Environmental predictors of walleye pollock recruitment in the Eastern Bering Sea

Franz J. Mueter Brenda L Norcross Research Scientist, Sigma Plus Consulting Professor, Fisheries Oceanography

697 Fordham Drive Institute ofMarine Science Fairbanks, Alaska, 99709 School ofFisheries and Ocean Sciences Ph: 1-907-479-8815 University of Alaska Fairbanks fax: 1-907-479-8815 Fairbanks, Alaska 99775-7220 e-mail: finueter@alaska.net Ph: 1-907-474-7990

fax: 1-907-474-1943 e-mail: norcross@ims.uaf.edu

Principal Investigators

Annual progress report

to the

Pollock Conservation Cooperative Research Center

Prepared by: Franz J. Mueter

Michael C. Palmer Brenda L. Norcross

December 2003

Table of contents Table ofcontents 2 List ofTables 3 List ofFigures 3 Abstract 4 Introduction 4 Project goals and objectives 5 Hypotheses linking pollock recruitment to environmental variables 6

1. Winter ice conditions and the cold pooL 6 2. Timing of ice retreat and the spring bloom 6 3. Mixed layer dynamics and summer production 7 4. Advection 8

Methods and progress to date 8 Walleye pollock data and biological indices 9

. Oceanographic indices 9 Atmospheric indices 12 Data analysis 13

Findings 14 Literature cited 15

2

List of Tables Table 1: Effects of ice retreat and predator abundance on recruitment ofwalleye pollock. The

expected response ranges from a strong negative effect (--) over neutral or ambiguous effects (+/-) to strong positive effects (++) 7

Table 2: Cross-correlations among various ice related indices, including southernmost extent of sea ice along 168°W (168W), spatial sea ice index (SSI), temporal sea ice index (TSI), cold pool index (CPI), and average bottom temperature (BT) 11

List of Figures Figure 1: Map of the southeastern Bering Sea. The NMFS survey area (heavy line), M2 mooring

location (56.8° N, 164.0° W) and three hydrographic domains (inner, middle and outer shelves) separated by the 50 m and 100 m isobaths are shown 18

Figure 2: Comparison of annual values of the temporal sea ice index (TSI =first week with < 20% ice cover, bars) and that of the ICT index from the Pacific Marine Environmental Laboratory (line) 19

Figure 3: Examples of cold pool extent in 4 different years. Lightly shaded area indicates estimated extent ofbottom waters below 2°C 20

Figure 4: Scatterplot of raw cold pool index (fraction of survey area < 2°C), 1973-2003, against mean sampling date with linear regression line. Standardized residuals from the regression line were used as the final cold pool index (see Figure 5) 21

Figure 5: Time series of the final Cold Pool Index, 1973-2003 (Note that 1974 and 1977 are missing). A positive index implies a large area ofbottom water below 2°C 21

Figure 6: Time series of the index of winter ice severity conditions, 1964-2003. A positive index implies cold winters with severe ice conditions and cold bottom temperatures in the following summer. Note mild conditions in the past 3 winters, similar to 1978/79 22

Figure 7: Time series of annual air temperature anomalies at St. Paul Island, 1950-2002, with smooth LOESS fit 22

Figure 8: Time series of average May-September wind stress at St. Paul Island, 1951-2002, with smooth LOESS fit. 23

Figure 9: Time series of average April-June cross-shelfwind speed at St. Paul Island, 1951-2002, with smooth LOESS fit. " 23

3

Abstract To develop a series of statistical models linking walleye pollock recruitment in the

Eastern Bering Sea to climatic and oceanographic conditions at regional and larger spatial scales~

predictor variables were carefully selected based on four general hypotheses that have been advanced to explain variations in pollock recruitment. Inthe first hypothesis, Wyllie-Echeverria and Wooster (1998) relate survival oflarval and/or juvenile pollock to the severity ofwinter ice conditions and to the size and temperature of the resulting pool ofcold bottom water, i.e., cold pool~ on the shelf. The second and third hypotheses incorporate aspects ofthe recently proposed oscillating control hypothesis ofHunt et al. (2002). They relate pollock survival to the presence or absence of an ice-related spring bloom (Hypothesis 2) and to summer stratification and temperatures on the middle shelf region (Hypothesis 3). The fourth hypothesis examines the advection/predation hypothesis ofWespestad et al. (2000), which relates pollock recruitment to the degree of spatial separation between juvenile and cannibalistic adults. The degree of separation, in tum, is believed to be related to the passive drift oflarvae into favourable or unfavourable areas. To examine the evidence for and against each of the proposed hypotheses~

thus far we have obtained relevant predictor variables based on the literature and from available data sets and constructed a limited set of statistical models of recruitment.

Our overall objective is to model relationships between the recruitment or survival of larval and juvenile walleye pollock (ages 0-2), stock size, and relevant environmental variables using linear and non-linear models. These models will be used to assess the performance ofeach predictor variable and to assess the strength ofevidence for a given hypothesis. The best models for each hypothesis will be combined into one or several models for predicting walleye pollock recruitment in the Eastern Bering Sea. The performance of the final predictive model(s) will be evaluated in a retrospective analysis and their use in stock assessment will be examined.

Introduction There is general agreement that changes in the climate and oceanography of the

Northeast Pacific influence the recruitment of walleye pollock and other groundfishes in the Bering Sea (Hollowed et al. 2001). Currently, relationships between environmental variability and recruitment in the Bering Sea are not well understood and are not incorporated into the assessment of groundfish stocks (NPFMC 2001). Although a number ofpotential relationships between large-scale climate changes, the oceanography of the Bering Sea, and pollock recruitment success have been reported (Ohtani and Azumaya 1995; Quinn and Niebauer 1995; Wespestad et al. 2000; North Pacific Fishery Management Council (NPFMC) 2001), reliable empirical relationships have not been established to date. There is a need for a comprehensive analysis ofenvironmental influences on pollock recruitment in the Bering Sea to improve our understanding of the processes that determine recruitment and to develop better predictive models (NPFMC 2001). Our research examines all of the major hypotheses that have been put forward to explain variations in pollock recruitment in a unified and rigorous statistical framework to identify variables that can serve as useful predictors ofpollock recruitment success.

Most of the variability in pollock recruitment occurs during the first years oflife, in particular during the early ocean life oflarvae and age-O fish. For example~ recruitment of walleye pollock in Shelikof Strait is largely determined by the end of the larval period (Kendall et al. 1996). In Shelikof Strait, survival is enhanced by strong freshwater runoff in the spring

4

prior to spawning, which enhances productivity, and by weak mixing in the summer, which enhances retention on the shelf (Megrey et al. 1995; Megrey et al. 1996). These relationships have been quantified and successfully included in forecasts of recruitment success in the Gulf of Alaska (Megrey et al. 1996). Processes determining recruitment of pollock in the eastern Bering Sea are not well understood, but ice extent (Ohtani and Azumaya 1995; Quinn and Niebauer 1995), timing of ice retreat (Hunt et al. 2002), advection on the shelf rvvespestad et al. 2000). water temperature (Bulatov 1989; Ohtani and Azumaya 1995), and summer stratification (Hunt et al. 2002; Ladd et al. 2002) all are believed to influence recruitment success of pollock in the Bering Sea.

