Inter-Hemisphere Decadal Variations in SST, surface wind and heat flux over the Atlantic basin

Transcription

1 Inter-Hemisphere Decadal Variations in SST, surface wind and heat flux over the Atlantic basin Decadal climate variations are examined using a new observed dataset of marine meteorological variables in the Pan- Atlantic basin. Leading empirical orthogonal functions (EOFs: spatial patterns) of sea surface temperature (SST) anomalies on the decadal band (8-16 years) conducted in independent two subdomains north and south of equator feature one center of action at 15 o N and another center with opposite polarity at the same latitudes in the Southern Tropics. The same EOF analysis for sea level pressure (SLP) anomalies also indicates two centers of action with opposing polarities around 30 degree latitudes, straddling the equator. The accompanying four principle components (PCs: time series) that contain three cycles of decadal variations are correlated well with one another, indicative of the existence of the tropical dipole mode in the ocean-atmosphere system. Composite anomaly maps of wind velocity and heat fluxes, based on the PCs of leading modes of SST and SLP anomalies, indicate that the latent heat flux induced by the cross-equatorial wind plays an important role in forcing the dipole mode of the decadal SST variability. Anomalies on either side of equator show comparable amplitudes in the SST field, but have quite different amplitudes in wind velocity and flux fields. The role of low-level clouds in forcing the SST anomalies is discussed. Key Words : Pan-Atlantic, decadal variability, ocean-atmosphere interaction 57

2 1. Introduction Decadal sea surface temperature (SST) variations in the tropical Atlantic are organized into dipole patterns with centers of action around degree latitudes (Servain 1991, Nobre and Shukla 1996, Chang et al. 1997, among others). At the same time, sea level pressure (SLP) anomaly field also displays a dipole structure, whose centers of action locate slightly more poleward at 30 degree latitudes and sign opposite to local SST anomalies (Tanimoto and Xie 1999). This air-sea coupled tropical dipole structure could be associated with the mid-latitudes decadal variations via the atmospheric teleconnections (Watanabe and Kimoto 1999). Rajagopalan et al. (1998) presented that high coherences and in-phase (out-of phase) relation were found between North Atlantic Oscillation (NAO, Hurrell 1995) index and SST time series in northern (southern) tropics. The extratropical North Atlantic also displays pronounced decadal SST variations, which are associated with a decadal change of mid-latitudes westerlies (Deser and Blackmon 1993, Kushnir 1994, Delworth, 1993, Halliwell and Mayer 1996). Similar situations occur in the SST and SLP fields in the Southern Hemisphere (Venegas et al., 1997). The dipole mode, however, is not the only mode of SST variability in the tropical Atlantic. The interference of different modes is considered to cause an apparent inter-hemispheric decorrelation in both SST and SLP field (Fig. 1). Previous empirical decomposition analyses and their variations performed in domains that include the whole tropics produce contradictory results as to whether the northern and southern tropical SST anomalies vary independently or can be decomposed into a pair of monopole and dipole modes. The leading rotated EOFs of tropical SST anomalies in recent four decades (Houghton and Tourre 1992) have one major center of action each confined to one hemisphere but show no substantial signal in the other hemisphere. Cross spectral analyses by Mehta (1998) and Enfield et al. (1999) with longer data records present little coherence of SST anomalies between northern and southern tropics on any frequency domain. In contrast, a singular value decomposition (SVD) analysis of SST, wind stress and heat flux anomalies over 40 years by Chang et al. (1997) showed that an SST dipole structure was maintained by equatorial anti-symmetric heat flux anomalies, in association with cross-equatorial atmospheric flows. The same conclusion is reached from a joint empirical orthogonal function (EOF) analysis of SST and zonal wind anomalies (Nobre and Shukla 1996). Most previous modal decompositions are applied to both northern and southern parts of the tropical Atlantic at the same time, and then tropics-extratropics linkages are examined. Here we will further reexamine this interhemispheric relation in the SST variability from a different perspective. First, we will perform EOF analysis sepa- Fig. 1 Scatter plot of zonal mean (a) SST (ºC) and (b) SLP (hpa) anomalies in the Northern tropics ( horizontal axis) and Southern tropics (vertical axis). Zonal mean anomalies are calculated form unfiltered boreal winter in 8-20º latitudes for SST and 20-40º latitudes for SLP. 58 JAMSTECR, 41 (2000)

