In the past few years, a new paradigm has begun to emerge concerning how Alaskan salmon production is regulated. Traditional research into variability in salmon production and survival has focused on freshwater (e.g., lake/river temperature, streamflow, freshwater predators) and life history factors (stock-recruitment relationship). Recognition that the marine environment may play an equal, if not greater, role has grown in parallel with the regime view of climatic variability. The idea of change occurring via abrupt shifts (as opposed to gradual change) in atmospheric/oceanic physics and biological populations is receiving widespread attention. The nature of 20th century variability in Alaskan salmon populations has been characterized as alternating production regimes (Chapter 3 of this dissertation, Francis and Hare 1994, Hare and Francis 1995, Francis et al. in review), with abrupt shifts in the mean level of production occurring in the mid 1920s (low to high), late 1940s (high to low) and late 1970s (low to high). In addition to the salmon, a mid to late 1970s shift in other biological populations of the North Pacific has been noted, including phytoplankton (Venrick et al. 1990), zooplankton (Brodeur and Ware 1992, Roemmich and McGowan 1995), groundfish (Hollowed and Wooster 1992, Beamish 1993), and reef fish populations in the Northwest Hawaiian Islands (Polovina et al. 1994).
In this chapter, I focus on the change in salmon production. Two assumptions are key to this analysis. The first is that the primary driving force behind the decadal-scale variability in salmon production is climate change. Certainly, this is not universally accepted, and I do not propose that all of the variability derives from environmental forcing. Other factors have been proposed as responsible for the observed changes, include cessation of high seas interceptions, hatchery production, changes in harvest and management practices and, particularly for west coast salmonids, habitat loss. Arguments for the dominance of the environment over other factors can be found in Francis and Sibley (1991), Pearcy (1992), Beamish and Bouillon (1993), and Mann (1993), and will not be reproduced here. The second key assumption, demonstrated in Chapters 2 and 3, is that the physics and biology described above do undergo periodic regime shifts that mark the transition from one mean state to another. Thus, these variables exhibit two characteristics: alternate stable mean levels and abrupt transitions in changing from one level to another. Note there is still variability around a mean level which may be random, autocorrelated, or driven by other factors.
Attempts to relate Alaskan salmon production variability to large-scale climate change in the North Pacific have led to development in the literature of a conceptual mechanistic model. Thus far, analyses have only been speculative, following the "historical-descriptive" method of science (Francis and Hare 1994), in their attempts to construct plausible explanations. The details of the model are not all agreed upon, and the relative importance of various factors is disputed. The goal of this chapter is to statistically test several of the competing and complementary mechanisms advanced to explain the coincident changes in the physics and biology of the Subarctic Pacific marine ecosystem. The hope is that certain key linkages worthy of more critical investigation can be identified.
The structure of this chapter is as follows. In the following section (4.2), I describe the conceptual model that has developed in the literature and summarize the key publications and their contributions. Next (Section 4.3), I describe the various data sets I obtained, and briefly revisit the statistical methods I use (fully outlined in Chapter 1). The main point here is to emphasize the necessity of "prewhitening" data prior to drawing any conclusions from historical time series and correlation analyses. In Section 4.4, I use the assembled data to statistically probe various aspects of the conceptual model. In the final section (4.5), I summarize the results and determine which mechanisms appear to hold promise for future investigation.
The model of how Alaskan salmon production is regulated is incomplete, but certain aspects are now generally agreed upon. In the regime shift view, there are alternating productive and unproductive regimes. The model is summarized below with the characteristics of the productive regime outlined. In most cases, an unproductive regime is defined by a reversal of the productive conditions. A diagram of the conceptual model is given in Figure 4.1.
Aspects of this conceptual model can be traced back to Wickett (1967) who first postulated that the annual concentrations of zooplankton off California are directly related to concentrations "upstream", at the diversion point of the West Wind Drift (Subarctic Current) the previous year. Chelton and Davis (1982) and Chelton (1984) proposed that the two branches of the eastward flowing Subarctic Current, i.e., the Alaska and California Currents, vary out-of-phase in response to interannual changes in surface winds. During years of an enhanced winter Aleutian Low, cyclonic wind stress results in increased transport into the Alaska Current at the expense of the California Current. Conversely, years with a weak Aleutian Low result in anomalously high transport into the California Current.
An overall increase in the productivity of the North Pacific since 1976 was first shown by Venrick et al. (1987). Interannual and interdecadal variations in North Pacific zooplankton biomass and distribution have been summarized by several authors. Fulton (1983) and Fulton and LeBrasseur (1985) illustrated how zooplankton distributions shifted north and southwards with the annual shifting of the Subarctic Front. A doubling in zooplankton biomass between the 1950s and the 1980s was documented by Brodeur and Ware (1992) though they were unable to explain the increase mechanistically. The large increase in copepod production at Ocean Station P beginning in 1976 was shown to be correlated with an index of the Aleutian Low (McFarlane and Beamish 1992). Recently, Polovina et al. (1995) have documented decadal-scale changes in the mixed layer depth in the central and north Pacific.
