Abstract

A significant part of the West African monsoon (WAM) interannual variability can be explained by the remote influence of El Niño–Southern Oscillation (ENSO). Whereas the WAM occurs in the boreal summer, ENSO events generally peak in late autumn. Statistics show that, in the observations, the WAM is influenced either during the developing phase of ENSO or during the decay of some long-lasting La Niña events. The timing of ENSO thus seems essential to the teleconnection process. Composite maps for the developing ENSO illustrate the large-scale mechanisms of the teleconnection. The most robust features are a modulation of the Walker circulation and a Kelvin wave response in the high troposphere.

In the Centre National de Recherches Météorologiques Coupled Global Climate Model, version 3 (CNRM-CM3), the teleconnection occurs unrealistically at the end of ENSO events. An original sensitivity experiment is presented in which the ocean component is forced with a reanalyzed wind stress over the tropical Pacific. This allows for the reproduction of the observed ENSO chronology in the coupled simulation. In CNRM-CM3, the atmospheric response to ENSO is slower than in the reanalysis data, so the influence on the WAM is delayed by a year.

The two principal features of the teleconnection are the timing of ENSO onsets and the time lag of the atmospheric response. Both are assessed separately in 16 twentieth-century simulations of the third phase of the Coupled Model Intercomparison Project (CMIP3). The temporal aspects of the ENSO teleconnection are reproduced with difficulty in state-of-the-art coupled models. Only four models simulate an impact of ENSO on the WAM during the developing phase.

1. Introduction

Over the Sahelian region, the livelihoods and even the lives of most people are directly tied to the amount of monsoon rainfall in the boreal summer. Any progress in the understanding and forecasting of the African climate is therefore of great importance. The causes of the interdecadal variability of Sahelian rainfall and the possible impacts of global warming are still open questions. In addition to the interdecadal variability, Sahelian rainfall also presents a strong interannual variability, which can lead to extremely dry years or to floods, with dramatic impacts on agriculture (Tarhule and Lamb 2003). A key factor of this interannual variability is the influence of El Niño–Southern Oscillation (ENSO). The influence of ENSO on seasonal rainfall in the tropics has been extensively examined since the pioneering studies by Walker in the 1920s, in which he tried to understand and predict the variations in the Indian monsoon. However, it is only in the 1990s that the influence of ENSO on Sahelian rainfall was firmly recognized (Semazzi et al. 1988; Rowell et al. 1992; Janicot et al. 1996). With the exception of Rowell (2001) and Janicot et al. (2001), very few authors document precisely the dynamical processes of the remote influence of ENSO on the West African monsoon (WAM). A better understanding of this interannual teleconnection seems, however, essential to improve seasonal forecasts over this sensitive region.

Whereas an atmospheric general circulation model (GCM) forced with a given field of sea surface temperature (SST) is quite useful to reproduce the observed variability and to perform various sensitivity experiments, using a coupled ocean–atmosphere GCM is essential to simulate and forecast ocean–atmosphere feedbacks that play an important role in the teleconnections. Several studies show that the coupling improves the impact of SST on the simulated tropical variability (e.g., Peña et al. 2004; Krishna Kumar et al. 2005; Bracco et al. 2007). However, coupled models still have large regional biases and do not reproduce perfectly the observed modes of variability. It is therefore important to further improve them. From this point of view, the recent simulations achieved in the framework of the third phase of the Coupled Model Intercomparison Project (CMIP3) for the fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR4) are a unique opportunity to better understand the variability of the WAM, and its biases, in state-of-the-art coupled models. Concerning the African climate, most papers based on the IPCC AR4 CMIP3 database focus on the decadal and long-term evolution of Sahelian rainfall (Lau et al. 2006; Biasutti and Giannini 2006; Hoerling et al. 2006; Cook and Vizy 2006). Whereas these studies emphasize the relevance of SST–monsoon relationships in the detection of the anthropogenic climate change over West Africa, there are fewer studies addressing the issue of the interannual variability of the WAM. In their paper that addresses the influence of the oceans on the WAM, Joly et al. (2007) show that most IPCC AR4 coupled models present a single SST–WAM teleconnection at the interannual time scale. This teleconnection always involves ENSO, but SST patterns are often different from those observed.

In the observations, SST anomalies in the equatorial Pacific are associated with rainfall anomalies of the opposite sign over the Sahel (Fig. 1a). The Météo-France Centre National de Recherches Météorologiques Coupled Global Climate Model, version 3 (CNRM-CM3) is an example of a coupled model with a strong ENSO–WAM teleconnection that differs from the observed one. The SST pattern in the model has its main loading in the western equatorial Pacific, but also involves significant SST anomalies both in the Atlantic and Indian Oceans (Fig. 1b). The precipitation pattern is concentrated in the eastern Sahel, which is the region of strongest rainfall variability in the model. Note that the teleconnection explains 43% of precipitation interannual variance in the simulation, against 24% in the observations.

Fig. 1.

MCA of WAM rainfall (0°, 20°N; 20°E, 35°W) and equatorial Pacific SST (10°S, 10°N; 160°E, 80°W) in JAS (a) for the observations (CRU and HadC) and (b) for the CNRM-CM3 simulation. The (left) rainfall homogeneous maps and the (right) SST heterogeneous maps display the regression (contours) and correlation (color shading) of the rainfall time series with the given field. The contour spacing is 0.2 mm day−1 for rainfall and 0.2°C for SST, and the thick gray contour is for the zero isoline. Only the correlations significant at the 95% level are color shaded (bootstrap procedure).

Fig. 1.

MCA of WAM rainfall (0°, 20°N; 20°E, 35°W) and equatorial Pacific SST (10°S, 10°N; 160°E, 80°W) in JAS (a) for the observations (CRU and HadC) and (b) for the CNRM-CM3 simulation. The (left) rainfall homogeneous maps and the (right) SST heterogeneous maps display the regression (contours) and correlation (color shading) of the rainfall time series with the given field. The contour spacing is 0.2 mm day−1 for rainfall and 0.2°C for SST, and the thick gray contour is for the zero isoline. Only the correlations significant at the 95% level are color shaded (bootstrap procedure).