3 rately for the North and South Atlantic, thus avoiding the criticism that the EOF analysis over the whole tropics might force artificial interhemispheric correlation. Second, we will include the extratropics into our analysis domains, recognizing the effects of atmospheric teleconnections linking the tropics and mid-latitudes, which in turn produce the Pan-Atlantic Decadal Oscillation (PADO; Xie and Tanimoto 1998). We will repeat the same analysis on the SLP variability that presumably most directly interacts with SST anomalies to see if these interacting fields lead to a coherent interhemispheric relation. Observational and model studies indicate that surface heat flux is crucial in forcing the dipole mode of SST variability in the tropical Atlantic. Wagner (1996) examined contributions to SST dipole mode from each surface heat flux component. His analysis method inevitably emphasizes variability on interannual time scales where significant positive contribution is seen only from windinduced latent heat variations. This latent heat flux contribution has been independently confirmed in ocean general circulation model (GCM) simulation (Carton et al. 1996), but it is still unclear if other heat flux components might also contribute on longer decadal time scales. Dipole-associated wind variability is weak in amplitude and not well organized in the South Atlantic and may not be able to account for all the local SST anomalies that have comparable amplitudes with those in the northern tropics (Chang et al. 1997, Nobre and Shukla 1996, Tanimoto and Xie 1999). In these analyses, wind anomaly amplitudes north of equator are generally two or three times larger than those south of equator even though SST anomaly amplitudes are comparable on either side of equator. Thus it is interesting to examine why this asymmetric amplitudes of the atmospheric field could be associated with the symmetric amplitudes of the SST field. We will show that low-level cloud variability is one of the missing forcings for SST anomalies in the southern tropics. Now two types of gridded SST datasets are available for the statistical analyses: One intends a complete spatial coverage via an optimal method. This benefits numerical experiments with a complete boundary condition, just like GISST (Folland and Parker 1995) and GOSTA (Bottomley et al. 1990). The other gridding method fills a grid point only when there are enough number of observations in a grid for monthly seasonal averaging. Previous empirical studies (Mehta 1998, Enfield et al. 1999, Xie et al. 1999) have employed the former dataset to capture the large-scale structure of Atlantic SST anomalies. Over data sparse region like the South Atlantic, however, heavy temporal and spatial interpolations can be a source of signal distortion. Here in order to ensure the fidelity of the gridded data, we require all grid points in the analysis to have uninterrupted seasonal means (see the next section for details). In the present study, we construct such a new observational dataset of marine meteorological variables including monthly SST, SLP, cloud amount and ocean heat fluxes, calculated only from observations by ship of opportunities. We did not use satellite-based observations in calculation to avoid possible system bias due to instrument changes. No spatial interpolation is allowed. Later we will compare the results from this cutto-the-bone dataset with more heavily interpolated datasets like the GISST and GOSTA. The rest of paper is organized as follows; Section 2 describes the datasets and the analysis procedures. Section 3 shows leading modes that explain the meridional gradients of the inter-tropical SST and SLP anomaly fields. Section 4 presents the Pan-Atlantic patterns for SST, SLP, wind velocity and ocean heat fluxes. Section 5 gives a discussion of a lower cloud effect on the asymmetric amplitudes of SLP and wind velocity, associated with the tropical SST dipole. Concluding remarks are given in Section Data set and analysis procedure A monthly 4-degree latitude-longitude dataset of marine meteorological variables is constructed from quality-controlled ship and buoy observations compiled in Long Marine Reports in fixed length records (LMRF) of comprehensive ocean-atmosphere dataset (COADS; Woodruff et al. 1987) for the North and South Atlantic (70ºN- 50ºS) from 1950 through The domain contains the Norwegian, North and Caribbean Seas, southern part of Labrador Sea, Gulf of Mexico and Mediterranean. In the present study, we examine SST, SLP, vector and scalar wind speed, sensible and latent heat flux fields. A higher resolution (2 degree x 2 degree) dataset of cloudiness is also constructed from ship reports in LMRF of COADS. The cloud amount is visually measuring a cloud coverage of the whole sky. Turbulent heat fluxes are calcu- JAMSTECR, 41 (2000) 59