Sea surface temperature has been identified as playing an important role in various aspects of salmon growth, survival and production. Rogers (1984) attributed the upswing in salmon production beginning in the late 1970s to changes in their return migrations that reduced their susceptibility to predation in their final season at sea. Higher growth rates (as indicated by scale increments) for Bristol Bay and Nushagak sockeye salmon are correlated with Alaska temperatures (Rogers and Ruggerrone 1993). In a recent intriguing paper, Welch et al. (1994) demonstrated that ocean distributions of salmon may be shaped by temperature barriers. Each species has a distinct upper thermal limit, which varies by season, above which they are not found; below the critical temperature their distribution is essentially uniform in regards to temperature
The notion of "shifts", as discussed earlier, to explain the swings in salmon production has been proposed by Francis and Hare (1994) and Hare and Francis (1995). They have identified shifts in sea level pressure and temperatures that indicate the linkage between physics and salmon biology occurs in the first year of marine life. Francis and Sibley (1991) illustrated that production of Alaska salmon varies inversely with salmon from Washington, Oregon and California. Hollowed and Wooster (1992) found shifts in recruitment patterns of Northeast Pacific groundfish that coincided with sea surface temperature anomalies. The duration of regimes, i.e., the period between shifts differed for salmon and groundfish. The average duration of salmon production regimes was 25 years (Hare and Francis 1995) compared to 6-12 years for groundfish.
At present, this conceptual model of how salmon production in the North Pacific is regulated by large scale climate variability has not been field-tested. Such an endeavor will require careful thought and planning, and is in fact a major goal of a joint PICES and GLOBEC project proposal. Prior to the design and implementation of experiments, it would be worthwhile to carefully analyze statistically some of the assertions and hypotheses that form the model using historical data. In this section, I examine the evidence for the model and attempt to identify those relationships that appear to hold the most promise for detailed investigation.
The tests are grouped into four categories. I explore the role of 1) the Aleutian Low; 2) temperature; 3) zooplankton abundance; 4) marine growth of salmon. In most cases, I examine the relationship to salmon production, though other relationships are discussed and examined.
To test the model, I rely on linear correlation and cross-correlation analysis. Correlation techniques have several shortcomings and their use is generally restricted to exploratory investigations such as this. Cross-correlation analysis is used to test for lead-lag relationships between time series. To reduce the possibility of spurious correlations, I rely heavily on a technique from the field of time series analysis called prewhitening. The effect of prewhitening is to reduce unassociated autocorrelation and/or trends within time series prior to computation of their cross correlation function. It is well established that autocorrelation within time series results can produce spurious correlation (Newton 1988, Katz 1988b, Box et al. 1994).
There are two methods of prewhitening that are used to remove the effect of autocorrelation (Wei 1990). Both methods involve fitting Box-Jenkins ARIMA model to the time series and then conducting correlations with the model residuals. The first method, termed simple prewhitening, is used when there is a clear unidirectional influence such as between a physical (i.e., the explanatory) and biological (response) variable. An example would be the effect of temperature on salmon growth. In this situation an ARIMA model is fit to the explanatory variable and this model is used to prewhiten both the explanatory and response variables. The second method, double prewhitening, is used when there is potential feedback between the series or if the direction of influence is unknown. An example of this might be the relationship between atmospheric and oceanic temperatures. In such a situation, separate ARIMA models are fit to the time series and used for prewhitening.
In addition to prewhitening for autocorrelation, I also prewhiten for interventions (Francis and Hare 1994). The reason for this, which was explored extensively in Chapter 1, can be summarized as follows. In an ecosystem driven by regime shifts, many variables may jump simultaneously and, as a result, will appear to be highly related to each other, even if there is no actual link between them. These uncorrelated variables that both shifted mean levels at approximately the same time will appear to be positively correlated if they changed in the same direction, and negatively correlated if they changed in opposite directions. Removing the intervention effect aligns both time series to a zero mean, but does not affect any direct connection between the series. In other words, if the series are in fact related, say in a cause-effect way, the intervention prewhitening has no effect on the appropriate-lag correlation coefficient. One can also think of a step intervention as being similar to an autocorrelated process with a large lag-one coefficient.
In many cases, as it will be shown, the prewhitening process reduces
seemingly significant physical/biological correlations to insignificance.
There have been numerous references in the literature to fisheries
oceanography correlations that, while significant at time of publication,
do not hold up with time (Walters and Collie (1988), Tyler (1992),
Sheperd et al. (1984), Mann (1993)). I attribute the large majority
of these spurious correlations to the autocorrelation/regime shift
effect. Most time series encountered in nature are autocorrelated:
natural processes simply do not vary randomly from one time step
to another. Failure to take account of this fact usually results
in the correlation breakdown just mentioned. Finally, the existence
of a significant correlation after prewhitening does not guarantee
a relationship, merely a heightened probability thereof and candidacy
for more critical investigation.
