Bjorndal et al.: Somatic growth function for immature Caretta carettci 
241 
Length-frequency analysis has been used for many years 
to estimate growth rates, age structure, and mortality in 
marine fish and invertebrate populations (Ricker, 1975; 
Pauly and Morgan, 1987; Hilborn and Walters, 1992). More 
recently, length-frequency analyses have been used to eval- 
uate somatic growth rates in sea turtles. The accuracy 
of four length-frequency analysis programs (ELEFAN I, 
[Holden and Bravington, 1992 1 Shepherd’s length compo- 
sition analysis [SLCA, Holden and Bravington, 1992], pro- 
jection matrix method [Holden and Bravington, 1992], and 
MULTIFAN) for predicting growth rates was tested in a 
population of green turtles, Chelonia mydas , in the south- 
ern Bahamas for which growth rates had been measured in 
a long-term mark and recapture study (Bjorndal and Bol- 
ten, 1995; Bjorndal et al., 1995). MULTIFAN successfully 
estimated growth rates in this population, SLCA was par- 
tially successful, and ELEFAN I and the projection matrix 
method were not successful. In young, pelagic-stage logger- 
head sea turtles (Caretta caretta), estimates of growth rates 
generated by MULTIFAN were consistent with results from 
recaptures of tagged turtles (Bjorndal et al., 2000). 
In our study, we generated a growth model for imma- 
ture loggerhead sea turtles in southeastern U.S. waters 
between the size at which they begin to recruit in substan- 
tial numbers to neritic habitats (46 cm curved carapace 
length [CCL]) and minimum size at sexual maturity (87 
cm CCL). The duration of the growth interval between 
46 and 87 cm CCL is critical information for developing 
management plans and demographic models for this sea 
turtle, which is listed as a threatened species in the U.S. 
Endangered Species Act of 1973. This size range includes 
the large juvenile and subadult lifestages defined in the 
stage-based population model developed for North Atlan- 
tic loggerhead sea turtles (Crouse et al., 1987; Crowder et 
al., 1994). This stage-based population model has identi- 
fied survivorship in the large juvenile lifestage as the most 
critical for population recovery. We based our length-fre- 
quency analyses on data collected from hundreds of log- 
gerhead sea turtle carcasses that were measured by the 
Sea Turtle Stranding and Salvage Network from Florida, 
Alabama, Mississippi, Louisiana, and Texas between 1988 
and 1995. 
Methods 
Length-frequency data 
The Sea Turtle Stranding and Salvage Network (STSSN) 
is an organized network of individuals who monitor the 
shoreline and record data on each stranded sea turtle, 
including date and location of stranding, species, and cara- 
pace length (curved or straight carapace length, or both). 
Carapace length is measured from the anterior point at 
midline (nuchal scute) to the posterior tip of the supra- 
caudals. The stranding data are compiled and verified by 
state coordinators and archived at the Southeast Fish- 
eries Science Center (SEFSC) Miami Laboratory (Teas, 
1993). We received data on stranded turtles for 1988 
through 1995 for Alabama, Mississippi, Louisiana, and 
Texas from SEFSC and data for 1988 through 1995 for 
Florida from the Florida Department of Environmental 
Protection, Florida Marine Research Institute. All turtles 
known to have been “head-started” (that is, raised in cap- 
tivity before being released into the wild) were excluded 
from the analyses because growth rates in captivity, and 
therefore length-at-age, may be quite different for head- 
started turtles. We divided the data into two geographic 
regions: the Atlantic coast of Florida and the U.S. coast of 
the Gulf of Mexico (Florida, Alabama, Mississippi, Louisi- 
ana, and Texas). The Florida coast was divided between 
Atlantic and Gulf by the STSSN at 80.5°W. 
Because most of the data on carapace length were over- 
the-curve measurements, straight carapace lengths (SCL) 
were converted to curved carapace lengths (CCL) by using 
the conversion equation (n= 932, r 2 =0.97, P<0.001) in Teas 
(1993) 
SCL = (0.948 x CCL ) - 1.442. 
After conversions were completed, all CCL data were 
rounded to the nearest cm. 
We wanted to limit our analyses to the subadult, nerit- 
ic lifestage that inhabits the coastal waters of the south- 
eastern U.S.; therefore we limited the size range of logger- 
head sea turtles from 46 to 87 cm CCL. The lower value 
was based on length-frequency distributions (Bolten et al., 
1993; Bjorndal et al., 2000) that indicated that 46 cm CCL 
is the size at which these sea turtles begin to recruit to 
neritic habitats in substantial numbers. The largest sub- 
adult size was taken as 87 cm CCL based on Withering- 
ton (1986), who reported that 88 cm CCL was the size of 
the smallest nesting loggerhead sea turtle at Melbourne 
Beach, Florida. This value is a very conservative division 
between subadults and adults; many with CCLs greater 
than 87 cm are still immature. For length-frequency anal- 
yses, however, it is better to exclude some subadults than 
to include many adult animals. Any factor that acts to 
obscure the modal structure of the sample — such as ces- 
sation or near-cessation of growth in older age classes — 
will decrease the potential for successful length-frequen- 
cy analysis. If older age classes cannot be distinguished, 
K (intrinsic growth rate) will be overestimated and the 
number of age classes underestimated (Terceiro et ah, 
1992). Because loggerheads essentially stop growing at 
sexual maturity and because they attain sexual maturity 
at a range of sizes (Frazer and Ehrhart, 1985), the age 
classes — or modes — above the minimum size at sexual ma- 
turity are obscured and cannot be distinguished in length- 
frequency analyses. 
Kolmogorov-Smirnov analyses were conducted with SPSS 
software (version 9.0 SPSS, 1996). 
Length-frequency analysis 
We used MULTIFAN (version 32(f), Otter Research Ltd., 
1992) modified to include 30 age classes by Fournier ( Otter 
Research Ltd., 1992). MULTIFAN simultaneously ana- 
lyzes multiple samples of length-frequency data (Otter 
Research Ltd., 1992) and uses nonlinear statistical mod- 
