The Predictive Skill and the Most Predictable Pattern in the Tropical Atlantic: the Effect of ENSO

1786 MONTHLY WEATHER REVIEW VOLUME 135

The Predictive Skill and the Most Predictable Pattern in the Tropical Atlantic: The Effect of ENSO

ZENG-ZHEN HU Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland

BOHUA HUANG Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland, and Department of Climate Dynamics, College of Science, George Mason University, Fairfax, Virginia

(Manuscript received 24 March 2006, in final form 28 August 2006)

ABSTRACT

This work investigates the predictive skill and most predictable pattern in the NCEP Climate Forecast System (CFS) in the tropical Atlantic Ocean. The skill is measured by the sea surface temperature (SST) anomaly correlation between the predictions and the corresponding analyses, and the most predictable patterns are isolated by an empirical orthogonal function analysis with a maximized signal-to-noise ratio. On average, for predictions with initial conditions (ICs) of all months, the predictability of SST is higher in the west than in the east. The highest skill is near the tropical Brazilian coast and in the Caribbean Sea, and the lowest skill occurs in the eastern coast. Seasonally, the skill is higher for predictions with ICs in summer or autumn and lower for those with ICs in spring. The CFS poorly predicts the meridional gradient in the tropical Atlantic Ocean. The superiority of the CFS predictions to the persistence forecasts depends on IC month, region, and lead time. The CFS prediction is generally better than the corresponding persistence forecast when the lead time is longer than 3 months. The most predictable pattern of SST in March has the same sign in almost the whole tropical Atlantic. The corresponding pattern in March is dominated by the same sign for geopotential height at 200 hPa in most of the domain and by significant opposite variation for precipitation between the northwestern tropical North Atlantic and the regions from tropical South America to the southwestern tropical North Atlantic. These predictable signals mainly result from the influence of the El Niño–Southern Oscillation (ENSO). The significant values in the most predictable pattern of precipitation in the regions from tropical South America to the southwestern tropical North Atlantic in March are associated with excessive divergence (convergence) at low (high) levels over these regions in the CFS. For the CFS, the predictive skill in the tropical Atlantic Ocean is largely determined by its ability to predict ENSO. This is due to the strong connection between ENSO and the most predictable patterns in the tropical Atlantic Ocean in the model. The higher predictive skill of tropical North Atlantic SST is consistent with the ability of the CFS to predict ENSO on interseasonal time scales, particularly for the ICs in warm months from March to October. In the southeastern ocean, the systematic warm bias is a crucial factor leading to the low skill in this region.

1. Introduction Saharan Africa (Lamb 1978; Folland et al. 1986). Wang (2002) further demonstrated that both the Atlantic Early studies have demonstrated the impact of the equatorial mode and the meridional gradient mode are tropical Atlantic sea surface temperature (SST) on the linked with large-scale atmosphere circulation, such as climate variability in the surrounding regions, such as the Walker and Hadley cells. In addition, the impor- the precipitation over northeast Brazil (e.g., Hastenrath tance of air–sea interaction on the variability and pre- and Heller 1977; Moura and Shukla 1981) and sub- dictability of tropical Atlantic climate has also been well demonstrated (e.g., Zebiak 1993; Chang et al. 1997; Xie 1999; Huang and Shukla 1997, 2005; Hu and Huang Corresponding author address: Zeng-Zhen Hu, Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 2006a,b; and references therein). Dynamical modeling 302, Calverton, MD 20705. and predictions of the tropical Atlantic variability E-mail: [email protected] (TAV) have received increasing attention over recent