Marine populations respond to environmental variability in complex, nonlinear ways that cannot be adequately described by linear correlations (Cury et al. 1995). For example, it has been suggested that both warm winters (high sea surface temperature) and strong summer wind mixing must occur to produce strong pollock year-classes in the Bering Sea (James Overland, pers. comm.). Such complex relationships are unlikely to be captured by linear regression models (Quinn and Niebauer 1995). Nonlinearities, strong interactions, and threshold relationships may drive much of the dynamics in the Bering Sea, as evidenced by the highly unusual conditions in 1997 (Napp and Hunt Jr. 2001; Overland et al. 2001; Stockwell et al. 2001). Hare and Mantua (2000), in an analysis of 100 time series from the North Pacific, conclude that "the large marine ecosystems of the North Pacific and Bering Sea appear to filter climate variability strongly and respond nonlinearly to environmental forcing.". The statistical approach we take here takes into account potential nonlinearities, interactions, and thresholds in environment-recruitment relationships for Bering Sea pollock.

This study relates directly to several ofthe research priorities identified by the Pollock Conservation Cooperative Research Center, in particular to "Climatic regime shifts and interannual variability in the Bering Sea ecosystem and their impacts on Bering Sea sp~cies,

particularly pollock and marine mammals". Understanding relationships between environmental variability and recruitment can improve the management ofpollock stocks in several important ways. First, it can help to explain why Bering Sea pollock stocks are fluctuating in particular ways and help managers devise management strategies that account for such fluctuations. Second, it can lead to improved stock assessment models that incorporate such relationships. Third, it can lead to predictive hypotheses that can be explored andlor tested through rigorous research programs andlor adaptive management procedures. Ultimately, improved understanding of environmental processes will reduce biases and uncertainties in parameter estimates and hence in setting harvest limits.

Project goals and objectives The primary goal of this study is to develop statistical models linking walleye pollock

recruitment in the Eastern Bering Sea (Fig. 1) to climatic and oceanographic conditions at regional and larger spatial scales. Models are being developed to reflect our current understanding ofpotential mechanisms underlying these links and will be used to assess the strength ofevidence for and against the major hypotheses that have been proposed to explain variability in pollock recruitment and survival. Specifically, we will model the proposed relationships between recruitment or survival of larval and juvenile walleye pollock (ages 0-2),

,""-" stock size, and relevant environmental variables using a flexible approach that will accommodate non-linearities, interactions, and threshold relationships. Statistical model selection criteria will be used to identify the best model or models for each proposed mechanism.

5

A secondary goal is to compile and / or develop new environmental indicators that best reflect the potential mechanisms driving pollock recruitment. A large number ofatmospheric and oceanographic data sets, as well as a number of specific indicators are available in the literature or through various websites. However, available data sets may require extensive processing or appropriate spatial and temporal averaging to extract relevant indicators. Furthermore, there is much room for developing additional indicators that more appropriately reflect the most important aspects of environmental variability. We are updating existing indicators and develop new ones to reflect the most recent conditions.

Hypotheses linking pollock recruitment to environmental variables A number of empirical relationships between oceanographic variability and pollock

recruitment have been previously established or proposed by various authors. Based on these, we hypothesized four major sources ofenvironmental variability that may affect recruitment of walleye pollock in the Bering Sea. Here we briefly summarize the hypothesized mechanisms linking variability in the physical environment to pollock recruitment and discuss relevant environmental variables that will be used as potential predictors ofrecruitment.

1. Winter ice conditions and the cold pool The extent of ice cover on the eastern Bering Sea shelf has previously been shown to be

related to pollock recruitment (Fair 1994; Quinn and Niebauer 1995) and the cold-pool hypothesis ofWyllie-Echeverria and Wooster (1998) suggest one potential mechanism. Following cold winters with extensive ice cover, a cold pool ofwater remains on the middle shelf and tends to concentrate juvenile pollock on the outer shelf. As a result, age-1 pollock and cannibalistic older pollock co-occur on the outer shelf (Ohtani and Azumaya 1995; Wyllie­Echeverria and Ohtani 1999), resulting in increased cannibalism. Cannibalism is known to be a major determinant of recruitment variability in walleye pollock, hence increased cannibalism is associated with weak year classes (Wespestad and Quinn 1996; Wespestad et al. 2000). In addition, cold temperatures affect prey availability (see below) and can delay the maturing and growth of fish eggs and larvae, thus affecting their survival (Houde 1987).

To test the cold-pool hypothesis we recalculated and updated an index of the spatial extent of the cold-pool, estimated average summer bottom temperatures on the shelf, assembled and updated various indices of sea ice extent, and compiled several climate indices that reflect winter atmospheric conditions. In addition, we constructed measures of cannibalism potential to examine potential relationships between the size of the cold pool and the intensity ofcannibalism on juveniles. These indices are described in detail below. Specific questions that will be addressed with respect to winter ice conditions and the cold pool include: Are recruitment and / or survival rates of pollock related to the severity ofice conditions or the size of the cold pool? Is cannibalism more likely to occur if the cold pool is large? What life stages (age-O, age-I, or age­2), if any, are most strongly affected by ice conditions and the cold pool?

2. Timing of ice retreat and the spring bloom In addition to the spatial extent of sea ice, the timing of ice retreat has important

implications for productivity on the shelf. It is related to the timing of the spring bloom and may regulate growth of forage fish and their pelagic predators through mechanisms that were recently summarized in the oscillating control hypothesis (Hunt Jr. et al. 2002). Late ice-retreat results in an early, ice-associated bloom in a cold, low-salinity layer. Cold temperatures limit zooplankton

6

growth and much of the phytoplankton production may sink to the bottom. Hence prey limitation is likely to limit survival of larval and juvenile pollock. High abundances of predators at the beginning of a cold period will further reduce survival ofjuveniles. In contrast, if ice retreats early, light limitation does not allow for an ice-associated spring bloom; the main spring bloom occurs later in the season when summer stratification commences. Hence the bloom occurs in relatively warm water and zooplankton is not limited by temperature, providing abundant prey for larval and juvenile pollock. According to the hypothesis, growth and survival ofpollock is enhanced during the onset ofa warm regime if abundance ofpiscivores is low, but survival and recruitment is controlled through top-down processes once the biomass ofpiscivore predators has increased to high levels.

The oscillating control hypothesis implies a threshold relationship between the timing of ice retreat and survival ofpollock that is modified by the abundance ofpotential predators such as cannibalistic adults, which are responsible for the majority ofpredation on juvenile pollock. Expected responses of recruitment to the timing of retreat and different levels ofpredator biomass are summarized in Table 1. To examine the effects of ice retreat and to test the oscillating-control hypothesis we constructed several indices that reflect spring ice conditions and the timing of ice retreat, as well as indices ofpredator biomass and predation mortality. Specific questions that will be examined include: Is survival ofpollock enhanced ifthere is a late spring bloom? Is recruitment reduced at high levels ofpredator biomass? Does the effect of predator abundance depend on the timing of the bloom and conversely, does the effect of ice retreat depend on predator abundance?

