Read sardine_spawn_habitat.pdf text version


Pacific sardine (Sardinops sagax) spawning habitat in the California Current ecosystem

S. McClatchie, M.C. Ferguson, and R. Charter

Running head: Pacific sardine spawning habitat

Abstract: We used a Generalized Additive Model (GAM) to relate sardine (Sardinops sagax) egg and larval densities in the California Current ecosystem to environmental variables. In the GAMs we found significant relationship between temperature, salinity, mixed layer depth, zooplankton displacement volume and sardine egg and larval densities, but the results for eggs were biased by autocorrelation. When autocorrelation was accounted for we found no significant correlations between environmental variables and sardine egg densities. In contrast for sardine larvae, there remained significant (p < 0.001) correlations with mixed layer depth. Our analysis indicates that the spatial distribution of mixed layer depth and sardine larvae holds promise for improving sardine stock assessments incorporating environmental parameters.

Key words: sardine spawning habitat spatial autocorrelation.


The biomass of Pacific sardine (Sardinops sagax) in the California Current is characterized by high variability, part of which is caused by their response to the environment (MacCall, 2009). Variation in spawning habitat mediated by environment varying at different time and space scales ultimately impacts Pacific sardine recruitment and spawning stock biomass (Bakun, 1996). Which environmental variables have the most influence on Pacific sardine spawning habitat is currently uncertain. Various authors have emphasized the effects of temperature (Lluch-Belda et al., 1991), temperature and salinity (Checkley Jr et al., 2000), temperature and chlorophyll (Reiss et al., 2008), temperature and zooplankton (Lynn, 2003), and wind stress (Rykaczewski and Checkley Jr., 2008). It has been inferred that prey particle size structure is important (Rykaczewski and Checkley Jr., 2008). Pacific sardine spawn over a wide range of temperatures off the North American west coast (12 - 25o C)(Lluch-Belda et al., 1991; Lynn, 2003). Spawning is distributed bimodally in relation to temperature. Spawning around the 15o C mode occurs from central California to northern Baja California (Mexico), while spawning from southern California to southern Baja California (Mexico) is distributed around a 23o C mode (Lluch-Belda et al., 1991; Lynn, 2003). Lynn (2003) found the inshore limit for

S. McClatchie1 and R. Charter. Southwest Fisheries Science Center, National Marine Fisheries Service, 8604 La Jolla Shores Drive, La Jolla, California 92037 - 1508, U.S.A. M.C. Ferguson. Alaska Fisheries Science Center, National Marine Fisheries Service, 7600 Sand Point Way N.E., Seattle, Washington 98115, U.S.A.


Corresponding author (e-mail: [email protected]).

DOI: 10.1139/Zxx-xxx c 2009 NRC Canada

unknown 99: 1­19 (2009)


unknown Vol. 99, 2009

Pacific sardine spawning in the Southern California Bight (SCB) to be 12o C, and the offshore to be limited by zooplankton displacement volume in the range of 20 - 40 ml 1000 m-3 . We studied the spatial distribution of Pacific sardine eggs and larvae in the California Current ecosystem to define spawning habitat in relation a suite of environmental variables that have been measured over decades (Figure 1). Previous studies have focussed on two (Checkley Jr et al., 2000) to four years (Lynn, 2003), and sometimes as many as eight years (Reiss et al., 2008). We analyzed the CalCOFI dataset over 56 years from 1950 to 2005 and determined which environmental variables could be used to predict primary spawning habitat over the much longer time scale. The environmental space defining Pacific sardine spawning habitat is likely to be multivariate, but there are few environmental variables measured over the entire CalCOFI time series. The variables that are available for the whole series are bottle cast measurements of temperature, salinity, and oxygen, and net tows analyzed for for ichthyoplankton counts and zooplankton displacement volume. Recruitment appears to be more related to the survival of one year old fish rather than to survival of eggs and larvae (Watanabe et al., 1995; Rykaczewski and Checkley Jr., 2008; Takahashi and Checkley, 2008), but we lack a recruitment index for Pacific sardine, while we have extensive data on the distribution and abundance of the ichthyoplankton. Consequently the eggs and larvae are regarded as the only field-based proxy for recruitment, since all other recruitment indices are model-based (Hill et al., 2008). In this paper we are not attempting to predict Pacific sardine recruitment. To understand the effect of environmental forcing on spawning variability it is important to quantify the association between environmental variables and the distribution and abundance of Pacific sardine eggs and larvae, and to understand what constitutes primary spawning habitat. Our focus is on determining which environmental variables are related to sardine eggs and larval density after accounting for spatial autocorrelation. In this study we develop Generalized Additive Models (GAMs) relating the abundance of Pacific sardine eggs and larvae to environmental variables. We then determine the effect of the spatial field on the relationships derived from the GAMs and used the spatial correlations to parse out the nonspatial relationship from the spatial effects. We were interested in which environmental variables were correlated with densities of eggs and larvae given the complex spatial variability in the Southern California Bight. Ultimately this information is expected to help focus the development of an improved environmental index for incorporation in the Pacific sardine harvest guideline.