DOI: 10.1175/MWR3393.1

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC

MWR3393 MAY 2007 H U A N D HUANG 1787 years. However, the analyses of predictability of TAV North Atlantic Ocean is substantially affected by are few and the predictive ability of current climate ENSO, with enhancement from local coupling in their models in the tropical Atlantic Ocean is still unclear. model. Merkel and Latif (2002) demonstrated a south- Most prediction studies focused on the variability of the ward shift of the North Atlantic low pressure systems in North Atlantic Oscillation (e.g., Hu and Huang 2006c) the winter season during El Niño events. Wang (2002) or on the influence of prescribed SST anomalies suggested that an El Niño can affect the tropical North (SSTA) on the atmosphere. Prediction experiments of Atlantic through the Walker and Hadley circulations, Chang et al. (1998) using a regional coupled ocean– favoring the tropical North Atlantic warming in the atmosphere model of the tropical Atlantic Ocean indi- subsequent spring of the El Niño year. Wang (2002) cated that the SST decadal variability in the subtropics also indicated that the tropical South Atlantic does not of the northern Atlantic Ocean is predictable several significantly correlate with Pacific El Niño. Recently, years ahead with modest levels of skill. They suggested Tourre and White (2005) examined the space–time evo- that atmosphere–ocean interactions enhance the pre- lution of the dominant ENSO signal of SST, upper- dictability of low-frequency SST variation beyond the ocean heat storage, zonal surface wind, and sea level time scale of persistence. They showed that the low- pressure (SLP) in 3.4–5.7-yr band period in the tropical frequency SST variability in the tropical North Atlantic eastern Pacific–Atlantic domain. They found that slow has significant predictability using statistical models at SST–SLP coupled waves propagate eastward from the short lead times (less than 1 yr), and using a coupled eastern Pacific Ocean into the tropical Atlantic Ocean atmosphere model for long-range forecasts with lead basin at a speed of about 20 cm sϪ1. As a result, a peak times up to about 3 yr. Using linear inverse modeling, signal in the equatorial Atlantic lags that in the eastern Penland and Matrosova (1998) quantified the predict- equatorial Pacific by 12–18 months. It has also been ability of tropical Atlantic SST on seasonal to interan- noted in some previous studies that the influence of nual time scales. They found that predictability of the ENSO on TAV is seasonally dependent. For example, Caribbean Sea and tropical North Atlantic SSTA is Huang (2004) showed a distinct progression of the enhanced when using global tropical SSTA as predic- ENSO signals in the tropical Atlantic Ocean from sea- tors compared with using only the tropical Atlantic son to season with different mechanisms. During the Ocean as predictors. This predictability advantage does boreal winter of a maturing El Niño year, a basinwide not carry over into the equatorial and south tropical warming occurs in the tropical Atlantic with a major Atlantic Ocean where persistence becomes a competi- center in the southern subtropical Atlantic. In following tive predictor in those regions. boreal spring, the warming occurs mainly in the north- It has been noted previously that TAV is closely ern tropical Atlantic Ocean. linked to the El Niño–Southern Oscillation (ENSO) The robustness of these results needs to be further phenomenon. The global temperature and precipita- examined, although it is clear that knowledge of the tion response to ENSO have been well documented ENSO condition ahead of time would provide reason- (e.g., Yulaeva and Wallace 1994; Tourre and White able opportunities to predict TAV. These studies have 1995; Chiang and Sobel 2002). In an El Niño year, shown promise for TAV predictions. However, there is warming dominates the whole tropical troposphere no consensus regarding how ENSO affects the predic- (Chiang and Sobel 2002). Previous investigations have tive skill, and what the most predictable pattern in also demonstrated the impact of ENSO on TAV. For TAV is in a real climate prediction system of a coupled example, Tourre et al. (1985) and Katz (1987) pointed general circulation model (CGCM). This work exam- out that tropical Atlantic wind forcing induced by ines the predictive skill and the most predictable pat- ENSO is linked with equatorial Atlantic warm events. tern in the tropical Atlantic Ocean in a real climate Carton and Huang (1994) showed the dependence of prediction system. The paper is organized as follows. In enhanced equatorial Atlantic trade winds during ENSO section 2, the data and analysis strategy are briefly de- warm phases on the thermocline in the western equa- scribed. Section 3 examines the predictability of TAV, torial Atlantic. Delecluse et al. (1994) suggested that including the dependence on lead month and initial ENSO may force the Atlantic equatorial mode. Enfield condition (IC) month. Section 4 investigates the most and Mayer (1997) indicated that ENSO is a major re- predictable patterns and their connection with ENSO. mote forcing of TAV anomalies. Latif and Grötzner Section 5 is a summary and gives further discussion. (2000) identified an internal equatorial Atlantic oscillation, which is strongly influenced by ENSO with the 2. Data and analysis strategy equatorial Atlantic SST lagging by about 6 months. The predictions, analyzed in this work, were con- Huang et al. (2002) found that the SSTA in the tropical ducted by the National Centers for Environmental Pre-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1788 MONTHLY WEATHER REVIEW VOLUME 135 diction (NCEP) Climate Forecast System (CFS), which SST dataset (Reynolds and Marsico 1993; Reynolds cover predictions initialized in all calendar months from and Smith 1994; Reynolds et al. 2002). This SST dataset 1981 to 2003. In a given start month, 15 predictions of is referred to as OIv2-analyzed SST in the following 9-month length from lead month 0 to 8 are produced text. The precipitation fields are a combination of sat- with different oceanic and atmospheric ICs. The oce- ellite retrievals, in situ rain gauge stations, and atmo- anic ICs are chosen as the three Global Ocean Data spheric model products (Xie and Arkin 1996, 1997). All Assimilation System (GODAS) analyses at the 11th the data used this work span the period from January and 21st of lead month 0 and 1st of lead month 1 (Saha 1981 to December 2004, except that OIv2-analyzed SST et al. 2006). The atmospheric ICs are from the NCEP are from November 1981 to December 2004. reanalysis II (Kanamitsu et al. 2002). In addition to the We first examine the evolution of averaged SST pre- atmospheric ICs at the same time as the oceanic ICs, dictability and signal-to-noise ratio of SST variance as a each of the three oceanic states is also paired with other function of lead time and also the predictability for four atmospheric ICs, namely, those within two days some typical regional mean SST indices. The most before and after at a daily interval to form a five- predicatable patterns of SST, geopotential height at 200 member local cluster. Therefore, there are total 15 hPa (H200), and precipitation, as well as the effect of ocean–atmosphere ICs spread from the ninth of lead ENSO on these patterns, are investigated. The predi- month 0 to the third of lead month 1. Details of the ICs catable patterns are isolated by applying an empirical are given in Saha et al. (2006) and Huang et al. (2007). orthogonal function (EOF) analysis with maximized The datasets of these monthly mean predictions of the signal-to-noise ratio (MSN EOF hereafter) to the pre- 15 individual members as well as their ensemble mean dicted time series of 1981–2003 with given lead time of the predictions. are available from the NCEP Web site (online at http:// The MSN EOF is a method to derive patterns that nomad6.ncep.noaa.gov/cfs/monthly). optimize the signal-to-noise ratio from all ensemble The atmospheric component of the CGCM used to members. This approach was developed by Allen and conduct the predictions is a lower-resolution version of Smith (1997) and discussed in Venzke et al. (1999). It the global weather forecast system of NCEP. The hori- has also been used by Sutton et al. (2000), Chang et al. zontal resolution is T62, and there are 64 vertical levels. (2000), and Huang (2004) in extracting the dominant The subscale physical parameterizations have been MSN EOF patterns in ensemble GCM integrations. modified from the model used to produce the NCEP– This method minimizes the effects of noise in a mod- National Center for Atmospheric Research (NCAR) erate ensemble size. An ensemble mean is supposedly reanalysis (Kalnay et al. 1996), as described in Saha et composed of a forced and a random part, which may be al. (2006). The oceanic component is the Geophysical attributed, respectively, to the prescribed external Fluid Dynamics Laboratory Modular Ocean Model boundary conditions and the unpredictable internal (version 3; Pacanowski and Griffies 1998). The ocean noise. In our case, the forced part is associated with the model domain extends across the world oceans from predictable signals, which show certain consistency ϫ 74°Sto64°N with a horizontal grid of 1° 1° poleward among different members of the ensemble predictions of 30°S and 30°N and with gradually increased meridi- because of the memories contained in the ICs. The onal resolution to 1/3° between 10°S and 10°N. The leading MSN EOF mode is the one with the maximum ocean model has 40 levels vertically with 27 of them in ratio of the variance of the ensemble mean to the de- the upper 400 m. The atmospheric model is coupled viations among the ensemble members. In this work, with the oceanic component on a daily basis without we only analyze the leading MSN EOF mode, which is flux adjustment or correction. ENSO and its associated defined as the most predictable pattern and is signifi- features are realistically reproduced by this model in cant at the 95% level using an F test. Details of this long-term simulations (Wang et al. 2005) and are well method were documented in Allen and Smith (1997), captured in predictions. Venzke et al. (1999), and Huang (2004). For the CFS prediction validation, we used the monthly mean data from the NCEP–NCAR reanalysis 3. Predictability of TAV I and II (Kalnay et al. 1996; Kanamitsu et al. 2002). We also used monthly SST data on a 2° ϫ 2° grid (Reynolds In this section, we analyze the evolution of the CFS and Marsico 1993; Reynolds and Smith 1994; Reynolds predictive skill as a function of lead time and its varia- et al. 2002), and precipitation field data on a 2.5° ϫ 2.5° tion with region in the tropical Atlantic Ocean. The resolution (Xie and Arkin 1996, 1997). The SST dataset predictive skill is defined as the correlation between the is the second version of the optimally interpolated (OI) CFS-predicted and OIv2-analyzed SST. The predictive