Table 1: Effects of ice retreat and predator abundance on recruitment ofwaUeye pollock. The expected response ranges from a strong negative effect (-) over neutral or ambiguous effects (+/-) to strong positive effects (++).

Ice retreat Early (+) Late (-)

Low (+) ++ +/­Predator biomass

High (-) +/­

3. Mixed layer dynamics and summer production While the advance and retreat of sea ice is one of the major features of the Bering Sea

shelf, conditions on the shelf during summer are strongly influenced by solar radiation, wind mixing, and tidal mixing (Overland et al. 1999; Stabeno et al. 2001). The stability of the mixed layer, as determined by temperature and salinity conditions, and the frequency and intensity of storms determine the supply ofnutrients to the surface layer, which modifies the productivity and species composition of the plankton community. Variability in these processes results in variations in primary productivity on the order of 30-50%, as well as large differences in the species composition of the plankton community (Sambrotto and Goering 1983), thereby affecting feeding conditions for walleye pollock (Napp et al. 2000). We hypothesize that recruitment success ofwalleye pollock will reflect these conditions. Survival is expected to increase with the amount of summer production, as determined by the amount ofnutrients mixed

7

into the surface layer during the summer. However, intense wind mixing may limit primary production and I or feeding success of larval pollock, which could result in a dome-shaped relationship between pollock recruitment and mixing. Such relationships have been observed in upwelling systems (Cury et al. 1995), laboratory studies (MacKenzie et al. 1994), and modelling studies (Megrey and Hinckley 2001).

To examine relationships between wind mixing, summer stratification, and pollock survival we obtained wind data from St. Paul, model-based wind indices, and estimates of stratification and entrainment ofnutrients in the surface mixed layer. In addition, we obtained indices of summer production and zooplankton abundances to directly examine relationships between physical conditions, primary and secondary productivity, and pollock survival. Some specific questions to be examined are: Does recruitment success increase with the amount of new production in the summer, as estimated by entrainment? Is recruitment related to other mixed­layer characteristics such as depth or temperature of the mixed layer? Is recruitment linearly or non-linearly related to wind stress? Are currently available measures ofprimary and secondary production related to entrainment and to the survival ofpollock? What life stages are primarily affected?

4. Advection Advective processes in the basin and advection onto the shelf exert a strong influence on

both the supply ofnutrients and the distribution ofeggs and larvae ofpollock (Napp et al. 2000). While little is known about interannual variability in sub-surface flows onto the shelf, model results suggest considerable interannual variability in surface layer flows (Ingraham et al. 1991; Wespestad et al. 2000). This variability leads to interannual variations in the food supply of larval pollock, particularly on the outer shelf (Napp et al. 2000), and therefore this variability affects the growth and survival oflarval walleye pollock. In addition, certain patterns of surface flow on the shelfmay separate larval pollock from adults, thus reducing cannibalism (Wespestad et al. 2000). Strong year classes of walleye pollock are produced in years with strong onshore transport (to the north and east) and weak year classes in years ofminimal onshore transport during the larval stage (Wespestad et al. 2000, Fig. 6).

Indices of surface currents and atmospheric conditions that relate to circulation on the shelf were obtained to examine relationships between advection and year class success. In addition, we modified and updated Wespestad et al. 's (2000) index of separation between juveniles and adults and constructed an alternative index ofcannibalism potential to examine whether advection influences survival by altering the amount ofcannibalism. Specific questions to be examined are: Is the relative distribution ofjuveniles and adults, and hence the potential for cannibalism, related to surface transport? Are surface currents (distance and direction offlow) in the spring and early summer related to the survival and recruitment ofwalleye pollock? If so, do these effects occur primarily at the larval stage or at later juvenile stages?

Methods and progress to date

The emphasis of the first 6 months of the project has been to assemble and process various data sets to construct a comprehensive set ofbiological and environmental indicators that relate directly or indirectly to the above hypotheses.

8

Walleye pollock data and biological indices Stock-assessment data: Estimates of age-1 recruitment, female spawning biomass, and

total adult (age 3+) biomass of walleye pollock in the Eastern Bering Sea from 1964-2003 were obtained from the most recent assessment (lanelli et al. 2003). An index of spawner-to-recruit survival rates (SR index) was obtained by fitting a Ricker spawner-recruitment relationship with first-order auto-correlated errors to the relationship between estimated recruitment and female spawning biomass. It should be noted that a similar stock-recruitment relationship is integrated within the assessment model.

Indices ofpredation / cannibalism: To examine hypotheses related to predation we obtained estimates ofpredation mortality for age-O and age-1 pollock from 1979-2002 based on an MSVPA model (Pat Livingston, APSC, pers. comm.). Two indices of cannibalism were constructed. First, we updated an index of spatial separation originally proposed by Wespestad et al. (2000) to include years through 2003. Our index differed from that of Wespestad et al. (2000) in that we only included a fixed set of 306 stations that were sampled consistently in all years from 1982-2003 to avoid biases due to differences in sampling locations among years. We found that the index of separation that we constructed had several shortcomings. Therefore, we constructed an alternative index that simply measures the degree of association between juveniles « 20cm) and adults (> 40cm) by computing the covariance between juvenile and adult catch-per-unit-effort for each year:

1= = t.(J, -t.J,InXA, -t,A, In)f where ~ and Ai are the CPUEs ofjuveniles and adults at station i, respectively and n is the number of stations.

Primary and secondary productivity: No index ofprimary productivity was available at this time but may be incorporated at a later date if data become available. An index of zooplankton abundance was derived from plankton samples collected since 1955 by the Japanese research vessel Oshoru Maru (Sugimoto and Tadokoro 1997). A variable number of stations were sampled and station locations differed among years. To obtain a reasonable consistent index of zooplankton abundance we computed the average wet weight of all zooplankton samples collected on the middle shelf (between the 50m and 100m isobaths) of the standard NMFS trawl survey area (Fig. 1). Stations with high concentrations ofphytoplankton, salps, or jellyfish, as well as stations with bottom sediment and zero wet weight were excluded prior to taking the average. The resulting zooplankton index was available for the years 1955/56, 1960, 1962, and 1964-2000.

Oceanographic indices Sea ice extent: While the southernmost extent and timing of sea ice retreat along 169°W has

been used in the past (Wyllie-Echeverria 1996) as well as observations of sea ice at mooring M2 (Stabeno et al., 1998, Hunt et al. 2002) these observations along a single transect or at a single point inadequately describe sea ice over the 463,000 km2 area of ocean contained within the NMFS survey area (Fig. 1). Therefore we also estimated the proportion of the survey area covered by sea ice, averaged over the winter season. Digital ice charts for the southeastern Bering Sea region were obtained (Arctic Climatology Project, National Ice Center, http://www.natice.noaa.gov) for the years 1972 to 2003. All digital ice charts had a resolution of at least 0.25 decimal degrees. Sea ice was defined as regions with ;;:: 30% ice concentration. The

9

average weekly percentage of ice-covered ocean was calculated for January through May (approximate ISO (International Organization for Standardization) week numbers 1 - 22) using ArcGIS geographic information system (GIS) software (EnviroIUIiental Systems Research Institute, Inc., Redlands, CA). Based on weekly averages two indices of sea ice for the region were developed (Palmer 2003). The first index was a measure of the spatial ice cover over the region during the January to May period and calculated as a simple average of all weekly concentrations between January and May. This index, termed Spatial Sea Ice Index (SSI), agreed closely with a recently developed ice cover index (lCI) developed at the Pacific Marine Environmental Laboratory, NOAA (see Appendix 1, correlation coefficient: r = 0.84). The ICI combines a number ofdisparate indices but has the advantage ofcovering a considerably longer time period (1954-2003).

