Water mass sampling and otolith chemistry
Similar methods were used in the central and southern California surveys (detailed in Wyllie Echeverria et al. 1990; Nishimoto and Washburn 2002; Sakuma et al. 2000) to collect the otoliths and oceanographic data for this study. Fish were collected at night with a modified Cobb mid-water trawl with a 9 mm codend towed at depth for 15 min at ~5 km h−1 covering ~1.5 km (Nishimoto and Washburn 2002). The opening of the net used in both surveys was approximately 10 m wide and 14 m high when trawling at a headrope depth of 20 m. The depth interval for each haul was estimated as the distance from the mean depth, d0, of the headrope (rope across the top of the net opening) to 14 m below d0 (Table 1). Vertical profiles of potential temperature T and salinity S (averaged into 1-m depth bins) were obtained either immediately before or after each haul to at least 200 m or a few meters above shallower bottom depths. A water mass where fish were sampled was defined by the profile within the depth interval fished.
Otoliths from 68 specimens of shortbelly rockfish sampled from the central and southern California collections were assayed using laser ablation inductively coupled plasma mass spectrometry (LA-ICPMS) (Fig. 1, Table 1). The fish ranged from 12.8 mm to 47.2 mm SL. Shortbelly rockfish were the most abundant of the pelagic juvenile rockfishes collected in the survey. Adults of this active, schooling species range from southern British Columbia to southern Baja California, and larvae are found as much as 400 km from shore (Love et al. 2002). Fifty-seven specimens represent all areas defined by Nishimoto and Washburn (2002) where shortbelly rockfish were collected during 3–15 June 1998 in southern California (Fig. 1b, Table 1). Eleven otolith samples from central California were collected during 11–27 May 1998 from the Farallon Islands west of San Francisco to Cypress Point, Monterey (stations 1–6 in Fig. 1a and Table 1). The otoliths of one to four specimens per station were assayed (Table 1).
The otolith samples from the two surveys were handled somewhat differently. Specimens from the 1998 southern California survey were subsampled in 2000 from an ETOH-preserved archive, measured, and the sagittae extracted. The fish had been frozen at sea, then thawed and archived within several months after collection. The otoliths were rinsed in deionized water, air dried and stored in plastic bags for several months until the time when each otolith was affixed to a plastic slide using epoxy resin (Epo-Thin, Buehler). The fish collected in the central California survey were frozen at sea. Within several months, the specimens were thawed and measured, and the otoliths extracted, rinsed, dried and affixed to glass slides using a clear nail polish epoxy (Sally Hansen’s Hard-as-Nails brand).
All otoliths were polished down to about 15 μm above the center of otolith nucleus using a lapping wheel and 9, 3, and 1 μm 3 M diamond polishing films. After polishing, decontamination steps described by Ruttenberg et al. (2005) and Warner et al. (2005) were performed in a clean laboratory on individually isolated otoliths.
Otolith material produced within 1 week preceding capture was analyzed on a Finnigan MAT Element 2-sector field ICPMS as described in Warner et al. (2005) and Ruttenberg et al. (2005). A VG UV microprobe Nd:YAG (Neodymium-doped Yttrium Aluminum Garnet) 266 nm laser ablation system was outfitted with a helium aerosol carrier system to transfer the sample from the enclosed sample chamber to the ICPMS as described in Zacherl et al. (2003). The laser was set at 0.1 mJ at 3 Hz. The laser was used to ablate three sample spots in the same growth zone near the edge of the anterior rostrum of each otolith with care taken to avoid epoxy that overlaid one or two outermost increments. For the southern California otoliths, run in May 2002, the laser aperture was set to 1 μm which emitted a beam that ablated a pit diameter of about 20 μm. For the central California otoliths, run in September 2003, the laser aperture was set to 2 μm which ablated a pit diameter of about 30 μm.
Two samples were collected at each of the three spots to assay a suite of isotopes: Calcium 48 (48Ca), Strontium 86 (86Sr) Barium 138 (138Ba) and Lead 208 (208Pb) were determined by using low resolution mode (R = 300); Magnesium 24 (24 Mg), Calcium 48 (48Ca), Manganese 55 (55Mn), Iron 56 (56Fe), and Zinc 64 or Zn 66 (64Zn for SBC samples and 66Zn for CC samples) were determined in medium resolution mode (R = 4,000). Eight laser pulses were emitted on a targeted spot to collect each sample. Each targeted spot was pre-ablated with two laser pulses as a precautionary cleaning measure. The isotope intensities of each sample were blank-corrected by subtracting isotope intensities of a 1% nitric acid (HNO3) instrument blank preceding the sample sequence. The plasma conditions were kept similar for both ablated material and solution-based standards by aspirating a 1% nitric acid solution during the ablation process and keeping the gas flows constant. The abundance of an element was expressed as a ratio relative to the amount of calcium to control for the amount of material analyzed per sample spot. A solution standard of known element/Ca ratios was used to correct for instrument bias and convert to molar elemental ratios. Solution-based standards were used in an attempt to match the high calcium matrix of the otoliths. The currently available solid standard materials, such as NIST 612 glass, are very different in composition from the otolith (Pearce et al. 1997). Solid calcium carbonate pellets have been prepared to solve this issue, but there are problems associated with heterogeneity.
