutch initiation in 2004. All nests were accessible by boat.

Whiskered terns are known to lay two- and three-egg clutches, the latter being more common [37,38]. The length and width of 1079 eggs from 121 two-egg clutches and 279 three-egg clutches were measured with a Vernier calliper to the nearest 0.1 mm. Egg volume (cm³) was calculated using Coulson's equation [39]. Egg weight was measured using a Pesola^® to 0.2 g in Brenne and an electronic balance to 0.1 g in Grand-Lieu. Because the weight of an avian egg declines during incubation (i.e. egg age) as a result of water loss [40,41], we used a linear egg density/age relationship determined on egg data collected at Grand-Lieu for calculating egg age in clutches [38]. This relationship (R2=0.77) was established from 188 measurements: 70 different eggs of known laying date (32 were measured once, 30 twice and 8 three times totalling 116 egg measurements), plus 72 eggs (measured once) of known hatching date. Based on this relationship, we recorded whether clutches measured in Brenne, and notably two-egg clutches, were complete. We then back-calculated fresh egg weight (i.e. egg weight at laying) using that density/age relationship. The average daily decrease in egg weight was 0.66%±0.23SD of the egg weight (multiple measurements of eggs, n=62). Lastly, successive visits performed on Grand-Lieu allowed us to identify the last egg (c-egg) in 11 three-egg clutches, the laying sequence of the two other eggs (a- and b-eggs) being unknown. Finally, based on the density/age relationship, we determined the clutch initiation date (i.e. the laying date of the first egg, see [38]) of each colony of the two study sites as well as the average clutch initiation date (±95% CI, range).

We used Eadie's method [31] to detect parasitized clutches. The first step in using Eadie's method consists in standardizing all egg measurements (length, width and fresh weight) by colony (z-scores, i.e. mean of 0 and unit variance, see [42]). In the second step, z-scores are used to calculate the Euclidean distance (ED) for the pair of eggs within the two-egg clutches, and the three pairs within the three-egg clutches. Lastly, for each clutch, the largest ED value, MED (maximum Euclidean distance, i.e. the distance between the two most dissimilar eggs), is used to separate parasitized and non-parasitized clutches [31]. Eadie [31] used a cut-off point of 2.5 in MED to identify parasitized clutches for the barrow's goldeneye (Bucephala islandica). Since the within-clutch variance in egg size differs between bird species, a threshold value of MED must be calibrated for the whiskered tern. We used exceptionally large clutches (four-, and five-egg clutches), that were considered to result from parasitized clutches in whiskered tern as they do in similar species [3,11,43,44], for calibration. In the two study areas, 11 such clutches (10 four-egg clutches and one five-egg clutch, providing a total of 70 egg-pairs) were available. In each one of these clutches, the egg (or the two eggs in the single case of a five-egg clutch) showing the highest ED values in all pairs (six pairs within the four-egg clutches and 10 pairs within the five-egg clutch) was considered a parasitic egg. Nevertheless, some caution is needed for using this calculated ED threshold based on supernormal clutches, because the number of egg pairs is limited and we do not have any proof demonstrating that odd eggs in supernormal clutches have been laid by a parasitic female. So, an additional validation of the MED calibration was performed by simulating parasitized clutches (1) by moving one randomly chosen egg (i.e. the parasitic egg) from each three-egg clutch to another randomly chosen three-egg clutch within the data set, or (2) by adding one randomly chosen egg (i.e. the parasitic egg) from each two-egg clutch to another randomly chosen two-egg clutch.

Whiskered terns defend their territories vigorously against conspecifics, including parasitic females [45, unpublished data]. We assumed that parasitic females would lay their egg only when both members of the pair are away from the nest, so we focused on the time during observations where none of the mates were incubating or guarding the nest (“absence time” of both pair members). Since CNP corresponds to multiple eggs laid in a day or eggs laid several days after clutch completion, the study of absence time of both pair members was carried out during the laying and incubation periods. Nest attendance was monitored with a telescope from a distance of 50–100 m on Grand-Lieu in 2004–2005. A total of 135 different nests was monitored before incubation started, i.e. during the laying period (30-min focal each nest; 13 days; 9 May to 6 July) for a total observation time of 4050 min, and additionally 68 other nests were monitored during incubation (12 days; May 26 to July 6) for a total observation time of 2040 min. Nests were randomly selected in 10 colonies to have a representative sampling of bird activity. Observations were carried out between 8–12 a.m. and 3–8 p.m. Sometimes, several absence times were observed during a nest focal. Absence time was classified into five classes: 0 min, 0.01–1.00 min, 1.01–3.00 min, 3.01–5.00 min and 5.01–10.00 min.