4 lated using Kondo s (1975) aerodynamic bulk method for each of ship-buoy measurement. In regions like the South Atlantic where there are few ship observations, the COADS suffers sampling errors, and SLP and wind velocity may not even satisfy basic dynamic constraints like the geostrophic balance. We also complement the COADS with the NCEP (National Centers for Environmental Prediction, monthly 2 degree latitude-longitude resolution) reanalysis dataset over , which provides dynamically-consistent SLP and wind velocity data even with insufficient observed input in the Southern Hemisphere to an assimilation model. In contrast, grid point values of COADS are calculated from independent observations of SLP and wind velocity. For each of these variables we calculate a monthly climatological mean annual cycle based on the entire period of record (COADS: 46 years, NCEP: 38 years), and the monthly anomalies are defined as departures from the climatological means. Seasonal averages are used in the following sections. Averages are calculated only for boreal winters (December through March) only on those grid points with more than three monthly-mean values. This study is based on the modal decomposition (EOF and SVD analyses) of SST and SLP anomalies at those grid points that contain no missing boreal winter averages for the entire period of 1950/51 through 1994/95. Large variance is found on interannual and decadal time scales (Servain 1991, Huang and Shukla 1997, Mehta 1998) with a clear spectral gap between the two time scales (Mehta and Delworth 1995, Mehta 1998). Based on the above spectral structure of tropical Atlantic SST anomalies, Tanimoto and Xie (1999) applied a time-scale separation method to 51-year SST time series and found negative (positive) correlation on decadal (interannual) time scale in SST anomalies across the equator. A band-pass filter is accepted to extract decadal variations when substantial variances are found in a frequency band. Explained variance of the leading EOFs depends on the number of the grid points for which a modal decomposition is performed. The grid points employed in the modal decomposition are 312 (103) for the North (South) Atlantic SST field, and 314 (113) for the North (South) Atlantic SLP field. Composite anomalies of SST, SLP, lower cloud amount and heat fluxes are calculated based on the time series of the leading modes to examine regional signals of the PADO. 3. Inter-hemispheric mode of decadal climate variability Large variance appears in a decadal frequency domain of cycle per year in the tropical Atlantic SST anomalies (Mehta 1998, Rajagopalan et al. 1998). Indeed, Tanimoto and Xie (1999) showed that the crossequatorial gradient index of annual mean SST anomalies varied on decadal time scales, often having an anti-symmetric dipole-like spatial pattern. Regressed SST, SLP and wind vector anomalies onto the cross-equatorial gradient index revealed a PADO pattern with the dipole in the tropics. In this section, we present collaborative evidence for the PADO based on different analysis methods. Before we perform an EOF analysis, we divide the Atlantic basin into two independent parts north and south of the equator to eliminate an artificial-correlation problem in modal decomposition of tropical SST anomalies (Houghton and Tourre 1992). Then, SST anomalies are averaged for boreal winter and filtered through the decadal band (8-16 year). The upper two curves in Figure 2 show normalized principle components (PCs) of the leading mode of SST anomalies in the North and South Atlantic, respectively. Figure 3 depicts the SST regressions onto the PCs instead of EOF eigenvectors. Grid points used in the EOF analysis are shaded in the background. These leading EOFs explain 37.0% and 54.9% of band-passed variance of the used grid point values in the Northern and Southern hemisphere, respectively. The second modes explain only 19.0% and 13.2% of the variance in the Fig. 2 Upper curves: the normalized principal components (PCs) of SST anomalies in the North (solid) and South (dashed line) Atlantic. Lower curves: PCs of SLP anomalies. The vertical axis of SST anomalies is reversed. 60 JAMSTECR, 41 (2000)

5 Fig. 3 The first SST EOFs for the North and the South Atlantic, which explain 37.0% and 54.9% of decadal band-passed (8-16 years) boreal winter SST anomalies, respectively. Regressed SST anomalies onto the PCs for the North (South) Atlantic are shown in the upper (lower) panel. Negative values are dashed. Contour interval is 0.1ºC. Fig. 4 Same as Figure 3, but for the SLP anomalies. They explain 36.4% and 50.2% of decadal band-passed (8-16 years) boreal winter SLP anomalies for the North and the South Atlantic, respectively. Contour interval is 0.2hPa Northern and Southern Hemisphere, respectively, ensuring a fair separation of leading EOFs. The leading SST patterns feature two tropical centers of action with opposing polarities across the equator with maximum regression of 0.3º C in the Northern Tropics. SST anomalies display a PADO pattern, with centers of action lining up meridionally in the North Atlantic. Significant negative regressions, with a magnitude of -0.4ºC, extend from south of Newfoundland through Gulf of Mexico and Caribbean Sea, while small positive regressions of about 0.1ºC appear between South Greenland and south-western Europe. Negative regressions cover the whole