Decadal climate variability in the eastern Caribbean



[1] Rainfall variability in the eastern Caribbean during the 20th century is analyzed using principal component analysis and singular value decomposition. In contrast to earlier studies that used seasonal data, here we employ continuous signal processing. The leading mode is a decadal oscillation related to third and fourth modes of sea level pressure (SLP) and sea surface temperatures (SST) which together identify three zones of action in the Atlantic: 35°N–20°N, 20°N–5°N, and 5°N–20°S. The ability of the ECHAM4.5 model to simulate this signal is investigated. Its decadal variability is also represented through lower-order SLP and SST modes that comprise an Atlantic tripole pattern with lower pressure east of the Caribbean. Composite analysis of high and low phases of the decadal mode reflects a cool east Pacific and a more active Atlantic Intertropical Convergence Zone during boreal summer, conditions that favor the intensification of African easterly waves. The decadal signal has strengthened since 1970, yet the three centers of action in Atlantic SST are relatively unsynchronized.

1. Introduction

[2] Observational studies have revealed decadal variability in the Atlantic climate [Deser and Blackmon, 1993; Kushnir, 1994; Czaja and Marshall, 2001] associated with a SST tripole structure comprising zonal bands of action extending from northern midlatitudes across the tropics. This pattern has been related to air-sea flux patterns of the North Atlantic Oscillation (NAO) [Cayan, 1992; Visbeck et al., 2003]; standing atmospheric Rossby waves [Frankignoul et al., 1997] and changes in oceanic advection [Saravanan and McWilliams, 1998; Krahmann et al., 2001]. These features are replicated in coupled models [Wu and Liu, 2005] and rely on adjustment of the ocean gyres to anomalous wind stress forcing, and subsequent response of atmospheric convection to the sea surface temperature (SST) anomalies produced [Eden and Greatbatch, 2003].

[3] The Atlantic Intertropical Convergence Zone (ITCZ) is located between the tropical centers of action and links the African and South American monsoons [Sultan and Janicot, 2003]. It spawns hurricanes that pass through the Caribbean and is affected by local SST and remote atmospheric circulations such as the Pacific El Niño–Southern Oscillation (ENSO), through changes in wind shear and tropical easterly waves [Magaña et al., 1999; Amador et al., 2000; Wang and Enfield, 2001, 2003; Wang et al., 2006; Wang and Lee, 2007]. The ENSO alternates with a period of 3–7 years [Kirtman, 1997] while the NAO imparts slower ocean-atmosphere oscillations [Barnston and Livezey, 1987; Jones et al., 1997; Chelliah and Bell, 2004; Charlery et al., 2006; Danabasoglu, 2008], and both modulate west Atlantic rainfall [Malmgren et al., 1998; Giannini et al., 2001; Chen and Taylor, 2002; Charlery et al., 2006; Jury et al., 2007]. Climatic extremes in the Caribbean Sea coincide with antiphase SST anomalies between the tropical Atlantic and Pacific [Hastenrath, 1978, 1984; Enfield, 1996; Enfield and Mayer, 1997; Enfield and Alfaro, 1999] and a convective dipole between Africa and South America associated with a zonal overturning Walker circulation [Cook et al., 2004; Yeshanew and Jury, 2007; Mestas-Nuñez et al., 2007]. Fluctuations in the meridional Hadley circulation have been found to contribute to decadal climate variability in West Africa and the Caribbean [Horel et al., 1989; Liebmann and Marengo, 2001; Nogués-Paegle et al., 2002; Higgins and Shi, 2001; Vera et al., 2006; Jury, 2009a], a feature that may be linked with the SST tripole pattern.

[4] Little work has been devoted exclusively to the eastern Caribbean Antilles island chain, though they are vulnerable to Atlantic hurricanes spawned from African easterly waves. While the western Caribbean has a bimodal rainfall distribution with spring rains modulated by ENSO phase [Jury et al., 2007], here we focus on the eastern Caribbean climate regime and evidence of its decadal variation through analysis of observed and simulated data. The following scientific questions are addressed: (1) What factors underlie the decadal cycle and where is its amplitude greatest? (2) How does the frequency of rainfall variability evolve in the 20th century? (3) How well is the decadal cycle represented in a coupled GCM simulation? In section 2 the data and methods are outlined. Section 3 provides results by analysis of interannual-filtered rainfall, SST and SLP modes, comparison with a model simulation, evaluation of interrelationships, a composite analysis of high- and low-phase conditions, and a statistical analysis in discrete frequency bands. A concluding discussion is provided in section 4.