Survey coverage by years The CalCOFI surveys can be grouped into sampling domains that were the most commonly used over the duration of the time series. The largest is the original, or near original, area spanning the US west coast from Oregon to Baja California, Mexico (surveyed in 1951, 1952, 1954, 1956, 1958 - 1960, 1969 and 1972). The next largest domain runs from San Francisco to southern Baja California (surveyed in 1953, 1955, 1957, 1961 - 1966, 1968, 1974, 1975, 1978, 1980 and 1981). The smallest of the common sampling domains is the SCB (sampled over the entire time series). Among the CalCOFI surveys there are also some less common sampling domains. These are San Francisco to mid-Baja California (1979, 1984), San Francisco to San Diego (1982, 1983), Point Conception to mid-Baja California (1977), and Point Conception to southern Baja California (1967, 1970). The less common sampling domains all were sampled in the interval 1967 to 1984 during the period of the Pacific sardine fishery collapse, when Pacific sardine eggs were rare. The number of cruises analyzed in our study was large (237), but sample sizes of net tows with positive Pacific sardine egg counts where mixed layer depth and Brunt-Vaisala frequency could be estimated greatly reduced the sample size. 1,383 stations were used to develop Generalized Additive Models.

c 2009 NRC Canada

McClatchie, Ferguson, Charter


Oceanographic variables To determine the environmental correlates of actual spawning habitat we used hydrographic data collected by Scripps Institution of Oceanography and ichthyoplankton data collected by Southwest Fisheries Science Center on joint CalCOFI cruises. The hydrographic dataset, collected with water bottles and interpolated to standard depths at 10m intervals, included temperature, salinity data (as well as oxygen, chlorophyll, phaeopigments, primary productivity, and nutrients after 1985 that were not used in this study). The CalCOFI fisheries dataset obtained from oblique net tows provided Pacific sardine egg and larval counts and zooplankton displacement volumes. The fish larval volumes are such a small fraction of the net catch that they can be neglected when estimating the zooplankton volume. The hydrographic dataset and the fisheries dataset are not exactly spatially coincident. This is due to ship drift, the sequence of sampling on station, or net and bottle samples having been collected by different vessels during the same survey. In addition the navigational fixes for early CalCOFI cruises were often approximate due to limitations of the technology of the day. To relate the two datasets, we spatially gridded both the hydrographic and the fisheries data to a common grid. We averaged the hydrographic data in the upper 50 m at each station in order to merge the environmental variables with the egg counts and zooplankton displacement volumes obtained from oblique net tows. This is based on evidence from early CalCOFI studies that most of the Pacific sardine eggs were found shallower than 40 m depth and none were found deeper than 70 m (Ohman and Smith, 1995). The horizontal spacing between CalCOFI stations used in this study was 40 to 70 km. The vertical depth over which we averaged (50 m) and the spatial resolution of the horizontal grid (40 70 km) provides the fundamental sampling unit for the analysis. We are not interested in predicting or investigating phenomena that occur on vertical and horizontal scales that are smaller than those used to build the model. We calculated a proxy for water column stratification, as well as mixed layer depth for each station over the 1950 - 2005 CalCOFI time series. Water column stratification was estimated as the Brunt1 Vaisala frequency, N 2 = -gE where E = - d (units are cycles h-1 ) and g is acceleration due to dz gravity. Brunt-Vaisala frequency was calculated as the maximum value of the profile of N2 using the Oce package for analysis of oceanographic data in the R statistical language. We calculated optimal mixed layer depth (MLD) (Kara et al., 2000) as the depth below a column of well-mixed water, determined by iteratively searching for depths where |t | < 0.125. Optimal mixed layer depth represents the depth to which turbulent mixing penetrates the surface ocean (Kara et al., 2000). Differences between nets Descriptions of the standard techniques and gear used for the ichthyoplankton component of the CalCOFI surveys are available elsewhere (Smith, 1977) and are not repeated here. Egg data were available from 0.71-m diameter Bongo after 1977 and from the 1-m diameter ring nets prior to that. Nets were towed obliquely from 140 m to the surface from 1951 to 1969, and from 210 m to the surface subsequently (Ohman and Smith, 1995). Bongo net data were from the 505 µm mesh nylon net. Prior to 1969 the ring net was 550 µm silk mesh but was subsequently changed to 505 µm mesh nylon (see Table 1 in Ohman and Smith (1995). We tested whether the 505 µm mesh nylon 1-m ring net and the 0.71-m Bongo net towed obliquely from 210 m to the surface could be treated as equivalent after applying the standard CalCOFI conversion factors based on volume filtered. Similar but not exactly equivalent tests had previously been done for krill (Brinton and Townsend, 1981), for anchovy larvae (Hewitt, 1980) and for zooplankton (Ohman and Smith, 1995), but not for fish eggs. Both the ring and Bongo nets were towed at the same stations on five cruises in 1977 and 1978 (CalCOFI cruises 7712, 7801, 7803, 7805, 7807). These were the same nets and cruises addressed in the third net comparison of zooplankton catches by Ohman and Smith (1995).