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1789 skill is compared with that of the corresponding persis- analyzed indices are shown in the left column of Fig. 3, tence forecast. We also examine the predictability for the corresponding ones for the persistence forecast some regional mean SST indices of TAV and its depen- are shown in the middle column, and the right column dence on IC month. The evolution of the signal-to- is the difference. The predictability is the largest for noise ratio of SST variance as a function of lead time is the TB index (Fig. 3a) and the smallest for the dipole also investigated. index (Fig. 3d). The predictability of the NB and SB The correlations between CFS-predicted and OIv2- indices is in between. This result suggests that although analyzed SST in the tropical Atlantic Ocean vary with CFS is successful in predicting the basinwide fluctua- regions and lead time (left column of Fig. 1). In a tion, the predictions of the anomalous meridional SST 1-month lead, the correlations between the predictions gradients are relatively poor. Figure 3 shows that the and analyses are generally larger than 0.5. The predic- predictive skill is dependent on when the predictions tions in the current work are the ensemble mean of 15 are initiated. The predictive skill is highest when ICs individual prediction members. Beyond 5-month lead, are in boreal spring and summer for the TB index (Fig. the correlations are below 0.5 in most regions of the 3a) and the SB index (Fig. 3c), from boreal summer to domain, except for some areas in the western ocean. winter for the NB index (Fig. 3b), and in boreal spring Spatially, the SST predictive skill of CFS is higher in the and early summer for the dipole index (Fig. 3d). The west than in the east. The highest predictive skill is near dependence of the skill on the season for the NB and the tropical Brazilian coast and in the Caribbean Sea, SB indices may imply that the skill is generally higher and the lowest skill occurs in the eastern coast and in for prediction with ICs in a cold season than in a warm the regions near the northern and southern boundary of season. the domain. As we will see later, the higher skill in the Although the correlation patterns are different be- west is associated with a stronger ENSO influence tween the CFS prediction and persistence forecast, and there. In contrast to that ENSO is the major source of also vary with the indices, there are some similarities in predictability of TAV in CFS; the model systematic the differences of the predictive skill between the CFS error is one of the factors that limits the model predic- prediction and persistence forecast (right column of tive skill. For example, the large positive SST bias in the Fig. 3). When the lead is less (greater) than 3 months, southeastern Atlantic (Fig. 2) may modify the west– the CFS prediction is worse (better) than the persis- east SST gradient and change the trade winds in the tence forecast. The superiority of the CFS prediction to region. This affects the atmosphere–ocean interaction the persistence forecast enlarges with increase of the features in this region. lead month. That is consistent with Fig. 1 (right col- On average, the correlation patterns of the CFS pre- umn). For the dipole index, mainly because of the very diction (left column of Fig. 1) are similar to those of low persistency, the predictability difference of the di- persistence forecast of SSTA (middle column of Fig. 1). pole index between the CFS prediction and persistence The correlations of persistence forecast are larger than forecast is high when ICs are in boreal spring and early those of the CFS prediction when the lead time is less summer, although the correlation between the CFS than 3 months (right column of Fig. 1). The CFS pre- predicted and analyzed dipole index is small. It is diction is superior to the persistence forecast in most shown in Fig. 3 that persistence beats the CFS skill in regions of the tropical Atlantic when the lead time is the first three months. This is probably because the SST beyond 3 months (see right column of Fig. 1), although from the forecast IC are different than the verifying the differences of the corrections are small between the SST analysis; it probably would not happen if the CFS prediction and the persistence forecast. The supe- NCEP GODAS (Behringer et al. 1998; Behringer and riority of the CFS prediction to the persistence forecast Xue 2004) SST analysis were used as verifying dataset. grows when the lead month increases. Therefore, there is room for improving the forecast The features of the spatial and temporal evolution of skill in the first three months by improving the ocean the predictability of TAV in Fig. 1 are elucidated by initialization. After all, it is interesting to note that high examining the predictability of some indices that char- predictive skill up to 8 months can be achieved for acterize TAV. Following Servain (1991), the TB index those indices. In the next section, we will identify the is the averaged SSTA from 20°Sto30°N, the NB index source of this high predictability. represents the averaged SSTA in the north (5°–28°N), The decay of the correlation between the CFS- the SB index expresses the averaged SSTA in the south predicted and OIv2-analyzed SST (left columns of Figs. (20°S–5°N), and the dipole index is defined as NB–SB. 1 and 3) with lead time is also demonstrated by the The correlations between the CFS-predicted and decline in the ratio of the CFS-predicted signal-to-RMS

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1790 MONTHLY WEATHER REVIEW VOLUME 135

FIG. 1. Correlations between (left) OIv2-analyzed and predicted SST anomalies, (middle) persistence correlations of OIv2-analyzed SST anomalies, and (right) the differences at different lead months for all ICs in 1981–2003. The contour interval is 0.1, and the shading represents values greater than 0.5 or less than 0.3 in the left and middle columns, and positive values in the right column.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1791

FIG. 2. Mean errors of SST (predicted–OIv2 analyzed) for all ICs in different lead months. The contour interval is 0.4°C, and the shading is for values greater than 0.4°C or less than Ϫ0.4°C.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1792 MONTHLY WEATHER REVIEW VOLUME 135