The SLP time series I use was derived in Chapter 2. It is the dominant rotated principal component (SLP RPC1) for winter (Dec.-Feb.) over the period 1900-1992. The winter SLP RPC1 was shown in Chapter 2 to be an index of Aleutian Low activity, and is highly correlated (r > 0.9) with two other commonly used indices of the Aleutian Low, Trenberth and Hurrell's (1994) North Pacific Index and Beamish and Bouillon (1993) Aleutian Low Pressure Index. The SLP RPC1 time series is illustrated in Figure 2.28, the corresponding EOF map is illustrated in Figure 2.10.
Several time series were initially examined as indicators of trends in Alaskan temperatures. Sea surface temperatures are generally available only from 1950, however air temperature records are available dating back to 1900 for several Alaskan cities. From the COADS database, I prepared three SST time series. The first is the dominant spring SST principal component derived in Chapter 2. The other two were created by averaging COADS data over spring months for two regions, one representing the northern Gulf of Alaska and the other Bristol Bay (Figure 4.2). For longer term indicators of Alaskan temperatures, I obtained winter (Nov.-March as explained below) air temperature records for two cities: King Salmon (as a proxy for Bristol Bay) for 1900-1992 and Kodiak (proxy for northern Gulf of Alaska) for 1919-1992. The five temperature time series are illustrated in Figure 4.3.
One reason for selecting several temperature time series is that there can be surprisingly low correlation among Alaska temperature series. Several factors can lead to inconsistent behavior in indices that one might blindly assume to be highly related. These factors include local-scale variability, measurement error, different averaging schemes, a lagged relationship, etc. This problem was highlighted in Chapter 2 with the SST EOF maps which indicated that different areas of Alaska have responded differently to the climatic regime shifts.
The pattern of correlations among the five time series described above is illustrated in Table 4.1. In this situation, prewhitening of the series (for autocorrelation) is probably unnecessary as the series are being tested for how indicative they are of a single phenomenon. While most of the correlations are significant, only the two air temperature series might be considered interchangeable. The two air temperature series were formed using a "long winter" definition, i.e. November to March. This was done because the correlation between the 3-month winter and 3-month spring series in each region were extremely low (r < 0.3). The "long winter" time series were well correlated with both the "short winter" and spring (r > 0.6 for Bristol Bay and r > 0.7 for Kodiak) time series. Perhaps most disturbing is the almost zero correlation between the northern Gulf of Alaska and Bristol Bay SST time series: they could conceivably generate opposite hypotheses on how temperature affects a response variable. It is encouraging to note that the SST principal component has the most consistent pattern of correlation with the other time series. This result is likely due to the derivation of the principal component. It is essentially an average of the dominant trends among all the Alaskan sea surface temperature stations.
Four zooplankton datasets series are analyzed below. The data come from Fulton (1983), McFarlane and Beamish (1992), Brodeur and Ware (1992), and Roemmich and McGowan (1995). More complete descriptions than those given below can be obtained from the source papers.
The Brodeur and Ware (1992) time series is a measure of the overall zooplankton biomass within the Subarctic and Transition regions of the North Pacific. Their index is derived from two sets of research surveys, one covering the years 1956-1962, the other covering the years 1980-1994 (1986 is missing). The unit of measurement for their data is grams dry weight of zooplankton per cubic kilometer of sea water (g/1000 m3). The data for each year were collected between June 15 and the end of July.
The McFarlane and Beamish (1992) time series is a measure of zooplankton biomass at Ocean Station P (50°N, 145°W). Their index is a count of the average number of copepods per m3 during March-May from 1965-1980.
The Fulton (1983) dataset is a measure of zooplankton biomass (mg/m3) from 24 years (1956-1980) of vertical net sampling at Ocean Station P. Data were collected at alternating six week periods prior to 1969 and continuously after that time. A change in gear type occurred in 1966, which Fulton (1983) corrected for based on a series of intercalibration tows. Reanalysis of that data, however, indicate the correction factor was too low (Waddell and McKinnell 1995, Brodeur et al. in press) and the newly corrected data are used in this study. The data are available on a monthly basis; I formed a spring time series by averaging the months of March, April and May within each year.
The Roemmich and McGowan (1995) data are zooplankton displacement volumes (ml/1000m3) from the central part of the CalCOFI grid (Lines 77-93) over the period 1951-1994. Because I use these data to analyze advective processes, the data set was trimmed to include only those stations > 60 km offshore. Like the Fulton (1983) data, collections were made on a monthly basis, though there is a great deal of missing data. A spring time series was also formed by averaging data for March, April, and May within each year. The four spring time series are illustrated in Figure 4.4
The final set of data I analyze are growth indices for five Alaskan and British Columbia sockeye salmon stocks (Figure 4.5). The growth indices are scale measurements taken from escaped fish. The measurements are the average width of the growth zone (in mm) for each year of marine residence, with larger values indicating faster growth. In the three Alaskan stocks I examined (Nushagak, Black Lake, and Chignik Lake), there are three years of marine growth, which I term SW1, SW2, and SW3. Details on the collection and measurement of the scales can be found in the following publications: Zimmerman (1990) for the Nushagak data and Bumgarner (1992) for the Black Lake and Chignik Lake data. For each of these three stocks, I selected and utilized only the dominant age group. For Nushagak and Black Lake, this was 1.3 fish (i.e., 1 year freshwater, 3 years marine residence); for Chignik Lake this was 2.3 fish. For the two British Columbia stocks (Adams Lake and Weaver Creek), I had only first year marine growth data. These data were provided to me by Dr. Michael Henderson of the Dept. Fisheries and Oceans, Vancouver, B.C., Canada. The time series of growth increments for the five stocks are illustrated in Figure 4.6.