The present paper aims at understanding the processes that make the teleconnection simulated by state-of-the-art coupled models so different from the observed one. In the first part, the atmospheric processes of the ENSO–WAM teleconnection are reviewed in the atmospheric reanalyses. The importance of the timing of ENSO events is highlighted, and the main large-scale atmospheric processes are detailed. In the second part, to explain the behavior of the CNRM-CM3 coupled model, an original sensitivity experiment is conducted. The ocean component is forced with a reanalyzed wind stress over the tropical Pacific and fully coupled with the atmospheric model over the rest of the globe. The aim is to improve the timing of the simulated ENSO and to assess the consequences for the ENSO–WAM teleconnection. In the third part, the study is extended to a larger set of coupled simulations produced for the IPCC AR4. This brings us to interesting concluding remarks on ENSO teleconnections in state-of-the-art coupled models.

2. Statistical tools

a. Data filtering

In this study, the focus is on the interannual variability. It is therefore necessary to filter any long-term change or interdecadal variability in both the observed and simulated time series. The series are first detrended linearly, and then filtered using the fast Fourier transform (Press et al. 1992). A low-pass filter with a cutoff at 8.8 yr is applied to the yearly seasonal means, while a bandpass filter is used for the monthly data to filter out both the annual cycle and the low frequency.

b. Principal components analysis

The principal components analysis (PCA; Hannachi et al. 2007) is employed to define monthly varying ENSO indexes. In the observations, the obtained ENSO time series is almost identical to the usual Niño-3.4 index. But in the models, as the shape and the evolution of the simulated ENSO differ sometimes significantly from the observed one (van Oldenborgh et al. 2005; Capotondi et al. 2006; Joseph and Nigam 2006; Leloup et al. 2007), the PCA is undoubtedly a more refined method than the rough average of SST anomalies in a fixed box.

c. Maximum covariance analysis

The maximum covariance analysis (MCA; Bretherton et al. 1992) can be considered a generalization of the PCA. In the present study, the MCA is employed to identify the regions in West Africa where rainfall covaries with ENSO and to extract the corresponding time series. Each pair of spatial patterns describes a fraction of the covariance [quantified by the squared covariance fraction (SCF)], and the correlation between the time series of the two variables (Corr) indicates how strongly related the coupled patterns are. Significance levels are estimated using a moving block bootstrap procedure (Wilks 1997).

d. Composite maps

To gain deeper insight into the processes of the ENSO–WAM teleconnection, composite maps are computed in section 3. The significance of the composites is assessed with a bootstrap technique, which is more appropriate than the ordinary Student’s t test (Nicholls 2001). For further details, see the appendix 1 of Terray et al. (2003).

3. The ENSO–WAM teleconnection in two atmospheric reanalyses

a. Observation and reanalysis data

Two observation datasets are used in this study:

  • The Climate Research Unit (CRU2) precipitation climatology (hereafter, CRU), provided at the 0.5° resolution for the 1901–2002 period, and based only on rain gauge measurements (Mitchell and Jones 2005).

  • The Hadley Centre Sea Ice and Sea Surface Temperature dataset (HadISST1) climatology (hereafter, HadC), provided at the 1° resolution for the 1870–2002 period (Rayner et al. 2003).

The quality of the data greatly increased over the twentieth century. This study concentrates on the period 1958–2001, which is also covered by atmospheric reanalyses. Two reanalysis products are used in the following:

  • The National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis-1 (hereafter, NCEP; Kalnay et al. 1996).

  • The 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005).

The assimilation of satellite observations considerably improves the quality of reanalyses over the last three decades (Sturaro 2003) but does not guarantee their temporal homogeneity (dell’Aquila et al. 2005). Despite our prior filtering of the low frequency in all datasets, the lack of reliability of reanalysis data before 1968 might have an impact on our results over Africa (Poccard et al. 2000). Both the NCEP and ERA-40 reanalyses are therefore used in parallel all along the study to try to assess the robustness of the results. Note that both observation and reanalysis data are interpolated onto a 128 × 64 horizontal grid (2.8° resolution).

b. Selecting the appropriate years

Figure 1a illustrates in the observations the well-established relationship between the July–September (JAS) SST anomalies in the equatorial Pacific and JAS rainfall anomalies over Africa. Whereas the WAM takes place in the boreal summer, ENSO events generally peak in late autumn. Therefore, our first question is whether the SST pattern corresponds to the beginning, the end or involves both stages of ENSO. No study has as yet addressed this issue precisely, despite the fact that it has been extensively documented for the Asian monsoons (Chang et al. 2004). To find out which stage of ENSO events influences the WAM most, it is first necessary to characterize each year, both in terms of rainfall and ENSO. The use of the PCA and MCA instead of averaging variables in a fixed box is justified by the fact that the procedure is also intended for the model simulations, in which the main patterns of variability are sometimes displaced compared with the observations (Fig. 1).

  • ENSO events: As mentioned in section 2b, the ENSO index is defined as the first time series of the PCA of filtered monthly SST anomalies in a large equatorial Pacific region (10°S, 10°N; 160°E, 80°W). Based on Trenberth (1997), ENSO events are defined as periods of at least six months during which the index exceeds half of its standard deviation. Compared with Trenberth (1997), we get a good agreement. In two cases however (1967 and 1981), our procedure detects a long-lasting La Niña event that is absent in Trenberth (1997). This is not due to the use of a PCA but rather to the filtering of the data. Fedorov and Philander (2001) argue for low-frequency filtering, which emphasizes the cold phases in the last decades of the twentieth century.

  • Rainfall events: The index chosen here is the first rainfall time series calculated with the same MCA as in Fig. 1. JAS seasons that outmatch two-thirds of the standard deviation are selected as positive or negative years. With the observations, an ordinary Sahel rainfall index gives similar results. However, a procedure based on the MCA is more versatile, as it can be applied to the simulations independently of the climatology of the model.

An important feature of the observed ENSO–WAM teleconnection is that Sahelian rainfall anomalies are correlated with SST anomalies of the opposite sign in the equatorial Pacific (Fig. 1a). Therefore, let us select the years with co-occurrence of a rainfall event (in JAS) and an ENSO event of the opposite sign (during the year). Our procedure automatically discards the years with an ENSO onset that is too late (i.e., during or after JAS) to have a marked influence on the WAM. Figure 2 shows the time evolution of the ENSO index for the years selected over the period 1958–2001.

Fig. 2.

Time evolution of the normalized ENSO index (HadC) for each of the years selected with the following criteria. (left) Co-occurrence of a weak monsoon in JAS and an El Niño event during part of the year. (right) Co-occurrence of a strong monsoon in JAS and a La Niña event during part of the year. The horizontal dashed lines indicate the 0.5 level used to detect the onset months.

Fig. 2.

Time evolution of the normalized ENSO index (HadC) for each of the years selected with the following criteria. (left) Co-occurrence of a weak monsoon in JAS and an El Niño event during part of the year. (right) Co-occurrence of a strong monsoon in JAS and a La Niña event during part of the year. The horizontal dashed lines indicate the 0.5 level used to detect the onset months.

