Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 This study documents the variability of noise in the equatorial Pacific associated with the El Niño and La Niña events based on the ensemble retrospective forecasts of the Climate Forecast System. It is found that the noise (measured by the ensemble spread) in the western equatorial Pacific zonal wind stress is enhanced around and before the peak of El Niño events and weakened around 2–3 months after the peak of El Niño events. The change in the wind stress noise is communicated to the thermocline depth in the eastern equatorial Pacific with about 1 month time lag. The eastern equatorial Pacific SST noise, however, decreases during the decay stages of El Niño events while the corresponding noise in the surface zonal wind and heat flux increases. This decrease in the SST noise is related to the weakening of the SST front and the suppression of the tropical instability waves along the flanks of the equatorial Pacific cold tongue that is associated with the SST warming due to El Niño events. Similar relations are seen during La Niña events except that the changes are in opposite sense. As such, the signal-to-noise ratio and, thus, the predictability for the eastern equatorial Pacific SST is relatively high during warm events compared to cold events.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 There is an important distinction between the typical prescription of stochastic forcing (i.e., noise) as from Jin et al.  and the relatively high-frequency variability associated with WWB or the MJO discussed above. For example, white noise has a uniform distribution; in other words it has power on all space and time scales. No distinction is made between high frequencies and low frequencies. The stochastic forcing used by Jin et al.  is white although its amplitude is modulated by the ocean state, whereas if we interpret the MJO and WWB as noise, it has more power at relatively high frequencies as compared to ENSO.
 In the conceptual model of Jin et al. , the effects of stochastic atmospheric variability on the eastern equatorial Pacific (EEP) sea surface temperature (SST) anomalies are simply represented by a noise forcing term in which the state-dependent part is proportional to the magnitude of SST anomalies. As such, the state-dependent noise forcing is stronger when the SST anomalies are large than when the SST anomalies are small. The resulting ensemble spread for the EEP SST is larger for El Niño events than for La Niña events in their conceptual model. Can this contrast be corroborated using ensemble coupled model forecasts? Another question is how does the stochastic atmospheric variability feed back on ENSO? It is well known that the interannual wind anomalies in the western equatorial Pacific (WEP) can feed back on the EEP SST variations by generating equatorial Kelvin waves that modify the thermocline depth and then the SST in the EEP [Zebiak and Cane, 1987; Neelin et al., 1994; Wang et al., 1999]. Can the change in the ensemble spread of WEP winds lead to a change in the ensemble spread of EEP thermocline depth in a similar manner? If so, can the change in the ensemble spread of thermocline depth be transmitted to the SST in a manner similar to that in the interannual variations?
 The present study addresses these issues on the basis of ensemble retrospective forecasts of the Climate Forecast System (CFS) [Saha et al., 2006]. Our strategy is to use the ensemble forecast spread as an indicator of the noise. As such, we attempt to make no assumptions regarding statistics of the noise. Note that the noise defined using the ensemble spread, which includes impacts of the WWB and MJO, differs from other traditional definitions of noise. We also recognize that as lead time increases, a significant fraction of, say, the atmospheric spread is due to spread in the SST and vice versa. Spread in forecasts from a coupled model does not provide an ideal definition of noise as compared to, say, AMIP-type simulations. Nevertheless, our results indicate that variations in the WEP wind noise are directly related to WEP wind anomalies and changes in the WEP wind noise can lead to changes in the EEP thermocline noise. However, changes in the EEP SST spread cannot be explained by changes in the EEP thermocline spread. The activity of tropical Pacific instability waves (TIWs) that is modulated by ENSO seems to be the main factor for changes in the EEP SST spread.