Timing ofice retreat: The second index derived from digital ice charts used the first week during which the percentage of the NMFS survey area covered by ice was less than 20%. This temporal sea ice index (TSI) correlated well with a similar index of ice retreat developed at the Pacific Marine Environmental Laboratory (PMEL), NOAA (lCT index, Appendix 1, r = 0.73) from a 2° by 2° region centered on 54°N, 164°W (Fig. 2). Some obvious differences, notably in the late 80's to early 90's, most likely reflect the restricted spatial extent of the ICT compared to the TSI.

Cold pool index (CPl): Based on bottom temperature measurements, the CPI is a standardized estimate of the fraction of the NMFS survey area that is covered by bottom water with a temperature below 2°C during annual summer bottom trawl surveys. For each NMFS survey from 1973-2003 (data were insufficient in 1974 and 1977) we contoured bottom temperature using all measurements available between May 1 and August 31 of a given year. Contours were fit to the data using a universal kriging approach implemented in ArcGIS (Fig. 3). Because the surveys did not always occur at the same time ofyear (mean sampling date ranged from June 13 in 1976 to July 21 in 1985), we adjusted for the date of sampling as follows. We fit a linear regression of the raw cold pool index (fraction ofsurvey area < 2°C) on the mean sampling date each year (R2

= 0.68, Fig. 4). The negative slope of the regression line reflected the average decrease in the area of the cold pool as the season progresses. To remove the seasonal warming pattern, we used the standardized residuals from this regression line as our final index of the annual cold pool area (Fig. 5).

Summer bottom temperatures: Like the cold pool index, average bottom temperatures in the NMFS survey area were estimated from temperature measurements taken during summer trawl surveys. Annual means ofbottom temperature across the survey area were estimated by fitting a generalized additive model ofmeasured bottom temperatures on latitude, longitude, depth, and Julian day as continuous covariates and year as a categorical covariate. Because Julian day was included as a covariate, the estimated annual means are implicitly adjusted for differences in sampling dates across years. The model fit was satisfactory and the index reflected annual means more accurately than a simple area-weighted average oftemperatures across survey stations, which can be severely biased in years when sampling occurred earlier (e.g. 1999) or later (e.g. 1985) than average.

Combined index ofwinter ice conditions: All ice and bottom temperature indices, including the southernmost extent of sea ice, the SSI, TSI, CPI, and average bottom temperature anomaly were highly correlated (Table 2). Therefore we derived a single index for the severity of sea ice conditions using a principal components analysis based on the correlations among these 5 variables (Table 2). Years included in the analysis were 1975 and 1979-2003. The first principal

10

component accounted for 74% of the overall variability in these variables, while the second PC accounted for an additional 16%. A high value of PC 1 corresponded to cold bottom temperatures, large average ice concentrations that extend far to the south, a late ice retreat, and a large spatial extent of the cold pool. The second PC loaded most heavily on the southernmost ice extent. High values ofPC 2 corresponded to years where the ice stayed far north and had a relatively small spatial extent, yet the ice retreated relatively late and the cold pool was large in spite of above average bottom temperatures (for example 1982, 2001). Because 3 of the 5 series are available for all years from 1972-2003 (see Appendix 1), we computed a combined index of the severity of ice conditions for this time period as a simple variance adjusted average of the 5 variables (after inverting signs for bottom temperature and southernmost ice extent), using the combined standardized index method of Boyd & Murray (2001). The index was further extended back in time based on a multiple regression of the 1972-2003 time series on St. Paul air temperatures, the Siberian-Alaskan Index (Appendix 1), and the Aleutian Low Pressure Index (Appendix 1). The regression had a coefficient ofdetermination of 87% and was used to reconstruct the ice severity index for the period 1964-1971. The resulting reconstructed index for 1961-2003 was our final index of winter sea ice severity (Fig. 6).

Table 2: Cross-correlations among various ice related indices, including southernmost extent of sea ice along 168°W (l68W), spatial sea ice index (881), temporal sea ice index (T81), cold pool index (CPI), and average bottom temperature (BT).

168W SSI TSI CPI SSI -0.796 TSI -0.549 0.790 CPI -0.505 0.734 0.820 BT 0.464 -0.628 -0.754 -0.942

Sea surface temperature: Indices of winter and summer sea-surface temperature (SST) were based on ship-of-opportunity observations, satellite observations, and measured temperatures from summer bottom trawl surveys. We used NOAA reconstructed (1900-1981) and NOAA Optimum Interpolation SST version 2 data (1982-2003) provided by the NOAA­CIRES Climate Diagnostics Center, Boulder, Colorado, USA, from their Web site at http://www.cdc.noaa.gov/. A yearly winter SST index was based on average SSTs from December to February over an area on the Eastern Bering Sea shelfbetween 57-590 Nand 163­171 0 W. Similarly, a summer index was based on May - September averages over the same area. An alternative summer SST index was derived from summer bottom trawl survey data (1975, 1979-2003) using the same approach that was used to estimate annual means ofbottom temperature described above. This survey-based index agreed well with the satellite-based index (correlation coefficient: r =0.87).

Bloom date and SST at bloom date: Model-based estimates of the date ofonset of the (non-ice-related) spring bloom at Mooring 2 on the middle shelf (Fig. 1), as well as SSTs at the time of the bloom, were obtained from Carol Ladd, PMEL, NOAA (Ladd et al. 2002, Carol Ladd, PMEL, pers. comm.).

Stratification and entrainment: Other indices based on the same mixed-layer depth model (Ladd et al. 2002) were the average depth of the mixed layer during the month of June as an

11

indicator ofearly summer conditions, an estimate of the strength of stratification as an index of summer water column stability, and total July-August entrainment as a proxy for potential new summer production.

Surface transport and larval drift: Indices of surface transport during the spring and early summer were based on simulated ocean surface currents (OSCURS model, Appendix 1). We updated the ocean surface current simulation (OSCURS; Ingraham and Miyahara 1988) model described in Wespestad et al. (2000), which simulates a 90-day drift of a particle released at the surface at 55.2°N, 164.5°W, the historical center ofpollock spawning, on April 1 (median date of spawning) for each year from 1967-2003. We used the northernmost latitude ofthe 90-day drift trajectory and the final latitude and longitude ofthe trajectories on June 30 as indices of egg and larval drift during their first 3 months of life.