  • El Niño teleconnection: Except in 1987, the teleconnection always occurs at the beginning of an El Niño event (1963, 1965, 1972, 1976, and 1997).

  • La Niña teleconnection: If we discard the 1974 outlier season, the selected years can be divided into two groups:

  • In four cases (1964, 1970, 1975, and 1988) the strong monsoon is associated with a La Niña that appears in April, May, or June.

  • In four cases (1967, 1971, 1981, and 1999) there is a long-lasting cold event in the Pacific that began the preceding year.

These results are the starting point of our study and show the relevance of taking into account the timing of ENSO events when studying the influence on the WAM. This will be further illustrated with the coupled simulations. The three groups of years highlighted in italic above are the three categories of ENSO teleconnections that will be considered hereafter. Based on this selection, it is possible to compute composite maps for each of the three cases. For most of the atmospheric variables that we studied, results turned out very symmetric in the developing El Niño and developing La Niña cases, with, however, weaker statistical significance in the cold case. This might be due to the weaker SST anomalies during La Niña events (and thus weaker atmospheric anomalies) and to the number of years for the significance testing (only 4 yr). Because our purpose is not to assess in detail the nonlinearities of the ENSO–WAM teleconnection, it has been decided to present in this paper only two kinds of maps: (i) the developing ENSO 9-yr composites (El Niño minus La Niña), and (ii) the mature La Niña 4-yr composites.

c. “Developing ENSO” teleconnection

The “developing ENSO” composites are computed by averaging the five El Niño anomalies together with the four La Niña anomalies multiplied by −1. This is different from calculating the difference between the average El Niño anomalies and the average La Niña anomalies, which would give the same weight to both subsamples and would approximately double the amplitude of the anomalies.

The SST composite (not shown) is identical to the heterogeneous map from the MCA (Fig. 1a). Figures 3a,b compare the rainfall composites with two observed datasets: CRU and Global Precipitation Climatology Centre (GPCC; http://gpcc.dwd.de/). In both datasets, the WAM anomaly is associated with significant anomalies over the north of South America and over India. Over West Africa, the negative pattern extends further to the north with GPCC. Over East Africa, the anomaly is stronger with CRU.

Fig. 3.

Developing ENSO JAS composites (mean of the El Niño and the minus La Niña anomalies). The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is precipitation 0.3 mm d−1, OLR 5 W m−2, KHI200 0.5 m2 s−1, U200 2 m s−1, T 0.1°C, T200 0.3°C, Z 2 mgp, Z200 5 mgp, Z850 2 mgp, PSI925 1 m2 s−1. OLR is taken so that positive anomalies indicate enhanced convection. The longitude–height cross sections are along the equator (10°S–10°N average), and the dashed lines correspond to the Greenwich meridian, the date line, and the 850- and 200-hPa pressure levels. Pressure levels range linearly from 1000 to 200 hPa.

Fig. 3.

Developing ENSO JAS composites (mean of the El Niño and the minus La Niña anomalies). The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is precipitation 0.3 mm d−1, OLR 5 W m−2, KHI200 0.5 m2 s−1, U200 2 m s−1, T 0.1°C, T200 0.3°C, Z 2 mgp, Z200 5 mgp, Z850 2 mgp, PSI925 1 m2 s−1. OLR is taken so that positive anomalies indicate enhanced convection. The longitude–height cross sections are along the equator (10°S–10°N average), and the dashed lines correspond to the Greenwich meridian, the date line, and the 850- and 200-hPa pressure levels. Pressure levels range linearly from 1000 to 200 hPa.

Figure 3 shows a selection of composites based on reanalysis data. The outgoing longwave radiation (OLR) is used here as an estimate of the convection in the ITCZ. Over the equatorial Pacific, the SST anomaly causes a convective anomaly in the ITCZ, which has a different shape in NCEP and ERA-40 (Figs. 3c,d). Over Africa, the response of the OLR is weak in both reanalyses and is part of a negative pattern that extends from India to Africa. The positive anomalies on both sides, especially over the Mediterranean, are coherent with observed rainfall (Figs. 3a,b). Some strong temperature and geopotential anomalies are also observed near the tropopause (Figs. 3i,j), which might indicate a modulation of the regional Hadley cell.

The main result here is that both the NCEP and ERA-40 reanalyses have difficulties in simulating the observed influence of ENSO on the convection over the Sahel. With the reanalyses available for the period 1958–2001, it is thus not possible to study the regional details of the WAM convective response to ENSO. In the following, we concentrate therefore on the large-scale mechanisms that link ENSO with Africa. Only the boreal summer (JAS) maps computed with ERA-40 are discussed, because on such a large scale the two reanalysis products are in good agreement. For the sake of simplicity, the El Niño minus La Niña maps will be discussed in terms of response to El Niño (which corresponds to the sign of the anomalies on the maps). It should be kept in mind that the signal is very linear, so all the conclusions are valid for the La Niña case but in the opposite sense.

1) The large-scale circulation

A common explanation for ENSO teleconnections is that, during El Niño, there is an anomalous ascending motion over the equatorial Pacific, which results in the Walker circulation being displaced to the east. This is illustrated in Fig. 3e by the composite of the velocity potential (KHI) at 200 hPa: the strong seesaw anomaly reflects the shift in the equatorial Walker circulation. The negative anomaly over the central and eastern Pacific corresponds to the divergent response to the enhanced convection, and the positive anomaly over the Indian Ocean corresponds to the anomalous convergence, and thus anomalous subsidence. Note that this anomalous convergence extends over Africa. Therefore, the first possible mechanism for the teleconnection is that the anomalous subsidence favors droughts in the Sahel at the beginning of El Niño events. This basic description of the ENSO teleconnection is, however, insufficient. Velocity potential maps only bring out the very largest (i.e., planetary) scales in the divergence pattern. Sardeshmukh and Hoskins (1985) argue that the notion of the Walker cell as a simple divergent, thermally direct overturning in the vertical plane passing through the equator should be viewed with some suspicion.

Figure 3f shows the zonal wind anomaly at 200 hPa. The large positive (i.e., westerly) pattern between 90°W and 90°E reveals a strong weakening of the tropical easterly jet (TEJ), with a symmetric pattern about the equator, which suggests that this may be a Kelvin wave response (Rowell 2001). Nicholson and Grist (2003) suggest that east of 20°E, the location of the TEJ may promote convection, whereas west of 10°E it is rather the convection that contributes to the TEJ. Therefore, the influence of ENSO on the TEJ over eastern Africa is a second way of explaining the modulation of WAM rainfall at the beginning of ENSO events.