 These TIWs are important because of their potential to change local stability of atmospheric boundary layer (ABL) through changes in the surface winds. Previous studies have shown clear evidence of air-sea interactions associated with the TIWs in the eastern tropical Pacific [e.g., Xie et al., 1998; Chelton et al., 2001; Hashizume et al., 2001; Xie, 2004]. For example, Chelton et al.  established an unambiguous feedback mechanism operating between the SST front associated with TIWs and divergence and curl of the near-surface wind stress. The TIW-related SST front influences the wind stress divergence and convergence with these quantities attaining their maxima when the wind blows perpendicular to the isotherms set up by the above mentioned SST fronts. With the use of surface pressure data in the equatorial Pacific during the TIW activity, Cronin et al.  showed that the SST-induced pressure gradient force is of first-order importance in the ABL momentum budget and, hence, plays a role in the cross-equatorial wind structure. With the use of a more advanced Tropical Rain Measuring Mission (TRMM) satellite's Microwave Imager (TMI) and QuickSCAT remote-sensing observations of SST, wind structures, and cloud water content, Xie  showed the inherent coupling of atmosphere and ocean at the mesoscale. Similar coupling arguments were made with the high-resolution regional coupled model used by Seo et al. .
 While these localized air-sea interactions in association with the TIWs are unresolved by current global climate models, there is a detectable rectification effect in fairly low-resolution atmospheric general circulation models (AGCMs). For instance, Jochum et al. [2007a] performed simulations with a moderate resolution AGCM to reveal that the stochastic nature of the inclusion of TIWs result in an increase of wind and precipitation variability on climatic time scales in the equatorial Pacific. Kessler et al. , Jochum and Murtugudde , Menkes et al. , and Jochum et al. [2007b] indicated that the TIWs play an important role in the meridional transport of heat and momentum in the tropical Pacific Ocean. Zhang and Busalacchi  suggested a cooling effect of the TIW-induced atmospheric wind feedback on the eastern equatorial Pacific SST through advection.
2. Data and Methodology
 The analyses in the present study are based on outputs from the National Centers for Environmental Prediction (NCEP) CFS 24 year retrospective forecasts for the period 1981–2004. The atmospheric component of the CFS is the NCEP Global Forecasting System model [Moorthi et al., 2001], which has a spectral triangular truncation of T62 in the horizontal and a finite differencing in the vertical with 64 sigma levels. The oceanic component is the Geophysical Fluid Dynamics Laboratory Modular Ocean Model version 3 (MOM3) [Pacanowski and Griffies, 1998], which has a zonal resolution of 1°. The meridional resolution is 1/3° between 10°S and 10°N, gradually increasing until 1° at 30°S and 30°N. There are 40 layers in the vertical with 27 layers in the upper 400 m. There are 15 forecasts for a specific month starting from 15 different atmospheric initial conditions that are partitioned into three segments. Each segment uses the same oceanic initial conditions. The atmospheric initial conditions are from the NCEP-Department of Energy Reanalysis 2 [Kanamitsu et al., 2002]. The oceanic initial conditions are from the NCEP Global Ocean Data Assimilation System (GODAS) (D. Behringer et al., personal communication, 2005). The GODAS SST is also used as a proxy for observations in calculating the forecast skill of SST. Each of the CFS retrospective forecasts covers a full 9 month period. These 9 months are denoted as target months and the first month is also denoted as the initial month in figures presented in this study. Details for the CFS retrospective forecasts are referred to Saha et al. .
 The analysis is performed for ensemble forecasts at different leads. We focus on SST and the 20°C isotherm depth (thermocline depth, d20) in the NINO3.4 region (5°S–5°N, 170°W–120°W), and surface zonal wind stress (tx) in the western equatorial Pacific (WEP) (5°S–5°N, 130°–170°E). These are critical quantities in the ENSO evolution [Zebiak and Cane, 1987; Neelin et al., 1994; McPhaden et al., 1998]. The ensemble spreads calculated as standard deviations of the 15 retrospective forecasts with respect to their ensemble means are used to represent the noise. Spread anomalies are obtained by removing mean seasonal cycle of spread. This is done separately for forecasts at different leads. Composite analysis is performed for six El Niño events (1982–1983, 1986–1987, 1991–1992, 1994–1995, 1997–1998, and 2002–2003) and four La Niña events (1984–1985, 1988–1989, 1998–1999, and 1999–2000) during the study period. The above analysis is mainly based on the CFS monthly means that are archived at 2° × 1° grid. We also use daily CFS forecasts of SST archived at 1° × 1° grid to calculate the spread and subseasonal standard deviation of SST. The subseasonal variations are obtained by removing a 61 day running mean from original daily means.