FIG. 3. Correlations of regional mean SST indices between (left) OIv2-analyzed and predicted, (middle) persistence correlations of OIv2-analyzed anomalies, and (right) the differences at different lead months and all ICs for 1981–2003 in the Atlantic. (a) TB (SST averaged in 20°S–30°N), (b) NB (SST averaged in 5°–28°N), (c) SB (SST averaged in 20°S–5°N), and (d) dipole (NB–SB) indices. The contour interval is 0.1, and the shading represents values greater than 0.7 or less than 0.5 in the left and central columns, and positive values in the right column. The small black squares in the left and right columns of (a) are the target month of March in the following analyses.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1793 error. The ratio is calculated as the standard deviation 4. The most predictable patterns of TAV and the (std dev) of the OIv2-analyzed SST multiplying the cor- effect of ENSO relation between the CFS ensemble mean and the analyzed SST then divided by the RMS error. Figure 4a In this section, the most predictable patterns in the (first row) is an example, showing the predicted signal- tropical Atlantic Ocean are extracted by applying MSN to-RMS error ratios of SST in March with 0- (left EOF, and the effect of ENSO on these patterns is ex- panel), 4- (middle panel), and 8-month (right panel) amined. In this section, March is chosen as the targeted leads. It is noted that almost all the ratios are positive, month for the most predictable pattern study, since it is consistent with the positive correlation between the the month with largest variability and strongest lag re- CFS-predicted and the analyzed SST (see left column lation between ENSO and TAV (Tourre and White of Fig. 1 for the situation of the tropical Atlantic and all 2005). Anomalous atmospheric circulation patterns ICs). The ratio is generally smaller than 1 without maxi- over the north tropical Atlantic are phase locked to the mum center in the tropical central and eastern Pacific, seasonal circle, and boreal spring is the season having resulting from a similar distribution of RMS error (Fig. the strongest signal of out-phase variation between the 4d) and the std dev of the forecast ensemble mean (Fig. northern and southern tropical Atlantic (Nobre and 4e) with a maximum center in the tropical central and Shukla 1996). Moreover, for the target month in March, eastern Pacific. Except for some regions in the south- on average in the tropical Atlantic, there are the high- eastern oceans, the similar spatial distribution patterns est correlations between the CFS predictions and the of Figs. 4d and 4e, particularly in the tropical Pacific, OIv2 analyses (see the small black squares in the left may suggest that CFS RMS error evolves similarly to panel of Fig. 3a), and also the superiority of the CFS that of the std dev of the forecast ensemble mean. Be- prediction to the persistence forecast is generally larger cause the amplitudes in Fig. 4d are slightly larger than for March than for any other target months, except for that in Fig. 4e and the correlation between CFS en- ICs in January, February, and March (see the small semble mean and the analyzed SST is less than 1.0, the black squares in the right panel of Fig. 3a). Separating ratio of the CFS-predicted signal-to-RMS error is calculations indicate that the most predictable pattern smaller than 1.0. is generally similar for different target months (not shown), implying independence of the results to the The evolution pattern of the ratio of the predicted target month. The MSN EOF is calculated in this work signal to ensemble noise (Fig. 4a) is different from that based on forecast ensemble mean and forecast en- of the signal-to-noise ratio (Fig. 4b), in both the spatial semble spread. It should be noticed that each member pattern and amplitudes. The ensemble signal and noise in an ensemble prediction is treated equally in the MSN are calculated following Rowell et al. (1995). The ratio EOF calculation. This means that there is no consider- is a measurement for the skill and spread relationship in ation given to the fact that some members are better the model. Figure 4b shows a clear maximum center in predictions than others. the tropical central and eastern Pacific, indicating that the CFS ensemble spread (noise) is smaller than the signal (std dev of forecast ensemble mean) in these a. The most predictable pattern of SST regions. From Figs. 4c and 4d, we see that the model Figure 5 shows the first mode of MSN EOF (MSN ensemble noise and the RMS error are largest in the EOF1) of SST in the tropical Atlantic Ocean for (a) a 4-month lead prediction and smallest in the 0-month March IC at a 0-month lead, (b) a November IC at a lead prediction, and also the RMS error is larger than 4-month lead, and (c) a July IC at an 8-month lead. It the model ensemble noise. Thus, the decline of the should be pointed out that the predicted anomalies are model signal-to-noise ratio in Fig. 4b may be mainly the ensemble mean, which are calculated with respect due to the decrease of the signal, suggesting that the to different monthly climatology means for the two pe- decline of the CFS predictive skill is mainly due to the riods 1981–90 and 1991–2003 to remove the influence of decrease of the signal instead of the increase of the the systematically warm bias in CFS during 1981–90 ensemble spread (noise). In addition, the evolution of (Peng et al. 2005). It has been found that the warm bias the RMS error in the southeastern Atlantic is consistent was caused by a coding problem, which results in about with the result of Huang et al. (2007). They found that 0.5°C warmer on average in 1981–90 than in 1991–2003 the error of the CFS prediction in the southeastern in the tropical and subtropical SST of the GODAS tropical Atlantic grows much faster for the prediction ocean analysis. Its impact on the CFS is a discontinuity starting in boreal summer and autumn than that start- of the tropical SSTA prediction, which was systemati- ing in late winter and spring. cally warmer in 1981–90 than in 1991–2003. It is rea-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1794 MONTHLY WEATHER REVIEW VOLUME 135

FIG. 4. (a) Predicted signal-to-RMS error, (b) signal-to-ensemble noise ratio of standard deviation, (c) CFS ensemble noise, (d) CFS RMS error, and (e) the standard deviation of forecast ensemble mean for March SST averaged over 1981–2003. (left) 0-month lead and the March IC; (middle) 4-month lead and the November IC; and (right) 8-month lead and the July IC. The contour interval is (a) 0.3, (b) 0.5, (c) 0.25°C, and (d), (e) 0.5°C. The shading represents values greater than (a) 0.6, (b) 1.0, (c) 0.5°C, and (d), (e)1.0°C.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1795

FIG. 5. (left) Time series and (right) spatial patterns of MSN EOF1 of March SST: (a) 0-month lead and the March IC, (b) 4-month lead and the November IC, and (c) 8-month lead and the July IC. The contour interval is 0.2, the zero lines are omitted, and the shading is for values greater than 0.2 or less than Ϫ0.2 for the spatial patterns. The real magnitude of the SST anomalies (°C) can be restored by multiplying the values in the spatial patterns with the corresponding time series. The percentage of the explained variance for the ensemble mean anomalies is indicated in each panel.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1796 MONTHLY WEATHER REVIEW VOLUME 135 sonable and necessary to compare the predictions with neous correlations between the time series in Fig. 5 and the corresponding GODAS data, which were used to the corresponding Niño-3.4 index in the predictions make the predictions. (see the second row in Table 1). However, for the OIv2- There are obvious similarities among the leading analyzed SST, the strongest regressions occur in the modes derived from different leads for both the pat- preceding December (right panel of Fig. 6a); then the terns and time series (Fig. 5). The percentages of the regressions are weakened continuously in the following variance of the ensemble mean SSTA explained by the months (right panels of Figs. 6b–e). From the regres- MSN EOF1 are 45%, 46%, and 50% in Figs. 5a, 5b, and sions in Fig. 6, we also note that the association of the 5c, respectively. In general, positive values dominate most predictable patterns of TAV with the predicted most of the domain with some negative signals existing SST variations in the tropical Indian Ocean is similar to in the northwestern corner of the domain (Fig. 5). It is but stronger than the association with the OIv2- interesting to note that the MSN EOF1 for the July IC analyzed SST. is the only one that has a maximum in the Gulf of In the Atlantic, the differences of the regressions for Guinea (Fig. 5c), which may be due to the relatively using the predicted and OIv2-analyzed SST are clear, small ensemble noise in this region (see right panel of particularly in the South Atlantic. In the preceding De- Fig. 4c). For the time series, large positive values cor- cember (Fig. 6a), the significant signals are confined to respond to El Niño years, such as 1983, 1988, 1998, and the equatorial and South Atlantic for the predicted 2003, and large negative values generally correspond to SST, but there are almost no significant regressions for La Niña years, such as 1989 and 1999–2001. This result the OIv2-analyzed SST. In the following months (Figs. is consistent with the ENSO influence on TAV indi- 6b–e), the significant regressions occupy the whole cated by previous studies mentioned in the introduction tropical Atlantic for the predicted SST. In contrast, the [e.g., Wang (2002), Chiang and Sobel (2002), etc.]. significant regressions mainly exist in some regions of However, the relationship between ENSO and the the subtropical North Atlantic for the OIv2-analyzed MSN EOF1 mode does not always hold for all warm SST. The area with significant regressions in the tropi- and cold years of ENSO. The relationship is more rep- cal Atlantic Ocean reaches its maximum in simulta- resentative for strong ENSO years, such as 1983, 1998, neous and 1-month-lag regressions for both the OIv2- and 1989, than for weak ENSO years. For instance, analyzed and predicted SST (Figs. 6c,d). After that the 1992 is distinct in that it has a little warming in the regression in the tropical Atlantic Ocean weakens with tropical Atlantic in boreal spring in 0- (Fig. 5a) and a reduced area of significance (Fig. 6e). 4-month (Fig. 5b) lead predictions and a substantial Besides the differences of the regressions between warming in 8-month lead predictions. The complicated the predicted and OIv2-analyzed SST in Fig. 6, the dif- relationship between ENSO and the most predictable ferences are also seen between the most predictable patterns may be because ENSO is just one of the fac- patterns in Fig. 5 and the regression patterns with the tors affecting TAV. OIv2-analyzed SST in the Atlantic in Fig. 6 (right col- The relationship between the most predictable pat- umn), particularly for the southeastern tropical Atlan- terns of TAV and global SST is demonstrated by cal- tic Ocean. In contrast to the most predictable pattern culating the regressions of the CFS-predicted (left col- with an almost whole domain positive values (Fig. 5), umn) and OIv2-analyzed (right column) global SST which is similar to the regression patterns using the onto the time series of MSN EOF1 with the November predicted SST (left column of Fig. 6), for the regres- IC at 4-month leads (Fig. 6). It is seen that the general sions using the OIv2-analyzed SST (right column of Fig. regression patterns for CFS-predicted (left column) and 6), the significant positive signals are mainly in the OIv2-analyzed (right column) SST are similar, particu- tropical North Atlantic and the regression has little sig- larly in the Pacific and Indian Oceans, although the nificance in the South Atlantic. The discrepancy be- regressions are stronger and more significant in the tween the CFS predictions and the OIv2 analyses may former. The regressions in the Pacific are dominated by reflect a flaw in the prediction system; that is, although ENSO in both the predicted and OIv2-analyzed SST. the Atlantic meridional SST gradient is influenced by That suggests that ENSO, both in the real world and in ENSO in reality, the ENSO response in the CFS in the the model, is related to the TAV predictability. For the Atlantic sector suggests a symmetric mode with respect predicted SST, there are little changes for the connec- to the equator. This is consistent with the evidence in- tion between the most predictable patterns of TAV and dicated in the previous section that the model is unable ENSO from preceding December until April (left col- to predict the meridional gradient (dipole index) very umn of Figs. 6a–d). This is consistent with the simulta- well. This distinction may impact the atmospheric cir-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1797