Four of the time series of salmon production used to test the conceptual model were introduced in Chapter 3. They are the historical (corrected for high seas interception) catch from 1925-1992 for western and central Alaska sockeye salmon, and central and southeast Alaska pink salmon. In addition to these series, run size (catch+escapement) for the three Alaska sockeye stocks for which I obtained growth data are also utilized (Figure 4.7). These data are the number of fish of the particular life history that returned in a given year, i.e., Black Lake and Nushagak 1.3 fish and Chignik Lake 2.3 fish.
In order to compute cross-correlations between physical and biological time series that are unaffected by the regime shift, I constructed time series models for the physical time series described above. The prewhitening models were intervention/time series ARIMA models, with 1925, 1947 and 1977 used as the intervention years for the physical variables. In all cases, no autocorrelation remained following removal of the intervention effects. Therefore, the time series models contain only intervention parameters and an intercept. In only one case was an intervention parameter not statistically significant at the 0.05 level. The 1925 intervention parameter for the Kodiak air temperature, while of the same sign (positive indicating a shift to warmer temperature) as the Bristol Bay air temperature, was poorly resolved since the series did not begin until 1919. There were, therefore, insufficient data to sufficiently resolve the intervention. Nevertheless, I still removed the intervention effect so that all physical time series had a constant mean throughout the whole time series. The prewhitening models for all physical time series are given in Table 4.2.
To prewhiten the biological time series, interventions were removed in the year of physical effect. Thus, for the zooplankton and growth series, the interventions are the same year as for the physical series (though for reasons discussed below, the zooplankton data could not be prewhitened). For the salmon production series, the interventions are applied in the year when the particular run/stock entered the ocean. The prewhitening models for western and central Alaska sockeye, central and southeast Alaska pink were derived in Chapter 3. The prewhitening models for the growth time series and the other salmon production series are given in Table 4.6 and discussed below in the appropriate sections.
As noted above, one of the most commonly accepted aspects of the conceptual model is the role of the Aleutian Low as the engine of change in salmon production variability. As the driving force behind the observed changes, there should be a direct relationship between the Aleutian Low and observed air and sea surface temperature variability. Correlations between my index of the Aleutian Low - SLP RPC1 - and the 5 seasonal air and sea surface temperature series, before and after prewhitening, are shown in Table 4.3. With the exception of Bristol Bay air temperature, all the temperature series are strongly negatively related to the Aleutian Low index. Prewhitening for interventions has very little effect on the magnitude of the correlation coefficients. Note that the tests with SST incorporates a seasonal lag allowing for the SST response to develop. However, the relationship between the SLP PC1 and the winter (Dec-Feb) SST PC1 is nearly as strong with before and after correlation coefficients of -0.70 and -0.64, respectively. The idea that SLP drives SST, and not vice versa, was first demonstrated by Davis (1976) who also used the method of PCA. In his study he did not correct, however, for the autocorrelation or regime effect. These results confirm his conclusions and demonstrate that SLP and SST are connected at both the interannual and decadal time scales.
To date, the most thorough exploration of the link between the Aleutian Low and salmon production has been Beamish and Bouillon (1993). They developed an index of the Aleutian Low, based on area encompassed by the 1000 millibar isobar during winters from 1900-1992, and illustrated how a smoothed version of their index bore a remarkable similarity to smoothed time series of salmon production. However, cross-correlations between the unsmoothed time series were very low, a factor they attributed to climate effects being spread out over a number of years, and/or interannual variability unassociated with large-scale climate effects. They noted that correlations between the smoothed time series were considerably higher, but correctly refrained from reporting the values due to statistical inference problems.
I examined the statistical relationship between my index of the Aleutian Low and the Alaskan salmon production time series and the how existence of regime shifts in these time series influences the apparent relationship. In Figure 4.8, the cross correlation functions (CCFs) before and after prewhitening are illustrated. In the absence of prewhitening, all four of the salmon production time series at negatively correlated with the Aleutian Low index at multiple lags. For western Alaska sockeye, lags 2 and 3 are significant, for central Alaska sockeye lags 3 and 5, for central Alaska pink lags 1, 3, and 4; and for southeast Alaska pink lags 3 and 5 are significant. With the exception of southeast Alaska pink, the earliest lag that is significant coincides the first year of marine residence for each stock. For instance, western Alaska sockeye generally spend two years in the ocean, while central Alaska sockeye spend three years. Pink salmon spend one year in the ocean and the lack of a significant lag-1 cross-correlation coefficient possibly indicates somewhat different mechanisms at work for that stock - though the temperature CCFs later are consistent with the other stocks.