Average detection limits for each element per sequence (n = 20 sequences), calculated as 3 × SD of the intensity of 1% HNO3 blanks and expressed as ratios of the isotope intensity and mean otolith Ca48 intensity, were 24 μmol·mol−1 for Mg/Ca, 0.95 μmol·mol−1 for Mn/Ca, 12.74 μmol·mol−1 for Fe/Ca, 6.77 μmol·mol−1 for Zn/Ca, 59 μmol·mol−1 for Sr/Ca, 0.06 μmol·mol−1 for Ba/Ca, 0.06 μmol·mol−1 for Pb/Ca.
We analyzed solid glass standard reference material (NIST 612) at the beginning and end of each workday to check overall analytical precision or repeatability for the entire study. Average percent relative standard deviations (n = 10 workdays) were 21% for Mg/Ca, 13% for Mn/Ca, 26% for Fe/Ca, 37% for Zn/Ca, 39% for Sr/Ca, 10% for Ba/Ca, 10% for Pb/Ca. The NIST 612 was ablated using the same method parameters as that used for the otolith samples. The constraint of using small 20 μm and 30 μm spot sizes resulted in low signal to noise for Sr/Ca estimates in the NIST 612 standard. An alternative reproducibility estimate for the study was determined for Sr/Ca by ablating material (three spots) along a randomly selected band of growth increments between the core and the edge of an otolith at the end of each sequence. The average percent relative standard deviation was 8% for Sr/Ca (n = 19 sequences).
We excluded negative blank-corrected intensity values from the sample spot dataset. We also excluded outliers from the sample spot dataset. Outliers were identified by examining frequency distribution and normal probability plots of the log-transformed element/Ca sample spot data. The outlier values were at least 1.5 times the interquartile range from the median of the log-transformed data (SPSS, Inc 2002). The exclusions reduced the original dataset of three samples of seven isotopes per otolith to 0–3 samples of each isotope per otolith. The abundance of each element from the otolith edge was estimated from either a single sample spot or the mean of two or three sample spots after the molar ratio dataset was log-transformed (i.e., log(μmol element/mol Ca)). A specimen lacked an abundance estimate for a given element if the readings from all three sample spots at the otolith edge were excluded.
Relating otolith chemistry to in situ water masses
We examined the T-S distribution within the trawling depth ranges to resolve water mass groups. Principal component analysis (PCA) of T (15 1-m bin averaged variables, Td = 0, 1, 2, …, 14) and S (15 variables, Sd = 0, 1, 2, ..., 14) profile data over the depth interval of each haul was used to confirm the water mass groups. T and S values from each profile were normalized by subtracting the mean of all profiles and dividing by the standard deviation of all profiles.
Otoliths were assigned to the water mass groups by pairing hauls with vertical T-S profiles. All analyses were performed on log-transformed otolith element ratio data. Prior to the analyses, we assessed the normality and homoscedacity of the otolith data among water mass groups (SPSS, Inc 2002). We evaluated whether fish size was a confounding factor influencing the effect of water mass type on otolith element concentration. Rather than use log(element/Ca) in our analyses when the relationship between an element and fish size was significant (Pearson correlation coefficient r, α = 0.01), we used the residual values from the least squares regression of log(element/Ca) as a function of log(SL) to remove the fish size effect on the concentration of an element in otoliths (Systat Software Inc 2006).
We performed one-way analysis of variance (ANOVA) to test for the effect of the water mass group on individual log-transformed element ratios in otoliths (α = 0.01) (SPSS, Inc 2002). Although we would have preferred to use a nested ANOVA to account for within-group variability in individual elemental abundances due to station (i.e., water mass defined by the CTD cast data at an individual station) differences, the dataset was unsuitable because 10 of the 24 stations were represented by only one otolith, including all but one of the stations from central California. The Tukey’s honestly significant difference (HSD) test was used to detect a posteriori differences among means (α = 0.01).
We used PCA to examine whether variability among the 7-element signatures of otoliths showed natural groupings by water mass. Otoliths with excluded data from all three sample spots for at least one of the seven elements were not included in the PCA.
To test whether the elemental signature of otoliths identified water mass membership, we used canonical discriminant function analysis (DFA) (SPSS Inc 2002). The DFA included only the elements that showed significant differences among water masses as determined in the ANOVA. In comparison to the PCA, the DFA was run on an expanded subset of otoliths that had readings of the elements which showed a water mass grouping effect regardless of whether abundances of the other elements could be estimated.
The classification accuracy of each DFA was evaluated by leave-one-out cross-validation, also called jack-knife reclassification (SPSS Inc 2002; White and Ruttenberg 2007). The prior probabilities of group membership were assumed to be uniform. We used the randomization test of White and Ruttenberg (2007) to estimate the probability (p-value) that the observed jack-knife reclassification success rate was drawn from a null distribution of jack-knife values (n = 5,000 randomizations) given no difference among groups of samples.
We examined relationships between element abundances in the otolith and relationships between otolith elemental properties and T and S. We used the Pearson correlation coefficient, r, to determine whether otolith element abundances that were identified as the best discriminators of the water mass groupings co-varied (SPSS, Inc 2002). We estimated the correlation between the average otolith element abundance at a station and the average T or S. If the correlation was significant, we used least squares regression to model the relationship between T or S, the environmental predictor, and otolith log[element/Ca], the dependent variable, by fitting a linear equation to the observed data (Systat Software Inc 2006).