FIG. 6. (left) CFS-predicted and (right) OIv2-analyzed SST regression onto SST MSN EOF1 time series for a 4-month lead and November IC (Fig. 5b) at different leading and lagging months. The contour interval is 1.0°C per std dev of the MSN EOF1 time series, the zero lines are omitted, and the shading is for significant regression at the 95% level.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1798 MONTHLY WEATHER REVIEW VOLUME 135

TABLE 1. Simultaneous correlations between the predicted Niño-3.4 index and the MSN EOF1 time series in Figs. 5, 8, and 9. Correlations are significant at the 95% level using a Student’s t test, when it is larger than 0.41.

Indices Mar IC, 0-month lead Nov IC, 4-month lead Jul IC, 8-month lead Niño-3.4 and SST MSN EOF1 0.41 0.76 0.92 Niño-3.4 and H200 MSN EOF1 0.84 0.90 0.92 Niño-3.4 and precipitation MSN EOF1 0.86 0.95 0.89

culation over the Atlantic. The possible reasons for the lyzed total precipitation onto Niño-3.4 (Fig. 7c). The differences will be further discussed in next section. main differences are in the tropical Americas and the To further analyze the relationship between the most tropical western Atlantic. The regression value in this predictable pattern of TAV and ENSO, Fig. 7 shows region is small (Fig. 7c) and strong negative values the simultaneous regressions of global SST, H200, and dominate this region for the most predictable pattern total precipitation in the tropical Atlantic Ocean onto (Fig. 9). The negative regression is confined to the the Niño-3.4 index in March by using reanalysis and northeastern coast of Brazil (Fig. 7c), but in the model other observational data. The regression patterns in the negative is shifted northward and the area with Fig. 7a are almost identical to Fig. 11 of Chiang and negative regressions is enlarged (Fig. 9). The patterns Sobel (2002). The basic features in Fig. 7a are also simi- of Figs. 7c and 9 agree in showing positive and signifi- lar to that in Fig. 6c with the OIv2-analyzed SST, con- cant values along the equatorial eastern Pacific and firming the connection between the most predictable around the Caribbean Sea and the Gulf of Mexico, pattern of TAV and ENSO. The major difference be- which implies that the rainfall variation in these regions tween Figs. 7a and 6c (right panel) is that the regres- may be associated with ENSO. The regression pattern sions in the Atlantic are more significant in Fig. 6c is similar when calculating the regressions of the ana- (right panel) than in Fig. 7a. lyzed precipitation onto Niño-3.4 or onto the time series of MSN EOF1 (Figs. 7c, 10a). It is also the case b. The most predictable patterns of H200 and when using the CFS-predicted precipitation (Figs. 9, precipitation 10b). However, the clear difference is seen in the re- The simultaneous regression pattern of H200 onto gression between using the analyzed (Figs. 7c, 10a) and Niño-3.4 (Fig. 7b) is similar to the MSN EOF1 of H200 predicted precipitation (Figs. 9, 10b). (Fig. 8). The percentages of the variance of the en- Comparing with Figs. 7c and 10a, we found that the semble mean H200 anomalies explained by the MSN CFS overestimates the contrast of precipitation be- EOF1 are 68%, 82%, and 70% in Figs. 8a, 8b, and 8c, tween regions from tropical South America to the respectively. Both Figs. 7b and 8 are dominated by posi- tropical southwestern North Atlantic and the tropical tive values symmetric about the equator, implying that eastern Pacific (Figs. 9, 10b). The overestimation of the the model reproduces this symmetric response to contrast of the precipitation variation (Figs. 9, 10b) is ENSO in the upper atmosphere quite realistically. This associated with too-strong divergence and convergence result seems to tell a quite different story from the SST; in the CFS. On average, the CFS has excessive diver- that is, that the upper-atmospheric response to ENSO gence (convergence) at low (high) levels over the re- is symmetric about the equator in the Atlantic sector. gion from tropical South America to the tropical south- This is consistent with the previous finding of Yulaeva western North Atlantic and excessive convergence (di- and Wallace (1994) based on the temperature in the vergence) at low (high) levels in the tropical eastern upper atmosphere. They showed that warming (cool- Pacific (Figs. 10d,f), compared with the corresponding ing) in the tropical Atlantic Ocean and the tropical Pa- fields from the reanalyses data (Figs. 10c,e). It seems cific corresponds to positive (negative) H200 over the that the influence of ENSO on the tropical Atlantic Tropics. precipitation is overestimated and misplaced. The spatial pattern and time series of MSN EOF1 of The large values in the time series of Figs. 8 and 9 are total precipitation are shown in Fig. 9. The percentages connected with ENSO. This is consistent with the high of the variance of the ensemble mean total precipita- correlation between the time series and the Niño-3.4 tion anomalies explained by the MSN EOF1 are 45%, index in the prediction (third and fourth rows of Table 44%, and 60% in Figs. 9a, 9b, and 9c, respectively. 1). However, it is interesting to note that the correlation There are clear differences between the spatial patterns is insignificant at the 95% level between the time series (Fig. 9) and the simultaneous regression pattern of ana- of MSN EOF1 of SST (Fig. 5a) and that of precipitation