After prewhitening the series as described earlier, all significant cross correlations vanish. The implication here is that the Aleutian Low and salmon production are linked on decadal time scales and the linkages are consistently timed in such a way as to indicate that salmon production is affected in the first year of marine residence, thus adding more support to thoughts on this subject summarized by Pearcy (1992). The lack of consistent covariation at the interannual time scale implies that there is no direct mechanistic relationship at this scale between the dominant climate feature (Aleutian Low) and salmon production. It is quite likely that the link between the Aleutian Low and salmon production is mediated through several intermediate, yet unidentified, processes. Several of those potential processes are examined in subsequent sections.
In fisheries oceanography studies, temperature is usually the first factor cited as a possibly important variable in the observed variability of the biological property under investigation. However, of all environmental variables measured, SST is likely the most highly autocorrelated of all due to the intrinsic heat storage capacity of water. Unfortunately, this factor is often not taken into account, with the resultant artificial inflation of significance values. Air temperature is also autocorrelated, though generally not as strongly as SST.
Given the qualifications cited above, any reported significant correlation with temperature in a non-experimental study must be interpreted cautiously. With that proviso, I examined the relationship between the Bristol Bay air temperature series and the salmon production time series. The Bristol Bay series was selected because it is of equal duration as the salmon series, while the SST time series are shorter. I did, however, repeat the analysis using the shorter salmon time series and SST PC1 and obtained the same results as with the air temperature analysis. Those results are not presented for the sake of brevity.
Prior to prewhitening, the production time series appear to be very strongly related to temperature over a large number of lags (Fig. 4.9). Examination of the cross-correlation function (CCF) would seem to indicate, for example, that Alaskan sockeye salmon catch is related to SST at all lags between -3 years (i.e., 3 years after the catch) and +6 years (i.e., six years prior to the catch). For pink salmon, the lag relationships are all significant between -2 and +6 years. A negative lag effect would have the interpretation that salmon production leads (drives) temperature.
As with the Aleutian Low correlations, removal of the intervention/regime shift effect almost completely eliminates all significant lag correlations. The large number of significant correlations is due almost exclusively to the fact that all six time series show decadal scale shifts in their mean. There is no evidence of an annual, lagged relationship between temperature and salmon production, with the possible exception of western Alaska sockeye salmon. Removal of the intervention effect had little impact on the lag 2 relationship between Bristol Bay air temperature and western Alaska sockeye salmon. This suggests that, at the interannual time scale, winter air temperature may be related to sockeye production 2 years later. I examined several other CCFs between the various temperature and salmon production time series above and, in all cases, found the same result.
Zooplankton constitute a major component of the salmonid diet, particularly during the juvenile phase (Burgner 1991). In searching for a bottom-up controlling mechanism of salmonid production, food availability should be considered a strong candidate for influence. There are several aspects to the food/production relationship that could be influential. These include availability at different life stages, quantity and quality of food, annual relative distributions of food and prey, etc. It is also more straightforward to envision a link between large-scale climate variability and zooplankton production.
However, despite the likely importance of zooplankton, the available data for Gulf of Alaska have little statistical utility due to their incompleteness, brevity, and disparate collection years (Figure 4.4). For instance, the Fulton (1983, hereafter called "Z1") time series has 16 common years with the McFarlane and Beamish (1992, hereafter "Z2") time series, but both have seven common years with the Brodeur and Ware (1992, hereafter "Z3") time series; and Z2 and Z3 have only one common year. The longest of the three time series - Z1 has 24 data points, Z2 has 16 data points, and Z3 has 21. Perhaps more problematic to this analysis is that none of the three time series satisfactorily captures the late 1970s regime shift. The Z3 series contains data from the late 50s, early 60s and 1980-94 (excluding 1986), but nothing in between. Both Z1 and Z2 terminate in 1980, thus providing insufficient sampling of the post regime period. Finally, the series do not provide evidence of a single signal. In particular, Z1 and Z3 show almost opposite trends in the late 50s period (r = -0.55 based on the seven years of overlap). During the common period of 1965-1980, Z1 and Z2 have a correlation of 0.60. For all of these reasons, I decided not to attempt to prewhiten the zooplankton series for a late 1970s regime shift. The magnitude of the shift could, at best, be estimated only very poorly.
The non-prewhitened correlations between the three indices of zooplankton production and climate and salmon production indices are given in Table 4.4. I used the SLP and SST principal components for measuring climatic influence on zooplankton production. All three time series were negatively related to SLP and positively related to SST, though not all significantly. The strongest relationship was between Z2 and SLP, which is essentially the same relationship as reported by McFarlane and Beamish (1992).