2. Data and Methods

[5] Monthly gridded sea surface temperature (SST) and sea level pressure (SLP) fields are drawn from the International Research Inst. (IRI) Climate Library website,, in the period January 1901 to December 2002 in the latitudes 20°S and 35°N and longitudes 100°W to 20°E, to cover influences of Atlantic/Pacific Oceans and African/American monsoons on Caribbean climate [Wang et al., 2006]. The grid resolution is 2° for the NOAA extended SST [Smith et al., 2008] and Kaplan optimally interpolated SLP [Kaplan et al., 1998] from ships data. Monthly gridded rainfall is analyzed across the eastern Caribbean (10°N–20°N, 68°W–57°W) at a resolution of 0.5° from gauge interpolated fields by the University of East Anglia Climate Research Unit [Mitchell and Jones, 2005]. There are no gaps in these data sets, but observations increase in density with time.

[6] The consecutive monthly fields were subjected to principal component (PC) analysis to explore the climate signals. Our work here is in contrast to earlier studies which isolated specific seasons and did not regard continuous signals. The PC analysis involves an eigenvector decomposition of the covariance matrix within a single input field that generates a spatial loading pattern and fluctuating time score for each mode. Prior to the application of PC analysis, all data were standardized, linearly detrended, the annual cycle was removed (anomaly), and an 18 month running mean was imposed to focus on interannual oscillations. These methods are often used to cluster and deduce the climate signals [Plaut and Vautard, 1994; Cheng et al., 1995]. With evenly spaced, standardized data, we minimize the chance for artificial Buell patterns [Mason, 2010]. Our filtering reduces the degrees of freedom to ∼66 in the 20th century observations. Similarly, PC modes for rainfall, surface temperature and SLP fields were calculated from a hindcast 20-member ensemble simulation of the ECHAM4.5 coupled model at 2.8° resolution. The model outputs differ from observed in that the data start in 1950 and surface temperature (Ts) is used instead of SST. Although many coupled numerical models could be considered, the ECHAM4.5 has been run in forecast mode (in 2010) by the IRI using NOAA SST fields as boundary conditions. It is based on the ECMWF atmospheric model and a parameterization package developed at Max Planck Institute for Meteorology which enables operational coupled climate simulations. The model physics are described by Roeckner et al. [1996], it is a spectral transform type with 19 atmospheric layers. The static vegetation cover is based on work by Wilson and Henderson-Sellers [1985], and the dynamic ocean model is updated from work by Oberhuber [1993].

[7] Covarying oscillations in the filtered PC time scores were analyzed using wavelet spectrum analysis in the Web site based on the methods of Torrence and Compo [1998]. The wavelet spectrum analysis helps to establish the statistical significance of covariance between eastern Caribbean rainfall and hemispheric SST and SLP. The degree of association between the leading PC time scores was also studied using cross correlation and multivariate regression. Statistical significance was evaluated using the Pearson product-moment test that depends on the degrees of freedom (df). For model intercomparisons since 1950 with df ∼ 33, r > ∣0.33∣ is significant at 95% confidence; while for observations since 1901 with df ∼ 66, r > ∣0.25∣ is significant at 95% confidence. Multivariate regression is used to combine the individual PC modes into a tripole pattern; it was optimized by backward stepwise elimination of predictors with lower influence or colinearity.

[8] To understand some of the processes underlying decadal oscillations in the reanalysis era, the five July–October seasons with high (1966, 70, 79, 81, 96) and low (1973, 74, 83, 87, 94) PC time scores for observed rain and associated SST were selected and composite difference maps were calculated. The July–October season is chosen because it is the seasonal peak for rainfall in the eastern Antilles. The composite difference maps include: CDC gauge-satellite merged (CMAP) rainfall, NOAA extended SST, NCEP reanalysis [Kalnay et al., 1996] winds, vertical motion and velocity potential, CPC soil moisture, ECMWF marine wind stress, and upper ocean salinity and currents from SODA reanalysis [Carton and Giese, 2008]. To study differences in the vertical structure in opposing phases of the decadal cycle, composite height sections in specific latitudes and longitudes were calculated for vertical motion, wind and geopotential height. In addition, the National Hurricane Center (NHC) storm database was queried and named storm tracks were plotted for the top five high- and low-phase seasons.

