106 



Fishery Bulletin 92(1). 1994 



where f- is the proportion of females in the ith 

 length class, p is the proportion of the sample 

 spawning, F is the fecundity of a fish of that length 

 taken from the fecundity-length regression, SR t is 

 the sex-ratio of the ith length class and W i is the 

 total weight of fish in the ith sample. CPUE was 

 estimated from the monthly catch returns of the 

 commercial fleet. We chose this method of estimat- 

 ing egg production because S. delicatulus have de- 

 mersal eggs (Leis and Trnski, 1989) and the eggs of 

 A. sirm and H. quadrimaculatus are difficult to 

 sample adequately in the large areas of suitable 

 habitat in each lagoon. 



For comparison with adult spawning data, we 

 back-calculated the distribution of birthdates offish 

 collected in each length-frequency sample by using 

 the growth equations of Milton et al. (1993). Fre- 

 quencies in each age class were adjusted for mor- 

 tality by using the estimates of Rawlinson et al. 

 (1992). The distribution of birthdates was also back- 

 calculated for H. quadrimaculatus and S. delica- 

 tulus length-frequency samples from previous stud- 

 ies at one site (Tarawa) January 1976 to February 

 1977 (R. Cross, 1978 4 ) and May 1983 to April 1984 

 (McCarthy, 1985 1 ). We used age distribution in these 

 earlier studies and those of the present study to ex- 

 amine seasonal, annual, and site-related differences 

 in the reproductive life span of each species. 



Statistical analyses Inter- and intra-specific differ- 

 ences in fat index, HSI and K were examined with 

 Fisher's r-tests to account for unequal sample sizes. 

 Seasonal and site-related differences in fecundity 

 (expressed as oocytes per gram) were examined by 

 analysis of covariance with weight as the covariate. 

 Hydrated oocyte weight and reproductive life span 

 were examined by one-way analysis of variance. 



We examined the relative influence of exogenous 

 and endogenous factors on the fecundity of each 

 species at each site by stepwise regression (Sokal 

 and Rohlf, 1981). We included the following: length, 

 weight, age, sea-surface temperature (°C), wind 

 speed (in knots), moon phase (expressed by fitting 

 a sin/cosin curve to the number of days since the last 

 full moon before the sample was taken divided by 

 the number of days in a lunar month (29.5) (Milton 

 and Blaber, 1991), fish condition (K: weight/length 3 ), 

 fat, and HSIC7r ). We retained only those variables 

 that significantly improved the fit of the model 

 (P<0.05). Because several of these variables were 

 correlated, we did a partial-correlation analysis be- 

 tween these variables and fecundity, and the results 



4 Cross, R. 1978. Fisheries research notes. Fisheries Division, 

 Ministry of Commerce and Inductry, Tarawa, Kiribati, 58 p. 



of the two approaches were compared. If the variable 

 most related to fecundity in the stepwise regression 

 was not the one most related to fecundity in the par- 

 tial-correlation analysis, the stepwise regression model 

 was discarded and no relationship was assumed. 



In order to estimate egg production (Eq. 1), we 

 estimated the proportion of females in each 5-mm 

 length class from the total sample of each species. 

 The variance of these estimates was calculated by 

 using the normal approximation to the binomial dis- 

 tribution (Walpole, 1974). We assessed whether the 

 monthly percentage of annual egg production was 

 related to the proportion of annual recruitment in 

 the same month by rank-correlations (Conover, 

 1980). 



The average age of the potential spawning popu- 

 lation in each sample was compared by a nested 

 analysis of variance with month of sampling nested 

 within year. Significant differences between treat- 

 ments were identified from comparison of the least- 

 squares means of each treatment, as sample sizes 

 differed between cells (Sokal and Rohlf, 1981). 



Results 



Environmental parameters 



Sea-surface temperature in Kiribati varied little 

 throughout the year. During the study period, tem- 

 peratures varied between 29°C and 32°C (Table 2). 

 Rainfall varied along the Gilbert Island group; rain- 

 fall was higher in Butaritari than at the other sites. 

 Some rain fell throughout the study period but was 

 more intense during 1990 at all sites. Rainfall dur- 

 ing 1989 was below the long-term average at all 

 sites and was 16-50% that of 1990. The highest 

 rainfall fell during the north-east monsoon (Decem- 

 ber-April) at all sites. Winds were mostly light, and 

 varied in direction seasonally, blowing from the east 

 during the monsoon, but from the south-south-west 

 for the rest of the year (Table 2). 



Reproductive biology 



Maturation The length and age at first maturity 

 of A. sirm varied between sites (Table 3). Ambly- 

 gaster sirm matured younger and smaller in Kiribati 

 than elsewhere. Length and age at first spawning 

 were much greater than the length or age when fish 

 reached sexual maturity, but this size was similar 

 to that of fish from northern Australia (Table 3). 

 Herklotsichthys quadrimaculatus matured and were 

 capable of spawning at 70 mm length and 4 months 

 of age (Table 3). The relative size and age at which 

 fish matured (as a proportion of maximum size and 