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1799

FIG. 7. Simultaneous regressions of (a) OIv2-analyzed global SST, (b) NCEP–NCAR re- analyzed tropical Atlantic H200, and (c) Xie–Arkin tropical Atlantic precipitation in March onto the Niño-3.4 index. The contour interval is (a) 0.2°C, (b) 5 m, and (c) 1.0 mm dayϪ1 per std dev of the Niño-3.4. The zero lines are omitted, and the shading is for significant regressions at the 95% level using a t test.

(Fig. 9a) for the predictions of the March IC at 0-month significant at the 95% level (second row of Table 1). lead (third row of Table 2). The simultaneous correla- The low correlations for the predictions in 0-month tion between the predicted Niño-3.4 index and the time lead may be a symptom of a nonbalanced initialization series of MSN EOF1 of SST for the predictions of the in the model. From Table 2, we also note that the cor- March IC at 0-month lead (Fig. 5a) is only marginally relations between the model most predictable patterns

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1800 MONTHLY WEATHER REVIEW VOLUME 135

FIG. 8. Same as in Fig. 5, but for H200. The contour interval is 1, the zero lines are omitted, and the shading is for values greater than 4 or less than 0 for the spatial patterns. The real magnitude of the H200 anomalies (gpm) can be restored by multiplying the values in the spatial patterns with the corresponding time series. The percentage of the variance of the ensemble mean anomalies explained by the MSN EOF1 is given in each panel.

in different fields generally grow with lead months. This tween the Niño-3.4 index and the model most predict- suggests an increase of the ENSO contribution to the able patterns with lead time (Table 1). This interesting ensemble mean anomalies with lead time. This is con- phenomena may be explained as follows: the model firmed by the general increase of the correlations be- usually produces more partitions unrelated to ICs for

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1801

FIG. 9. Same as in Fig. 5, but for total precipitation. The contour interval is 2, the zero lines are omitted, and the shading is for values greater than 2 or less than Ϫ2 for the spatial patterns. The real magnitude of the precipitation anomalies (mm dayϪ1) can be restored by multiplying the values in the spatial patterns with the corresponding time series. The percentage of the variance of the ensemble mean anomalies explained by the MSN EOF1 is given in each panel. the prediction with a longer lead time. The IC-unrelated 5. Summary and discussion partition will be largely erased by averaging the members in the ensemble. As a result, the ENSO-related This work investigates the predictive skill and most signal becomes more dominant in the prediction with a predictable pattern in the CFS in the tropical Atlantic longer lead time. Ocean. The skill is measured by SSTA correlation be-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1802 MONTHLY WEATHER REVIEW VOLUME 135

FIG. 10. Simultaneous regression on the time series in Fig. 9b. (a) Xie–Arkin precipitation, (b) CFS- predicted ensemble mean precipitation, (c), (e) divergence at 200 and 850 hPa, respectively, of NCEP– NCAR reanalysis data, and (d), (f) divergence at 200 and 850 hPa, respectively, of CFS-predicted ensemble means. The contour interval is 4 mm dayϪ1 per std dev of the time series for precipitation and 3 ϫ 10Ϫ6 sϪ1 per std dev for divergence at 200 hPa and 2 ϫ 10Ϫ6 sϪ1 for divergence at 850 hPa, the zero lines are omitted, and the shading is for significant regression at the 95% level.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1803

TABLE 2. Simultaneous correlations among the MSN EOF1 time series in Figs. 5, 8, and 9. Correlation is significant at the 95% level using a Student’s t test, when it is larger than 0.41. Insignifi- cant correlations are in italics.

Mar IC, Nov IC, Jul IC, 0-month 4-month 8-month Indices lead lead lead SST and H200 0.64 0.93 0.98 SST and precipitation 0.20 0.78 0.86 H200 and precipitation 0.84 0.94 0.86

tween the predictions and the corresponding analyses, and the most predictable patterns are isolated by an EOF analysis with maximized signal-to-noise ratio. On average, for the predictions with ICs of all months, the predictability of SST is higher in the west than in the east. The highest skill is near the tropical Brazilian coast and in the Caribbean Sea, and the lowest skill occurs in the eastern coast. Seasonally, the skill is higher for predictions with ICs in summer or autumn and lower for those with ICs in spring. CFS poorly predicts the meridional gradient in the tropical Atlantic Ocean. The superiority of the CFS predictions to the persistence forecasts depends on the IC month, region, and lead time. The CFS prediction is generally better than the corresponding persistence forecast when the lead time is longer than 3 months. The most predictable pattern of SST in March has the same sign in almost the whole tropical Atlantic. The corresponding pattern in March is dominated by the same sign for geopotential height at 200 hPa in most of the domain and by significant opposite variation for precipitation between the FIG. 11. (a) Correlations between analyzed and CFS-predicted northwestern tropical North Atlantic and the regions Niño-3.4 (5°S–5°N, 170°–120°W) index at different lead months from tropical South America to the southwestern tropi- and all ICs for 1981–2003 in the Atlantic, (b) the corresponding cal North Atlantic. These predictable signals mainly re- correlations of persistence forecast, and (c) the difference of (a) sult from the influence of ENSO. The significant values minus (b). The contour interval is 0.1, and the shading represents in the most predictable pattern of precipitation in the values greater than 0.7 or less than 0.5 in (a) and (b), and greater than 0.0 in (c). regions from tropical South America to the southwestern tropical North Atlantic in March are associated with excessive divergence (convergence) at low (high) may also lead to excessive atmosphere convection in levels over these regions in CFS. the model. According to Chiang and Sobel (2002), the The predictive skill of the CFS in the tropical Atlan- existence of atmospheric convection is crucial for tem- tic Ocean is largely determined by its ability to predict perature variations to propagate from the troposphere ENSO. This is because of the strong connection be- to the surface. This is confirmed by the zonal and ver- tween ENSO and the most predictable patterns in the tical cross section averaged between 12°S and 8°S for tropical Atlantic Ocean in this model. Fortunately, the temperature regression onto the time series of MSN CFS predictions for ENSO are quite skillful on in- EOF1 of precipitation (Fig. 12). The figure shows that terseasonal time scales (Fig. 11). On average, the CFS the precipitation pattern in Fig. 9b is associated with prediction is better than the persistence forecast be- temperature variations at all levels over the tropical yond a 1-month lead, particularly for the ICs from southeastern Atlantic in CFS (Fig. 12b). In contrast, in January to July. the reanalysis data, the association over the tropical The warm bias in the southeastern Atlantic (Fig. 2) southeastern Atlantic is constricted to the upper tropo-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1804 MONTHLY WEATHER REVIEW VOLUME 135