[9] To test the statistical significance of decadal covariance between observed Caribbean rainfall and hemispheric SST fields, a singular value decomposition (SVD) analysis was made in discrete frequency bands: 7–9 years, 9–11 years, and 11–13 years by wavelet filtering as in work by Tourre et al. [1999] and evaluation of covariance as in work by Tootle et al. [2008]. Maps of the heterogeneous correlation were calculated by considering the opposing covariance matrix and matched time score as in work by Enfield and Alfaro [1999]. Our wavelet filter artificially segregates the decadal climate variability, and also reduces the df to ∼9, so r > ∣0.60∣ is significant at 95% confidence using the Pearson product-moment test. An advantage of this approach is that small shifts in frequency can be tracked through the 20th century, and the temporal amplitude and spatial associations can be analyzed. Based on the ranked combined SVD time scores, we construct composite maps of NCEP precipitable water and ECMWF marine wind stress per frequency band, to study the corresponding features. As in the earlier PC analysis, the composites are based on the top and bottom five years in each band, and maps represent high- minus low-phase differences. For 7–9 years the composite is: high (1971, 79, 80, 89, 99) and low (1975, 76, 83, 84, 92), for 9–11 years the composite is: high (1960, 69, 70, 79, 88) and low (1955, 65, 74, 75, 84), and for 11–13 years the composite is: high (1954, 55, 56, 66, 67) and low (1948, 49, 60, 61, 73). In the composites, areas of consensus (dispersion) achieve higher (lower) amplitude and thus guide our interpretation.

3. Results

3.1. Observed and Simulated PC Modes

[10] The leading mode of eastern Caribbean rainfall (rain1) which explains 54% of total interannual variance exhibits highest loading across the southeastern Antilles islands (Figure 1a). The second rainfall mode (not shown) exhibits a north-south dipole explaining 18% of interannual variance. The leading rain mode is significantly associated with the third and fourth modes of SST and the third mode of SLP (Table 1). The spatial loading patterns for these modes (Figures 1b, 1c, and 1d) represent a broad tropical trade wind zone, and opposing centers of action in the subtropical anticyclones near Bermuda (north) and St. Helena (south) as identified in earlier research [Deser and Timlin, 1997; Chang et al., 1997; Tourre et al., 1999; Ruiz-Barradas et al., 2000]. In high phase, the corresponding Atlantic SST anomalies are cool 35°N–20°N/warm 20°N–5°N/cool 5°N–20°S, while southeast Pacific SST are cool. The SLP anomalies in high phase are low in North Atlantic latitudes 10°N–35°N and high over both southeast Pacific and South Atlantic sectors.

Figure 1.

(a) Spatial loading pattern for SE Caribbean rain mode 1 from interannual filtered gauge data. Spatial loading patterns for NOAA (b) SST mode 3 and (c) SST mode 4. (d) Spatial loading pattern for Kaplan marine SLP mode 3, with color bar from blue –3 σ to red +3 σ that is valid for all maps. Time scores for the SST modes combined as in Table 2a, and (e) rain mode 1 and (f) its wavelet cospectrum shaded at 25% power intervals with the darkest shading corresponding to 95% significance.

