Butler et a\ Biology and population dynamics of Sebastes levis in the southern California Bight 



265 



Moreover, the geographic area of the CalCOFI survey and 

 cowcod stock are both centered in the SCB. 



Samphng gear, samphng procedures, and standardiza- 

 tion of numbers of larvae caught in CalCOFI tows are 

 described by Moser et al. (1993) and Ohman and Smith 

 (1995). Our analysis used data from all tows within the 

 "current" sampling area in the SCB during calendar years 

 1951-98 because it was the largest region sampled consis- 

 tently since 1951 (Hewitt, 1988; Moser et al., 1993; 1994), 

 and because cowcod larvae are most common there (Moser 

 et al., 1994). CalCOFI data from the current sampling pat- 

 tern included a total of 46 "seasons" used in modeling (e.g. 

 the 1951 season was July 1951-June 1952) and 12,274 

 bongo or ring net tows of which 120 (0.98'^) contained at 

 least one cowcod larva. Almost all positive tows (116 or 

 97%) were inshore of CalCOFI station 67.5 (Moser et al., 

 1994). Almost all positive tows ( 117 or 98%) were made dur- 

 ing January-June (Moser et al., 1994). Numbers of positive 

 tows ranged from 5 to 32 per month between January and 

 June and only 1 to 2 per month otherwise. Based on these 

 preliminary results, we used data for all tows («=5003) 

 collected inshore of CalCOFI station 67.5 during Janu- 

 ary-June for the remainder of our analysis. 



We used a logistic model to derive a standardized index 

 of larval presence for cowcod from CalCOFI data. The lo- 

 gistic model was a generalized linear approach (McCullagh 

 and Nelder, 1989) that accommodates zeroes (tows catching 

 no cowcod larvae). It was fitted to tow-by-tow CalCOFI data 

 by logistic regression (assuming a binomial distribution 

 for statistical errors). The dependent variable was (if no 

 cowcod larvae were observed in a tow) or 1 (if larvae were 

 observed). Independent variables included years, months, 

 and a dummy variable that was 1 if the tow was in the 

 "inshore" area (Butler et al., 1999) and otherwise. The 

 best model for cowcod CalCOFI data was identified by us- 

 ing a step-wise procedure and Mallow's C statistic. The 

 index of abundance was the expected probability that a 

 CalCOFI tow is positive for cowcod larvae in each year 

 for an arbitrary reference month and arbitrary reference 

 location. 



Trawl surveys 



Two sets of trawl survey data were available for cowcod. 

 Trawl survey data collected by the Los Angeles City Sani- 

 tation District and Orange County Sanitation District 

 (LAOCSD) off southern California were used as an index 

 of presence for juvenile cowcod. Beginning in 1973, the 

 Los Angeles City Sanitation District sampled twelve sta- 

 tions along four cross-shelf transects and at three depths 

 (23 m, 61 m, and 137 ml twice each year (Stull, 1995; 

 Stull and Tang, 1996). Beginning in November 1970, the 

 Orange County Sanitation District sampled a fixed grid of 

 8 stations at 20-170 m quarterly (Mearns, 1979). Juvenile 

 cowcod in these trawls ranged from 3 to 38 cm in length. 

 Catch rates were highly variable; therefore we used a 

 simple average of the proportion of positive tows in both 

 surveys as a single index (LAOCSD) of juvenile cowcod 

 presence-absence in the SCB during the 1972-94 seasons 

 (Mangel and Smith, 1990). 



CPFV catch per unit of effort (CPUE) 



We calculated a habitat-area-weighted (Hilborn and Wal- 

 ters, 1992) average recreational catch per unit of fishing 

 effort (CPUE) index for cowcod from CPFV logbook data 

 for trips in the SCB during 1963-97. As described in the 

 discussion section, the index measured catch rates while 

 accounting for important changes in the spatial distribu- 

 tion of fishing effort over time and spatial differences in 

 abundance trends. Data for trips before 1964 were not 

 available because cowcod catches were combined with rock- 

 fish in early years. Each record contained total number of 

 rockfish caught, number of cowcod caught, and total angler 

 hours from logbooks for one month and one "block" ( lO'x 10' 

 area, Fig. 1). 



We assumed total angler hours reported on CPFV logs for 

 sampling blocks with rockfish catches during November- 

 April was a measure of relative fishing effort for cowcod. 

 CPUE was in units of numbers of fish per angler hour 

 (fish/h). 



We used CPFV logbook records for November-April to 

 model trends because the CPFV fishery tends to target 

 rockfish during the winter when migratory game fish are 

 seldom caught. Data from blocks in U.S. waters south of 

 Point Conception (blocks 651-897, Butler et al., 1999) were 

 used in the analysis so that CPUE was measured for the 

 entire SCB. Logbooks for 1979 were not summarized by 

 month in logbook records and were therefore excluded. We 

 excluded records for blocks 600, 699, 700, 799, 800, and 899 

 because these codes are used for data of uncertain origin. 

 We excluded a few records that reported cowcod catches 

 larger than total rockfish catches, and records with high 

 catches from blocks with no cowcod habitat as likely data- 

 base errors. We also excluded data for the 1979 and 1998 

 seasons because data for some months were missing and 

 the number of blocks with logbook reports was low (<150). 



It was necessary to have at least one logbook record for 

 each spatial stratum during each season, but many blocks 

 had missing data for some seasons. We therefore strati- 

 fied CPFV logbook data based on "pseudo-blocks." In some 

 cases, pseudo-blocks were the same as fishing blocks. In 

 other cases, pseudo-blocks were composed of many fishing 

 blocks with similar average catch rates. 



The first step in stratifying the data was to delete data 

 for blocks with mean CPUE (over the entire time series) 

 that was zero or in the first quartile (<0.05 cowcod per 

 angler day). Blocks with zero CPUE values were from 

 areas where cowcod had never been reported and where 

 there was probably no habitat. Blocks with very low mean 

 CPUE provided little information about trends in cowcod 

 abundance. Of 190 blocks (covering 19,000 nmi^), there 

 were 102 blocks with mean cowcod CPUE greater than 

 the first quartile. 



In the second step, we calculated the number of seasons 

 with rockfish catch and effort data for each block. Twenty- 

 seven blocks had complete time series and were assigned 

 to a pseudo-block that was the same as the original fishing 

 block. 



The third step was to assign blocks with incomplete time 

 series to pseudo-blocks based on mean CPUE. Specifically, 