2) The temperature response

Another description of ENSO teleconnections has recently gained importance in the community. Because the tropical atmosphere cannot support large temperature gradients, the anomalous convective heating over the tropical Pacific spreads rapidly through the entire troposphere. Chiang et al. (2002) explain that such free-troposphere temperature anomalies can lead to a stabilization of the atmospheric column in the remote tropical regions climatologically affected by deep convection. According to Giannini et al. (2005), such a “stabilization mechanism” could explain the reduction of Sahelian rainfall during El Niño events. Figure 3g shows that, at the beginning of El Niño events, the low layers of the troposphere get warmer over the equatorial Pacific (up to 2°C in our composite). The warmer and moister boundary layer enhances the local convection, which leads to a warm anomaly in the higher levels (around 1°C), because of the anomalous latent heat release. However, the most interesting feature in Fig. 3g is that the anomaly in the high troposphere extends eastward right over Africa. The horizontal composite at 200 hPa confirms this result (Fig. 3i). Over the Pacific, there are two warm cores away from and straddling the equator, and the warm anomaly extends along the equator, as predicted by the linear theory (Gill 1980). This is consistent with a Kelvin wave response: the convective heating anomaly over the tropical Pacific is communicated to the large-scale flow over Africa via planetary waves. As a consequence of this direct atmospheric influence of ENSO, the hypothesis is that rainfall may be reduced as a whole over Africa (Giannini et al. 2005). However, the temperature anomaly in Fig. 3i is centered on the equator and does not affect the Sahelian band. Is there really an impact on Sahelian rainfall? Neelin et al. (2003) discuss in detail the response of tropical convection to ENSO. They claim that, while the warming is indeed at the heart of the teleconnection, the moist feedbacks yielding the rainfall anomaly differ from region to region. A warm troposphere increases the value of surface boundary layer moisture required for the convection to occur. In regions of plentiful moisture supply, moisture simply rises to maintain precipitation, but this increases the moisture gradient relative to neighboring subsidence regions. Reductions in rainfall then result for those margins of convection zones that have a strong inflow of air from the subsidence regions and less frequently meet the increased “ante” for convection. This is what is called the “upped-ante mechanism,” and is actually a plausible mechanism for Sahel droughts.

3) The geopotential response

Figure 3h displays the same vertical cross section as for the temperature, but for the geopotential. Over the Pacific, associated with the convective anomaly, there is a negative anomaly in the low troposphere and—by “mirror effect”—the associated positive anomaly in the high troposphere. As for the temperature, the anomaly extends eastward at 200 hPa (Fig. 3j), which is reminiscent of the Kelvin wave response.

The vertical cross section of the geopotential anomalies in Fig. 3h holds another interesting feature. A strong geopotential anomaly over the Indian Ocean extends westward over Africa in the low and midtroposphere. This anomaly is certainly part of the large-scale shift of the Walker circulation (cf. the KHI200 dipolar anomaly). We would thus expect a negative anomaly aloft, which is actually neutralized by the counteracting positive anomaly because of the Kelvin wave response at these levels. The 850-hPa composite (Fig. 3l) shows that the compensating subsidence extends over Africa. According to Rowell (2001), the changes in low-level height over the Sahel change the circulation, so this reduces the moisture flux into the Sahel. To test this hypothesis, Fig. 3k shows the anomaly of the streamfunction at 925 hPa. Since the total moisture flux in the tropics is dominated by its rotational, stationary component (Chen 1985; D’Abreton and Tyson 1995), low-level streamfunction (PSI) anomalies should capture the main aspects of the moisture anomalous transport during El Niño (Lyon and Mason 2007). Over the Indian Ocean and the Gulf of Guinea, the PSI925 climatologically displays a maximum centered on the equator (not shown), which corresponds to the anticyclonic circulation of the transequatorial trade winds in the Indian and African monsoons. Therefore, the significant negative anomalies in that region indicate that the rotational component of the trade winds is weakened, which leads to a reduced moisture transport toward the monsoon system. This confirms the results from Janicot et al. (2001) that show that the extension of the moisture advection over West Africa is modified during ENSO events.

In conclusion, the vertical cross section of geopotential anomalies along the equator is quite interesting, as it provides a signature of the two main mechanisms of the ENSO teleconnection: the first one (baroclinic) in the high troposphere (the Kelvin wave response) and the other one (barotropic) in the low troposphere (that originates from the Indian Ocean). Jury et al. (2002) show that, at a 6-months lead, surface pressure in the north Indian Ocean is a key factor in determining the West African climate. Interestingly, surface pressure over the Indian Ocean is also crucial for the Indian monsoon (Huang and Shukla 2007). This explains the large pattern in the OLR composites that extends from India to Africa: there may be a link between the Indian and West African monsoons at the interannual time scale.

4) Summary

There can be no simplistic explanation for the ENSO–WAM teleconnection. In particular, the attracting “stabilization mechanism” with the sole action of planetary waves seems insufficient. The 200-hPa temperature anomaly over equatorial Africa might impact the convection over the Sahel, either via the “upped-ante mechanism” (Neelin et al. 2003) or via the wind anomaly that modulates the TEJ. However, the main mechanism of the teleconnection after the onset of ENSO events seems to be the anomalous large-scale subsidence, with different pathways: either directly via a modulation of the convection or indirectly via a modification of the low-layer moisture flux.

d. “Mature La Niña” teleconnection

For the four selected years, the averaged SST anomaly hardly reaches 0.5°C in the equatorial Pacific (Fig. 4a), which is not significant in regard to the high variability of SST in that region. The rainfall anomaly is strong (Fig. 4b); however, the significance level is hardly reached, probably because of the small size of the sample and the strong dispersion in it. The mature La Niña case is therefore a difficult issue. For such long-lasting events, all the dynamical fields indicate a lack of convective response over the Pacific (not shown). There is no significant modulation of the Walker circulation. Nevertheless, there are significant temperature and geopotential anomalies in the free troposphere (Figs. 4c,d). Therefore, the likeliest mechanism for this teleconnection is that WAM rainfall is enhanced because of a destabilization of the vertical column over Africa. In a sense, this corresponds to the “equilibrium” described in Giannini et al. (2007) that might be reached after La Niña mature phases. These results confirm the necessity to separate the developing La Niña from the mature La Niña. The response of the atmosphere is indeed not the same.

Fig. 4.

