One problem regarding regime period definition as given by Lluch-Belda et al. (1989, 1992) is that sardine and anchovy stocks over the world, although fluctuating during the same gross periods, show year to year differences between areas which make it hard to establish when a given global regime starts or finishes. One reason for this is that stock abundances must be inferred from catch data, since more accurate indices of abundance are lacking for all areas and periods. While this is not a problem in demonstrating the existence of regime changes, the characterization of regime periods in terms of the ENSO phenomenon requires these periods to be defined as accurately and objectively as possible.
We first approached the problem using the annual global air surface temperature (GSAT) to build an "index" of global climate regimes, keeping in mind that sardines tend to be abundant during sustained, long-term high-temperature periods and anchovies are abundant during sustained, long-term low-temperature periods (except for the Benguela System, where the opposite situation seems to be the case). A second approach was to build a composite series for the sardine and anchovy regimes. This will be explained later.
Figure 1. Global Surface Air Temperature (GSAT)
GSAT series, as made available by the NCAR/NWDC/NOAA, is presented in Figure 1. This series consists of yearly anomalies computed as departures from the 1951 to 80 year average. Since regime variability is over the decadal time scale, some transformations have to be made to these data before they are evident. First, the series shows a clear long term upward trend which should be removed, since we are not interested in this time scale variability. Similarly, the high frequency should be removed for the same reason. Finally, the period considered for computing the average year temperature implies that the average of the entire series is not zero, which makes it difficult to classify the years as either warm or cold. To avoid these problems, the original GSAT series was first detrended by substracting the simple regression line and then smoothed by means of the following simple exponential transformation:
St = 0.1 * Xt + (0.9) * St-1
where St is the value of the transformed series at time t, and St-1 is the value of the transformed series at time t-1, Xt is the value of the untransformed series at time t. Finally, data were standardized to make the average value equal to zero. To show the correspondence between climate regime periods as identified within this series and those shown by the small pelagic catch, Tables 1 and 2 (page 14) present a synthesis of the sardine and anchovy regime periods as identified by Lluch-Belda et al. (1989, 1992).
For the analysis of the spatial coverage of the climate regime signals, we used the large coverage air temperature series provided by the NCAR/NWDC/NOAA. These are 12 series of yearly mean air temperature, grouped into three spatial resolution levels. The first level includes two series, one for each hemisphere, which will be referred as NH and SH. The second level includes three series for the following latitudinal bands: 24 to 90 N (referred as N1), 24S to 24N (referred as E1) and 24 to 90S (referred as S1). The third includes the following eight latitudinal bands: 64 to 90N (N11), 44 to 64N (N12), 24 to 44N (N13), 0 to 24N (N14), 0 to 24S (S11), 24 to 44S (S12), 44 to 64S (S13) and 64 to 90S (S14). For comparison to the identified regime trends and periods, these series were treated the same way as the GSAT was (i.e., detrended, smoothed and standardized).
To test the relation between each area and the global regimes, we computed the correlation coefficients between the eight latitudinal bands and the transformed GSAT, for 1904 to 1995. Before doing this, normality was tested for all the series using the Shapiro-Wilks W test. All series were found to behave normally except Nil, thus its correlation coefficient is not to be considered.
For the analysis of El Niño events, we built series of relative strength and frequency content of interdecadal variations, based on the 20th Century data of years in which large-scale El Niño events occurred, as stated by Quinn (1992). These include not only the year in which the event started, but also a gross estimate of its strength based on different sources of historical evidence (see Quinn et al. 1987, Quinn and Neal 1992 and Quinn 1992).
To show some possible changes in the frequency of events over the interdecade time scale, the number of events were summed over moving 11-year periods. A similar process was applied to the relative-strength data, using zero for those years in which no event started. Because each event is associated with the year in which it started, it is considered only once regardless of its duration.
These data were smoothed and then compared to the periods identified within the transformed GSAT series. The unsmoothed frequency and strength values were also grouped by the two periods defined from the transformed GSAT series and distributions for each period were carried on. When looking for differences between these periods, we used a Kolmogorov-Smirnov two-way mean comparison, since their distributions proved to be non-normal. Correlations between the transformed GSAT and all the unsmoothed frequency and relative strength data were determined after successfully testing the three variables for normality by means of the Shapiro-Wilk W-test.
Since the GSAT series consists of annual means, statistical comparisons between this series and the monthly Tahiti-Darwin SOI series was difficult. Therefore, we looked at the correlations between SOI annual and seasonal means series (as computed from the monthly values), in order to define which series were to be compared against the yearly GSAT series. As can be noted (Figure 12), there are close and statistically significant correlations between the yearly averages (YR) and all seasonal series except winter (W), which varies rather independently of the others. Thus, we decided to use both the yearly mean series (as representative of the SOI variations from spring through fall) and the winter series. For this last series the SOI value for December of the preceding year was used when computing the seasonal average.
One additional problem when trying to analyze the ENSO variability is that all the available time series (Tahiti-Darwin, Australian, COADS-derived, etc.) show some gaps during the present century. However, they are not for the same periods in all of the series, and since the different indexes are usually well correlated, it is possible to combine them to build a continuous time series. Therefore, we correlated the Tahiti-Darwin SOI to the COADS SOI, and used the regression equation to "fill the gaps" by calculating the missing values of the first series from the observed values of the second series. This procedure allowed us to complete most of the missing values. However, a two years gap (1914 and 1915) persisted because the COADS SOI index has no observations for those years. Because this is a minor gap, those values were interpolated within each series. The resulting "composite" SOI series is almost continuous for the present century.