served monthly mean cloud cover data, which were produced by the International Satellite Cloud Clima- tology Project using a series of satellite radiance mea- surements (Rossow and Dueñas 2004). They demonstrated the impact of a cloud–radiation–SST feedback on the interannual variability of climate in the southeastern tropical Atlantic Ocean. It is found that the leading pattern of the low-cloud anomalies over the southeastern tropical Atlantic Ocean from June to Au- gust (JJA) is a modulation of its climatological center off the Angola and Benguela coasts on both interannual and longer time scales. The anomalous JJA low- cloud pattern is influenced by SST anomalies off the western coast of Africa near 15°S in January and Feb- ruary. In an anomalous warm SST event, the slow ex- pansion of warmer surface water into the open ocean is conducive to deficit low-cloud cover in the subsequent JJA. The reduced low-cloud in turn increases the amount of the local solar radiation reaching the sea surface and forces a positive SST tendency in the southeastern ocean. This moves the center of the SST anomalies away from the coast and closer to the equator. This feedback affects the evolution of the southeastern tropical Atlantic anomalous events. By comparing this observed cloud–SST feedback with the CFS simulation, Hu et al. (2007, manuscript submitted to J. Climate) found that this cloud–radiation–SST feedback is not well simulated in CFS. The major problem is that the model underestimates low-cloud cover in the southeastern tropical Atlantic Ocean, causing an overestimated amount of the local solar radiation reaching FIG. 12. Zonal and vertical cross section averaged between 12°S the sea surface and resulting in the warm SST bias in and 8°S of simultaneous regression of temperature in (a) the Fig. 2. NCEP–NCAR reanalyses and (b) the CFS predictions onto the time series of precipitation MSN EOF1 for a 4-month lead and November IC (Fig. 9b). The contour interval is 0.5°C per std dev Acknowledgments. The authors appreciate the com- of the MSN EOF1 time series, the zero lines are omitted, and the ments and suggestions of B. Kirtman, as well as S. Yang shading is for significant regression at the 95% level. and D. Straus, which significantly improved the manuscript. We also acknowledge the help of J. D. Doyle and I. Buffone, and the constructive comments and sugges- sphere and the regressions become negative for the tion from two anonymous reviewers. This work was lowest levels in the southeastern tropical Atlantic (Fig. supported by the NOAA CLIVAR Atlantic Program 12a). Therefore, the theory of Chiang and Sobel (2002) (NA04OAR4310115). can explain why the most predictable pattern of SST in the CFS is an almost homogeneous pattern in the tropi- REFERENCES cal Atlantic Ocean (see Fig. 5), different from that in the southeastern Atlantic regions shown in Fig. 7a of Allen, M. R., and L. A. Smith, 1997: Optimal filtering in singular this work and in Fig. 11 of Chiang and Sobel (2002). spectrum analysis. Phys. Lett., 234, 419–428. This result suggests that the systematic warm bias in the Behringer, D., and Y. Xue, 2004: Evaluation of the global ocean southeastern ocean is a crucial factor leading to the low data assimilation system at NCEP: The Pacific Ocean. Pre- prints, Eighth Symp. on Integrated Observing and Assimila- skill and unrealistic most predictable pattern in this re- tion Systems for Atmosphere, Oceans, and Land Surface, Se- gion. The impact of this error on the interannual vari- attle, WA, Amer. Meteor. Soc., CD-ROM, 2.3. ability has also been demonstrated by Huang (2004). ——, M. Ji, and A. Leetmaa, 1998: An improved coupled model Recently, Huang and Hu (2007) analyzed the ob- for ENSO prediction and implications for ocean initializa-

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC MAY 2007 H U A N D HUANG 1805