the Southern Hemisphere domain except in the southeastern corner where poor data coverage generates an apparently spurious center of action. The spatial structure of SST anomalies will be discussed in Section 4. These coherent patterns in fact fluctuate almost in phase on the decadal time scales (Fig. 2, simultaneous correlation coefficient is 0.63). Note again that these leading EOFs are derived from independent fields so that there is no a priori reason for them to be correlated. Decadal variability is pronounced after mid 1960s while the agreement of two PCs may be insignificant in the first 15 years and the last 5 years. Simultaneous correlation maps onto two PCs (not shown) present a similar pattern to regression maps. All centers of action have statistically-significant correlations above 0.8 in the Pan-Atlantic domain. The same analysis performed with the GISST dataset gives a similar PADO pattern (Xie et al. 1999). We perform the same analysis to the boreal winter SLP anomalies in the two subdomains (Figure 4). The leading EOFs of SLP fields explain 36.4% and 50.2% of bandpassed variance of used grid point values in the Northern and Southern Hemisphere, respectively. A subtropical SLP center of action appears in 20-40ºN band in the central North Atlantic, which is degrees poleward of the tropical SST center and has the opposite polarity to the tropical SST anomalies. Another extratropical center west of Europe also has significant regressions. The South Atlantic EOF does not have much spatial structure, with the SLP varying more or less uniformly over the whole subdomain. The accompanying PCs of the SLP field (the lower two curves of Figure 2), correlate to one another. More strikingly, the SST and SLP pairs of PCs are well correlated among themselves, despite the fact that they are all derived from independent samples (Table 1). Two minima in early 1970s and 1980s and three maxima in late 1960s, late 1970s and early 1990s are shown up in all JAMSTECR, 41 (2000) 61

6 Table 1 Correlation between leading PCs of SST and SLP fields in the two subdomains north and south of equator SLP N.Atl. SLP S.Atl. SST N.Atl. SST N.Atl. SLP N. Atl SLP S. Atl SST N. Atl four time series (Figure 2). These results indicate that an air-sea coupled PADO dominates the recent three decades. Scatter plots of zonal mean values in the 20-40ºN and 20-40ºS bands, calculated from bandpass filtered SLP anomalies (not shown), confirm this out-of-phase relationship, with a tilted elliptical track much like the scatters of decadal SST anomalies (see Fig. 5 in Tanimoto and Xie 1999). We also perform an SVD analysis to examine coupled modes of SST and SLP fields in the combined Atlantic. The leading SVD mode explains 60.1% of total squared covariance. The heterogeneous regression maps of SST and SLP fields (not shown) are similar to the pieced up regression fields. All decomposed modes (EOFs and SVDs) feature distinct NAO and associated SST patterns (Deser and Blackmon 1993, Kushnir 1994) in the Northern Hemisphere, but show spatially uniform patterns in the Southern Hemisphere. The agreement between four time series of PCs is fairly pronounced during three cycles of the decadal variability from 1966 through Before and after this period, however, the correlation among the PCs does not hold so well. Although this could be due to the end effect of the band-pass filter, similar EOF analyses of SST and SLP fields for a shorter period of reproduce the three cycles of a decadal oscillation, raising the more explained variance by about 15% (not shown). This result seems to suggest the nonlinear relationship between this decadal variability and lower frequency fluctuations. But a nonlinear diagnosis is out of scope in the present study. 4. Decadal climate variability in Pan-Atlantic To examine features common to three distinct cycles of the distinct PADO, we made composite maps of meteorological variables based on the PCs of leading SST and SLP modes. Compositing helps us to see the PADO signature outside areas of EOF analyses. Six years each are chosen to represent the positive phase ( , 79-81) and the negative phase ( , 84-86) of the PADO. 62 JAMSTECR, 41 (2000)

7 The positive phase corresponds to a northward SST or a southward SLP gradient in the tropics, and vice versa. We will show difference maps of climate anomalies between the two phases. Figures 5 shows the difference map of unfiltered SST and SLP anomalies between the two phases. The polarity of tropical SST and SLP anomalies is zonally uni- Fig. 6 Same as Figure 4, but for wind velocity anomalies (vectors). The wind velocity scale (5.0ms -1 ) are indicated on the bottom of panel. Dark (light) shades are the regions in which scalar wind speed anomalies are more than 0.5ms -1 (less than -0.5ms -1 ). form between 20 degree latitudes on either side of equator, while SST amplitudes increase to the east, exceeding 0.8ºC in the Northeastern Tropics. The composite SST anomalies south of the equator exceed 0.6º C, but the spatial pattern is