 We use 3 day mean SST from version 4 of the TRMM Microwave Imager (TMI) [Wentz et al., 2000] for the period 1998–2008. The original TMI data are on 0.25° × 0.25° grid. We have averaged the TMI data to 1° × 1° grid for comparison with the CFS. A 61 day running mean has been removed from the 3 day mean SSTs to obtain the subseasonal variations. The subseasonal SST standard deviation for TMI is calculated on the basis of both the 0.25° × 0.25° and 1° × 1° grids. The composite is made for three El Niño events (2002–2003, 2004–2005, and 2006–2007) and three La Niña events (1998–1999, 1999–2000, and 2007–2008) available during the TMI data period. We only present results from the 1° × 1° grid. The results from the 0.25° × 0.25° grid are similar except for the magnitude of the standard deviations is somewhat larger.
Figure 1 shows the composite of spread anomalies for the WEP tx (Figure 1a), NINO3.4 d20 (Figure 1b), NINO3.4 SST (Figure 1c), and NINO3.4 net surface heat flux (Figure 1e) in the CFS for El Niño events. The spread anomalies are calculated for each initial and target months. Then, the composite is constructed by averaging spread anomalies for the six El Niño events and the four La Niña events, respectively. The composite of the spread anomalies for La Niña events (figures not shown) tends to be opposite to that for El Niño events except for d20 whose spread anomalies are small around the mature phase of La Niña events because d20 cannot be well defined when the cold tongue is strong and the 20°C isotherm outcrops. Figure 1 also includes the composite of anomalies for the WEP tx (Figure 1d) and NINO3.4 SST (Figure 1f).
 There are several features of note in Figure 1 in terms of the spread. First, the WEP tx spread is largest in fall and winter (Figure 1a) during the developing and mature stages of El Niño events when there are warm SST anomalies in the NINO3.4 region (Figure 1f) and westerly wind anomalies in the WEP (Figure 1d). The spread is smallest in spring during the decaying stage of El Niño events when the WEP wind anomalies are easterly but the NINO3.4 SST anomalies are still positive. Second, the NINO3.4 d20 spread displays clear transitions from larger spread to smaller spread during late winter to early spring (Figure 1b). There appears to be a 1 month time lag between the NINO3.4 d20 and WEP tx spread variations. Third, the NINO3.4 SST spread anomalies are large and negative in February–May during the decaying stage of the El Niño events (Figure 1c). Fourth, the NINO3.4 net surface heat flux (NHF) spread anomalies are large and positive during the peak and decaying phases of the El Niño events (Figure 1e). Another feature to note is that the magnitude of negative anomalies in the NINO3.4 SST spread is larger in the decaying phase than in the developing phase of El Niño events (Figure 1c). In sections 3.1, 3.2, and 3.3, we discuss the relationship among WEP tx spread variations, WEP wind, and NINO3.4 SST anomalies; the relationship between WEP tx spread and NINO3.4 d20 spread variations; and variations of the NINO3.4 SST spread, respectively.
3.1. Relationship of Wind Spread Changes With Wind and SST Anomalies
 The relation of wind spread changes with interannual SST and wind anomalies has been discussed in previous studies. Kirtman et al.  showed that the zonal wind stress in the western-central equatorial Pacific has a larger spread during warm events. Consistently, Wu and Kirtman  identified larger rainfall spread in the western-central equatorial Pacific during warm events. Kug et al.  showed that the high-frequency zonal winds over the WEP display the largest variance in the developing stage of the EEP warm SST anomalies and the high-frequency wind variance over the central equatorial Pacific tends to vary in phase with the EEP warm SST anomalies. These studies indicate the modulation of the noise in the western-central equatorial Pacific winds by interannual SST anomalies. Recent studies suggest that the wind noise variations (at least limited to high frequencies) are more tied to the local wind anomalies than to ENSO-related SST anomalies in the tropical Pacific. For example, Seiki and Takayabu [2007a] showed that WWBs (considered as high-frequency variability compared to interannual low-frequency anomalies) tend to occur frequently under low-frequency westerlies. Sooraj et al.  found a better correlation of high-frequency atmospheric wind variability with local wind anomalies than SST anomalies.