Mature La Niña JAS composites. The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is (a) SST 0.5°K, (b) precipitation 0.3 mm d−1, (c) Z 2 mgp−1, and (d) Z200 5 mgp−1. Other aspects of the figure are as in Fig. 3.

Fig. 4.

Mature La Niña JAS composites. The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is (a) SST 0.5°K, (b) precipitation 0.3 mm d−1, (c) Z 2 mgp−1, and (d) Z200 5 mgp−1. Other aspects of the figure are as in Fig. 3.

4. The ENSO–WAM teleconnection in CNRM-CM3

It is shown above that in the observations the ENSO–WAM teleconnection generally happens just after the onset of ENSO events in the boreal summer, when SST anomalies are still rather weak and concentrated to the east of the date line. Therefore, we can wonder if the observed and simulated SST patterns in Fig. 1 correspond to the same stage of ENSO development. Figure 5 clearly answers this question:

  • In the observations, JAS rainfall anomalies are significantly correlated with ENSO from May to December, with the highest values in JAS. Correlations are not significant from January to April before the monsoon. This confirms that Sahelian rainfall is statistically influenced during the developing phase of ENSO (before the autumn peak).

  • In the coupled simulation, correlations are highest from January to June, which means that in the model the teleconnection happens during the decay of ENSO events. This explains the SST pattern in Fig. 1b: at the end of ENSO events, the SST anomaly first erodes in the eastern equatorial Pacific in the model, and the links with the Atlantic and Indian tropical basins are at a maximum.

Fig. 5.

Correlations between the JAS yearly rainfall time series from the MCA (Fig. 1) and each of the 12 yearly SST time series extracted from the monthly ENSO index by taking successively the January values, February, and so on. Small bullets highlight the correlations that are significant at the 95% level (bootstrap procedure).

Fig. 5.

Correlations between the JAS yearly rainfall time series from the MCA (Fig. 1) and each of the 12 yearly SST time series extracted from the monthly ENSO index by taking successively the January values, February, and so on. Small bullets highlight the correlations that are significant at the 95% level (bootstrap procedure).

Figure 5 reveals that in CNRM-CM3 an important part of the interannual variability of the WAM (43% of the variance) is strongly predictable from the knowledge of SST anomalies in the Pacific during the preceding winter. The teleconnection does not happen at the same stage of ENSO life cycle, so the atmospheric processes may be different from the reanalyses. The ENSO simulated in CNRM-CM3 has several faults (Achutarao and Sperber 2006; Capotondi et al. 2006; Guilyardi 2006; van Oldenborgh et al. 2005; Leloup et al. 2007). Therefore, we can wonder if improving the simulated ENSO would improve the realism of the teleconnection. To test this hypothesis, the sensitivity experiment described hereafter is conducted.

a. Experimental design and validation

We use the same version of the Météo-France CNRM coupled model as for the IPCC AR4. All the information on CNRM-CM3 can be found on the Program for Climate Model Diagnosis and Intercomparison (PCMDI) Web site (online at http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php). In the sensitivity experiment, the sole modification concerns the tropical Pacific (Fig. 6): over that region, the ocean model uses ERA-40 reanalyzed daily wind stress instead of the wind calculated by the atmospheric model. The geographical area for this forcing is shown in Fig. 6, with two tapering zones at 30°–40°S and 30°–40°N. After a 10-yr spinup using the wind stress of the year 1960 (neutral from the point of view of ENSO), the simulation spans the period 1960–2001.

Fig. 6.

Design of the CNRM+ws sensitivity experiment.

Fig. 6.

Design of the CNRM+ws sensitivity experiment.

Globally, the SST mean state is unchanged in that experiment. The widespread cold bias—characteristic of CNRM-CM3—is only slightly weakened in the equatorial Pacific (not shown). Concerning the interannual variability, SST standard deviations become more realistic in the equatorial Pacific (not shown), but the ENSO pattern still extends too far in the west of the basin, which is a common bias of state-of-the-art coupled models (e.g., Leloup et al. 2007).

Figure 7 compares the simulated ENSO index with the observed one. Strikingly, the simulation catches all the ENSO events of the 1960–2001 period, which is of course not the case for the “free” simulation. Therefore, the main interest of this experiment is that using reanalyzed wind stress over the Pacific reproduces accurately the observed ENSO chronology in the coupled model. Such a protocol allows the year-by-year comparison of the observed and simulated ENSO teleconnections, which will provide further insight into the CNRM-CM3 coupled system. However, note the weak but systematic delay of the ENSO index in CNRM+wind stress (ws) (Fig. 7). This slight shift in the ENSO life cycle is not specifically due to the wind stress forcing: it is actually characteristic of the ENSO simulated in CNRM-CM3 (see section 5).

Fig. 7.

ENSO normalized index in the observations, in the standard CNRM-CM3 simulation (CNRM), and in the wind stress–driven CNRM-CM3 simulation (CNRM+ws).

Fig. 7.

ENSO normalized index in the observations, in the standard CNRM-CM3 simulation (CNRM), and in the wind stress–driven CNRM-CM3 simulation (CNRM+ws).

b. The simulated ENSO–WAM teleconnection

To make a basic assessment of the teleconnection in CNRM+ws experiment, Fig. 8 displays the same MCA as in Fig. 1b. The sensitivity experiment yields the same patterns as in the fully coupled simulation, but with weaker loadings for the SST heterogeneous map. The explained precipitation variance is 2 times weaker in the new simulation, and the correlation between the time series is only 0.53. The teleconnection is thus statistically weaker but remains physically the same.

Fig. 8.

Same MCA as in Fig. 1, but with CNRM+ws sensitivity experiment.

Fig. 8.

Same MCA as in Fig. 1, but with CNRM+ws sensitivity experiment.

As the experiment captures the observed chronology of ENSO, Fig. 9 shows the averaged annual evolution of ENSO events for the years selected in section 3. There is good agreement between the observed and simulated ENSO events, except that SST anomalies are—on average—delayed in the model (by about 2 months). In addition, Table 1 clearly demonstrates that the model produces a rainfall anomaly 1 yr later than in the observations. For the mature La Niña, there is no clear result.

Fig. 9.

Composite evolution of the ENSO normalized index for the observations (thick) and the CNRM+ws simulation (thin).

Fig. 9.

Composite evolution of the ENSO normalized index for the observations (thick) and the CNRM+ws simulation (thin).

Table 1.

Success ratio of the CNRM+ws experiment in simulating a precipitation anomaly of the expected sign over West Africa for each category of ENSO teleconnection. Yr(0) are the years selected in the observations, and Yr(1) the following years.