less coherent. In the eastern boundary regions, seasonal variations are largely due to the development of the Guinea dome -the shallow domelike thermocline feature in subsurface ocean- and the Angola dome -a counterpart of Guinea dome in the Southern Hemisphere-, respectively (Yamagata and Iizuka, 1995). These domes develop due not only to an one-dimensional surface heat flux, but also to active divergence of horizontal heat transport in subsurface. Further investigations into such subsurface variations in response to anomalous wind stresses associated with cross-equatorial SST gradient are desired. In the difference map of wind vectors (Fig. 6), anomalous southwesterlies in geostrophic balance with the SLP difference reduce the climatological northeasterly trade. This leads to a reduction in scalar wind speed by up to 1.5ms -1 (light shade in Fig. 6), suppressing the latent heat flux release (negative anomalies) in the same region (Fig. 7a). The sensible heat flux (Fig. 7b) also depends on the wind speed, but there is no substantial difference between the two phases in the tropics. In the Southern Tropics, the SLP difference field suggests anomalous southeasterlies, which enhance the heat release (positive anomalies) from the ocean surface. This is consistent with the negative SST anomalies south of the equator. Fig. 7 Same as Figure 4, but for (a) latent and (b) sensible heat flux anomalies. Contour interval is 10Wm -2. JAMSTECR, 41 (2000) 63

8 Wind velocity differences between the two phases display southeasterlies in the Southern Tropics. But the amplitudes in the Southern Tropics are about one-third of those in the Northern Tropics. Note that anomalous southerlies on the equatorial grid points are stronger than those further north in 8-12ºN. These strong southerly anomalies are associated with a northward shift of the ITCZ, a manifestation of interhemispheric interactions. The heat flux anomalies show incoherent structures in the Southern Tropics, becoming even worse south of 40ºS. Few observations result in large sampling errors, especially in those higher order fields such as wind velocities, heat and momentum fluxes. The atmospheric composites (right panel in Fig. 5 and Fig. 6) in the Southern Hemisphere indicate a noisy anticyclonic SLP pattern, small speeds and disorganized directions of wind velocities. Composite maps of SLP and wind vectors calculated from the NCEP reanalysis datasets (Figure 8) are quite similar to those from COADS in an overall sense. But the anticyclone centered on 30ºS is better defined and associated wind pattern is well organized in the Southern tropics. Inter-hemispheric flows within 10 degree latitudes have comparable amplitudes both sides of the equator. Although the SLP and wind velocity difference fields now have coherent structure, their magnitudes poleward of 20ºS are still smaller than those in the Northern subtropics. In the extratropical North Atlantic, an atmospheric NAO-like SLP and SST patterns (Figs. 5 and 8) are dominant (Deser and Blackmon 1993, Kushnir 1994). Subtropical SLP anomaly pattern in the North Atlantic shows an intensification of the climatological Azores high, and is sandwiched by negative SST anomalies off east coast of United States and positive ones around Newfoundland and south of Greenland. This association indicates an ocean surface response to the atmospheric forcing by the intensified westerlies to the south of the SLP center and by weakened ones to its north. Positive SLP anomalies in the Norwegian Strait are cooperative in weakening the westerlies. The latent flux is one of the major component in forming SST anomalies over most of the extratropics, but the sensible heat flux makes comparable contribution in the southern Labrador Sea. Such a change in heat inputs into the extratropical atmosphere may in turn maintain the polarity of SLP anomalies (Peng et al. 1997, Nakamura and Yamagata 1999, Rodwell et al. 1999), but a controversial issue needs further investigation. Positive SST anomalies in 30-40ºS, 10-40ºW have a somewhat coherent structure, but the SLP and wind velocity anomalies do not. Because large sampling errors are likely involved, we will not further discuss anomalies in the extratropical South Atlantic. Fig. 8 Same as Figure 5, but for SLP and wind velocity anomalies from the NCEP reanalysis dataset. The reference wind velocity (5.0ms -1 ) are indicated on the bottom of panel. Contour interval is 0.5hPa for the SLP field. Positive (negative) values are represented by the solid (dashed) lines. 5. Discussion The previous studies of decadal variability have revealed characteristic in the SST and SLP anomaly patterns in the North Atlantic (Deser and Blackmon 1993, Kushnir 1994, Peng and Fyfe 1996, Robertson 1996, Bojariu 1997, Zorita et al. 1992, Watanabe et al. 1999) and the South Atlantic (Venegas et al. 1997). These anomaly patterns seem components of a single PADO mode in the air-sea coupled field, as is demonstrated in the Sections 3 and 4. Investigations of tropics-extratropics connections (Xie and Tanimoto 1998, Tanimoto and Xie 1999) indicated that change in cross-equatorial SST gradient, showing a distinct decadal variations, was strongly 64 JAMSTECR, 41 (2000)