Table 1. Pair-Wise Correlation Values for Model-Simulated (m) and Observed PC Modes in the Period 1950–2002, With Values <95% Significance Omitted Based on the Pearson Product-Moment Test and Degrees of Freedom
SLP1m 0.510.81          
SLP2m 0.43           
SLP3m−0.34  −0.81         
SLP4m 0.33           
SST1 0.98   0.450.44      
SST2  0.88  0.84       
SST30.32  0.81   −0.78     
SST4−0.49   0.89        
SLP1 0.380.56  0.70   0.330.71  
SLP2 −0.49−0.60  −0.69   −0.46−0.54  
SLP3−0.31−0.50 −0.530.33  0.74 −0.48 −0.63 
SLP4   0.41       0.50 

[11] Pair-wise correlations between observed PC time scores for rain1 and SST3, SST4 and SLP3 are +0.40, −0.32, −0.31, respectively (Table 1), all significant at 95% confidence. These lower-order modes, which individually account for about 13, 6 and 9% of variance, are distinct from ENSO signals represented by the leading SST and SLP modes. Time scores are plotted in Figure 1e for rain1 and (+.45*SST3 −0.52*SST4) from multivariate regression: Table 2a. The correlation between the two is statistically significant +0.51 and the time scores remain in phase through most of the 20th century. The combined SST modes do not lead rainfall: the correlation at −3 to −9 months is actually lower (+0.43). The wavelet spectral covariance (Figure 1f) reveals some energy near 5 years that evolves to lower frequency from 1930 to 1970. The main spectral energy is around 10 years, evolving to 12–14 years by 1950 and returning to 8–10 years near the end of the record. Statistically significant covariance is therefore indicated in the decadal frequency band.

Table 2a. Multivariate Regression Statistics for Observed PC Modes Fitting Observed Eastern Caribbean Rain Mode 1 in the Period 1901–2002a
  • a

    Regression statistics: multiple R is 0.51; standard error is 0.20. Abbreviations: df, degrees of freedom; SS, sum of squares; MS, mean sum of squares or RMS; F, F statistic.

 CoefficientStandard Errort StatisticP Value

[12] The leading mode of ECHAM4.5 model-simulated eastern Caribbean rainfall explaining 74% of model variance is uniformly distributed across the Antilles islands (Figure 2a), but only achieves a 35% correlation with observed rainfall. The second rain mode (not shown) represents a north-south dipole with 16% of variance, similar to observed. The leading modes of model-simulated temperature (Ts) and SLP (presenting east-west dipole), like the observed modes, are unrelated to the leading Caribbean rain mode (Table 1). It is the fourth mode of Ts and the third mode of SLP that are significant, although individually they account for 6% of variance. The spatial loading for these modes (Figures 2b, 2c, and 2d) together constitute three centers of action in the Atlantic. The fourth mode of simulated SLP has an influence in multivariate regression (Table 2b) and its positive loading extends from the Sahara in the latitudes 35°N–20°N, suggesting a waveguide that may intensify tropical convection. Cross correlations between observed rain1 and model-simulated Ts4m and SLP3m are −0.39 and −0.34, respectively (Table 1), both significant at 95% confidence.

Figure 2.

(a) Spatial loading patterns for ECHAM simulated rain mode 1, (b) surface temperature mode 4, and SLP (c) mode 3 and (d) mode 4. Time scores for the ECHAM modes combined as in Table 2b, and (e) observed rain mode 1 and (f) its wavelet cospectrum shaded at 25% power intervals with the darkest shading corresponding to 95% significance.

Table 2b. Multivariate Regression Statistics for Model PC Modes Fitting Observed Eastern Caribbean Rain Mode 1 in the Period 1950–2002a
  • a

    Regression statistics: multiple R is 0.51; standard error is 0.22. Abbreviations: df, degrees of freedom; SS, sum of squares; MS, mean sum of squares or RMS; F, F statistic.

 CoefficientStandard Errort StatisticP Value

[13] Time scores are intercompared in Figure 2e for observed rain1 and model PC modes (−0.52*Ts4m −0.56*SLP3m +.67*SLP4m) from multivariate regression: Table 2b. The correlation between the two is +0.51, and statistically significant at 95% confidence. The wavelet spectral covariance exhibits significant energy around 10 years that remains stable across the period 1960–1995 (Figure 2f). Thus the model simulates the decadal oscillation quite well. The ECHAM simulation captures modes of observed temperature variability with Ts - SST correlations > +0.80 as seen in Table 1. For SLP, modes 1 and 3 are well replicated by the model. An overarching interpretation is that the leading environmental modes (representing the east-west dipole linked with ENSO) exert a secondary influence on eastern Caribbean rainfall variability. It is the third and fourth SST and SLP modes (together representing an Atlantic tripole) that provide the primary decadal modulation. This interpretation is tested by SVD analysis in section 3.3.

