Luthy el al Identification of larval sailfish, white marhn, and blue marhn in the western North Atlantic Ocean 



593 



-6 Axis 1 6 



Figure 4 



Canonical variates analysis with the reduced set of variables. Arrows 

 indicate the direction of increase in continuous variables and may 

 be extended backward through the origin of the graph to show a 

 decrease in the value of the character. Variables that extend far- 

 thest from the origin are most useful in the separation. SN = snout 

 length; JD = difference in the lengths of the jaws; ED = eye diameter; 

 PRO = length of preopercular spine; p (number i = presence of pig- 

 ment in lower jaw grid cell (number). 



through September, white marlin (4.5 mm-20.3 mm SLi 

 were collected from March through June, and larval blue 

 marlin (3.8 mm-22.1 mm SL) were collected from June 

 through September. Month of capture closely matched the 

 reported spawning seasons for these species in the west- 

 ern North Atlantic: April through October for sailfish, 

 March through June for white marlin, and July through 

 October for blue marlin (de Sylva and Breder, 1997). 

 Because blue marlin larvae were also caught in June, the 

 blue marlin spawning season was expanded to include 

 that month for the purposes of the identification key. 



Canonical variates analysis 



In the CVA with all variables included, separation of the 

 three species was achieved with little overlap. Sailfish 

 larvae were separated from the marlins along canoni- 

 cal axis 1 (eigenvalue = 5.45). The separation was driven 

 mainly by ED, OD, and lower jaw pigmentation. White 

 marlin larvae separated from blue marlin primarily 



along canonical axis 2 (eigenvalue = 0.79), largely by 

 month of capture, as well as SN, SN-E, and JD . The 

 overall ordination was significant at P=0.002. 



The forward selection process, along with the re-addi- 

 tion of counterpart pigment grids and the full spawning 

 season, yielded the following 21 out of 32 variables: 

 March, April, May, June, July, August, September, SN, 

 JD, ED, PRO, and pigment grids 1-4, 6-9, 11, and 12. 

 The following variables were ultimately excluded from 

 the data set: SL, SN-E, OD, HL, PTS, and pigment 

 grids 5, 10, and 13-16. The degree of species overlap 

 was similar to that in the full model (Fig. 4). This 

 overall ordination was also significant at P=0.002. The 

 eigenvalue of the first canonical axis was 4.71, whereas 

 the eigenvalue of the second canonical axis was 0.71. 

 Coordinates obtained from the canonical coefficients 

 and character values, standardized by reference set 

 character means and standard deviations (Table 1), 

 accurately placed test "unknowns" in the ordination of 

 the reference larvae. 