9 linked with extratropical decadal variability. It remains controversial whether the meridional SST gradient variability in the tropics represents an intrinsic coupled mode (Chang et al. 1997, Xie and Tanimoto 1998, Tanimoto and Xie 1999) or arises from fortuitous superposition of two independent modes of SST variations confined on either side of the equator (Houghton and Tourre 1992, Mehta 1998). The results of our EOF analyses conducted separately in two subdomains north and south of the equator in Section 3 support the existence of an air-sea coupled inter-hemispheric interaction, as is indicated by the mutual temporal correlation among the four PCs of SST and SLP fields and by a tropical dipole configuration in their EOF spatial patterns. Recent theoretical analysis with a generalized coupled model (Xie et al. 1999) that includes both Bjerknes and wind-evaporation-sst feedbacks provides some physical basis for our empirical modal decompositions. In an ocean of Atlantic zonal size, arbitrary initial disturbances in the model disperse into two sets of modes: equatorially symmetric and anti-symmetric, respectively. Furthermore, the dispersion relations of these coupled models also are such that their frequencies are well separated: the anti-symmetric mode prefers decadal and longer time scales, while the symmetric mode exhibits higher interannual frequencies. Simple/intermedium coupled models such as Xie et al. s (1999) produce a dipole mode of similar amplitudes of the equator. While the observed SST pattern indicates such a rough symmetry in amplitude, the SLP and wind speed anomalies in the Southern Hemisphere are only onethird of those in the Northern Hemisphere. While the SLP and wind anomalies can be understood as the baroclinic response to the SST dipole, the anomalous cyclonic circulation centered on 35 o N is largely a surface signature of a deep barotropic response. It appears at 500hPa with its center shifted slightly westward (Fig. 9). It will be interesting to include this barotropic response in the atmospheric component of simple/intermedium coupled models and see how coupled modes, particularly the dipole, will change. We note that this hemispheric difference in atmospheric response can be important for understanding upper ocean variability that is largely winddriven. We can expect larger subsurface decadal variability in the North than the South Atlantic as ocean GCM simulations seem to suggest (Huang and Shukla 1997). Fig. 9 Same as Figure 5, but for 1000hPa (thin contours) and 500hPa (thick contours) anomalies from the NCEP reanalysis dataset. Here we consider two factors which might cause asymmetric amplitudes of the dipole-associated atmospheric anomalies in the Atlantic. First, there are few observations outside of merchant ship routes in the Southern Hemisphere. Tropical SST anomalies usually have a persistence of more than one season. The persistence of atmospheric anomalies, by contrast, is much shorter than that of SST anomalies. Few observation gives rise to larger sampling errors in the atmospheric fields than in the ocean, and in turn might mask climate signals. Similarities between the results from NCEP and COADS datasets, however, suggest that sampling errors are not a critical problem in the tropics. Second, the cloud shielding of solar radiation may contribute to maintaining the tropical dipole mode of SST anomalies. In boreal spring when the SST field in the equatorial Atlantic is nearly uniform in both the zonal and meridional directions, the ITCZ is sensitive to the changes in an interhemispheric SST gradient associated with the dipole (Nobre and Shukla 1996). This meridional shift of the ITCZ causes the decadal variability in northeastern Brazil rainfall (Servain 1991, Mehta 1998, among others). Composites based on a 2 degree latitudelongitude higher-resolution dataset from the COADS capture this shift in ITCZ's latitude. Associated with an anomalous northward SST gradient is an increase (decrease) in cloudiness at 5ºN (5ºS; Fig. 10). These near- JAMSTECR, 41 (2000) 65

10 Fig.10 (a) Boreal spring (Feb.