Atmospheric indices A number of regional and large-scale indices were obtained that are believed to capture

the major atmospheric influences on the climate of the Eastern Bering Sea. Large-scale climate indices were obtained from various online sources (Appendix 1) and included the Southern Oscillation Index, a Multivariate ENSO index, a December-March average of the Arctic Oscillation, a December-March average of the Pacific Decadal Oscillation, the Aleutian Low Pressure Index, and a November-March average of the North Pacific Index. Large-scale indices that are more specific to the Bering Sea included the Siberian-Alaskan index designed to predict ice cover and the Bering Sea Pressure Index, which is the mean spring (Apr-Jun) sea level pressure averaged over the Bering Sea. Both indices were obtained form PMEL, NOAA (Appendix 1).

These climate indices were primarily used to examine large-scale influences on the local climate of the Bering Sea as evident in wind mixing, air and water temperatures, and ice conditions. In the initial hypothesis testing phase we did not directly model pollock recruitment as a function oflarge-scale climate indices. Instead, relationships with climate indices were used to improve measures ofoceanic variability such as the ice index discussed above. The use of large-scale indices will be explored in the final stages ofbuilding a predictive model for pollock recruitment, primarily in cases where other, more regional indices do not cover the full range of available pollock data (1964-2002).

Regional climate indices that directly influence ocean conditions on the Southeast Bering Sea shelf include measures ofair temperature and wind mixing.

Air temperature: Monthly averages ofmeasured air temperatures at the St. Paul airport, as well as at stations in Nome, Bethel, and King Salmon, were obtained from the Western Regional Climate Center (Appendix 1) and covered the period from 1950-present (except King Salmon: 1955-present). Air temperatures tended to follow a regular seasonal pattern with higher variability in the winter months. Therefore, two missing values in the 81. Paul time series were estimated by fitting a seasonal, auto-regressive time series model to the monthly air temperature series and substituting predicted values. Annual anomalies were computed as weighted annual averages of the January through December means ofeach year, with more weight on the winter months. Weights used were the loadings from a principal components analysis of the annual series ofJanuary, February, March, etc. means. The index showed a series of exceptionally low temperatures in the early- to mid-70s at St. Paul Island (Fig. 7), as well as an increasing linear trend in air temperatures from 1950 to 2002 at the rate of0.17°F per decade. Air temperatures at S1. Paul Island on the Bering Sea shelfwere highly correlated with coastal station records at

12

,______ Nome, Bethel, and King Salmon. Because the latter to some extent reflect continental influences we used St. Paul temperatures in the analysis rather than averaging across stations.

Wind forcing: Indices ofwind mixing, along-shelfwind stress, and cross-shelfwind stress were obtained from modelled wind fields and from daily measured winds at St. Paul airport. Bond and Adams (2002) derived several indices based on the NCEPINCAR reanalysis data set. Updated versions of three oftheir indices through 2002 were obtained from Nick Bond, PMEL (pers. comm.). 1. Winter (Dec-Apr) average of along-peninsula wind stress at 53°N, 173°W, which provides a measure of the direct wind forcing of the Aleutian North Slope Current and is related to the flow onto the southern Bering Sea shelf (Bond and Adams 2002); 2. Summer (May-Sep) average of along-shelf wind stress near Pribilof Canyon (56°N, 169°W), which provides an index of water exchanges between the slope region and the shelf. 3. May-September average wind stress at Mooring 2 (Fig. 1), which provides an index of summer wind mixing on the middle shelf.

In addition to model-derived indices we obtained measured wind data from the National Climatic Data Center (courtesy ofDave Kachel, PMEL, pers. comm.) covering the period January 1, 1951 to August 31, 2002. Wind speeds were recorded every three hours through July 1981 and hourly thereafter. Missing values are prevalent throughout the data set, but are usually restricted to missing hours within a day. Despite missing values, daily averages could be calculated continuously over the time period with the exception of the period from September 1 to December 31, 1955 (no record) and three individual days, which were ignored in computing monthly or seasonal means. We constructed a summer wind mixing index by computing daily averages ofwind stress (cube of the measured wind speeds), and averaging daily values from

,r--, May 1 through September 30 (Fig. 8). Measured wind stress at St. Paul was strongly correlated with the model-based index of summer wind mixing at Mooring 2 (r =0.59). Finally, we computed an index ofspring and early summer winds in the SW-NE direction (cross-shelf winds) by averaging wind speeds for the April-June period, which may affect larval drift patterns (Fig. 9).

Data analysis All variables or indices used in this study are listed in Appendix 1. As a first step, we

examined the distribution of all variables to look for extreme outliers or strongly skewed distributions that may require appropriate transformations prior to modelling. To examine linear relationships among the variables, we computed pairwise correlations (Pearson product moment correlations) among all indices, using all years for which data were available. Pairwise scatter plots between variables were examined visually to screen for obvious non-linear relationships.

For each of the relationships and hypothesized mechanisms described above we selected a limited subset ofpotentially important variables from the indices listed in Appendix 1. We only included variables that either set up conditions for or reflect conditions during the larval and juvenile periods (ages 0-2). In the next step we selected the statistically best model for predicting walleye pollock recruitment or survival by following a series of steps similar to those outlined in Burnham and Anderson (1998): • We first developed a "global" model(s) that includes all of the selected explanatory variables

for a given hypothesis. This typically only included two or three independent variables, for example the combined ICE index and an index ofpredator abundance were used to examine the cold-pool hypothesis. Where several alternative indices were available (for example two similar indices ofwind mixing, two indices of summer SST), we fit and examined models

13

using one of the indices at a time and chose the most appropriate index based on the length of the available time series, goodness-of-fit criteria, and other considerations. Ifno objective choice can be made among several alternative indices we used data reduction techniques such as Principal Components Analysis to combine similar indices into a single index for modelling.

• In this first exploratory step, we used Generalized Additive Models (GAM), which can accommodate non-linear relationships between the predictor variables and pollock recruitment or survival (Hastie and Tibshirani 1990). The model has the following general form:

y, = 2:h(Xi,,-k)+&, i

where Yt was either log-transfonned recruitment or log-survival (log(RJS)) of the walleye pollock year class spawned in year t, and Ai.t+k is the ith predictor variable affecting juvenile pollock during year t+k (k = 0, 1, or 2), fi denotes a smooth function of the ith predictor variable, and et is the residual error. GAMs use a non-parametric regression approach that fits smooth functions (e.g. smoothing splines) of the predictor variables in place oflinear functions. A cross-validation approach was used to choose the degree ofsmoothing for each variable (Wood 2000) and an approximate F-test was used to decide whether the relationship was significantly different from linear (Hastie and Tibshirani 1990). As appropriate, linear or quadratic relationships were substituted for the smooth fits.

• The fit of the global models was carefully examined (residual patterns, R2, other goodness­

of-fit criteria) and we proceeded only if the global model provided an acceptable fit. Otherwise, the general hypothesis was rejected.

• We then derived a set of plausible sub-models under each hypothesis, which were simplifications of the global model. The relative weight of evidence for each of the models in this a priori set of candidate models was evaluated using suitable model selection criteria. Where appropriate, we used the Akaike Infonnation Criterion as our model selection criterion of choice (Burnham and Anderson 1998).

Model fitting and model selection is currently in progress. The best models for each of the four general hypotheses will be combined into one or several models for predicting walleye pollock recruitment in the Eastern Bering Sea. The perfonnance of the final predictive model(s) will be evaluated in a retrospective analysis.