c 2009 NRC Canada


unknown Vol. 99, 2009

Pacific sardine eggs were scarce on these cruises and there were insufficient Pacific sardine eggs in the ring net tows to make any comparison between nets, so we used northern anchovy (Engraulis mordax) eggs to compare the nets. We compared 82 pairs of net tows made at the same time and station using a 2-tailed Wilcoxon rank sum test for differences between the ranked observations. Generalized Additive Models We defined Pacific sardine spawning habitat in relation to a set of environmental variables for the California Current ecosystem using Generalized Additive Models (GAM). We determined which variables satisfied the following three criteria: (1) were available for the entire dataset, (2) were spatially co-located with the ichthyoplankton data, and (3) would affect the quality of spawning habitat. Temperature, salinity, oxygen and nutrients generally exhibit characteristic ranges in water masses in the California Current system, despite the non-conservative nature of oxygen and nutrients (Lynn and Simpson, 1987). We considered water masses to be important, given the observed spatial separation of inshore northern anchovy and more offshore Pacific sardine spawning. We chose temperature and salinity as predictors in the model because Pacific sardine have known temperature and salinity preferences for spawning (Checkley Jr et al., 2000). Recent work (Rykaczewski and Checkley Jr., 2008) indicated that Ekman pumping and wind stress curl influence the size structure of planktonic prey, and the distribution and relative success of Pacific sardine juveniles. The inferred mechanism is the effect of mixing on zooplankton production, and so we calculated a mixed layer depth from density profiles. The only index of prey resources available for the entire CalCOFI time series is zooplankton displacement volume. This variable includes zooplankton captured with the 505 or 550 µm mesh nets (e.g larger copepods, but not nauplii) with invertebrate predators such as chaetognaths and euphausiids. We expected components of the physical environment to interact to produce favorable conditions for zooplankton. However, zooplankton are patchily distributed, so we suspected that zooplankton displacement volume would not be present in high concentrations at all stations where the physical parameters produced favorable conditions. Covariates used in the GAMs were plotted against each other and the Spearman rank correlation coefficients calculated to determine if the predictors were sufficiently independent to be used in the GAM. Salinity and oxygen (r2 = -0.51) showed the highest correlations, indicating that there was some redundant information in these variables (Table 1), so oxygen was dropped as a predictor in the GAM. There was only low correlation between zooplankton and mixed layer depth (-0.24), or zooplankton and Brunt-Vaisala frequency (-0.14)(Table 1). The distribution of the dependent variable (Pacific sardine eggs) was highly skewed, and in addition was zero inflated. Zero inflation poses considerable problems for regression analyses, as discussed in a recent review (Ciannelli et al., 2008). For this study, we included the zeros (i.e. stations with no eggs) in a single phase GAM using a quasipoisson distribution, with variance proportional to the mean, and a logarithmic link function. Other methods for dealing with zero inflation (such as a twophase conditional probability GAM) also suffer limitations, and there is no established method for adequately dealing with zero inflation (Ciannelli et al., 2008). The model was developed using only the better sampled Southern California Bight data. Analyses were performed with the mgcv package (ver. 1.5-6) in the R statistical language (ver. 2.8.1) (Ihaka and Gentleman, 1996). We formulated the Generalized Additive Models as: ¯ ¯ µi E(Eggsi ) = f1 Ti + f2 Si + f3 log(Z)i + f4 M LDi + f5 Ni2 + Y, M + i