-Apr.) composite of cloud cover anomalies based on a higher resolution (2x2) COADS (left panel; heavy shading < -3.0% & light shading > 3.0%). (b) Zonal mean anomalies of SST (upper right) and cloud cover (%; solid) along with surface wind divergence (10-6 s -1 in dotted line; lower right panel). equatorial changes in cloudiness indeed seem associated with convective activity as indicated by the convergence (divergence) of anomalous winds at the surface (dotted line in Fig. 10b). These near-equatorial changes in high clouds act as a negative feedback to diminish the crossequatorial SST gradient. Additional cloudiness anomalies are found in off-equatorial tropics poleward of 10 degree latitudes, which have not been noted previously to our knowledge. These offequatorial cloudiness anomalies are negatively correlated with local SST anomalies, but apparently not associated with significant changes in surface wind convergence. Thus they most likely correspond to changes in low-level stratiform clouds. These low-level cloud anomalies are spatially organized (Fig. 10). The cloud anomaly pattern in the southern tropics is particularly robust, seen in all the seasons. In annual mean, the southern cloudiness anomalies are twice as large as the northern ones (not shown). Two mechanisms are possible for causing changes in low-level clouds. First is a top-down mechanism: the shifts in the ITCZ change the subsidence in off-equatorial tropics, affecting the height of and stratification at the top of the planetary boundary layer (PBL). The other is bottom-up: negative (positive) SST anomalies increase (decrease) the stratification across the top of the PBL provided temperatures in the free atmosphere do not change. Enhanced (weakened) capping of the PBL leads to an increase (decrease) in stratiform cloud cover trapped near the top of the PBL. Increased (reduced) cloud cover will in turn cause a further cooling (warming) in SST through insolation change, completing a positive feedback loop between local SST and stratus clouds. In a coupled ocean-atmosphere model, the temporal spectrum of the dipole mode response to a white-noise forcing is sensitive to thermal damping rate in the SST equation (Xie and Tanimoto 1998, Xie 1999). The SST dependence of surface evaporation provides a major mechanism for thermal damping, which can be linearized as a Newtonian cooling term with a damping rate of (1 year) -1 (Xie 1999). The rate of SST change due to cloud shielding is -0.62(1-A)S0C'/(cpρH), where Reed's (1977) formula for shortwave radiation has been used, S0=300 Wm -2 is the solar radiation in clear sky, A=0.96 the albebo of sea surface, C the perturbation cloud amount, cp and ρ are the specific heat at constant pressure and density of sea water, and H=50m is the depth of the mixed layer. Assuming that low-level clouds in offequatorial tropics vary with local SSTs, we have C'=-αT with α =0.1 K -1 from the right panels of Fig. 10. This leads to an SST-stratus feedback coefficient, b = 0.62α 66 JAMSTECR, 41 (2000)

11 (1-A)S0/(cpρH)=(3.5 years) -1. Thus the local SST-stratus feedback can reduce the Newtonian cooling rate for SST by as much as 30%. 6. Concluding Remarks We have examined dominant modes of the decadal variability in the Pan-Atlantic basin, using the new observational datasets of marine meteorological variables calculated from LMRF of COADS. An EOF analysis was performed for decadal SST anomalies in the North Atlantic and in the South Atlantic separately, avoiding the artificial correlation between the northern and southern tropics. The leading EOFs featured a dipole structure in the tropics across the equator and were associated with substantial extratropical signals. The SLP EOF featured a similar dipole albeit with subtropical centers of action on either side of the equator. Time series of these leading modes correlated well with one another and presented three cycles of distinct decadal oscillations during , indicative of the existence of an inter-hemispheric dipole mode that involved ocean-atmosphere interaction. Composite maps, based on the positive and negative phases of the tropical dipole mode, clearly showed that the latent heat flux induced by anomalous wind vector anomalies played a major role in coupling the atmosphere and ocean. The composite SST anomalies had comparable amplitudes on either side of the equator. However, the magnitude of the SLP and wind velocity anomalies to north of equator were three times larger than those to the south. Furthermore an NCEP reanalysis dataset showed a deep barotropic atmospheric pattern in the Northern Hemisphere associated with the NAO pattern, but not in the Southern Hemisphere at all. Whereas this asymmetric amplitude structure might be a consequence of large sampling errors due to insufficient observations in the Southern Hemisphere, low-level clouds might play a role in keeping SST anomalies having comparable amplitudes across the equator. The response of cloud fields to the SST dipole differed near and off the equator. Within 10 degrees latitude, it involved north-southward shift in the ITCZ and changes in deep convective clouds, acting to dampen the cross-equatorial SST gradient. Outside this equatorial zone, low-level clouds responded to and positively fed back onto the local SST, reducing the thermal damping rate by 30%. Acknowledgment The authors are grateful to Prof. T. Yamagata, Dr. H. Nakamura, Dr. Iwasaka and Prof. Matsuno for stimulating discussion. We also thank the NOAA/NCEP for providing the reanalysis data and the NCAR for providing the COADS/LMRF dataset. This work was partly supported by Frontier Research System for Global change. References 1) Bojariu R (1997) Climate variability modes due to ocean-atmosphere interaction in the central Atlantic. Tellus 49A: ) Bottomley M., Folland CK, Hsiung J, Newell RE, Parker DE (1990) Global Ocean Surface Temperature Atlas (GOSTA). Her Majesty s Stationery Office. 