Douglas et al : Geographic variation in cranial morphology of Stenella longirostns 



59 



S. attenuata or both; since these 

 two dolphin species have broad- 

 ly overlapping distributions in 

 the eastern tropical Pacific, the 

 blocks used are representative of 

 areas inhabited by S. longirostris. 



We conducted a principal com- 

 ponents analysis of the 13 envi- 

 ronmental variables for the 51 

 blocks in order to obtain sum- 

 mary variables that reflect over- 

 all environmental trends. Indi- 

 vidual blocks were projected onto 

 the resulting environmental prin- 

 cipal components based on stan- 

 dardized data. These block vari- 

 ables were used as composite 

 environmental variables for 

 comparisons with morphological 

 characteristics. 



In addition to using matrix cor- 

 relations and the Mantel proce- 

 dure to test for local and regional 

 patterning of variation in individ- 

 ual morphological characters, we 

 compared difference patterns of 

 selected morphological measures 

 with those of environmental vari- 

 ables. In these tests, differences 

 between each pair of blocks for a morphological vari- 

 able were compared with those for an environmental 

 variable. 



Sources for environmental data are expanded over 

 those used by Schnell et al. (1986: table 2) so as to ac- 

 commodate the broader geographic representation 

 resulting from increased numbers of specimens. Values 

 for depth of the oxygen minimum layer were taken for 

 all blocks from Levitus (1982). Data for sea surface 

 temperatures and thermocline depths were not avail- 

 able in the previously used source for blocks west of 

 120°. Data for these and other blocks north of 5°S were 

 taken from Robinson (1976). Overlapping blocks from 

 the two sources for each environmental variable were 

 used to produce regression equations. Previous data 

 for blocks south of 5°S were converted using these 

 regression equations. Overall, agreement of data for 

 overlapping blocks from the two sources was relative- 

 ly good. Correlations for sea surface temperatures 

 were: January, 0.956; July, 0.951; annual variation, 

 0.929. Thermocline depth in winter had a correlation 

 of 0.840, while that for summer values was lower 

 (0.767). All correlations were statistically significant 

 (P<0.001), and the associations of values from the two 

 sources were basically linear. 



1 00 





I 14 



1 Condyiobasal L. 



2 L Roslfum(frm Base) 

 30 L Ramus 



3 L Rosl(um(tfm Plerygoid) 



22 L Up Toothrow 



27 L- Low Toolhrow 



23 No Teeth{Up Lf ) 



24 No Teelh(UpRl) 

 J— 25 No Teelh(LowL() 

 1—26 No Teelh(LowRI ) 

 — 19 Oibital L 



4 W Roslfum(al Base) 



5 W Rosltum{al 1/4 L) 



6 W Rostrum(al 1/2L) 

 9 PreofbitalW. 



Poslotbital W- 



Skull W (al Zygomatic P } 



L Braincase 



28 HI Ramus 



20 L Antorbilal P. 

 1 8 W Temporal Fossa 



15 Max W Premax. 



12 Skull W.(alPafJelals) 



13 HI Braincase 



21 W Internal Nares 



7 W Premax (al 1/2L) 



8 W Roslrum{al 3/4 L.) 

 17 L Temporal Fossa 



16 W Exlernal Nares 



29 Tooth W. 



Figure 2 



Correlations among characters based on character means for 25 blocks. Clustering per- 

 formed using UPGMA on absolute correlations among characters (i.e., negative signs 

 removed). Cophenetic correlation coefficient is 0.74. 



Results 



Sexual dimorphism 



In the two-way ANOVA for block and sex, only three 

 measurements showed a significant interaction for 

 block and sex (W. Rostrum [at Base], L. Temporal 

 Fossa and No. Teeth [Up.Lf.]). All characters exhibited 

 significant variation by block (i.e., geographic varia- 

 tion), and 15 of the 30 characters displayed significant 

 sexual dimorphism (Table 1). For most characters, 

 males are larger than females. Character differences 

 between sexes range up to 6.34% (see Table 1), with 

 the most dimorphic character being W. Rostrum (at 

 3/4 L.). 



Correlation, ordination and clustering 



Figure 2 summarizes associations among characters 

 based on means for the 25 blocks. Virtually all of the 

 intercharacter correlations were positive in sign; a few 

 indicated weakly negative associations. For the cluster 

 analysis, absolute character correlations were analyzed 

 (i.e., sign of correlation ignored), because we wanted 

 to assess simply the degree of covariation. The char- 

 acter showing the most distinctive pattern relative to 