Findings Predictions ofwalleye pollock recruitment for the 2003 year class are not yet estimated.

Detailed findings will be presented in the final report.

14

Literature cited Beamish~ R.I. (ed)~ 1995. Climat~ change and northern fish populations. Can. Spec. Publ. Fish.

Aquat. Sci. 121. National Research Council of Canada~ Ottawa. Bond~ N.A.~ and Adams~ J.M. 2002. Atmospheric forcing of the southeast Bering Sea Shelf

during 1995-99 in the context of a 40-year historical record. Deep Sea Research Part II: Topical Studies in Oceanography 49:5869-5887.

Boyd~ I.L.~ and Murray~ W.A. 2001. Monitoring a marine ecosystem using responses of upper trophic level predators. J. Anim. Ecol. 70:747-760.

Bulatov~ O.A. 1989. The role of environmental factors in fluctuations of stocks ofwalleye pollock (Theragra chaleogramma) in the eastern Bering Sea. In Effects ofocean variability on recruitment and an evaluation ofparameters used in stock assessment models. Edited by RJ. Beamish and G.A. McFarlane. Can. Spec. Publ. Fish. Aquat. Sei. pp.353-357.

Burnham~ K.P.~ and Anderson~ D.R. 1998. Model selection and inference: a practical information-theoretic approach. Springer-Ver1ag~ New York.

Cury~ P.~ Roy~ C.~ Mendelssohn~ R.~ Bakun~ A.~ Husby~ D.M.~ and Parrish~ R.H. 1995. Moderate is better: Exploring nonlinear climate effects on the California northern anchovy (Engraulis mordax). In Effects of ocean variability on recruitment and an evaluation of parameters used in stock assessment models. Canadian Special Publications ofFisheries and Aquatic Sciences. National Research Council ofCanada~ Ottawa. pp. 417-424.

Fair~ L.F. 1994. Eastern Bering Sea walleye pollock: revised estimates ofpopulation parameters~

relation of recruitment to biological and environmental variables~ and forecasting. M.S. Thesis~ University of Alaska Fairbanks~ Juneau~ Ak.

Hare, S.R, and Mantua, N.J. 2000. Empirical evidence for North Pacific regime shifts in 1977 and 1989. Prog. Oceanogr. 47:103-145.

Hastie, T.I., and Tibshirani, RJ. 1990. Generalized Additive Models. Monogr. Stat. Appl. Probab. 43. Chapman and Hall, London.

Hollowed, A.B., Hare, S.R~ and Wooster~ W.S. 2001. Pacific basin climate variability and patterns ofNortheast Pacific marine fish production. Prog. Oceanogr. 49:257-282.

Houde, E.D. 1987. Fish early life dynamics and recruitment variability. American Fisheries Society Symposium 2:17-29.

Hunt, G.L., Jr., Stabeno, P., Walters, G., Sinclair, E., Brodeur, R.D., Napp, J.M., and Bond, N.A. 2002. Climate change and control of the southeastern Bering Sea pelagic ecosystem. Deep Sea Research Part II: Topical Studies in Oceanography 49:5821-5853.

Hunt Jr., G.L., Stabeno, P.~ Walters, G.~ Sinclair, E., Brodeur, RD., Napp, J.M.~ and Bond, N.A. 2002. Climate change and control of the southeastern Bering Sea pelagic ecosystem. Deep Sea Research Part II: Topical Studies in Oceanography 49:5821-5853.

Ianelli, J.N., Barbeaux~ S.~ Walters~ G.~ and Wi1liamson~ N. 2003. Eastern Bering Sea walleye pollock stock assessment. In Stock assessment and fishery evaluation report for the groundfish resources of the Bering Sea/Aleutian Islands regions. North Pacific Fishery Management Council~ 605 W. 4th Ave.~ Suite 306~ Anchorage~ AK 99501.

Ingraham~ W.I.~ Jr.~ Reed, RK.~ Schumacher~ J.D., ~d Macklin~ S.A. 1991. Circulation variability in the Gulf of Alaska. EOS~ Trans. Am. Geophys. Union 72:257-264.

Kenda11~ A.W.J., Schumacher, J.D., and Kim, S. 1996. Walleye pollock recruitment in Shelikof Strait. Fish. Oceanogr. 5:4-18.

15

Ladd, C., Hunt Jr., G.L., and Stabeno, P.J. 2002. Climate, mixing, and phytoplankton on the southeast Bering Sea shelf. unpubl. manuscript.

Mantua, N., Hare, S., Zhang, Y., Wallace, J., and Francis, R. 1997. A Pacific interdecadal climate oscillation with impacts on salmon production. Bull. Am. Meteorol. Soc. 78:1069-1 080.

MacKenzie, B.R., Miller, T.J., Cyr, S., and Leggett, W.C. 1994. Evidence for a dome-shaped relationship between turbulence and larval ingestion rates. Limnol. Oceanogr. 39: 1790­1799.

Megrey, B.A, Bograd, S.J., Rugen, W.C., Hollowed, AB., Stabeno, P.J., Macklin, S.A., Schumacher, J.D., and Ingraham, W.J., Jr. 1995. An exploratory analysis of associations between biotic and abiotic factors and year-class strength of GulfofAlaska walleye pollock (Theragra chalcogramma). In Climate change and northern fish populations. Edited by R.J. Beamish. Can. Spec. Publ. Fish. Aquat. Sci. pp. 227-243.

Megrey, B.A, and Hinckley, S. 2001. Effect of turbulence on feeding oflarval fishes: a sensitivity analysis using an individual-based model. ICES J. Mar. Sci. 58:1015-1029.

Megrey, B.A., Hollowed, A.B., Hare, S.R., Macklin, S.A, and Stabeno, PJ. 1996. Contributions ofFOCI research to forecasts of year-class strength ofwalleye pollock in Shelikof Strait, Alaska. Fish. Oceanogr. 5:189-203.

Napp, J.M., and Hunt Jr., G.L. 2001. Anomalous conditions in the south-eastern Bering Sea, 1997: linkages among climate, weather, ocean, and biology. Fish. Oceanogr. 10:61-68.

Napp, lM., Kendall, A.W.J., and Schumacher, J.D. 2000. A synthesis ofbiological and physical processes affecting the feeding environment of larval wall eye pollock (Theragra chalogramma) in the Eastern Bering Sea. Fish. Oceanogr. 9:147-162.

North Pacific Fishery Management Council (NPFMC) 2001. Stock assessment and fishery evaluation report for the groundfish resources of the Bering Sea/Aleutian Islands regions. North Pacific Fishery Management Council, 605 W. 4th Ave., Suite 306, Anchorage, AK 99501

NPFMC 2001. Stock assessment and fishery evaluation report for the groundfish resources of the Bering Sea/Aleutian Islands regions. North Pacific Fishery Management Council, 605 W. 4th Ave., Suite 306, Anchorage, AK 99501

Ohtani, K., and Azumaya, T. 1995. Influence of interannual changes in ocean conditions on the abundance ofwalleye pollock (Theragra chalcogramma) in the eastern Bering Sea. In Climate change and northern fish populations. Edited by RJ. Beamish. Can. Spec. Publ. Fish. Aquat. Sci. pp. 87-95.