3) Carton JA, Cao X, Giese BS, da Silva AM (1996) Decadal and interannual SST variability in the tropical Atlantic Ocean. J Phys Oceanogr 26: ) Chang P., Ji L, Li H (1997) A decadal climate variation in the tropical Atlantic ocean from thermodynamic air-sea interaction. Nature 385: ) Delworth TL (1993) North Atlantic interannual variability in a coupled ocean-atmosphere model. J Clim 9: ) Deser C, Blackmon ML (1993) Surface climate variations over the North Atlantic Ocean during winter: J Clim 6: ) Enfield DB, Mestas-Nunez AM, Mayer DA, Cid- Serrano L (1999) How ubquitous is the dipole relationship in the tropical Atlantic SST. J Geophys Res: submitted. 8) Folland CK, Parker DE (1995) Correction of instrumental biases in historical sea surface temperature data. Q J Roy Meteor Soc 121: ) Halliwell GR, Mayer DA (1996) Frequency response properties of forced climatic SST anomaly variability in the North Atlantic. J Clim 9: ) Hasternath S, Heller L (1977) Dynamics of climatic hazards in northeast Brazil. Q J Roy Meteor. Soc 110: ) Houghton RW, Tourre YM (1992) Characteristics of low-frequency sea surface temperature fluctuations in the tropical Atlantic. J Clim 5: ) Huang B, Shukla J (1997) Characteristics of the interannual and decadal variability in a general cir- JAMSTECR, 41 (2000) 67

12 culation model of the tropical Atlantic ocean. J Phys Oceanogr 27: ) Hurrell JW (1995) Decadal trends in the North Atlantic Oscillation: Regional temperature and precipitation. Science 269: ) Kondo J (1975) Air-sea bulk transfer coefficient in diabatic conditions Bound -Layer Meteor 9: ) Kushnir Y (1994) Interdecadal variation in North Atlantic sea surface temperature and associated atmospheric circulation. J Clim 7: ) Mehta VM (1998) Variability of the tropical ocean surface temperatures at decadal-multidecadal timescales. Part I: the Atlantic ocean. J Clim 11: ) Mehta VM, Delworth T (1995) Decadal variability of the tropical Atlantic ocean surface temperature in shipboard measurements and in a global ocean-atmosphere model. J Clim 8: ) Nakamura H, Yamagata T (1999) Recent decadal SST variability in the northwestern Pacific and associated atmospheric anomalies. Navarra A, ed. Beyond El Niño: Decadal Climate Variability Springer-Verlag: in press. 19) Nobre P, Shukla J (1996) Variations of sea surface temperature, wind stress, and rainfall over the tropical Atlantic and South America. J Clim 9: ) Peng S, Robinson WA, Hoerling MP (1997) The modeled atmospheric response to mid-latitude SST anomalies and its dependence on background circulation states. J Clim 10: ) Peng S, Fyfe J (1996) The coupled patterns between sea level pressure and sea surface temperature in the midlatitude North Atlantic. J Clim 9; ) Rajagopalan B, Kushnir Y, Tourre YM (1998) Observed decadal midlatitude and tropical Atlantic climate variability. Geophsy Res Lett 25: ) Reed RK (1977) On estimating insolation over the ocean. J Phys Oceanogr 7: ) Robertson AW (1996) Interdecadal variability over the North Pacific in a multi-century climate simulation. Clim Dyn 12: ) Rodwell MJ, Rowell DP, Folland CK (1999) Oceanic forcing of the wintertime North Atlantic oscillation and European climate. Nature 398: ) Servain J, (1991) Simple climatic indices for the tropical Atlantic Ocean and some applications. J Geophys Res 96: ) Tanimoto Y, Xie SP (1999) Ocean-Atmosphere Variability over the Pan-Atlantic basin. J Meteor Soc Japan 77: ) Venegas SA, Mysak LA, Straub DN (1997) Atmosphere-ocean coupled variability in the South Atlantic. J Clim 10: ) Wagner RG (1996) Mechanisms controlling variability of the interhemispheric sea surface temperature gradient in the tropical Atlantic. J Clim 9: ) Watanabe M, Kimoto M (1999) Tropical-extratropical connection in the Atlantic atmosphere-ocean variability. Geophys Res Lett.: submitted. 31) Watanabe M, Kimoto M, Nitta T, Kachi M (1999) A comparison of decadal climate oscillations in the North Atlantic detected in observations and a coupled GCM. J Clim: in press. 32) Woodruff SD, Slutz RJ, Jenne RL, Steurer PM (1987) A comprehensive ocean-atmosphere dataset. Bull Amer Meteor Soc 68: ) Xie SP (1999) A dynamic ocean-atmosphere model of the tropical Atlantic decadal variability. J Clim 12: ) Xie SP, Tanimoto Y (1998) A Pan-Atlantic decadal climate oscillation. Geophys Res Lett, 25: ) Xie SP, Tanimoto Y, Noguchi H, Matsuno T (1999) How and why climate variability differs between the tropical Atlantic and Pacific. Geophys Res Lett: in press. 36) Yamagata T, Iizuka S (1995) Simulation of the tropical thermal domes in the Atlantic: A seasonal cycle. J Phys Oceanogr 25: ) Zorita E, Kharin V, von Stroch H (1992) The atmospheric circulation and sea surface temperature in the North Atlantic area in winter: their interaction and relevance for Iberian precipitation. J Clim 5: (Manuscript received 15 December 1999) 68 JAMSTECR, 41 (2000)