Fregetta, Bonaparte, 1855

Bretagnolle, Vincent, Flood, Robert L., Gaba, Sabrina & Shirihai, Hadoram, 2022, Fregetta lineata (Peale, 1848) is a valid extant species endemic to New Caledonia, Bulletin of the British Ornithologists’ Club 142 (1), pp. 111-130 : 114-117

Multivariate analyses of Fregetta and suspected F. lineata specimens

We conducted a series of multivariate analyses on six biometrics: bill width, bill depth at gonys; and culmen, wing, tail and tarsus lengths. The sample (n = 261) comprised adults collected at colonies, held at museums worldwide, and measured by VB. Mean measurements are shown in Fig. 3 . Analyses were conducted using the packages FACTORMINE and MASS from R software. Biometric characters were centred and scaled prior to analyses.

Principal Component Analyses.—Principal Component Analysis ( PCA) was performed as a descriptive multivariate analysis. In the first PCA, specimens from breeding archipelagos were grouped and averaged, including F. tropica (11 populations), small White-bellied Storm Petrels F. grallaria (six populations), the large F. [g.] titan (from Rapa Island), and F. maoriana (from New Zealand). The two Brisbane specimens and the specimens ‘ lineata ’ from Samoa and ‘ guttata ’ from Ua Pou were treated individually. Results are shown in Fig. 4A . The first principal component (Dim 1; Fig. 4A ) was a body size axis (especially wing length; see Fig. 5). Dim 1 clearly separates the smallest F. maoriana (purple) from the largest F. [g.] titan (blue). F. tropica (green) and F. grallaria (orange and red) were fairly evenly spread, with F. grallaria having an overall smaller size and notably shorter tarsi, although some individuals (mainly from the western Pacific, but also the South Atlantic) had tarsi nearly as long as F. tropica ( Fig. 3 ). QM 31216 (light green) clumped with F. tropica (dark green), whilst QM 14391 clumped with ‘ lineata ’ and ‘ guttata ’ (black). All three were at the small end of the size range of F. grallaria , having slightly shorter wings, a shorter tail than other Fregetta (but longer than F. maoriana ), and tarsus length intermediate between F. tropica and F. grallaria ( Fig. 3 ). This morphometric analysis supports the separation of F. maoriana , highlights the distinctiveness of the ‘ lineata , guttata, QM 14391’ specimens within F. grallaria , and aligns QM 31216 with F. tropica .

These results were confirmed by two further PCAs run on individual specimens. For the F. maoriana , ‘ lineata ’ and ‘ guttata ’ specimens, we used measurements taken by various researchers, as if they were independent samples, yielding, respectively, seven and 11 ‘specimens’ (complete list and measurements in Appendix 2). In the PCAs, individuals were grouped by breeding locality ( Fig. 4B ) or by taxonomy ( Fig. 4C ). Both PCAs indicated that ‘ lineata ’, ‘ guttata ’ and QM 14391 clumped between F. grallaria and F. tropica , overlapping slightly with both. Also, many F. grallaria specimens overlapped with F. tropica . Lastly, once again, F. maoriana and F. [g.] titan map as extreme opposites ( Figs. 4B‒C ).

Discriminant Analyses.—Discriminant Analysis ( DA) maximises multivariate differences between groups (taxa). It establishes classification of individuals from their biometrics based on canonical discriminant functions. DA was performed on individuals grouped by nine breeding localities ( Fig. 4D ), five taxa ( Fig. 4E ), and a reduced set of three taxa ( Fig. 4F ). The five taxa were ‘ F. lineata ’ (a regrouping of ‘ lineata , guttata , and QM 14391’), F. grallaria , F. tropica (including QM 31216), F. maoriana , and F. [g.] titan. Error-classification rates are derived as apparent rates when using all individuals as a train dataset, or more robust estimates of error rates when using the cross-validation and leave one out procedure. The same six centred and scaled variables from PCA were used in DA.

First, a stepwise DA indicated that all six variables entered the model (with threshold significance level for dropping a variable established at 0.01), thus all six variables were retained (data not shown). Then, using Linear Discriminant Analysis ( LDA), we found

apparent classification-error rate was only

10.34% (n = 261) when applied to all breeding

localities as groups ( Fig. 4D ). Five of seven

F. lineata were correctly assigned, as were all

F. maoriana and all F. [g.] titan, and nearly

all F. tropica and F. g. segethi. Using a cross-

validation procedure for error-rate estimation

led to a slight increase, as expected, of 13.0%

error rate.

