668 



Fishery Bulletin 93(4), 1995 



winter (23 January to 22 February), throughout the 

 entire second and fourth winters, and for the one 

 summer sampling period. Current meter data were 

 not available for the entire third winter. For analy- 

 sis, hourly wind and current data were averaged over 

 24 hours to generate estimates of mean flow, which 

 were then decomposed into along-shore and cross- 

 shelf components of motion. 



Data analysis 



The A. vulpes leptocephali from all six nets for each 

 night during the winter sampling periods were 

 summed for time-series analysis. To analyze the re- 

 lationship between recruitment patterns and lunar 

 phase, time series of fish abundances for the four 

 winters were joined according to lunar month. Nine 

 days were deleted from the beginning and six days 

 from the end of 1991-92 (winter 2), and three days 

 were deleted from the beginning of 1993-94 to en- 

 sure continuity with respect to the lunar month. 

 Analyses were carried out on the resultant 277-day 

 period. The time series was log 10 (x+1) transformed 

 to minimize the influence of several large peaks in 

 the data. All time-series analyses were completed by 

 using the statistical package Mesosaur (Kuznetsov 

 and Khalileev, 1991). 



A periodogram of the recruitment data was con- 

 structed to identify dominant periodicities within the 

 277-day time series. An autocorrelation function was 

 then plotted to describe more accurately the cyclical 

 patterns of the data. Finally a cross-correlation was 

 run between the abundance time series and the cor- 

 responding "hours of dark flood tide" of each night, 

 which previous studies had identified as an impor- 

 tant variable affecting recruitment patterns of a 

 number of taxa (Shenker et al., 1993; Thorrold et 

 al., 1994b). The hours of dark flood tide is a measure 

 of the total number of flood tide(s) that occur between 

 sunset and sunrise under moonless conditions. This 

 variable differs from lunar phase in a subtle but sig- 

 nificant manner: as the time of moonrise becomes 

 progressively later over consecutive nights, greater 

 portions of the evening flood tide occur during dark- 

 ness prior to moonrise. Because significant auto- 

 correlations exist in both the larval abundance and 

 the tidal time series, the resultant correlation coeffi- 

 cients could not be assigned statistical significance, 

 and therefore confidence limits are shown only for 

 reference. These plots can be used to center periodi- 

 cities in recruitment with respect to the hours of dark 

 flood tide. 



Cross-correlations were used to examine responses 

 of recruiting larvae to wind and current patterns. 

 Because wind and current data were collected incon- 



sistently over the four winters, each year was ana- 

 lyzed separately. Significant autocorrelations present 

 in both recruitment and environmental data meant 

 that standard correlation coefficients would be arti- 

 ficially high (Chatfield, 1979). To remove the effects 

 of these autocorrelations, ARIMA (Auto Regressive 

 Integrated Moving Average) models were fitted to all 

 data. Residuals generated from the ARIMA models 

 were then used in the cross-correlations. Only those 

 correlations that identified a response of larvae to 

 particular wind and current patterns on a lag of up to 

 three days (i.e. fish moving onshore up to three days 

 after a specific wind or current pattern) are presented. 

 Occasional correlations at lags of greater than three 

 days were observed but are difficult to interpret in a 

 biological sense and may be statistical artifacts. 



Although the one set of summer data was not long 

 enough to permit rigorous analysis for cyclical pat- 

 terns, and the level of recruitment was generally too 

 low for correlation with environmental conditions, 

 the data were examined for resemblance to patterns 

 identified by the time series and correlation analy- 

 ses of the winter data sets. 



Otolith analysis 



Otoliths were removed from 150 of the 875 larvae 

 collected throughout the 1993-94 winter season. All 

 fishes collected during this winter were preserved in 

 70% ethanol. Specimens selected for analysis were 

 chosen from each day when recruits were captured. 

 However, preservation problems in some larger 

 samples prevented a more detailed examination of 

 the hatching patterns of large pulses of recruits. 



Sagittae were dissected and mounted in cyanurate 

 glue on a labelled glass slide. After curing for 24 

 hours, otoliths were polished down to the midplane 

 with a graded series of lapping papers with grits 

 ranging from 9 to 0.3 microns. A circle etched into 

 the slide with an electronic engraver prevented the 

 cyanurate glue from dislodging from the slide dur- 

 ing polishing. Prepared slides were projected onto a 

 computer screen at lOOOx by using a Sony TR81 cam- 

 era integrated through a Macintosh 2CI computer 

 equipped with Media-Grabber. This greatly facili- 

 tated counting and allowed two readers indepen- 

 dently to view the otolith(s) back- to-back under simi- 

 lar lighting conditions. Otoliths were randomized and 

 each otolith was read twice by each reader; if the 

 different readers obtained mean counts within five 

 increments of each other, the mean values were av- 

 eraged and used for analysis. The otolith was dis- 

 carded if differences in mean counts were greater 

 than five increments. A total of 87 of the 150 mounted 

 otoliths met this readability criterion. 