Success ratio of the CNRM+ws experiment in simulating a precipitation anomaly of the expected sign over West Africa for each category of ENSO teleconnection. Yr(0) are the years selected in the observations, and Yr(1) the following years.
Success ratio of the CNRM+ws experiment in simulating a precipitation anomaly of the expected sign over West Africa for each category of ENSO teleconnection. Yr(0) are the years selected in the observations, and Yr(1) the following years.

Why such a delay in the simulated response to developing ENSO events? Figure 10 shows the time evolution of the teleconnection in ERA-40 and CNRM+ws. This figure is based on the vertical cross section of the geopotential anomalies. Such maps indeed illustrate the main features of the teleconnection (see section 3). Warm minus cold composites are computed for the same years as in Fig. 3.

Fig. 10.

Developing ENSO composites (mean of the El Niño and the minus La Niña anomalies). Yr(0) corresponds to the selected year and Yr(1) to the following year. The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is 0.5°C for the SST and 2 mgp for the geopotential. Other aspects of the figure are as in Fig. 3.

Fig. 10.

Developing ENSO composites (mean of the El Niño and the minus La Niña anomalies). Yr(0) corresponds to the selected year and Yr(1) to the following year. The contours correspond to the composite anomaly, and the color shading shows the anomaly normalized by the standard deviation in each grid point where it is significant at the 95% level (bootstrap procedure). The thick gray contour is for the zero isoline, and the contour spacing is 0.5°C for the SST and 2 mgp for the geopotential. Other aspects of the figure are as in Fig. 3.

In the observations, there is a strong and widespread geopotential positive anomaly along the equator in January–March (JFM). Later on, in JAS, SST and geopotential anomalies are very weak, and there is no impact on WAM rainfall (not shown). In the CNRM+ws experiment, the ENSO SST pattern is narrow and confined to the equator in JAS of year(0), and there is no cold anomaly in the western Pacific contrary to the observations. Interestingly, the geopotential shows no response of the tropical troposphere to the onsetting ENSO. This explains why there is no teleconnection with the WAM in JAS of year(0). In JFM of year(1), the ENSO pattern is thicker, with a weak influence on the other basins, and the geopotential response is the same as in ERA-40. Finally, in JAS of year(1) the SST anomalies correspond to the pattern diagnosed by the MCA (Fig. 9), and there remains a positive geopotential anomaly in the whole equatorial free troposphere. These SST and geopotential anomalies, as well as the associated temperature anomalies (not shown) are responsible for the postponed teleconnection that occurs at the end of ENSO events in the model.

The physical reasons for the delayed atmospheric response in the simulation require further examination. It is mentioned above that ENSO onsets are late in the model. How long does it take for the geopotential anomalies to be set up over Africa? Figure 11 answers this question. Based on the first yearly time series from the PCA of October–December (OND) SST anomalies in the equatorial Pacific, correlations are calculated with the time series extracted from different monthly indexes (in the same way as in Fig. 5). It is shown in section 3 that geopotential anomalies in the high and low troposphere characterize the main large-scale processes of the ENSO–WAM teleconnection. Therefore, in the following the atmospheric response over Africa is assessed at the 850- and 200-hPa levels.

Fig. 11.

Correlations of the OND yearly SST time series yielded by the PCA of SST anomalies in the equatorial Pacific (0°, 20°N; 20°E, 35°W) with each of the 12 yearly time series extracted from the following monthly indexes: the ENSO index and the monthly geopotential anomalies averaged at 850 hPa over Africa (5°, 20°N; 15°E, 35°W), and at 200 hPa over Africa (10°S, 10°N; 15°E, 35°W). Small bullets highlight the correlations that are significant at the 95% level (bootstrap procedure). (a) HadC/ERA-40 and (b) CNRM+ws.

Fig. 11.

Correlations of the OND yearly SST time series yielded by the PCA of SST anomalies in the equatorial Pacific (0°, 20°N; 20°E, 35°W) with each of the 12 yearly time series extracted from the following monthly indexes: the ENSO index and the monthly geopotential anomalies averaged at 850 hPa over Africa (5°, 20°N; 15°E, 35°W), and at 200 hPa over Africa (10°S, 10°N; 15°E, 35°W). Small bullets highlight the correlations that are significant at the 95% level (bootstrap procedure). (a) HadC/ERA-40 and (b) CNRM+ws.

Figure 11a shows that in ERA-40 the lag between ENSO SST anomalies and the response over Africa is about 2.5 months in the low troposphere and 4 months in the high troposphere. In CNRM+ws experiment (Fig. 11b), these lags are, respectively, 4 and 5 months. The delay in the simulated response adds to the delay in ENSO onsets, so that the total delay (2–3 months) is sufficient to postpone the teleconnection to the following year, when correlations are still significant in the model, both in the low and high troposphere.

Section 3 emphasized the importance of the timing of ENSO events for the observed teleconnection. Section 4 shows that these temporal aspects are poorly reproduced in CNRM-CM3, which leads to a different ENSO–WAM relationship. Hence the following questions: is it the case only for the Météo-France model? Can such temporal aspects explain the diversity of the teleconnection patterns in state-of-the-art coupled models (Joly et al. 2007)? Both issues are addressed in the following section.

5. The ENSO–WAM teleconnection in IPCC AR4 simulations

In light of the above results, it seems that, in order to carefully study the processes of the ENSO teleconnection in IPCC AR4 simulations, it is essential to take into account both the time evolution of ENSO events and the time lag of the atmospheric response over Africa. This will provide an original insight into the tropical interannual variability as simulated by state-of-the-art coupled models.

a. IPCC AR4 coupled models

The purpose here is not to document the ENSO–WAM teleconnection in all the simulations available in the IPCC AR4 CMIP3 database, but rather to illustrate the importance of the time sequence of this teleconnection in some simulations of the coupled climate system. Our set of simulations is thus neither comprehensive, nor based on any type of a priori consideration. The 15 twentieth century (20C3M) simulations presented in Table 2 include runs performed with one model at different resolutions (which might have an impact on the teleconnection) and also runs from two models from the same research group. With this selection (15 simulations from 13 research groups in 10 different countries), we expect to have a representative sample of the performance of state-of-the-art coupled models, both in terms of ENSO and WAM variability. All the information about the IPCC AR4 can be found on the Web site (online at http://www.ipcc.ch/), and the coupled models are described in chapter 8 of the fourth Assessment Report (Randall et al. 2007). For further information, see the CMIP3 Web site (online at PCMDI: http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php).

Table 2.