tion. Part I: The ocean data assimilation system. Mon. Wea. tion and its response to ENSO. Climate Dyn., 16, 213–218. Rev., 126, 1013–1021. Merkel, U., and M. Latif, 2002: A high resolution AGCM study of Carton, J. A., and B. Huang, 1994: Warm events in the tropical the El Niño impact on the North Atlantic/European sector. Atlantic. J. Phys. Oceanogr., 24, 888–903. Geophys. Res. Lett., 29, 1291, doi:10.1029/2001GL013726. Chang, P., L. Ji, and H. Li, 1997: A decadal climate variation in Moura, A. D., and J. Shukla, 1981: On the dynamics of the the tropical Atlantic Ocean from thermodynamic air-sea in- droughts in northeast Brazil: Observations, theory and nu- teractions. Nature, 385, 516–518. merical experiments with a general circulation model. J. At- ——, ——, ——, C. Penland, and L. Matrosova, 1998: Prediction mos. Sci., 38, 2653–2675. of tropical Atlantic sea surface temperature. Geophys. Res. Nobre, P., and J. Shukla, 1996: Variations of sea surface tempera- Lett., 25, 1193–1196. ture, wind stress, and rainfall over the tropical Atlantic and ——, R. Saravanan, L. Ji, and G. C. Hegerl, 2000: The effect of South America. J. Climate, 9, 2464–2479. local sea surface temperatures on atmospheric circulation over the tropical Atlantic sector. J. Climate, 13, 2195–2216. Pacanowski, R. C., and S. M. Griffies, 1998: MOM 3.0 manual. Chiang, J. C. H., and A. H. Sobel, 2002: Tropical tropospheric NOAA, 668 pp. [Available from NOAA/Geophysical Fluid temperature variations caused by ENSO and their influence Dynamics Laboratory, Princeton, NJ 08542.] on the remote tropical climate. J. Climate, 15, 2616–2631. Peng, P., and Coauthors, cited 2005: GODAS data discontinuity Delecluse, P., J. Servain, C. Levy, K. Arpe, and L. Bengtsson, and impact on CFS forecast bias. [Available online at http:// 1994: On the connection between the 1984 Atlantic warm www.nws.noaa.gov/ost/climate/STIP/CFS_news_070805.htm.] event and the 1982–83 ENSO. Tellus, 46A, 448–464. Penland, C., and L. Matrosova, 1998: Prediction of tropical At- Enfield, D. B., and D. A. Mayer, 1997: Tropical Atlantic sea sur- lantic sea surface temperatures using linear inverse modeling. face temperature variability and its relation to the El Niño- J. Climate, 11, 483–496. Southern Oscillation. J. Geophys. Res., 102 (C1), 929–945. Reynolds, R. W., and D. C. Marsico, 1993: An improved real-time Folland, C., T. Palmer, and D. Parker, 1986: Sahel rainfall and global sea surface temperature analysis. J. Climate, 6, 114– worldwide sea surface temperatures. Nature, 320, 602–606. 119. Hastenrath, S., and L. Heller, 1977: Dynamics of climate hazards ——, and T. M. Smith, 1994: Improved global sea surface tem- in Northeast Brazil. Quart. J. Roy. Meteor. Soc., 103, 77–92. perature analyses using optimum interpolation. J. Climate, 7, Hu, Z.-Z., and B. Huang, 2006a: Air–sea coupling in the North 929–948. Atlantic during summer. Climate Dyn., 26, 441–457. ——, N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, ——, and ——, 2006b: Physical processes associated with the 2002: An improved in situ and satellite SST analysis for cli- tropical Atlantic SST meridional gradient. J. Climate, 19, mate. J. Climate, 15, 1609–1625. 5500–5518. Rossow, W. B., and E. N. Dueñas, 2004: The International Satel- ——, and ——, 2006c: On the significance of the relationship lite Cloud Climatology Project (ISCCP) Web site: An online between the North Atlantic Oscillation in early winter and resource for research. Bull. Amer. Meteor. Soc., 85, 167–172. Atlantic sea surface temperature anomalies. J. Geophys. Res., 111, D12103, doi:10.1029/2005JD006339. Rowell, D. P., C. K. Folland, K. Maskell, and M. N. Ward, 1995: Huang, B., 2004: Remotely forced variability in the tropical At- Variability of summer rainfall over tropical North Africa lantic Ocean. Climate Dyn., 23, 133–152. (1906–92): Observations and modelling. Quart. J. Roy. Me- ——, and J. Shukla, 1997: Characteristics of the interannual and teor. Soc., 121, 669–704. decadal variability in a general circulation model of the tropi- Saha, S., and Coauthors, 2006: The NCEP Climate Forecast Sys- cal Atlantic Ocean. J. Phys. Oceanogr., 27, 1693–1712. tem. J. Climate, 19, 3483–3517. ——, and ——, 2005: Ocean–atmosphere interactions in the tropi- Servain, J., 1991: Simple climatic indices for the tropical Atlantic cal and subtropical Atlantic Ocean. J. Climate, 18, 1652–1672. Ocean and some applications. J. Geophys. Res., 96, 15 137– ——, and Z.-Z. Hu, 2007: Cloud-SST feedback in southeastern 15 146. tropical Atlantic anomalous events. J. Geophys. Res., 112, Sutton, R. T., S. P. Jewson, and D. P. Rowell, 2000: The elements C03015, doi:10.1029/2006JC003626. of climate variability in the tropical Atlantic region. J. Cli- ——, P. S. Schopf, and Z. Pan, 2002: The ENSO effect on the mate, 13, 3261–3284. tropical Atlantic variability: A regionally coupled model study. Tourre, Y. M., and W. B. White, 1995: ENSO signals in global Geophys. Res. Lett., 29, 2039, doi:10.1029/2002GL014872. upper-ocean temperature. J. Phys. Oceanogr., 25, 1317–1332. ——, Z.-Z. Hu, and B. Jha, 2007: Evolution of model systematic errors in the tropical Atlantic basin from the NCEP coupled ——, and ——, 2005: Evolution of the ENSO signal over the hindcasts. Climate Dyn., in press. tropical Pacific-Atlantic domain. Geophys. Res. Lett., 32, L07605, doi:10.1029/2004GL022128. Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Re- analysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. ——,M.Déqué, and J. F. Royer, 1985: Atmospheric response of Kanamitsu, M., W. Ebisuzaki, J. Woollen, S.-K. Yang, J. J. Hnilo, a general circulation model forced by a sea surface tempera- M. Fiorino, and G. L. Potter, 2002: NCEP–DOE AMIP-II ture distribution analogous to the winter 1982–1983 El Niño. Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 1631–1643. Coupled Ocean-Atmosphere Models, J. C. J. Nihoul, Ed., Katz, E., 1987: Seasonal response of the sea surface to the wind in Elsevier, 479–490. the equatorial Atlantic. J. Geophys. Res., 92, 1885–1893. Venzke, S., M. R. Allen, R. T. Sutton, and D. P. Rowell, 1999: The Lamb, P. J., 1978: Large-scale tropical Atlantic surface circulation atmospheric response over the North Atlantic to decadal patterns associated with sub-Saharan weather anomalies. Tel- changes in sea surface temperature. J. Climate, 12, 2562–2584. lus, 30, 240–251. Wang, C., 2002: Atlantic climate variability and its associated at- Latif, M., and A. Grötzner, 2000: The equatorial Atlantic oscilla- mospheric circulation cells. J. Climate, 15, 1516–1536.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC 1806 MONTHLY WEATHER REVIEW VOLUME 135

Wang, W., S. Saha, H.-L. Pan, S. Nadiga, and G. White, numerical model outputs. Bull. Amer. Meteor. Soc., 78, 2539– 2005: Simulation of ENSO in the new NCEP coupled fore- 2558. cast system model (CFS03). Mon. Wea. Rev., 133, 1574– Xie, S.-P., 1999: A dynamic ocean–atmosphere model of the tropi- 1593. cal Atlantic decadal variability. J. Climate, 12, 64–70. Xie, P., and P. A. Arkin, 1996: Analyses of global monthly pre- Yulaeva, E., and J. M. Wallace, 1994: The signature of ENSO in cipitation using gauge observations, satellite estimates, and global temperature and precipitation fields derived from the numerical model predictions. J. Climate, 9, 840–858. microwave sounding unit. J. Climate, 7, 1719–1736. ——, and ——, 1997: Global precipitation: A 17-year monthly Zebiak, S. E., 1993: Air–sea interaction in the equatorial Atlantic analysis based on gauge observations, satellite estimates, and region. J. Climate, 6, 1567–1586.

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC AUGUST 2007 CORRIGENDUM 3009

CORRIGENDUM

ZENG-ZHEN HU Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland

BOHUA HUANG Center for Ocean–Land–Atmosphere Studies, Calverton, Maryland, and Department of Climate Dynamics, George Mason University, Fairfax, Virginia

In Hu and Huang (2007), the values of the regression coefficients shown in Figs. 6 and 10 are incorrect because of a coding error. The correct values should be divided by the square root of the sample size. For the sample size of 23, the amplitudes of the regression coefficients in these figures should be smaller by a factor of 4.8 than shown in the figures. All the regression patterns and significance tests remain the same. The conclu- sions are not affected by this error.

REFERENCES

Hu, Z.-Z., and B. Huang, 2007: The predictive skill and the most predictable pattern in the tropical Atlantic: The effect of ENSO. Mon. Wea. Rev., 135, 1786–1806.

Corresponding author address: Zeng-Zhen Hu, Center for Ocean–Land–Atmosphere Studies, 4041 Powder Mill Road, Suite 302, Calverton, MD 20705. E-mail: [email protected]

DOI: 10.1175/MWR3554.1

Unauthenticated | Downloaded 09/25/21 09:14 AM UTC

MWR3554