 According to Figures 1a and 1f, the WEP tx spread variance is not in phase with the NINO3.4 SST anomalies for the entire duration of the warm events. For instance, while the WEP tx spread is large in the developing phase of El Niño events, it is small in the decaying phase. This is consistent with Kug et al.  who showed that strong high-frequency atmospheric wind variability over the WEP tends to occur prior to the mature phase of El Niño events. Comparison of Figures 1a and 1d indicates that the WEP tx spread varies in phase with the WEP tx anomalies. This in-phase variation is likely because in the westerly phase, the atmospheric convection is strong and internal atmospheric dynamics leads to large spread. These results suggest that the WEP tx spread variations are more closely related to WEP wind anomalies than with the EEP SST anomalies.
 The above relationship is further demonstrated in Figure 2 that compares the spatial distributions of composite tx spread anomalies with composite tx and SST anomalies in November–January (NDJ) and March–May (MAM). NDJ and MAM correspond to positive and negative tx spread anomalies in the WEP, respectively (Figure 1a). The warm SST anomalies in the equatorial Pacific are similar during the two periods though there is some difference in the magnitude (Figures 2c and 2f). The westerly anomalies show an obvious eastward shift from NDJ to MAM (Figures 2b and 2e). Before NDJ, positive tx spread anomalies are present over the western North Pacific, as are the westerly anomalies (figures not shown). In NDJ, positive tx spread anomalies are seen over the WEP (Figure 2a). In MAM, the positive tx spread anomalies move to the central equatorial Pacific and negative tx spread anomalies appear west of the date line (Figure 2d). The switch of tx spread anomalies from positive to negative over the WEP corresponds to the appearance of large easterly anomalies there (Figure 2e). The eastward move of positive tx spread anomalies is also consistent with the eastward extension of the warm pool (Figures 2e and 2f). This agrees with previous hypothesis that the warm pool extent drives the WWBs [Kessler et al., 1995; Eisenman et al., 2005; Gebbie et al., 2007].
Figures 3a and 3b show the frequency of occurrence (number of months) of the WEP tx spread anomaly with respect to the NINO3.4 SST anomaly and the WEP tx anomaly, respectively. Figures 3a and 3b are constructed by counting the number of months with the tx spread anomaly and the SST (tx) anomaly falling in the respective bins for all the forecasts with different initial months and different leads. Comparison of Figure 3a and Figure 3b supports the argument based on the results of Figures 1–2; i.e., the WEP tx spread anomalies have a better correlation with WEP tx anomalies than with NINO3.4 SST anomalies. For example, the largest (smallest) WEP tx spread tends to occur when the WEP tx has the largest positive (negative) anomalies (Figure 3b), whereas the WEP tx spread anomalies are small when the NINO3.4 SST has the largest positive anomalies, and the largest WEP tx spread is seen when the NINO3.4 SST anomalies are small or weak positive (Figure 3a). Figures 3c (3d) shows the simultaneous correlation between the WEP tx spread and NINO3.4 SST (WEP tx) anomalies for forecasts at different leads. In Figure 3c, the correlation is positive during June–July and September–February but is negative during March–April. In contrast, Figure 3d shows that the correlation between WEP tx spread and WEP tx anomalies is positive and large throughout the year except for August–September. The differences between Figures 3c and 3d indicate that the WEP tx spread variations are more robustly related to the WEP tx anomalies than the NINO3.4 SST anomalies.
 While the above analyses emphasize the importance of the WEP mean wind anomalies in the WEP wind spread variations, there is possibility that the EEP SST anomalies can indirectly affect the WEP spread anomalies by modulating the WEP mean wind changes. This indirect effect is likely larger in the developing phases than in the decaying phases of El Niño and La Niña events (Figure 3c).