Normality and homoscedasticity were tested using Kolmogorov–Smirnov and Bartlett's tests, respectively. When deviations from normality were detected, non-parametric tests (Kruskal–Wallis) were used. Differences in egg morphology across years in Brenne, and between study sites were tested with multiple analysis of variance (MANOVA and Wilks' λ statistic). Between- and within-clutch variation in egg morphology for each clutch size class was analysed with one-way ANOVA. Fisher's exact test was used to compare %CNP among clutch size classes and sites. The likelihood-ratio chi-square test was used to compare absence times between periods. All statistical analyses were performed using STATISTICA (version 6.0), and 0.05 was taken as the level of significance.

There were no differences in any of the egg variables (length, width and fresh weight) between years in Brenne (MANOVA, Wilks' λ=0.97, F3,1243=1.34, p=0.21) or between the two study sites (MANOVA, Wilks' λ=0.99, F3,1075=1.15, p=0.33). Therefore, all data on egg biometrics were pooled for subsequent analyses. The third egg in three-egg clutches was not smaller than the other two eggs (Table 2). In addition, the between-clutch variation of each egg biometrics was significantly larger than within-clutch variation (see F-statistics in Table 3). From 71.6 to 74.6% and 80.9 to 86.6% in total variation in egg size was accounted for by between-clutch variation in three- and two-egg clutches respectively (Table 3). Therefore, within-clutch variation in egg size was relatively limited compared to between-clutch variation, allowing egg morphology to be used to investigate CNP in the whiskered tern.

Based on values obtained from the supernormal clutches (> three-egg clutches), we found that ED values between non-parasitic egg pairs never exceeded a value of 2.5 (Fig. 1). ED values between non-parasitic egg pairs in the two simulated data sets were similar to those obtained in the observed data set (supernormal clutches), with very few ED values exceeding 2.5 (2.9%, n=375 pairs, Fig. 1). Based on these findings, we retained a cut-off point of 3.0 in MED for detecting parasitized clutches because no ED values in non-parasitic egg pairs, whatever the data set used (simulated or observed), exceeded this threshold. Frequencies of parasitized clutches using egg biometry represent therefore minimum estimates for the whiskered tern.

Out of the 121 two-egg and 279 three-egg clutches, at least two (1.7%) and 25 (9.0%) were parasitized (Table 1). %CNP was significantly higher in three-egg than in two-egg clutches (Fisher's exact test, p=0.008). There was no effect of site on %CNP in two-egg clutches (a single case in each site, i.e. 1.6% and 1.7% respectively in Brenne and Grand-Lieu), conversely to three-egg clutches that occurred significantly more often in Brenne (19 clutches, i.e. 14.5%) than at Grand-Lieu (6 clutches, i.e. 4%; Fisher's exact test, p=0.003). CNP rates also varied strongly between colonies, in both sites, ranging from 0 to nearly 30% (Table 1). Lastly, no additional egg was ever noted after clutch completion at Grand-Lieu during successive visits to colonies.

Nests were left unattended on average no more than 1 min out of 30 (89.0% of absence time lasted less than 1.00 min whatever the periods, Fig. 2). However, absence time varied between periods (likelihood-ratio chi-square test, χ2=15.99, df=4, p=0.003). Absence time never exceeded 3 min during the incubation period (n=91 absence times), whereas it was more than 3 min in 4.2% of absence times during the pre-incubation period (n=282 absence times, Fig. 2), and still 2.8% of absence times exceeded 5 min (maximum reached: 10 min). When totalling absence times during each nest focal, 9% ranged between 5 and 10 min during the pre-incubation period (n=135 nests), while it never happened during the incubation period (n=68 nests). Synchrony in laying was very high in all colonies (Fig. 3), with CI of average laying date varying from ±0.90 to 4.08 days according to colonies.