¯ ¯ µi E(Larvaei ) = f1 Ti + f2 Si + f3 log(Z)i + f4 M LDi + f5 Ni2 + Y, M + i


c 2009 NRC Canada

McClatchie, Ferguson, Charter


¯ ¯ µi E(Larvaei ) = f1 Ti + f2 Si + f3 log(Z)i + f4 M LDi + f5 Ni2 + f6 Eggsi + Y, M + i


where Eggsi are sardine egg densities and Larvaei are sardine larval densities (counts 10m-2 ). fi are tensor product smooths with thin plate regression spline basis designed so that unimportant terms ¯ ¯ can be zeroed (Wood, 2006). Ti and Si are the mean temperature (o C) and salinity (psu) in the upper 50 m, Zi is zooplankton displacement volume (ml 1000m-3 ), M LD is the mixed layer depth (m), and N 2 is the maximum Brunt Vaisala frequency (Brunt-Vaisala frequency) (h-1 ) in the mixed layer. i = ln(µi ) which is the log link function relating the expected value of the dependent variable to the linear predictor. Tensor product smooths are scale-invariant, meaning that they are appropriate for smoothing interactions of predictors measured in different units, or where different degrees of smoothing are necessary for different covariates (Wood, 2006). The model contained parametric variables for months nested within years (Y, M ) to account for monthly and interannual variation (but data were restricted to the spawning season, March through May).

Spatial structure analysis Spatial analyses were performed using the vegan package (ver. 1.15-4) in R. We tested for spatial autocorrelation in both the egg and the larval data. We were interested in the underlying spatial distribution of eggs and larvae, but our primary focus was to determine if autocorrelation was likely to bias (i.e. lead to over-estimation) the significance of relationships between environmental variables and the sardine eggs or larvae as determined by the GAM. In the presence of spatial autocorrelation the most significant variables may just be those whose underlying distribution is due to a latent (unobserved or unmeasured) varaible that also affects egg or larval distribution (Lennon, 2000). A dissimilarity vector was created from the egg and larval series and a distance vector was calculated from the geographic positions. Euclidean distance measures were used for both. The test was performed using a Mantel correlogram with either eggs or larval dissimilarity and the distance vector as the variables, and the Pearson correlation coefficient method. The test groups the egg or larval data into distance classes and calculates tests for significant correlation at each distance class. The pattern of the Mantel correlation by distance class plot provides information about the underlying spatial (or autocorrelation) structure. The correlations between eggs or larvae, the environmental variables and space were tested using simple and partial Mantel tests. The Spearman rank correlation was used to test correlations between eggs or larvae and environmental variables because the relationships were expected to be non-linear (Hsieh et al., 2005). Tests were based on the dissimilarity matrices for environmental variables and distance matrices of distances for the spatial variable. Mantel statistics were computed using 10,000 permutations and corrected for multiple comparisons.


Differences between nets We found no significant differences between the mean egg densities (standardized to eggs 10m-2 ) derived from the different net types, using all samples (p = 0.97, n = 82) or using only samples with positive egg counts (p = 0.66, n = 10). The lack of detectable difference in egg catches between the nets, due to low egg catches and high variability, prevents us from developing any correction factor for gear changes, as was done by Ohman and Smith (1995). For the zooplankon displacement volume data, we applied the correction factor of 1.36 to increase volumes of samples collected between 1969 and 1977, as recommended by Ohman and Smith (1995).

c 2009 NRC Canada


unknown Vol. 99, 2009

Generalized Additive Models The GAM model for sardine eggs (model 1) reported significant relationships between sardine eggs and several of the environmental variables (Table 2). The significance of these relationships is later shown to be undermined by autocorrelation. Highest densities of eggs in the model occured at temperature and salinitiy ranges known from published studies (Figure 2), and there was a temperature optimum at 14o C. The relationship with mixed layer depth and zooplankton volume is less well known, and an optimal mixed layer depth of 45 m has not been previously reported. The trend in egg densities with zooplankton volume showed an increasing trend at low volumes, and a decreasing trend at high zooplankton volumes. The GAM for sardine larvae also showed significant relationships with environmental variables (Table 2). We later show that the only variable that was significant after accounting for spatial structure was mixed layer depth. Larvae appear to be less related to salinity than are eggs, and there is no clear temperature optima (Figure 3). The relationship between larval densities and zooplankton volumes is complex. As was the case for eggs, there was an optimal mixed layer depth in the model, but the optimal mixed layer depth of 55 - 90 m was considerably deeper for larvae than for eggs (45 m).