3.2. Relationship of Wind and d20 Spread
 As seen in Figures 1a and 1b, the NINO3.4 d20 spread variations lag the WEP tx spread variations by about 1 month. A similar time lag between the WEP tx and NINO3.4 d20 prediction skill and signal-to-noise ratio in the CFS was identified by Wu et al. . The 1 month lag relationship is further demonstrated in Figure 4a, which shows the frequency of occurrence (number of months) of the WEP tx spread anomaly with respect to the NINO3.4 d20 spread anomaly with the latter lagging the former by 1 month. Figure 4c shows the correlation between the WEP tx spread anomalies and 1 month lag NINO3.4 d20 spread anomalies for forecasts starting at different months. The positive correlation is quite robust except between the forecasts of August–October WEP tx and September–November NINO3.4 d20.
 The 1 month time lag between the WEP tx and NINO3.4 d20 spread variations is consistent with that seen in the corresponding anomalies [Wu et al., 2009]. This suggests that variations in the WEP tx spread can be transmitted to the EEP d20 spread through equatorial Kelvin waves, similar to the interannual anomalies during the evolution of ENSO [e.g., Zebiak and Cane, 1987; Kessler et al., 1995].
 One question is whether there are contributions to the NINO3.4 d20 spread variations from the NINO3.4 tx spread anomalies. It turns out that the NINO3.4 tx spread anomalies vary in phase with the NINO3.4 SST anomalies (figure not shown), consistent with previous studies [Harrison and Vecchi, 1997; Tam and Lau, 2005; Kug et al., 2008; Seiki and Takayabu, 2007a]. The magnitude of the tx spread anomalies in the NINO3.4 region is only about half of that in the WEP region. The correlation between the NINO3.4 tx and d20 spread anomalies is generally weak with their correlation coefficients smaller than 0.3 except for forecasts of January–March NINO3.4 d20 (figure not shown). The frequency of occurrence of NINO3.4 d20 spread anomaly with respect to NINO3.4 tx spread anomaly does not show an obvious preference (figure not shown). The results suggest that the contributions of the local wind spread to NINO3.4 d20 spread variations are small.
3.3. Variations of the SST Spread
 The NINO3.4 SST spread is small in the warm phase, in particular, during February–June (Figure 1c). If the d20 spread variations contribute to the SST spread variations, small spread should be seen in the NINO3.4 d20 at a time lead of a few months. The NINO3.4 d20 spread, however, is larger than normal during November–March (Figure 1b). This suggests that the NINO3.4 SST spread variations cannot be explained by the NINO3.4 d20 spread variations. This is further demonstrated using Figure 4b, which shows the frequency of occurrence of the NINO3.4 d20 spread anomaly with respect to the NINO3.4 SST spread anomaly with the latter lagging the former by 4 months. This 4 month lag is chosen according to the time lag between interannual d20 and SST anomalies in the NINO3.4 region [Wu et al., 2009]. The other time lags display similar results. Apparently, there is no obvious correlation between the NINO3.4 d20 and SST spread variations. Figure 4d examines the correlation between NINO3.4 d20 spread anomalies and 4 month lag NINO3.4 SST spread anomalies for forecasts starting at different months. No large positive correlation is seen except between the forecasts of August d20 and December SST starting from May. For forecasts of November–February d20 and March–June SST, there is a notable negative correlation. This negative correlation suggests that the SST spread anomalies cannot be explained by d20 spread anomalies. This result is surprising in view of the close link between interannual variations of SST and thermocline depth in ENSO [e.g., Zebiak and Cane, 1987; McPhaden et al., 1998]. This will be investigated in the following.