A second LDA was run on the five taxa as

a priori groups ( Fig. 4E ). Apparent error-rate

classification was lower, at 8%, but just two

F. lineata were correctly classified. Quadratic

Discriminant Analysis ( QDA) was also used,

as it is less conservative (permitting different

variance matrices for different groups), while

Figure 5. Principal Component Analysis showing

using a quadratic classification decision contribution of variables to axes 1 and 2.

boundary. QDA significantly improved the

apparent error rate (3.8%), notably with all

seven F. lineata specimens now being correctly assigned to their group. However, using the

cross-validation calibration, the error rate was 8.8%. Restricting DA to F. lineata , F. grallaria

and F. tropica did not significantly alter the results, but placed F. lineata between F. grallaria

and F. tropica in multivariate space ( Fig. 4F ), notably due to intermediate values in tarsus

length ( Figs. 3 and 6 ).

The three F. lineata specimens are larger than F. maoriana in all measurements ( Fig. 3 ,

Appendix 2), albeit with very slight overlap in bill width. Within each taxon, measurements

vary to some extent between individuals, as reported by relatively high standard error

values in wing and tail measurements (Appendix 2). Intra-taxon variation, however, is

particularly pronounced for the three F. lineata specimens. Part of this variation is due to

observer measurement error, and possibly also to specimen age. The within-individual

range of variation (i.e., measurements made by three or four different observers) is actually

higher than the within-population variation of F. maoriana . For example, the two specimens

of F. maoriana held at the Muséum national d’Histoire naturelle, Paris, were measured

by four different observers, and standard errors in measurements vary between 0.13 mm

(culmen length in one specimen) and 2.25 mm (tail length in the other). Tarsus length also

varies both within and between the three F. lineata specimens. Streaks on their bellies are

also quite variable, as in F. maoriana (e.g., Flood 2003, Stephenson et al. 2008b), but are

completely unique to F. lineata and F. maoriana . Neither taxon shows an absence of streaks,

but their extent and arrangement are variable and distinctive (see below).

Figures

Figure 3. Means of six biometrics (bill width, bill depth at gonys, and culmen, wing, tail and tarsus lengths) of small White-bellied Storm Petrels Fregetta grallaria (all taxa aggregated), New Caledonian Storm Petrel F. lineata, New Zealand Storm Petrel F. maoriana, the large Titan Storm Petrel F. [g.] titan, and Black-bellied Storm Petrel F. tropica (for methodology see main text).

Figure 4. Top row: results of Principal Component Analyses performed on six biometrics (bill width, bill depth at gonys, and culmen, wing, tail and tarsus lengths) taken from specimens of Black-bellied Storm Petrel Fregetta tropica (11 populations), small White-bellied Storm Petrels F. grallaria (six populations), the large Titan Storm Petrel F. [g.] titan (Rapa Island) and F. maoriana (New Zealand). The two Brisbane specimens and the specimens (‘lineata’) from Samoa and (‘guttata’) from Ua Pou were treated individually. (A) All taxa, plus the four museum specimens. (B) Same analysis performed on individual skins showing distinct taxa and breeding localities. (C) Same analysis for only the five recognised taxa. Bottom row: results of Discriminant Analyses performed on individual specimens from New Caledonian Storm Petrel ‘Fregetta lineata’ (a regrouping of ‘lineata, guttata and QM 14391’), small White-bellied Storm Petrels F.grallaria, Black-bellied Storm Petrel F.tropica (including QM 31216), New Zealand Storm Petrel F. maoriana, and the large Titan Storm Petrel F. [g.] titan. (D) Grouped by nine breeding localities. (E) Same analysis performed on the five taxa as a priori groups. (F) Same analysis performed on a reduced set of taxa (lineata, grallaria and tropica). In all plots, each point represents a specimen, except in 4A.

Figure 6. Comparative measurements of New Caledonian Storm Petrel Fregetta lineata (black), White-bellied Storm Petrel F. grallaria (red) and Black-bellied Storm Petrel F. tropica (green). Data are presented as a correlation matrix for each pair of morphometric variables, with the diagonal in the matrix showing the distribution of each variable per taxon. Colour chart as in Fig. 4F.