3.2. Composite Analysis

[14] Analyzing composites for high- minus low-phase seasons based on the PC time scores in Figure 1e, a zonal band of enhanced rainfall describes an active Atlantic ITCZ in both early and late summer (Figures 3a and 3b). The composite difference rainfall map also exhibits reduced convection in the eastern Pacific ITCZ. Rainfall differences in both ITCZ reach 3 mm/d, a sizable amplitude relative to climatology (∼10 mm/d). The composite high- minus low-phase soil moisture (Figure 3c) indicates positive differences over the Guinea coastal plains of Africa, Venezuela and central America, and negative differences over the southern Amazon.

Figure 3.

(a, b) Composite high- minus low-phase CMAP rainfall; (c) soil moisture and ECMWF wind stress; (d) 150 mbar wind and velocity potential; (e) SST; (f) 1–100 m depth-averaged salinity; all are for July–October season, except Figure 3a, which is for April–July season. Differences are based on high years (1966, 70, 79, 81, 96) and low years (1973, 74, 83, 87, 94), except that years before satellite era are omitted in Figures 3a and 3b. Largest vector in Figure 3c is 0.03 N/m2 and in Figure 3d is 4 m/s. Dashed lines in Figure 3f identify height sections in Figure 4.

[15] The ECMWF wind stress pattern overlain in Figure 3c is distinctive. Northerly wind differences penetrate from the Gulf Stream into the Caribbean, then turn cyclonically into westerly wind differences that extend from Venezuela toward West Africa. Divergence is evident around the east Pacific cold tongue. In the upper level, 150 mbar wind differences (Figure 3d) reflect enhanced easterly flow over the central Atlantic and southern Caribbean, a wind shear favoring the intensification of easterly waves. In conjunction, the 150 mbar velocity potential differences reveal the intercontinental dipole, with upper level convergent and divergence centers of action over the Amazon and Sahel, respectively (Figure 3d). The zonal circulation dipole is further studied below.

[16] The composite SST difference field (Figure 3e) reveals a warm zone in the Atlantic (10°N–20°N) linking weakened upwelling off West Africa (Morocco) and Venezuela, but also emphasizes cool La Niña conditions in the eastern Pacific. Upper ocean salinity differences (Figure 3f) exhibit freshening east of the Caribbean that is ascribed to three factors: reduced evaporation, enhanced marine rainfall, and outflow from the Orinoco River. In the east Pacific, the upper ocean is saltier in high-phase years in relation to much reduced marine rainfall there. Ocean current differences between high and low phase of the decadal oscillation (not shown) indicate that equatorial currents are westward in the Atlantic and eastward in the Pacific. This is consistent with earlier work on ocean advection [Foltz et al., 2003], the thermocline [Bourlès et al., 2002] and air-sea interactions [Hastenrath and Merle, 1987], that modulate the Atlantic climate [Chang et al., 1997; Tourre et al., 1999].

[17] Composite atmospheric circulation differences are analyzed as north-south height sections averaged in the 30°W–50°W longitudes in Figures 4a and 4b. An anomalous southern Hadley overturning is seen with rising motion 5°N–25°N and sinking motion 5°S–10°S. Zonal wind differences reveal alternating bands of upper westerly (20°S), easterly (10°N), and westerly (25°N) winds, representing accelerated jets. Low-level wind differences across the latitudes 0°N–10°N are westerly and generate cyclonic vorticity for African easterly waves.

Figure 4.

(a) Composite high- minus low-phase north-south height sections averaged 30°W–50°W of vertical motion and meridional wind and (b) zonal wind (m/s). (c) Composite east-west height sections averaged 10°N–20°N for zonal wind and vertical motion and (d) geopotential height. All are for annual periods except Figure 4d, which is for July–October season. Vector keys are given in Figures 4a and 4c (m/s), with vertical motion × 10. Differences are based on high years (1966, 70, 79, 81, 96) and low years (1973, 74, 83, 87, 94).