Overland, J.E., Bond, N.A, and Adams, J.M. 2001. North Pacific atmosphere and SST anomalies in 1997: links to ENSO? Fish. Oceanogr. 10:69-80.

Overland, lE., Salo, S.A., Kantha, L.H., and Clayson, C.A. 1999. Thermal Stratification and Mixing on the Bering Sea Shelf In Dynamics of the Bering Sea. Edited by T.R. Loughlin and K. Ohtani. University of Alaska Sea Grant, AK-SG-99-03, Fairbanks. pp. 129-146.

Palmer, M.C. 2003. Environmental controls offish growth in the southeast Bering Sea. M.S. Thesis, University ofAlaska, Fairbanks.

Quinn, T.J., II, and Niebauer, H.J. 1995. Relation ofeastern Bering Sea walleye pollock recruitment to environmental and oceanographic variables. In Climate change and northern fish populations. Canadian Special Publications ofFisheries and Aquatic Sciences. Edited by R.J. Beamish. National Research Council of Canada, Ottawa. pp. 497-507.

16

Sambrotto, RN., and Goering, J.J. 1983. In From year to year. Edited by W.S. Wooster. University ofWashington Sea Grant, Seattle. pp. 161-177.

Stabeno, P,J., Bond, N.A., Kachel, N.B., Salo, S.A, and Schumacher, J.D. 2001. On the temporal variability of the physical environment over the south-eastern Bering Sea. Fish. Oceanogr. 10:81-98.

Stockwell, D.A., Whitledge, T.E., Zeeman, S.I., Coyle, K.O., Napp, J.M., Brodeur, RD., Pinchuk, AI., and Hunt, G.LJ. 2001. Anomalous conditions in the southeastern Bering Sea: nutrients, phytoplankton, and zooplankton. Fish. Oceanogr. 10:99-106.

Sugimoto, T., and Tadokoro, K. 1997. Interannual-interdecadal variations in zooplankton biomass, chlorophyll concentration and physical environment in the subarctic Pacific and Bering Sea. Fish. Oceanogr. 6:74-93.

Wespestad, V.G., Fritz, L.W., Ingraham, W,J., and Megrey, B.A 2000. On relationships between cannibalism, climate variability, physical transport, and recruitment success ofBering Sea walleye pollock (Theragra chalcogramma). ICES J. Mar. Sci. 57:272-278.

Wespestad, V.G., and Quinn, T.J., II 1996. Importance of cannibalism in the population dynamics ofwalleye pollock, Theragra chalcogramma. NOAA Technical Report NMFS 126:212-217. .

Wood, S.N. 2000. Modelling and smoothing parameter estimation with multiple quadratic penalties.J. R. Statist. Soc. B 62:413-428.

Wyllie-Echeverria, T. 1996. The relationship between the distribution ofone-year-old walleye pollock, Theragra chalcogramma, and sea-ice characteristics. NOAA Technical Report NMFS 126:47-56.

Wyllie-Echeverria, T., and Ohtani, K. 1999. Seasonal Sea Ice Variability and the Bering Sea Ecosystem. In Dynamics of the Bering Sea. Edited by T.R Loughlin and K. Ohtani. University ofAlaska Sea Grant, AK-SG-99-03, Fairbanks. pp. 435-452.

Wyllie-Echeverria, T., and Wooster, W.S. 1998. Year-to-year variations in Bering Sea ice cover and some consequences for fish distributions. Fish. Oceanogr. 7: 159-170.

17

18O"0'0·E 170·0'0·W 160"O'O·W ISO"O'O·W

.......

.....

\/Bering Sea

" ~ , .... ,," . " . o 100 :zoo z LLJ ~ ''''~''''''O KilcmMlen

~ " .

Figure 1: Map of the Eastern Bering Sea. The NMFS survey area (heavy line), M2 mooring location (56.8° N, 164.0° W) and three hydrographic domains (inner, middle and outer

shelves) separated by the 50 m and 100 m isobaths are shown.

18

3.00 .,...----------------------------, ­

2.00

1.00

-1.00

-2.00

-3.00 +-.....-.-...----.-_.___,..,.--.--r-.,...-,..,.--.---.--.,...-,.....,...__._~.,___,,......,__._~.,___,____r___r__.___,.....--+

1972 1977 1982 1987 1992 1997

Year I§1JTSI -leT ice index

Figure 2: Comparison of annual values of the temporal sea ice index (fSI =nrst week with < 20% ice cover, bars) and that of the leT index from the Pacinc Marine Environmental

Laboratory (line)

19

Figure 3: Examples of cold pool extent in 4 different years. Lightly shaded area indicates estimated extent of bottom waters below 2°e.

20

CPt \'5. average date of sampling

(.)

i' O.95 +.'-:-.;+'-'~ 'tJ N V

f 0.9 ~~~ III >­~ ; 0.85 +---':-~'+.-+'*

'0 r:: o t; 0.8~~~~

l! u.

O.75 ..p:c:.~~=

160 170 180 190 200 210

Julian day

Figure 4: Scatterplot of raw cold pool index (fraction of survey area < 2°C), 1973-2003, against mean sampling date with linear regression line.

Final Cold Pool Index

2.5 -r:==..".,.",.

2

1.5

1;··

0.5

o -0.5

-1

-1.5

-2

Figure 5: Time series of the ("mal Cold Pool Index (standardized residuals from the r" regression line in Figure 4), 1973-2003. A positive index implies a large area of bottom

water below 2°C.

21

x CD

"C .E .?;­.~

CD >CD en 0

~

-"­CDc: § ....

I

Figure 6: Time series of the index of winter ice severity conditions, 1964-2003. A positive index implies cold winters with severe ice conditions and cold bottom temperatures in the

following summer. Note mild conditions in the past 3 winters, similar to 1978/79.

"'C C\I

~ :e 0 0 0

m "'C c: m.... rJ)

~ c: Q)c:

0

0 a. E 8 ....

I

ro 0.·0 c:

"t: C\I

I 0.0 a. ~ u::: C? 0

1950 1960 1970 1980 1990 2000

Figure 7: Time series of annual air temperature anomalies at St. Paul Island, 1950-2002, with smooth LOESS fit.

22

o

o

o g

1950 1960 1970 1980 1990 2000

Figure 8: Time series of average May-September wind stress at St. Paul Island, 1951-2002, with smooth LOESS fit.

o

~1-o Q

0: o

o '?--L.j------i------;.----i------ir--------i----J

1950 1960 1970 1980 1990 2000

Figure 9: Time series of average April-June cross-shelf wind speed at St. Paul Island, 1951­2002, with smooth LOESS fit.