 The NINO3.4 NHF spread is above normal in the decaying phases of El Niño events (Figure 1e), which is opposite to the NINO3.4 SST spread. This suggests that the variations in the NINO3.4 SST spread are not related to those in local NHF spread. The NHF spread anomalies are contributed by both surface latent heat flux and shortwave radiation spread anomalies (figures not shown) that tend to vary in phase with the NINO3.4 SST anomalies. The large surface latent heat flux spread in the NINO3.4 region during El Niño peak and decaying phases may be related to the large spread in local winds.
 The NINO3.4 SST spread variations can be related to activities of the TIWs along the SST fronts of the equatorial Pacific cold tongue [Legeckis, 1977; Chelton et al., 2001]. Previous studies showed that the TIW activity in the EEP is enhanced during La Niña events but suppressed during El Niño events [Contreras, 2002; Yu and Liu, 2003]. Such changes in the TIW activity are related to the modulation of ENSO on the latitudinal SST gradient across the SST fronts [Yu et al., 1995; Yu and Liu, 2003]. During El Niño events, the EEP cold tongue is weakened. The reduced SST fronts are unfavorable for the TIW activity. In contrast, during La Niña events, the EEP cold tongue is strengthened. The stronger SST fronts enhance the TIW activity. Previous studies also indicated that the TIWs affect the EEP SST changes [Jochum and Murtugudde, 2006; Menkes et al., 2006; Jochum et al., 2007b; Zhang and Busalacchi, 2008]. Because of high-frequency nature of the TIWs, the TIW induced SST changes vary for forecasts from different initial conditions, and thus, there will be large spread in SSTs.
 The interrelationship between the EEP SST spread changes and the TIW activity is indicated in the contrasts of SST spread and subseasonal SST standard deviation between El Niño and La Niña years in both the developing and decaying phases of ENSO. In comparison, the spatial structure of spread and subseasonal standard deviation distribution has a better indication of the activity of the TIWs in November–February (NDJF) than in MAM because climatologically the EEP cold tongue is relatively weak in MAM. Thus, here, we show the distribution of spread and subseasonal standard deviation in NDJF. Figures 5a and 5b show the six El Niño and four La Niña composite of SST spread, respectively, averaged for NDJF that are calculated from daily means of CFS ensemble forecasts with the initial month of July. Figures 5c and 5d show, respectively, the six El Niño and four La Niña composite of standard deviations of subseasonal SST variations averaged for NDJF that are calculated from daily means of CFS forecasts initiated at July 1. Figures 5e and 5f show the three El Niño and three La Niña composite of standard deviations of subseasonal SST variations, respectively, averaged for NDJF that are calculated from the 3 day mean TMI data. Figure 6 is similar to Figure 5 except for one El Niño (2002–2003) and one La Niña (1998–1999) event, both of which are common to CFS and TMI. For both Figures 5 and 6, we impose corresponding total SSTs (in contours) for convenience of examining the relationship of spread and subseasonal standard deviation with mean SST.
 There are several prominent features of note on Figure 5. First, both the spread and subseasonal standard deviation in the eastern equatorial Pacific are larger in La Niña years than in El Niño years. Second, the spread and subseasonal standard deviation show similar spatial distributions. In La Niña years, the subseasonal standard deviation displays relatively large values on both sides of the equator (Figures 5d and 5f). Such a feature is also visible in spread (Figure 5b) although not as clear as in subseasonal standard deviation. In comparison, the spread and standard deviation are smaller south of the equator than north of the equator. The large standard deviation south of the equator is located more westward in the TMI than in the CFS. In El Niño years, a band of relatively large subseasonal standard deviation is seen near the equator (Figures 5c and 5e). The large spread, however, displays a clear meridional extension (Figure 5a). Third, both large spread and subseasonal standard deviation are found along relatively large latitudinal SST gradients on flanks of the cold tongue. In La Niña years, large latitudinal SST gradient is seen both north and south of the equator with the north one stronger than the south one. The south one is located more westward in the TMI than in the CFS. In El Niño years, the latitudinal SST gradient is relatively large only north of the equator. These features agree well with those seen in the standard deviation and subseasonal standard deviation.
