Secular Trends And The Global Enso Mode

One reason for considering secular trends and the global ENSO mode first is that they will be removed from the data prior to analysis of decadal-multidecadal variability. It is also appropriate to discuss them together because of current speculation that global warming may affect ENSO variability (Trenberth and Hoar, 1996). More generally, the characteristics of ENSO, or even its existence (or lack of it) during prehistoric periods, may be dependent on the background climate (Sun, 2000). We shall redirect our attention to these timescales when we discuss the paleoclimate implications of modern measurements later in Section 2.6.

2.3.1. The Global Distribution of Secular Trends

Figure 1A is a map of the globally distributed linear trend in SSTA over a 136-year time period (1856-1991), based on a simple, unweighted linear regression of SSTAat each 5° X 5° grid point on the monthly time vector. The associated temporal variation of the global mean SSTA (Fig. 1B) is relatively trendless prior to 1900 and has a more accentuated trend after that time. Therefore, while the trend map is representative of the true secular pattern after 1900, the magnitudes shown tend to underestimate the post-1900 trend by about 40%. We have tried alternate approaches for treating a a


FIGURE 1 (A) Distribution of the 1856-1991 linear (least-squares) trend (°C/century) in the smoothed Kaplan et al. (1998) sea surface temperature anomaly (SSTA) data set. The contour interval is 0.2°C / century, the heavy contour is zero, and the light solid (dashed) contours are positive (negative). Positive trends exceeding 0.2°C/ century are shaded. (B) Time variation of the global average SSTA.


FIGURE 1 (A) Distribution of the 1856-1991 linear (least-squares) trend (°C/century) in the smoothed Kaplan et al. (1998) sea surface temperature anomaly (SSTA) data set. The contour interval is 0.2°C / century, the heavy contour is zero, and the light solid (dashed) contours are positive (negative). Positive trends exceeding 0.2°C/ century are shaded. (B) Time variation of the global average SSTA.

the secular change in the data—e.g., division into two periods with different means—and the description of the secular variation is not sensitive to the representation. Therefore, we adopted the simplest scheme (linear trend) rather than making necessarily arbitrary judgments.

At the largest global scale, the overriding impression from the trend map is that warming trends dominate over cooling and that the rate of secular change shows high spatial inhomogeneity. Off the equator, the shaded regions of largest positive trend (>0.4°C/century) are in the South Atlantic, the subtropics of the central and eastern Pacific, the Gulf of Alaska, and the region north of Cape Hatteras where the Gulf Stream separates from the eastern coastline of the continental United States. Regions of negative trend (<— 0.2°C/centu-ry) are found in the midlatitudes of the east-central North Pacific and in the high-latitude region southeast of Greenland. The equatorial Pacific, of particular interest because of its importance for ENSO variations, shows only a null or weakly positive trend between the Galápagos Islands and the date line, strong cooling off Ecuador and Peru, and warming west of the date line. The equatorial Atlantic shows moderate warming that appears as a low-latitude extension of the stronger warming in the South Atlantic. Other than in the central South Pacific, all other regions are characterized by a mild warming trend of ca. 0.1°-0.3°C/century.

The linear trends in the instrumental record shown here are the closest thing, in timescale, to the slow climate fluctuations that might be detected in previous epochs by less direct, paleoclimatic methods. Examples that come to mind are the Little Ice Age (LIA) and the Medieval Warm Period (MWP). To the extent that the LIA and MWP imply globally warmer or cooler climates in the past, we must keep in mind that those pat terns also may have been quite heterogeneous and that certain regions may also have been cooling (warming) when the Earth as a whole was warming (cooling). This lack of homogeneity is confirmed by the 500-year atmospheric model simulation of Hunt (1998), which reproduces ~100-year-long periods of globally warmer or cooler air temperatures similar to the LIA and MWP. Extreme decadal averages in these periods, which occur as natural (externally unforced) climate variability in the model, exhibit large contiguous areas of ocean and land with anomalies opposite to the contemporaneous global mean. This pattern reinforces the likelihood that a period like the mid-Holocene would have had very nonuniform differences from our present climate. These differences could be especially significant if the ancient patterns of SST change were anything like those shown in Fig. 1, because the regions with trends counter to the global averages are in the North Pacific and the North Atlantic, where present-day water masses are formed and subducted (Nakamura et al., 1997; Curry et al., 1998). These processes are currently suspected of playing an essential role in the shifts between climate regimes. The existence of such inhomogeneities at the secular timescale implies that similar feedback mechanisms may play a future role in climate variations with centennial or longer timescales and may also have played a role in the past.

The linear trends were removed from the K98 data prior to the removal of the ENSO mode described in Section 2.3.2. The order in which these removals were done has no effect on the residual (trendless, non-ENSO) variability. All discussions of the SSTA variance explained by the various components of variability have been referred to the detrended data set.

2.3.2. The Global ENSO Mode

The interannual ENSO signal is distributed from the Pacific to other tropical oceans through its associated global tropospheric anomalies (Hastenrath et al., 1987; Latif and Barnett, 1995; Lanzante, 1996; Enfield and Mayer, 1997; Enfield and Mestas-Nunez, 1999). The characteristic lag of the extended ENSO response in the Atlantic and Indian Oceans is approximately one to three seasons, which is comparable to the lag of the high-latitude North Pacific with respect to the tropical Pacific. This is not a mere curiosity. According to Lau and Nath (1994), if the extended ENSO signal in other ocean basins is not included in a global coupled model, the model cannot accurately replicate the tropo-spheric response, such as the Pacific-North American (PNA) pressure pattern (Barnston and Livezey, 1987). On the other hand, if one wishes to understand the relationship between land climate and another climate mode such as the North Atlantic Oscillation (NAO), the existence of the ENSO signal in the Atlantic makes unequivocal attribution difficult.

To better characterize the global ENSO mode of variability, we have performed a complex empirical orthogonal function (CEOF) analysis of the monthly K98 data that correctly preserves the phase lags within the Pacific and other oceans. The detrended data are first band-passed to block periodicities larger than 8 years or shorter than 1.5 years. The leading mode of the decomposition is the one that represents the global ENSO. It explains 13.4% of the variability in the detrended, un-filtered data set for the December-January-February (DJF) season. For additional details on the methods used, the reader is referred to Enfield and Mestas-Nunez (1999). However, we have repeated the Enfield and Mestas-Nunez calculation here for the shorter but more reliable 1871-1991 period. The differences be tween the two analyses are small, and the conclusions are essentially unchanged.

Because the ENSO mode is based on a complex decomposition of the data, it has complex eigenvectors and expansion coefficients which can be transformed into amplitudes and phases in both the spatial and temporal domains. We show the spatial amplitude and phase in Figs. 2Aand 2B. The obvious reference rectangle to characterize the ENSO mode is the well-known Nino-3 region in the equatorial Pacific, bounded by 5°N-5°S, 90°-150°W (Fig. 2B). The spatial amplitude (Fig. 2A) is shown as a percent gain with respect to the reconstructed temporal amplitudes for Nino-3 (Fig. 2C). Thus, the average gain in the Nino-3 rectangle is 100, and the contour labeled 30 passes through regions where the amplitude is 30% of the Nino-3 amplitude. Significance tests on the correlation coefficients between the data and the temporal reconstruction show a a

Was this article helpful?

0 0

Post a comment