Spatial structure analysis There was significant autocorrelation of sardine eggs at short distances. The distribution of the larvae is less skewed than the distribution of the eggs. The shape of the correlogram indicates a gradient or step in the spatial pattern of the eggs (Legendre and Fortin, 1989) (Figure 4). Sardine eggs are more abundant offshore compared to inshore and this gradient is consistent with the observed pattern in the correlogram. In contrast to the eggs, the larvae were not spatially autocorrelated at the measurement scales of this study. We cannot make any statement about autocorrelation below our sampling length scales. Simple Mantel tests showed that eggs were not significantly correlated with the environmental variables mean temperature, mean salinity, MLD, or logged zooplankton volume (Table 3, Env · Eggs). Eggs were significantly correlated with space (p = 0.022), indicating that egg numbers were similar at closely located stations, as expected when there is a degree of autocorrelation (Table 3, Env · Eggs). In contrast to the eggs, larvae were highly correlated with mixed layer depth, and were not significantly correlated with space (Table 4, Env · Larvae). We were also interested in the relationship of each environmental variable with space. For the Mantel analysis of eggs we selected locations where eggs were present and for larvae we selected where larvae were present. Consequently the Mantel correlation between environmental variables and space differs for the egg and larval datasets. With the exception of logged zooplankton in the egg dataset, all of the environmental variables were significantly related to space (Tables 3 & 4, Env · Space). Correcting for autocorrelation is therefore likely to be critical to detect relationships between environmetal variables and the eggs or larvae. None of the correlations between sardine eggs and environmental variables were significant after correcting for autocorrelation (Table 3, Env · Eggs -- Space). For sardine larvae, the correlation between larval densities and mixed layer depth was still highly significant (p < 0.001) after correcting for autocorrelation (Table 4, Env · Eggs -- Space). Finally, we were interested in the correlation between eggs or larvae and environmental variables, correcting for other environmental variables. None of the correlations were significant for eggs (Table 3, Env · Eggs -- *). For the larvae, the only relationship that was significant (p < 0.001) was larval density with mixed layer depth, indicating that this relationship was independent of the relationship with other environmental variables (Table 4, Env · Eggs -- *).

c 2009 NRC Canada

McClatchie, Ferguson, Charter



Previously published studies provided a lot of information about the environmental variables that are associated with small pelagic fish spawning habitat in the California Current System. The spatial distribution patterns of spawning of Pacific sardine and northern anchovy is different (Lluch-Belda et al., 1991; Checkley Jr et al., 2000; Reiss et al., 2008). California Cooperative Oceanic Fisheries Investigation (CalCOFI) egg distribution maps generally show northern anchovy eggs inshore in the SCB, while the Pacific sardine distribution extends further offshore and out of the wind-forced coastal upwelling. Checkley et al. (2000) showed that high concentrations of Pacific sardine eggs (> 25 eggs m-3 ) were associated with 12.5 - 15.5o C water, and were separated from northern anchovy eggs found at higher salinities (33.35 - 33.48). Reiss et al. (2008) found Pacific sardine eggs at 12 - 15o C associated with low chlorophyll-a (< 1.2 mg m-3 ) whereas northern anchovy spawned over slightly wider ranges of temperature (12.5 - 16.5o C) and chlorophyll-a (< 0.5 - 1.5 mg m-3 ). Spawning habitat of Pacific sardine and northern anchovy seems to be separated mainly by salinity and some measure of production. Earlier studies have also shown high densities of Pacific sardine eggs to the south of Point Conception (Kramer, 1970), which is the area of the SCB that experiences greatest wind stress in spring (Nelson, 1977). There is another center of primary spawning habitat around Punta Eugenia, Baja California Mexico (Kramer, 1970). These primary spawning habitat areas seem to be related to higher zooplankton production and possibly to the temperature gradients that can produce wind stress curl around headlands (Chelton et al., 2007). Point Conception is an area where cool water intrudes into the SCB from the north. The area also commonly exhibits enhanced primary production. Higher concentrations of Pacific sardine eggs in this area may reflect spawning fish exploiting the higher production of the area within the temperature ranges that are favorable for spawning. Accounting for the effect of autocorrelation revealed a key relationship that has not been addressed by earlier work, namely the significant correlation between mixed layer depth and the distribution and abundance of the sardine larvae. The seasonal deepening and subsequent erosion of the mixed layer is fundamental to production processes in upper 50 - 70 m of the water column where sardine larvae are found. While the correlation does not infer causation, it is not surprising that mixed layer depth would be related to the distribution of sardine larvae because it will affect the spatial distribution and abundance both of larval prey, and of larval invertebrate predators. Higher densities of Pacific sardine eggs and larvae were found in areas with mixed layers ranging from 40 m to 90 m. It may seem paradoxical that sardine larvae can be associated with deeper mixed layers. However, the offshore waters of the California Current Ecosystem commonly have deep chlorophyll maxima at depths greater than the mixed layer. When the mixed layer deepens, the deep chlorophyll maximum and any associated zooplankton are effectively brought closer to the mixed layer containing the sardine eggs. In this way the potential food environment for both adults and larvae could be enhanced by deepening of the mixed layer. Based on current understanding of the relationship between larval fish survival and turbulence (Ware and Thomson, 1991; Bakun, 1996; MacKenzie and Kiorboe, 2000) we expect best larval survival at mid-levels of stability at both large and small scales. This implies that larvae should be more abundant in mixed layers of intermediate depth because a shallow mixed layer implies stability and a deep mixed layer implies greater mixing. One reason why sardine larvae are more abundant in mixed layers of 55 - 90 m, which are relatively deep, may lie in the unsuitability of shallower mixed layers. It is possible that shallower mixed layers support greater numbers of predators, which is an hypothesis that could be tested in future work. In our study we cannot distinguish the effects of food environment from predation environment on the distribution of the eggs. The net mesh sizes (515 and 550 µm) are too coarse to capture nauplii, which are thought to be important prey for the larval sardine. The adult sardine feed on zooplankton that would be captured by the nets. Further, the proportions of zooplankton taxa captured by the