[18] East-west height sections of composite wind differences averaged 10°N–20°N exhibit a broad Walker overturning circulation (Figure 4c). Low-level westerly winds are overlain by upper easterly winds, with enhanced ascending motion at 80°W, 50°W and east of 10°W. Analysis of geopotential height differences (Figure 4d) reveals the largest response over West Africa, where −6 m lower heights (10°W–15°E, below 400 mbar) correspond with +1 g/kg increase of specific humidity below 850 mbar. Upper level high-pressure differences are indicative of midlevel convective heating extending downstream over the tropical Atlantic. The composite analysis highlights how Hadley-Walker circulations and ocean-atmosphere interactions promote decadal climate variability in the Caribbean zone.

[19] The tracks of tropical cyclones passing through the eastern Caribbean are plotted in Figures 5a and 5b for the top five high- and low-phase seasons. There is a clear difference, with numerous systems in high phase tracking westward along 15°S. In low phase, tropical cyclones are infrequent and recurving tracks scatter across the latitudes. Considering the number passing 50°W south of 20°N, the high phase has 11 cases (4 hurricanes), while the low phase has 4 cases (1 hurricane). The key feature in high phase is a plethora of tropical cyclones tracking westward across the eastern Caribbean that suggest a ‘waveguide’ associated with the SLP modes (see Figures 1d and 2d).

Figure 5.

Tracks of NHC named tropical systems in July–October for (a) high phase and (b) low phase of the decadal oscillation (image covers 5°N–35°N, 80°W–25°W). Colors refer to intensity: green, depression; yellow, cyclone; red, hurricane.

3.3. SVD Analysis in Discrete Bands

[20] To test whether the decadal tripole pattern is a statistically significant source of climate forcing, a SVD analysis is made after wavelet filtering the observed rain and SST data in discrete 2 year bands in the range 7 to 13 years. In contrast to the earlier PC analysis, which was ‘free’ to determine climate signals across a wide range of frequencies, our SVD analysis is mathematically ‘confined.’ The SVD temporal amplitude evolves as illustrated in Figure 6a: 7–9 year oscillations are strong at the beginning and end of the 20th century, and out of phase in midcentury. 9–11 year fluctuations are limited in the first 50 years, and strong thereafter. The 11–13 year variability gradually declines through the record; yet all three frequency bands have similar amplitude, indicating an equal influence on eastern Caribbean climate.

Figure 6.

(a) Temporal amplitude for SST (solid line) and rainfall, and correlation fields for (b) SST and (c) rainfall based on SVD analysis in each decadal frequency band (rows). Correlation values > ∣0.60∣ significant at 95% confidence.

[21] The SVD-analyzed SST correlation maps (Figure 6b) reflect three centers of action, but significant positive values are found across all decadal frequencies only in the Atlantic trade wind zone. The northern center of action near Bermuda exhibits a significant negative value in 7–9 and 11–13 year bands and shifts east to west. The southern center of action near St. Helena has a significant negative correlation only in the 11–13 year band. Of interest is the east Pacific correlation: in phase (+) with the Atlantic trade wind zone at 9–11 years, weak at 7–9 years and strongly antiphase (−) only at 11–13 years. A different picture would emerge if 3–7 year ENSO frequencies and western Caribbean rainfall were considered.

[22] The SVD-analyzed rain correlation maps (Figure 6c) are positive in the eastern islands at 7–9 years but more strongly so at 11–13 years. Rainfall-SST covariance is statistically significant in the northern islands at 9–11 years as found earlier [Jury, 2009b]. Patchy negative correlations in the rainfall maps are most common across Venezuela and Puerto Rico longitudes 64°W–66°W, where a transition zone between NAO and ENSO influence has been noted [Jury et al., 2007]. With these caveats, the Atlantic decadal tripole pattern may be considered a statistically significant source of eastern Caribbean climate variability.