I computed the correlations between the three indices of zooplankton production and the two Alaskan salmon catch indices. Because zooplankton production could have a delayed effect on the catch series, I computed the lagged (i.e., cross-) correlations for up to one year later for pink salmon and three years later for sockeye salmon. None of the correlations between the salmon catches and either Z1 or Z2 were significant. Conversely, all correlations between catch and Z3 were significant. This last result is, to a large extent, expected since the variability in the series is dominated by a decadal change in the level of production that matches the change in the salmon series, i.e., low production in the 50s and 60s and high production in the 80s and 90s.
One final test was conducted using the zooplankton data. This test is more fully described in Brodeur et al. (in press), the original publication for which I conducted the analysis The relationship tested here was between Z1, the Ocean Station P zooplankton series, and Z4, the CalCOFI grid zooplankton data. I noted earlier the hypothesis that flow into the Alaska and California Currents fluctuate out of phase. One test of this hypothesis is that zooplankton production in the two regions is also inversely related. This would require, of course, that the zooplankton measurements in both regions reflect advection and not in situ production. For the California Current, I guarded against this possibility by only including stations more than 60 km offshore to remove coastal upwelled production (Chelton et al. 1982). Ocean Station P is not well placed to measure Alaska Current production as it stands somewhat closer to the separation of the two currents than would be ideal. Nevertheless, it is the best available measure. The correlation between the two spring biomass series is -0.62 (Figure 4.10), indicating out of phase production between the two regions during the spring period.
I examined the relationships between the eleven time series of marine growth and temperature, sea level pressure and zooplankton indices. Prior to prewhitening, there was a large number of statistically significant correlations (Table 4.5). In particular, the SST principal component (SST PC1) time series and Bristol Bay air temperature series (BB AT) were positively related to the Alaskan stocks' first and second year marine growth, while McFarlane and Beamish's (1992) zooplankton (Z2) series was positively related to the Alaskan stocks' second and third year marine growth. In general, the two Canadian stocks had negative relationships with both the physical and zooplankton indices, though only a few were significant at the 0.05 level.
For the nine Alaskan growth time series, there were 25 out of 81 correlations significant at the 0.10 level, or roughly three times the number that might be expected due to chance alone. For the Canadian growth series, 4 of 18 were significant, or double the number of expected random significant correlations.
I next determined whether the growth time series exhibited a step change in 1977 coincident in time with the 1977 physical regime shift. All eleven time series were subjected to an intervention analysis using 1977 as the intervention year (Table 4.6). Of the 11, seven contained a statistically significant intervention. Five of the significant interventions were positive: Chignik Lake and Black Lake first year marine growth, Black Lake, Chignik Lake, and Nushagak second year growth. The two Fraser River stocks - Adams Lake and Weaver Creek - both had negative interventions. Of the four non-significant interventions, three were the third year marine growth of the Alaska stocks and the other was Black Lake first year marine growth which was positive, but was significant only at a significance level of 0.11. In summary, the Alaska stocks showed increased marine growth in their first two years of marine residence and a decrease in the third year. The Canadian stocks, for which I had only first year growth, showed a decrease in growth coincident with the physical regime shift. I formed prewhitened series for each of the growth time series using the models reported in Table 4.6. To be consistent, I removed the intervention effects for the four series that had non-significant interventions so that all series had a stationary mean for the entire time period.
After prewhitening , no significant correlations between growth and the physical indices remained at a significance level of 0.05. In other words, the simple act of setting the pre and post 1977 intervention periods to a common mean eliminated all indication of significant correlation between temperature and growth indices. Somewhat interestingly, however, one set of correlations did remain strongly significant - third year marine growth of the Alaska stocks and zooplankton series Z2. This result is difficult to interpret and is clouded by the fact that the Z2 series ends in 1980 and was not prewhitened.
I next examined the evidence for a relationship between marine growth and run size. For this exercise, I used the three Alaska stocks. As noted earlier, only the predominant life history for each stock was utilized. To compare marine growth with eventual run size, the number of years between the year of growth and the year of return must be properly aligned. Thus, the tests for these sockeye stocks, which all spend three full years in the ocean, is between the first year of marine growth (SW1) and run size three years later, SW2 and run size two years later, and SW3 and run size the following year. The lagged correlations, as described, are shown in Table 4.7. There is a strong positive relationship between both SW1 and run size three years later, and between SW2 and run size two years later. In all, five of the six SW1 and SW2 growth series were significantly (p < 0.1) related to eventual run size. Run size and growth in the final full year at sea (SW) is negatively related for all three stocks, though at a significance level greater than 0.10.
The three run size time series were tested for a significant intervention in 1980, as this would be the return year for fish that entered the marine environment in 1977. All three contained a highly significant intervention (Table 4.6). All three runs increased at least 50% with the physical regime shift of 1977. After prewhitening the growth and run size time series, the lagged correlations were repeated. The prewhitened correlations, while somewhat reduced in magnitude, remained positive (4 of 6 had p < 0.1)
The goal of this chapter was to critically examine a conceptual model of regime-scale variability in salmon production and its link to climatic variability. The model was examined in the context of abrupt regime shifts reflected in the level of salmon production. This amounted to computing correlations between different biological and physical time series. Correlations were first computed between the original time series. The time series were then tested for evidence of a regime shift, i.e., a statistically significant jump in the mean level at the time periods noted above. The time series were then "detrended" by removing the effect of the regime shift and new correlations computed. There are both positive and inconclusive findings in this analysis. They are briefly summarized in relation to the mechanisms discussed in the Introduction.