c 2009 NRC Canada


unknown Vol. 99, 2009

nets has changed over several decades due to reduction in the numbers of gelatinous zooplankton (mainly salps) (Lavaniegos and Ohman, 2003). The zooplankton displacement volume index is admittedly rather coarse, but its use is justified since it is the only zooplankton index available over the entire time series that is not aggregated by cruise (i.e. it retains the spatial information that is removed by aggregating samples). The effect of spatial autocorrelation demonstrated in this study suggests that more attention should be directed to spatial structure of the environment. The quality of Pacific sardine spawning habitat is almost certainly affected by predation as well as by bottom-up processes. Studies of the spatial distribution of ichthyoplankton predators and prey and temporal evolution of the mixed layer may prove fruitful in understanding the distribution, abundance and ultimately the survival of sardine larvae.


Ed Weber, Nancy Lo, Paul Smith, Steven Bograd, Roy Mendelssohn, Akinori Takasuka and Andrew Leasing all provided stimulating input. We also thank James Wilkinson and Ralf Goericke for providing the Scripps Institution of Oceanography hydrographic data and advising us during exploratory analysis of the data. We appreciated editing of the manuscript by Barbara Javor. Two anonymous reviewers greatly improved the manuscript. Funding was provided by the National Marine Fisheries Service Fisheries and the Environment (FATE) program.


Bakun, A. 1996. Patterns in the ocean: Ocean processes and marine population dynamics. California Sea Grant College System, University of California Sea Grant, California, USA, in cooperation with Centro de Investigaciones Biologicas de Noroeste, La Paz, Baja California Sur, Mexico. 323 pp. ISBN 1-888691-01-8 Brinton, E. and Townsend, A. 1981. A comparison of euphausiid abundances from bongo and 1-M CalCOFI nets. CalCOFI Reports 22: 111­125 Checkley Jr, D., Dotson, R. and Griffith, D. 2000. Continuous underway sampling of eggs of Pacific sardine (Sardinops sagax) and northern anchovy (Engraulis mordax) in spring 1996 and 1997 off southern and central California. Deep-Sea Research 47: 1139­1155 Chelton, D., Schlax, M. and Samelson, R. 2007. Summertime coupling between sea surface temperature and wind stress in the California Current System. Journal of Physical Oceanography DOI: 10.1175/JPO3025.1 Ciannelli, L., Fauchald, P., Chan, K., Agostini, V. and Dongsor, G. 2008. Spatial fisheries ecology: Recent progress and future prospects. Journal of Marine Systems 71: 223­236 Hewitt, R. P. 1980. Distributional atlas of fish larvae in the California Current region: northern anchovy, Engraulis mordax Girard, 1966 through 1979. CalCOFI Atlas 28: 1­101 Hill, K., Dorval, E., Lo, N., Macewicz, B., Show, C. and Felix-Uraga, R. 2008. Assessment of the Pacific sardine resource in 2007 for U.S. management in 2008. Tech. Rep. 413, U.S Department of Commerce, NOAA Technical Memorandum, NOAA-TM-NMFS-SWFSC-386 Hsieh, C., Glaser, S., Lucas, A. and Sugihara, G. 2005. Distinguishing random environmental fluctuations from ecologcal catastrophes for the North Pacific Ocean. Nature 435: 336­340 Ihaka, R. and Gentleman, R. 1996. R: A Language for Data Analysis and Graphics. Journal of Computational and Graphical Statistics 5(3): 299­314