Acronym given to the 20C3M simulations selected from the IPCC AR4 CMIP3 database. Corresponding model identifier (ID) and country of the laboratory.

Acronym given to the 20C3M simulations selected from the IPCC AR4 CMIP3 database. Corresponding model identifier (ID) and country of the laboratory.
Acronym given to the 20C3M simulations selected from the IPCC AR4 CMIP3 database. Corresponding model identifier (ID) and country of the laboratory.

b. Onset of ENSO events

To make the analysis simple, a linear approach is adopted. The same correlations as in Fig. 11 are calculated for each simulation, taking the hundred years of the twentieth century. Correlations do not account for the amplitude of the anomalies as regressions would do. However, this ensures a more refined insight into the temporal aspects and the strength of the relationships. Figure 12a shows that the timing of ENSO is fairly well represented in state-of-the-art coupled models. However, regarding the onset, in most simulations {all except the Institute of Numerical Mathematics (INM); the Model for Interdisciplinary Research on Climate, high-resolution version [MIROC(hires)]; and MIROC, medium-resolution version (medres)} the 0.5 arbitrary threshold is reached one month later than in the observations. Though weak, such a delay at the onset may have an impact on the teleconnection.

Fig. 12.

Similar to Fig. 11, but for the IPCC AR4 simulations. Correlations with (a) the ENSO index, (b) the geopotential anomalies at 850 hPa, and (c) the geopotential anomalies at 200 hPa.

Fig. 12.

Similar to Fig. 11, but for the IPCC AR4 simulations. Correlations with (a) the ENSO index, (b) the geopotential anomalies at 850 hPa, and (c) the geopotential anomalies at 200 hPa.

Regression values (not shown) indicate that in some models ENSO SST anomalies are weak [e.g., Canadian Centre for Climate Modelling and Analysis (CCCMAt63), Hadley Centre Global Environmental Model version 1 (HadGEM1), MIROC(hires), and MIROC(medres)]. Besides, some models [CNRM, the third climate configuration of the Met Office Unified Model (HadCM3), MIROC(medres), and Max Planck Institute (MPI)] exhibit strongly different time evolutions for the El Niño versus La Niña events (not shown), with possibly completely different impacts on the WAM. However, the purpose here is not to scrutinize the ENSO simulated in each model but rather to highlight some shared behaviors in the coupled simulations.

c. Time lag of the atmospheric response over Africa

To characterize the large-scale response over Africa, we use the same procedure as in Fig. 11. The 850- and 200-hPa geopotential indexes provide an excellent benchmark to test the response of IPCC AR4 models to ENSO over Africa.

1) Response in the low troposphere

Figure 12b reveals that the response at 850 hPa can be very different from one model to another. Even between the ERA-40 and NCEP reanalyses, there is a noticeable difference in terms of correlation values with, nevertheless, a similar time sequence. Simulations can be classified as follows. In CCCMAt63 and MIROC(hires) the weakness of the correlations indicates that the expected mechanism is missing in the low troposphere. In seven simulations [Bjerknes Centre for Climate Research (BCCR), CNRM, Commonwealth Scientific and Industrial Research Organisation (CSIRO), Geophysical Fluid Dynamics Laboratory (GFDL), HadGEM1, Istituto Nazionale di Geofisica e Vulcanologia (INGV), and MPI] the 0.5 (arbitrary) threshold is reached at least two months after ERA-40 and NCEP, that is, after the monsoon. Note that the correlations remain sometimes significant until the following monsoon season.

2) Response in the high troposphere

In comparison, the response at 200 hPa (Fig. 12c) shows less spread than at 850 hPa. Note that in NCEP the response is faster than in ERA-40 (about one month). In seven simulations [BCCR, CNRM, CSIRO, HadGEM1, INGV, MIROC(medres), and MPI] the 0.5 threshold is reached at least one month after ERA-40 (two months after NCEP). As already mentioned, in some simulations the response remains important during the following JAS season.

d. Resulting ENSO teleconnection

To assess cautiously the ENSO–WAM teleconnection in IPCC AR4 simulations, we use the same approach as in Fig. 5 but adapted as follows. Four MCA are computed between JAS rainfall anomalies over Africa and equatorial Pacific SST anomalies for each season: JFM, April–June (AMJ), JAS, and OND. The MCA that explains the maximum squared covariance (SCF) is selected. In three simulations [CCCMAt63, HadGEM1, and MIROC(hires)] this procedure finds no significant teleconnection (based on the SCF significance). This is probably due to the weak ENSO anomalies in these simulations and to the lack of response at 850 hPa over Africa (see above). The other simulations are classified as follows:

  • Group A: Four simulations have the strongest link between WAM and OND SST, as in the observations. Figure 13a shows that in these simulations the WAM is apparently influenced at the beginning of ENSO events. Unsurprisingly, these four simulations are actually characterized by a fast response over Africa both in the low and high troposphere (Figs. 12b,c).

  • Group B: In INM and MIROC(medres), the procedure selects the JAS season. According to Fig. 12a, both models are characterized by ENSO events that are set up—on average—earlier than in the observations. ENSO–WAM correlations peak, therefore, at the beginning of the monsoon season (Fig. 13b).

  • Group C: In six simulations (including CNRM-CM3), the strongest link is obtained with JFM SST. Figure 13c shows that the WAM is indeed influenced at the end of ENSO events. In these simulations, the response in the low troposphere is late (Fig. 12b) and prevents any influence on the WAM during the developing phase of ENSO. However, the response over Africa remains important during the following monsoon season, so that the teleconnection is postponed to the end of ENSO events.

Fig. 13.

Same MCA as in Fig. 1, but after selecting the SST season that maximizes the covariance: JFM (group C), AMJ (no simulation), JAS (group B), or OND (group A). (a)–(c) Same procedure as in Fig. 5. (d)–(f) JAS heterogeneous maps for the SST (see Fig. 1 for the specification).

Fig. 13.

Same MCA as in Fig. 1, but after selecting the SST season that maximizes the covariance: JFM (group C), AMJ (no simulation), JAS (group B), or OND (group A). (a)–(c) Same procedure as in Fig. 5. (d)–(f) JAS heterogeneous maps for the SST (see Fig. 1 for the specification).