It seems clear that the climatic regime shift is directly related to variability in the Aleutian Low. In Chapter 2 of this dissertation, I showed how an index of the Aleutian Low - SLP RPC1 - was best characterized statistically as alternating between positive and negative signed regimes. The regimes shifted from one sign to another in 1925, 1947, and 1977. Alaskan temperatures, both air and sea surface, also shifted in mean level at the same time. The correlations between the index of the Aleutian Low and temperatures were virtually unchanged with removal of the interventions.
A cross correlation analysis of SLP RPC1 and the four regional salmon production indices indicated there was no direct interannual link between the Aleutian Low and salmon production. While all the time series had highly significant regime behavior, the lag correlation coefficients, negative as expected (i.e. more intense Aleutian Low linked to higher salmon production), were not statistically significant after prewhitening for a regime effect. The timing of the lags prior to prewhitening does give an indication of the timing in the salmon life history when the regime shift effect takes place - the first year of marine residence. I surmise that the variability in production is related to the Aleutian Low, but the mechanism is complex enough that the signal is clouded by measurement noise.
I next examined the evidence for a temperature effect on Alaskan salmon production. Prior to prewhitening, two indices of temperature - Bristol Bay winter air and spring SST PC1 - both showed significant positive correlation with production over a large number of lags. Interpretation of these lag correlations would indicate that temperature both led and lagged salmon production. This result, as illustrated with simulation in Chapter 1, derives from both sets of times series (physical and biological) responding to the same regime signal. Removal of that signal eliminated any relationship between temperature and salmon production.
The four most commonly available zooplankton datasets used to measure secondary production in the North Pacific are of uneven quality. It is particularly discouraging that none capture the critical period of the mid to late 1970s. Nevertheless, I did correlate behavior in the zooplankton indices with both physical and biological time series. The strongest set of relationships were between Brodeur and Ware's (1992) zooplankton productivity index of the Subarctic Pacific and sockeye and pink salmon production. All lag correlations were strongly positive but, like the temperature cross correlations, this result is expected since all the series jump from low to high with the regime shift. Also curious is that the correlations are strongest at lag 0 (zooplankton production in the year of catch) and decrease with lag. Based on the conceptual model, one would expect the opposite pattern with the lags corresponding to the first year of marine residence (lag 1 for pink, lag 2 for western Alaska sockeye, lag 3 for central Alaska sockeye). The other two zooplankton time series showed little relation to salmon production. Correlations with the physical indices indicated that zooplankton production was more closely related to SLP and SST variation than was the case with salmon production. All three zooplankton indices were negatively related to SLP and positively related to SST. The one other result using zooplankton data was the out of phase relationship between zooplankton production at Ocean Station P and the CalCOFI region. This relationship, which is more extensively analyzed in Brodeur et al. (in press), supports the notion of an inverse relationship between the Alaska and California Currents.
The final variable I analyzed was growth in several Alaskan and Canadian sockeye salmon stocks. Like almost all the other variables, marine growth of the stocks responded to the climatic regime shift. The three Alaskan stocks showed a positive step increase in 1977, while the two Canadian stocks showed a step decrease. For the Alaskan stocks, only the first two years of marine growth increases - there was a decrease in third year marine growth. The inverse relation between the Alaskan and Canadian stocks may be further evidence of the out of phase relationship between the Alaskan and California Currents. Such an interpretation is speculative and clouded by the fact that British Columbia actually occupies an intermediate location between the main stems of the two currents. Prior to prewhitening, the Alaskan stocks showed statistically significant, though weak, relationships between climatic variables and growth. Removal of the intervention effects eliminated essentially all significant correlations. The same result held for relationships between zooplankton production indices and growth. Zooplankton production and marine growth are positively correlated until the 1970s regime shift effect is accounted for (though only for growth since the zooplankton stocks could not be prewhitened).
The final set of relationships examined was between marine growth and eventual run size for three Alaskan sockeye salmon runs. The results of cross-correlation analysis indicate that Chignik Lake, Black Lake, and Nushagak sockeye salmon is positively related to first and second year marine growth. Cross correlations remained statistically significant - though slightly reduced in value - with removal of intervention effects.
In conclusion, there are three "take home" conclusions from the above results. First, there is overwhelming evidence that the effects of climatic regime shifts reverberate throughout many aspects of the North Pacific ecosystem. Abrupt changes are documented in such biological indicators as secondary productivity, marine growth, and salmon run size. Summarization of the 1976/77 climatic regime impact on non-salmonid components of the marine ecosystem is given in Francis et al. (in review).