We present the "composite" yearly and winter SOI series (see Figure 14), with both the original and the transformed values (i.e. 9-terms moving average smoothed, detrended and standardized). We calculated regressions between SOI (winter and yearly) and GSAT series as well as distributions of the transformed SOI series (yearly and winter) grouped by the trends and periods previously detected in the GSAT.
SOI is considered as an index of ENSO activity at the interanual time-scale. Positive values indicate that atmospheric pressure at the western tropical Pacific Ocean is above that of the eastern side. Conditions when this situation is revert, and the SOI becomes negative, are indicative of an ENSO event. However, relations at decadal time-scale are not so direct and thus not as well established and understood. Since our research focuses on regime time-scale variability, we have explored this relation by plotting together the series of frequency and strength of ENSO events with two series derived from the SOI: the 11 years moving average of yearly SOI values and a series of the positive to negative rate of SOI yearly values over periods of 11 years.
One problem regarding the sardine and anchovy abundance regimes as defined by Lluch-Belda et al. (1989, 1992) is that different stocks, although they fluctuate grossly simultaneously, do show some differences in timing from one region to another. At least to some extent these variations result from different fishery regimes. Therefore catch records, while they may be regarded as a gross index of large abundance changes, are by no means precise indicators of abundance changes, particularly of the year-to-year variability. Thus, when looking for correlations between worldwide abundance fluctuations and global climate series it is necessary to build one "composite" series that reflects the gross global tendencies rather than the regional short-term variations.
We approached this problem by combining the catch records of the main sardine and anchovy fisheries as provided by SCOR WG-98 (1996); those of the Japan, California, Humboldt, Benguela, and Canary systems. Individual annual catch series were first standardized and then summed as follows:
RIS = (JS + CalS + HS + BA + CanS) - (JA + CalA + HA + BS + CanA)
RIS: Annual composite value of the regime indicator series
JS: Standardized annual catch of the Japanese sardine
JA: Standardized annual catch of the Japanese anchovy
CalS: Standardized annual catch of the California sardine
CalA: Standardized annual catch of the California anchovy
HS: Standardized annual catch of the Humboldt sardine
HA: Standardized annual catch of the Humboldt anchovy
BS: Standardized annual catch of the Benguela sardine
BA: Standardized annual catch of the Benguela anchovy
CanS: Standardized annual catch of the Canary sardine
CanA: Standardized annual catch of the Canary anchovy
As defined, RIS values are highly positive when sardine stocks worldwide are abundant and anchovy stocks are at low levels, while strong negative RIS values are indicative of the opposite combination of abundance levels. The Benguela stocks were summed differently than the others. Benguela sardine (anchovy) catches were added to the anchovy (sardine) catches of the other systems. By doing so, we followed Crawford et al. (1991) and Lluch-Belda et al. (1989, 1992), who noted the Benguela stocks fluctuate "out of phase" with the rest of the stocks.
To investigate the relationships between RIS regimes and variability of the tropical-extratropical climate, we used the SOI composite series (yearly) and the Aleutian Low normalized November to March sea level pressure series (AL), after Polovina et al. (1995). Both series were smoothed by means of a 7-year moving average filter to eliminate the high-frequency variability, and then standardized for comparisons.
Relations to solar activity were investigated using the series on yearly sunspot numbers as provided by the Sunspot Index Data Center, Royal Observatory of Belgium. To eliminate variability not on the decadal-scale, the original data were detrended, smoothed with an 11-year moving average filter and scaled as departures from the average value. The 11-year signal of the original series was removed (on purpose) because of the selected filter. We have compared this series with the absolute differences of the SOI-AL by simple linear correlation.
Small pelagic regimes, as identified within the RIS, were used as reference periods when looking for the interdecadal variability of the following selected marine ecosystems: California, Japan, Humboldt, Eastern Australia, Eastern Tropical Pacific, Brazil, Canary, Benguela, and Somali. For each of these regions, a 10x10° box was selected, and their climatic information extracted from the following databases: sea surface temperature and sea level pressure from the COADS-CD ROM (Roy and Mendelssohn 1994) and temperature profiles from the NODC-CD ROM (NODC 1991). Selected areas are shown in Figure 2.
For each area, individual SST observations were averaged to generate a monthly time series for 1900 to 1990 period. Seasonal climatology values were computed for this entire period. Then, the same procedure was performed but for the following selected periods: 1941 to 1961 (a period of small SOI-AL differences, see Figure 20) and 1971 to 1977 (a period of high SOI-AL differences). Departures of each period values were also compared to the overall climatology.
A similar treatment was applied to the pressure data, except that the monthly series of each system (Figure 2, dark shaded boxes) were first substracted from a similar series built for the tropical areas within each ocean (Figure 2, light shaded boxes) to build individual pressure indices. These are intended to reflect the variations of the pressure gradient within each system.
Figure 2. Selected 10×10° boxes for the computation of regional climatic series (dark shaded), and tropical areas selected for computing of the pressure gradients (light shaded).
Temperature profiles were first averaged for each selected area on a monthly basis, and the resulting values used to build a monthly time series of thermocline depth, estimated as the mid-depth of the layer where the maximum change of temperature per unit depth takes place. The resulting series were then used to estimate the seasonal climatology values and the seasonal averages for the selected periods, as previously defined for the sea surface temperature.