 The contrast between El Niño and La Niña years, the consistency between spread and standard deviation distribution, and the close relationship of spread and subseasonal standard deviation with SST are further demonstrated in Figure 6 that is based on the 2000–2003 El Niño and the 1998–1999 La Niña for both the CFS and TMI. Apparently, the spread and standard deviation are larger in 1998–1999 than in 2002–2003. In 2002–2003, relatively large standard deviation is seen near the equator, so the latitudinal SST gradient (Figures 5c and 5e). The spread extends meridionally across the equator (Figure 6a). In 1998–1999, two relatively large bands of spread and subseasonal standard deviation are seen on both sides of the equator along large latitudinal SST gradients (Figures 6b, 6d, and 6f). These large spreads or standard deviations tend to merge across the equator. The magnitude south of the equator is much larger compared to the composite and comparable to that north of the equator. As in the composite, the large standard deviation south of the equator is located more westward in the TMI than in the CFS, and so is the large latitudinal SST gradient.
 The results from Figures 5 and 6 clearly indicate that the large subseasonal variability in the EEP is due to activity of TIWs along the SST fronts on the flanks of the eastern Pacific cold tongue. During El Niño years, the cold tongue is weakened, and so is the activity of the TIWs. As a result, the subseasonal SST variability is significantly reduced. In contrast, during La Niña years, the cold tongue and the associated activity of the TIWs are intensified, enhancing the subseasonal SST variability. The consistency between spread and subseasonal variability in the EEP suggests that they are closely related. When the subseasonal variability is large, there may be large differences among forecasts from different initial states. This presumably will lead to large ensemble spread. The large subseasonal variance in SST during La Niña years is not limited to the SST front region but extends to the equator (Figures 5d, 5f, 6d, and 6f). It appears that the high-frequency SST variance in relation to the TIW activity extends to the equator by advection of ocean currents and/or diffusion. In relation, the large ensemble spread appears across the equator (Figure 5b and 6b).
 In interannual variations of the EEP SST, the thermocline effect is the dominant term [e.g., Zebiak and Cane, 1987]. The effects due to the meridional transport of heat in relation to the TIWs may be secondary, presumably because these effects are mostly on relatively high frequency. In interannual variations of the EEP SST spread, the impacts due to the activity of TIWs are very large, which may be because the TIW related large spread anomalies are maintained for a relatively long period following interannual SST anomalies. As a result, the effects of the TIWs may overcome those due to the spread in the thermocline depth. As such, it appears that the variations in the EEP SST spread are not related to those in the thermocline depth spread.
 The contrasting SST spread anomalies between the El Niño and La Niña events indicates that the signal-to-noise ratio is higher for warm events than for cold events. This is confirmed by the composite of the ratio between the magnitude of SST anomalies and the total SST spread in the NINO3.4 region for El Niño and La Niña events (Figure 7). Apparently, the signal-to-noise ratio is much larger during El Niño years (Figure 7a) than during La Niña years (Figure 7b). This contrast is especially prominent during the decaying phases.
 The decrease in the NINO3.4 SST spread is larger in the decaying phase than in the developing phase of El Niño events (Figure 1c). This indicates that the impacts of TIW activity on the EEP SST spread are larger during spring. Climatological mean spread estimated from the CFS ensemble forecasts is larger in summer and fall than in spring (figures not shown). This leads to a smaller total spread in the decaying phase than in the developing phase of El Niño events. As a result, the signal-to-noise ratio for the NINO3.4 SST is high in the decaying phase than in the developing phase of El Niño events (Figure 7a). For La Niña events, the contrast in the signal-to-noise ratio is less obvious (Figure 7b).
 Consistent with the contrast in the signal-to-noise ratio, the NINO3.4 SST prediction skill in the CFS is not as good for La Niña events as for El Niño events. This is demonstrated in Figure 8a that shows the frequency of occurrence of the NINO3.4 SST anomaly deviation of CFS ensemble forecasts from observations with respect to the observed NINO3.4 SST anomalies. The deviation is apparently larger for normal and negative SST anomalies and smaller for positive SST anomalies. The composite root-mean-square errors for ensemble forecast NINO3.4 SST anomalies are larger during La Niña years than during El Niño years during most of the time from August to May (Figures 8b and 8c). During La Niña events, the largest root-mean-square errors are seen in March (Figure 8c), which are about double of those corresponding to El Niño events (Figure 7b).