23

) ) )

Appendix 1: List of all variables and indices used in this study. Name Description Begin End R Pollock recruitment (age-I) estimates by year 1963 2002

class S Female spawning biomass estimates 1964 2003 SR Survival rate index (residual from Ricker stock­ 1963 2002

recruit model) by year class B Total pollock biomass (Age 3+) 1964 2003 M.O MSVPA-based estimates of predation mortality 1979 2002

for age-O pollock M.1 MSVPA-based estimates ofpredation mortality 1979 2002

for age-1 pollock Zoop zooplankton wet weight in the middle domain of 1955 2000

the shelf (some missing years) Separation Index of separation between juveniles «20cm) 1982 2003

and adults (>40cm) Association Index ofassociation between juveniles and adults 1982 2003

(covariance between juvenile and adult CPUE across 306 standard survey stations)

airT.StPaul First principal component ofmonthly series (Jan­ 1950 2002 Dec) of air temperatures at St. Paul airport

RACE. SST Estimated mean sea-surface temperature during 1982 2003 summer based on trawl survey data, corrected for differences in date ofsampling

RACE.BT Estimated mean bottom temperature during 1982 2003 summer based on trawl survey data, corrected for differences in date ofsampling

SST.sum May-Sep average sea-surface temperature in 1900 2002 region 57-59°N, 163-17l oW

SST.win Winter (previous Nov-Feb) average sea-surface 1901 2003 temperature in region 57-59ON, l63-l7l oW

Ice.168W Southernmost ice extent along 168 degrees West 1972 2003 SSI Spatial Sea-ice Index: Average percentage of 1972 2003

NMFS trawl survey area covered by ice, Jan-May

Source Ianelli et al. (2003)

Ianelli et a1. (2003) this study, based on data in Ianelli et al. (2003)

Ianelli et al. (2003) Pat Livingston, pers. comm.

Pat Livingston, pers. comm.

Sugimoto & Tadokoro (1997), this study, based on data available at Institute ofMarine Science, UAF Wespestad et a1. (2000) modified and updated in this study this study

this study, data provided by Western Regional Climate Center: http://www.wrcc.dri.edu/surnmaI:y/climsmak.html this study, data provided by RACE division, Alaska Fisheries Science Center, NOAA, Seattle

this study, data provided by RACE division, Alaska Fisheries Science Center, NOAA, Seattle

this study, data provided by Climate Data Center, NOAA: http://www.cdc.noaa.gov/

this study, data provided by Climate Data Center, NOAA: http://www.cdc.noaa.gov/ Wyllie-Echeverria (1996), updated in this study Palmer (2003), modified and updated for this study

) ) ')

TSI Temporal Sea-ice Index: First ISO week during 1972 which average ice concentration in NMFS trawl survey area first drops below 20%

CPI Cold Pool Index: Spatial extent of the cold pool 1973 « 2deg. C) during summer trawl survey, corrected for differences in date of sampling (missing years with insufficient data)

ICE Index of ice severity combining RACE.BT, 1964 Ice.168W, SSI, TSL and CPI, reconstructed values prior to 1972

ICI Multivariate Ice Cover Index 1954 ICT Weeks after March 15 where ice cover in 56­ 1973

58°N, 163-165°W exceeds 10% Wind.AP Winter (Dec-Apr) average of along-reninsula 1959

wind stress at 53~, 173°W in N m­Wind.AS Summer (May-Sep) averages of along-shelf wind

stress at 56°N, 169°W in N m­ 2 1959

Wind.CS Average spring / early summer (Apr-Jun) cross­ 1951 shelf (SW-NE) wind speed at St. Paul

Wind.mix1 Summer (May-Sep) averages of absolute 1959 NCEPINCAR reanalysis wind stress at 57~,

164°W Wind.mix2 Summer (May-Sep) averages of absolute 1951

measured wind stress at St. Paul Island MLD Average June mixed layer depth at Mooring 2 1951

(57°N, 164°W) B100mdate Estimated spring (non-ice-re1ated) onset of 1951

bloom (Julian day) at Mooring 2 (57°N, 164°W) SST.bloom SST during estimated onset ofbloom at Mooring 1951

2 (57~, 164°W) Entrainment Estimated July-August entrainment ofnutrients 1951

(deep water) at Mooring 2 (57~, 164°W) BSPI Bering Sea Pressure Index: mean spring (Apr­ 1943

Jun) sea level pressure averaged over Bering Sea SAl Siberian-Alaskan Index: mean Dec-Mar pressure 1949

difference between Siberia and Alaska-Yukon

2003 Palmer (2003), modified and updated for this study

2003 this study

2003 this study

2003 http://www.beringclimate.noaa.gov/ 2003 http://www.beringclimate.noaa.gov/

2002 Bond & Adams (2002), updated series provided by Nick Bond, PMEL (pers. comm.)

2002 Bond & Adams (2002), updated series provided by Nick Bond, PMEL (pers. comm.)

2002 this study, data provided by National Climatic Data Center (via Dave Kachel, PMEL, pers. comm.)

2002 Bond & Adams (2002), updated series provided by Nick Bond, PMEL (pers. comm.)

2002 this study, data provided by National Climatic Data Center (via Dave Kachel, PMEL, pers. comm.)

2002 Ladd et al. (2003), data provided by Carol Ladd (pers. comm.)

2002 Ladd et al. (2003), data provided by Carol Ladd (pers. comm.)

2002 Ladd et al. (2003), data provided by Carol Ladd (pers. comm.)

2002 Ladd et al. (2003), data provided by Carol Ladd (pers. conun.)

2003 http://www.beringclimate.noaa.gov/

2003 hno:/lwww.beringclimate.noaa.gov/

) ) )

SOl

MEl

PNA

AO

NPI

ALPI

PD~

OT.N

OT.lat

OT.long

Southern Oscillation Index, (previous) May­ 1952 April average Multivariate ENSO Index: (previous) Dec-Jan 1950 average Pacific North American pattern: (previous) Dec- 1951 Mar average Arctic Oscillation: (previous) December - March 1951 average North Pacific Index: (previous) December­ 1900 March average Aleutian Low Pressure Index: Winter (previous 1900 Dec-Mar) intensity ofAleutian Low, mean area where sea level pressure is < =1005 hPa Winter (previous Dec-Mar) average ofmonthly 1901 PDO values Northernmost point ofApril 1 - June 30 1967 OSCURS trajectory started at 55.2 N, 164.5 W Ending latitude of April I - June 30 OSCURS 1967 trajectory started at 55.2 N, 164.5 W Ending longitude ofApril 1 - June 30 OSCURS 1967 trajectory started at 55.2 N, 164.5 W

2003 ftp://ftp.ncep.noaa.gov/pub/cpc/wd52dgldatalindices/soi

2003 http://www.beringclimate.noaa.gov/

2003 http://www.cpc.ncep.noaa.gov/productsiprecip/CWlinkipnai

2003

2003 http://www.cpc.ncep.noaa.gov/productsiprecip/CWlinkidaily_lao index.html http://www.cgd.ucar.edul-jhurrelVrw·html

2002 Beamish et aI. (1997), index available at http://www.pac.dfo­mpo.gc.calscilsa-mfudldownloads/alpi.txt

2003

2003

2003

2003

Mantua et aI. (1997), updated data available at http://tao.atmos.washington.edulpdo/ Wespestad et aI. (2000), http://www.pfeg.noaa.gov/productsllas/OSCURS.html Wespestad et aI. (2000), http://www.pfeg.noaa.gov/productsilas/OSCURS.html Wespestad et aI. (2000), h ://www. fe .noaa. ovl roductsilasiOSCURS.html

ao_index