Secondly, the growth and production data support two hypotheses of the conceptual model. First, survival of salmon and, ultimate run size, appears to be set in the first year of marine residence, likely the first few months. Increased survival results from an increased growth rate at a young age, the so-called "productivity-growth" (Pearcy 1992) hypothesis. The higher growth rate results from the wide spread availability of zooplankton, the distribution and production of which dramatically changed with the climate regime (Brodeur and Ware 1992). The regime shift significantly increased abundance of large calanoid copepods (Neocalanus plumchrus and N. cristatus) (McFarlane and Beamish 1992), a dominant component of sockeye salmon diet (Burgner 1991). Secondly, density dependent growth at sea occurring in the final year results in smaller returning adults, such as documented by (Bigler et al. in press). This is supported by the inverse relationship between abundance and final year growth. The decrease in final year growth more than offsets the increase in early ocean growth.
The third conclusion is that while there does appear to be support for the basic premises of the conceptual model, much remains to be explained. Temperature, for instance, is often cited as an important variable and it is clearly covarying with many components of the ecosystem. The basic question is whether it is a driving factor, or just another response variable. The multi-lag correlation (at both positive and negative lags) between temperature and salmon production has no simple explanation other than the common regime behavior. Finally, while early marine growth appears to be important to survival, is this due to reduction in starvation mortality or predator avoidance? Designing studies to experimentally test these and associated hypotheses will present a key challenge over the next decade as we attempt to better understand how climate affects salmon production.
| BB AT |
| ||||
| BB SST |
|
| |||
| KOD AT |
|
|
| ||
| KOD SST |
|
|
|
| |
| SST PC1 |
|
|
|
|
|
| Physical variable | ||||||||
| SLP RPC1 | ||||||||
| Bristol Bay AT | ||||||||
| Bristol Bay SST | ||||||||
| Kodiak AT | ||||||||
| Kodiak SST | ||||||||
| SST PC1 | ||||||||
Variable | Before prewhitening | After prewhitening |
| spring SST PC1 | -0.75 | -0.71 |
| Kodiak air temp. | -0.52 | -0.44 |
| Kodiak SST | -0.66 | -0.61 |
| Bristol Bay air temp. | -0.54 | -0.47 |
| Bristol Bay SST | -0.26 | -0.11 |
| |||||||||
| Z1 | |||||||||
| Z2 | |||||||||
| Z3 | |||||||||
| Z1 | ||||||
| Z2 | ||||||
| Z3 | ||||||
| AL 1 | |||||||||
| WC 1 | |||||||||
| BL 1 | |||||||||
| CL 1 | |||||||||
| NU 1 | |||||||||
| BL 2 | |||||||||
| CL 2 | |||||||||
| NU 2 | |||||||||
| BL 3 | |||||||||
| CL 3 | |||||||||
| NU 3 |
After prewhitening
| AL 1 | |||||||||
| WC 1 | |||||||||
| BL 1 | |||||||||
| CL 1 | |||||||||
| NU 1 | |||||||||
| BL 2 | |||||||||
| CL 2 | |||||||||
| NU 2 | |||||||||
| BL 3 | |||||||||
| CL 3 | |||||||||
| NU 3 |
| Growth (mm) | ||||
| Adams Lake SW1 | ||||
| Weaver Creek SW1 | ||||
| Black Lake SW1 | ||||
| Chignik Lake SW1 | ||||
| Nushagak SW1 | ||||
| Black Lake SW2 | ||||
| Chignik Lake SW2 | ||||
| Nushagak SW2 | ||||
| Black Lake SW3 | ||||
| Chignik Lake SW3 | ||||
| Nushagak SW3 | ||||
| Run size (millions) | ||||
| Black Lake 1.3 | ||||
| Chignik Lake 2.3 | ||||
| Nushagak 1.3 | ||||
Table 4.7. Lag correlations between sockeye
salmon scale growth and ultimate run size for three stocks of
sockeye salmon, for the years 1950-1990. "SW" refers
to saltwater year of growth and the lag is the number of years
later that fish return to spawn after the respective growth year.
Single underlining indicates significance at the 0.10 level, double
underlining at the 0.05 level.
| lag 1 | lag 2 | lag 3 | |
| Black Lake SW1 | 0.40 | ||
| Chignik Lake SW1 | 0.33 | ||
| Nushagak SW1 | 0.48 | ||
| Black Lake SW2 | 0.43 | ||
| Chignik Lake SW2 | 0.25 | ||
| Nushagak SW2 | 0.42 | ||
| Black Lake SW3 | -0.18 | ||
| Chignik Lake SW3 | -0.05 | ||
| Nushagak SW3 | -0.29 |
After prewhitening
| lag 1 | lag 2 | lag 3 | |
| Black Lake SW1 | 0.33 | ||
| Chignik Lake SW1 | 0.30 | ||
| Nushagak SW1 | 0.07 | ||
| Black Lake SW2 | 0.26 | ||
| Chignik Lake SW2 | 0.11 | ||
| Nushagak SW2 | 0.28 | ||
| Black Lake SW3 | -0.08 | ||
| Chignik Lake SW3 | 0.26 | ||
| Nushagak SW3 | -0.23 |