 The NINO3.4 SST prediction skill shows notable differences between the developing and decaying phases of ENSO. The root-mean-square errors are larger during the decaying phases than during the developing phases of both La Niña and El Niño events (Figures 8b and 8c). This suggests that the ENSO growth phase is more predictable than the decay phase, which is consistent with Jin and Kinter  and Jin et al. . The contrast of the prediction skill between the El Niño developing and decaying phases, however, is inconsistent with the corresponding contrast of the ensemble spread (Figure 1c) and the signal-to-noise ratio (Figure 7a). This inconsistency is partly because the seasonal variations of the root-mean-square errors for the NINO3.4 SST anomalies differ from the seasonal variations of the ensemble spread (figures not shown). After removing the seasonal cycle, the root-mean-square errors (Figures 8d and 8e) show a better correspondence with the ensemble spread and the signal-to-noise ratio for most forecasts with the time lead less than 6 months.
 Analysis based on the CFS ensemble retrospective forecasts indicates that the variability of noise in the WEP tx is closely related to that of the WEP mean tx. The WEP tx spread is large during the developing phase of El Niño events when the WEP tx anomalies are westerly and small during the decaying phase of El Niño events when the WEP tx anomalies switch to easterly. Opposite spread anomalies are seen during La Niña events.
 The variations of the NINO3.4 d20 spread lag those of the WEP tx spread by about 1 month. This time lag is the same as that corresponding to interannual anomalies. This suggests a connection between the WEP tx spread and NINO3.4 d20 spread variations through equatorial waves in a manner similar to the connection between interannual wind and thermocline depth anomalies [Zebiak and Cane, 1987; Kessler et al., 1995].
 The NINO3.4 SST spread is small during the El Niño events, especially during the decaying phase of El Niño events. Such spread variations cannot be explained by the equatorial Pacific wind and heat flux spread variations or thermocline depth spread variations. Instead, they are likely related to the activity of the TIWs along the SST fronts of the Pacific cold tongue. During El Niño events, the cold tongue is weakened, leading to weaker SST fronts. This suppresses the activities of the TIWs. As a result, the spread in the EEP SST is reduced. The SST fronts are strong during the decaying phase of La Niña events due to the enhanced SST fronts of the cold tongue, which strengthens the activities of the TIWs. The contrasting spread suggests a larger SST signal-to-noise ratio for El Niño events than for La Niña events. Correspondingly, the prediction skill for the NINO3.4 SST is higher for El Niño events than for La Niña events.
 The results of the present analysis based on the CFS retrospective forecasts disagree with Jin et al.'s  conceptual model. The conceptual model assumes that the state-dependent noise varies in phase with the amplitude of SST anomalies. In the CFS, the largest spread in the WEP tx leads the peak warm NINO3.4 SST anomalies, in agreement with the observational evidence [Kug et al., 2008]. The conceptual model indicates that the EEP SST spread is larger in El Niño events than in La Niña events. The CFS derived NINO3.4 SST spread is smaller in El Niño events than in La Niña evens. Jin et al.  suggested that the El Niño events are less predictable than the La Niña events due to enhanced ensemble spread. Our results indicate the opposite. This suggests that the small-scale and high-frequency features, such as the TIWs, need to be resolved, e.g., using high-resolution dynamical models, in order to capture their impacts on ENSO and its predictability.
 The authors appreciate helpful comments of Ben Cash. The constructive comments of three anomalous reviewers have greatly improved this work. This research was supported by grants from the NSF (ATM-0332910), the NOAA (NA04OAR4310034 and NA05OAR4311135), and the NASA (NNG04GG46G). B.P.K. acknowledges support from NOAA grants NA17RJ1226, NA080AR4320889 and NSF grants OCI0749165 and ATM0754341.