c 2009 NRC Canada

McClatchie, Ferguson, Charter


Kara, A., Rochford, P. and Hurlburt, H. 2000. An optimal definition for ocean mixed layer depth. Journal of Geophysical Research Oceans 105(C7): 16,803­16,821 Kramer, D. 1970. Distributional atlas of fish eggs and larvae in the California Current region: Pacific sardine, Sardinops caerulea (Girard), 1951-1966. CalCOFI Atlas 12 Lavaniegos, B. and Ohman, B. 2003. Long-term changes in pelagic tunicates of the California Current. Deep-Sea Research II 50: 2473­2498, DOI:10.1016/S0967­0645(03)00,132­2 Legendre, P. and Fortin, M. 1989. Spatial pattern and ecological analysis. Vegetatio 80: 107­138 Lennon, J. 2000. Red-shifts and red herrings in geographical ecology. Ecography 23: 101­113 Lluch-Belda, D., Lluch-Cota, D., Hernandez-Vazquez, S. and Salina-Zavala, C. 1991. Sardine and anchovy spawning as related to temperature and upwelling in the California Current system. CalCOFI Reports 32: 105­111 Lynn, R. 2003. Variability in the spawning habitat of the Pacific sardine (Sardinops sagax) off southern and central California. Fisheries Oceanography 12(6): 541­553 Lynn, R. and Simpson, J. 1987. The California current system: the seasonal variability of its physical characteristics. Journal of Geophysical Research Oceans 92(C12): 12,947­12,966 MacCall, A. 2009. Mechanisms of low-frequency fluctuations in sardine and anchovy populations. In D. Checkley, J. Alheit, Y. Oozeki and C. Roy, eds., Climate change and small pelagic fish, 285­299. Cambridge University Press MacKenzie, B. and Kiorboe, T. 2000. Larval fish feeding and turbulence: A case for the downside. Limnology and Oceanography 45(1): 1­10 Nelson, C. 1977. Wind stress and wind stress curl over the California Current. Technical Report NMFS SSRF-714, NOAA Ohman, M. and Smith, P. 1995. A comparison of zooplankton sampling methods in the CalCOFI time series. CalCOFI Reports 36: 153­158 Reiss, C., Checkley Jr, D. and Bograd, S. 2008. Remotely sensed spawning habitat of Pacific sardine (Sardinops sagax) and Northern anchovy (Engraulis mordax) within the California Current. Fisheries Oceanography 17(2): 126­136 Rykaczewski, R. and Checkley Jr., D. 2008. Influence of ocean winds on the pelagic ecosystem of upwelling areas. Proceedings of the National Academy of Sciences 105(6): 1965­1970 Smith, P. 1977. Standard techniques for pelagic fish egg and larva surveys. FAO Technical Paper 175, FAO Takahashi, M. and Checkley, D. M. 2008. Growth and survival of Pacific sardine (textitSardnops sagax) in the California Current region. Journal of Northwest Atlantic Fisheries Science 41(doi:10.2960/J.v41.m626): 129­136 Ware, D. and Thomson, R. 1991. Link between long-term variability in upwelling and fish production in the Northeast Pacific Ocean. Canadian Journal of Fisheries and Aquatic Sciences 48: 2296­2306 Watanabe, Y., Zenitani, H. and Kimura, R. 1995. Population decline of the Japanese sardine, Sardinops melanostictus owing to recruitment failures. Canadian Journal of Fisheries and Aquatic Sciences 52: 1609­1616

c 2009 NRC Canada


unknown Vol. 99, 2009

Wood, S. 2006. Generalized additive models. An introduction with R. Chapman and Hall/ CRC, Boca Raton, Florida, USA

c 2009 NRC Canada

McClatchie, Ferguson, Charter Table 1. Pair-wise Spearman correlations of predictor variables used in GAMs. MLD = mixed layer depth (m). Brunt-Vaisala Freq. = Brunt-Vaisala frequency (h-1 ). Mean temp = mean temperature in the upper 50 m of the water column (o C). Mean sal = mean salinity over the same depth range as temperature. Zoopl. DV = zooplankton displacement volume from oblique net tows (ml 1000 m-3 ).


MLD Brunt-Vaisala Freq. Mean temp. Mean sal. Zoopl. DV

MLD 1 0.45 -0.02 0.08 0.24

Brunt-Vaisala Freq. 1 0.47 0.07 -0.14

Mean temp.

Mean sal.

Zoopl. DV

1 0.34 -0.47

1 0.08


c 2009 NRC Canada


unknown Vol. 99, 2009

Table 2. Comparison of the GAMs relating sardine eggs (model 1) and larvae (models 2 & 3) to environmental variables. Model numbers refer to the model equations in the text. An approximate indicator of model fit is lower GCV (generalized cross validation function) and higher percent explained deviance. Significant differences denoted by ** = p<0.01 and *** = p<0.001. T = mean temperature (o C) in upper 50 m, S = mean salinity in upper 50 m (psu), MLD = mixed layer depth (m), N2 = Brunt-Vaisala frequency (h-1 ), Z = zooplankton displacement volume (ml 1000m-3 ), E is sardine egg densities (eggs 10m-2 ) and Month:Year is a nested linear term.