[23] Using composites based on the ranked combined SVD scores, we study the NCEP precipitable water and ECMWF wind stress differences between high and low phase, in each decadal frequency band (Figures 7a and 7b). A moisture surplus is found over West Africa in the 7–11 year bands, which shifts over the tropical Atlantic at 11–13 years. The Caribbean moisture surplus is strongest in the 7–9 year band, while strong deficits are seen over the southeast Pacific in 7–11 year bands. The precipitable water signal over South America is neutral in 7–11 year bands, but moisture surpluses are noted in the 11–13 year band. The ECMWF marine wind stress exhibits weak anticyclonic northeasterly anomalies in the West Atlantic at 7–9 years. Reduced trade winds (southwesterly anomalies) become active at 9–11 years in 5°N–20°N latitudes and resemble the pattern of Ruiz-Barradas et al. [2000, Figure 2]. This area shifts poleward to 20°N–35°N in the 11–13 year band and takes on the appearance of a standing cyclonic trough near 40°W, that draws anomalous airflow northward across the equatorial Atlantic.

Figure 7.

(a) Composite high- minus low-phase NCEP precipitable water and (b) ECMWF wind stress based on ranked combined SVD time scores in each decadal frequency band (rows). Scales are adjusted to range.

4. Conclusions

[24] Based on a covariance analysis between interannual filtered rainfall, SST and SLP, the leading mode of eastern Caribbean climate variability in the 20th century is a decadal oscillation with three centers of action (schematically in Figure 8), related to lower-order modes of SST and SLP that comprise an Atlantic tripole feature (cool 35°N–20°N, warm 20°N–5°N, cool 5°N–20°S). The decadal rainfall variability is strongest in the northern Antilles Islands in the 9–11 year band, and in the southeastern Antilles Islands in the 11–13 year band. Wet phases are associated with warm SST in the Atlantic trade wind zone and corresponding lower pressure over the North Atlantic, and cool SST and higher pressure over the East Pacific and South Atlantic. Lower pressure over the North Atlantic contributes to lower evaporation rates and higher SST in the zone frequented by African easterly waves. Enfield and Alfaro [1999] and Giannini et al. [2001] describe a similar climatic pattern between opposing centers of action in the East Pacific and tropical North Atlantic, and Wang et al. [2009] found a decadal component in the signal. The ability of the ECHAM4.5 model to simulate this variability was investigated. Although its mode 1 rainfall time score correlated at +0.35 with observed, its model-simulated Ts and SLP modes represented the observed decadal variability reasonably well, again with three centers of action in the Atlantic: most notably for Ts in the trade wind zone and SLP in the North Atlantic.

Figure 8.

Schematic representation of high-phase features underlying the decadal cycle of Caribbean rainfall.

[25] The wavelet cospectrum of Caribbean rain1 PC scores and associated SST modes exhibited quasi-decadal fluctuations (Figure 1f) consistent with findings of Wang et al. [2009]. We suggest these derive from interaction of the Atlantic tripole and ENSO. The shift of the wavelet cospectrum from 1930 to 1970 coincides with a transition to a warmer North Atlantic that inhibited ENSO amplitude [Dong et al., 2006]. Before 1930 and after 1970, both exerted similar influence on Caribbean climate, whereas in the period 1930 to 1970 we suggest the Atlantic tripole was more active in modulating the intensity of African easterly waves reaching the Caribbean. The decadal signal has strengthened since 1970 (see Figure 6a) in relation to changes in NAO and its interaction with Pacific ENSO.

[26] Our wavelet-filtered SVD analysis found that the three centers of action in Atlantic SST are relatively unsynchronized. The northern center was absent in the 9–11 year range, while the southern center was only present in the 11–13 year band, coincident with negative correlations in the east Pacific. The unsynchronized behavior of the three centers of action can be found in Hovmoller analysis of decadal SST anomalies. The decadal oscillation, as in work by Goldenberg et al. [2001] and Lee and Wang [2008], contributes to changes in the tracks and frequency of hurricanes (see Figure 5) through the vorticity budget of African easterly waves (see Figure 7b). Composite analyses of high and low phases of the decadal mode found a more active Atlantic ITCZ in boreal summer and suppressed rainfall over the east Pacific ITCZ. Another highlight of the composite analysis was the discovery of three centers of action in upper zonal winds (Figure 4b): the two subtropical jets and equatorial jet oscillate together as a key part of the decadal signal. The resultant ‘footprint’ corresponds with a tripole in surface temperature, but the processes by which these are related deserve further attention in both continuous and seasonal analyses.


[27] We thank Bernd Sing, Department Mathematics, University of the West Indies, Barbados, for assistance with the wavelet-filtered SVD analysis.