Model (1) (2) (3)

GCV 917 502 303

% Explained deviance 19.1 22.8 56.3

Significant predictors T***, S*, MLD**, Z***, Month:Year* T***, MLD***, Z** T***, MLD**, N2 *, Z***, E***

c 2009 NRC Canada

McClatchie, Ferguson, Charter Table 3. Mantel statistics and probabilities (in brackets) for tests of correlation between environmental variables (temperature, salinity, mixed layer depth and logged zooplankton displacement volume) and sardine eggs (Env · Eggs), environmental variables and geographical position (Env · Space), environmental variables and eggs acounting for spatial autocorrelation (Env · Eggs -- Space), and environmental variables and eggs accounting for all other environmental variables and space (Env · Eggs -- *).


Variable Temp Salinity log(Zoo) MLD Space

Env · Eggs 0.018 (0.085) -0.004 (0.569) 0.004 (0.361) 0.023 (0.052) 0.024 (0.022)

Env · Space 0.055 (0.003) 0.268 (<0.001) 0.026 (0.094) 0.299 (<0.001) -

Env · Eggs -- Space 0.008 (0.239) -0.012 (0.872) -0.002 (0.537) 0.008 (0.247) -

Env · Eggs -- * 0.006 (0.281) -0.008 (0.728) -0.005 (0.643) 0.016 (0.111) 0.024 (0.029)

c 2009 NRC Canada


unknown Vol. 99, 2009

Table 4. Mantel statistics and probabilities (in brackets) for tests of correlation between environmental variables (temperature, salinity, mixed layer depth and logged zooplankton displacement volume) and sardine larvae (Env · Larvae), environmental variables and geographical position (Env · Space), environmental variables and larvae acounting for spatial autocorrelation (Env · Larvae -- Space), and environmental variables and larvae accounting for all other environmental variables and space (Env · Larvae -- *).

Variable Temp Salinity log(Zoo) MLD Space

Env · Larvae -0.020 (0.967) 0.019 (0.046) 0.012 (0.163) 0.079 (<0.001) 0.007 (0.203)

Env · Space 0.068 (<0.001) 0.354 (<0.001) 0.080 (<0.001) 0.284 (<0.001) -

Env · Larvae -- Space -0.024 (0.995) 0.015 (0.068) 0.012 (0.155) 0.064 (<0.001) -

Env · Larvae -- * -0.043 (1) 0.002 (0.414) -0.006 (0.689) 0.058 (<0.001) -0.003 (0.605)

c 2009 NRC Canada

McClatchie, Ferguson, Charter


Figure legends

Figure 1: Map of the study area showing stations used in the analysis. Data were restricted to March through May in the Southern California Bight region, and included the years 1950 to 2005. SF = San Francisco, SCB = Southern California Bight, BCM = Baja California Mexico. Figure 2: Visualization of response surfaces of Pacific sardine egg densities (eggs 10m-2 ) from model (1). (A) Smoothed surface in terms of mean temperature (o C) and mean salinity (psu) in the upper 50 m of the water column. (B) Contour plot of the temperature and salinity surface. (C) Response surface in terms of mixed layer depth (m) and zooplankton displacement volume (ml 1000m-3 ). (D) Contour plot of the mixed layer depth and zooplankton displacement volume surface. Figure 3: Visualization of response surfaces of Pacific sardine larval densities (larvae 10m-2 ) from model (3). (A) Smoothed surface in terms of mean temperature (o C) and mean salinity (psu) in the upper 50 m of the water column. (B) Contour plot of the temperature and salinity surface. (C) Response surface in terms of mixed layer depth (m) and zooplankton displacement volume (ml 1000m-3 ). (D) Contour plot of the mixed layer depth and zooplankton displacement volume surface. Figure 4: Autocorrelation pattern for sardine eggs from a Mantel correlogram. Solid squares indicate significant (p < 0.05) autocorrelation.

c 2009 NRC Canada

16 Fig. 1.

unknown Vol. 99, 2009

c 2009 NRC Canada

McClatchie, Ferguson, Charter Fig. 2.


c 2009 NRC Canada

18 Fig. 3.

unknown Vol. 99, 2009

c 2009 NRC Canada

McClatchie, Ferguson, Charter Fig. 4.


c 2009 NRC Canada


19 pages

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate


You might also be interested in

Feeding the Fish
Coastal: Potential Consequences of Climate Change