It should be kept in mind that our procedure detects the main ENSO–WAM link and provides an averaged view of the teleconnection. In some models, the WAM might be influenced either at the beginning, at the end, or even at both stages of the ENSO life cycle without distinction. Moreover, in some models the teleconnection might be nonlinear, with different behaviors in the El Niño and La Niña cases.

e. Discussion

This study shows that temporal considerations explain in part the diversity of the simulated SST–WAM teleconnections. Depending on the phase of ENSO (onset, peak, or decay), the SST patterns (Fig. 13) and the involved atmospheric processes can be completely different in coupled models. A factor that is not discussed in this paper is the precipitation response to ENSO-related atmospheric anomalies. Joly et al. (2007) conclude that the pattern, and even the sign, of the precipitation anomalies over Africa are highly model dependent. Among the four models that reproduce the observed time evolution of the teleconnection (Fig. 13a), HadCM3 and MRI yield a quite realistic precipitation pattern (not shown), whereas L’Institut Pierre-Simon Laplace (IPSL) and NCAR do not simulate the expected negative response over the WAM region.

Some interesting remarks can be further inferred. In the MIROC model, the two different resolutions yield different responses over Africa (Figs. 12b,c). Surprisingly, the low resolution performs much better (on this specific issue). The SST patterns in Fig. 13 confirm that the resolution of the atmospheric model does not seem to condition the quality of the response. Besides, note that some models have the same atmospheric component. For example, BCCR and CNRM use Action de Recherche Petite Echelle Grande Echelle (ARPEGE)-climat-v3, while INGV and MPI use, respectively, ECHAM4.6 and ECHAM5. In these simulations, the atmospheric response is slower than in reanalyses. Furthermore, some models have the same oceanic component. For example, while INGV uses Océan Parallélisé (OPA) 8.2, CNRM and IPSL use OPA8.1. IPSL exhibits, however, a behavior that differs from that of the other two. It is indeed the whole coupled system that creates the teleconnection, not a single component.

Beside the wide variety of simulated ENSO—an issue discussed in many recent papers (see references herein)—it seems essential to understand why in half of the models the atmospheric response to ENSO SST anomalies is slower than in reanalyses. We will limit ourselves to a few suggestions. First, the correlations with the ENSO index (Fig. 12a) are insufficient to account for the variety of SST patterns. Depending on the shape of the pattern, the convective response is different (Su et al. 2003). Figure 14 displays the same correlations as in Fig. 12a, but with the Niño-4 index. The onset of Niño-4 anomalies exhibits a much larger spread than with the ENSO index (Fig. 12a). Ten simulations out of 15 have a delay greater than one month in their Niño-4 response, which may be critical for the timing of convective anomalies in the central-western Pacific. Interestingly, the four models that have the slowest Niño-4 response (CSIRO, CNRM, INGV, and MPI) are among those that have a strongly delayed response over Africa.

Fig. 14.

Similar to Fig. 12a, but for the Niño-4 index.

Fig. 14.

Similar to Fig. 12a, but for the Niño-4 index.

To pursue the investigation it would be interesting to distinguish between the time lag of the convective response over the Pacific and the time lag of the propagation toward Africa. Some studies (Lin 2007; Wu et al. 2006) indicate that the coupling between the ocean surface and the overlying troposphere is actually too strong in current coupled models. Therefore, it is the horizontal propagation that may be deficient. Focusing on the tropical intraseasonal variability, Lin et al. (2006) state indeed that the variance of rainfall Kelvin modes along the equator in IPCC AR4 simulations is half the magnitude of the observed. According to the authors, models do not have enough wave–heating feedback and have an excessively strong persistence of tropical (daily) precipitation. This may be an interesting piece of information to understand the slow response in some models.

6. Conclusions

Numerous studies demonstrate the statistical significance of the relationship between ENSO and the interannual variability of Sahelian rainfall. However, not all ENSO events lead to Sahelian rainfall anomalies and, conversely, not all droughts over the Sahel are necessarily associated with a SST anomaly in the equatorial Pacific. Each ENSO event has actually a different “flavor,” and in each case the large-scale background state is different. Therefore, assessing the dynamical processes of the ENSO–WAM teleconnection is a difficult task. The present study shows that, over the second half of the twentieth century, the main influence of ENSO on Sahelian rainfall occurs during the developing phase of ENSO or marginally during the decay of some long-lasting La Niña. This classification of teleconnection years enables us to investigate more thoroughly the atmospheric processes involved. The teleconnection that takes place during the developing phase of ENSO events is fast and direct (through the atmosphere). The surface Atlantic and Indian Oceans are not involved (e.g., Fig. 1a), despite significant anomalies in the mean flow in the low troposphere (Fig. 3). In the Indian Ocean, SST anomalies due to ENSO are set up in autumn on average (not shown), that is, after the monsoon season.

The intercomparison of the teleconnection processes in IPCC AR4 simulations is particularly useful. In most models, ENSO onsets are slightly late on average. More worrying is the strong delay of the atmospheric response over Africa in half of the studied simulations. Diagnosing the time sequence of the ENSO teleconnection helps to understand the processes of the simulated teleconnections. Because of the temporal biases, the influence of ENSO on the WAM occurs at the beginning of ENSO events in only four simulations out of 15. In six models (among which is the CNRM-CM3) the teleconnection happens at the end of ENSO events. These results support the idea that, because the WAM occurs during the boreal summer, the influence of ENSO strongly depends on the timing of ENSO onsets and on the time lag of the atmospheric response.

Several important issues are not addressed in this paper, especially the issue of the climate shift in the 1970s and the low-frequency modulation of the ENSO teleconnection. The onset of ENSO events is thought to have changed (Wang 1995; Wang and An 2002), which means possibly different processes and a different time evolution for the teleconnection. The available reanalyses are, however, too short to study that low-frequency modulation. Their inhomogeneity, due to the use of satellite data for the last decades, might blur the results (Poccard et al. 2000).

Realistically simulating the processes of such a complex phenomenon as the ENSO–WAM teleconnection appears particularly challenging for state-of-the-art coupled models. Nevertheless, Douville et al. (2006) suggest that model deficiencies in simulating ENSO teleconnections with tropical rainfall could be partly responsible for the spread in the future projections of global land precipitation. Besides, we believe that diagnosing precisely the time sequence of boreal summer ENSO teleconnections may help to improve the seasonal forecasting of Sahelian rainfall.

Acknowledgments

This work was supported by the AMMA program (http://www.amma-international.org), and by the ENSEMBLES European project (Contract GOCE-CT-2003-505539). We acknowledge the modeling groups for making their simulations available, and the PCMDI for the build up of the CMIP3 database. Thanks are also due to Serge Janicot and Pascal Terray for their helpful comments and to the three anonymous reviewers who helped improving the original manuscript. Note that the figures have been prepared using the GrADS software.

REFERENCES

REFERENCES
Achutarao
,
K.
, and
K. R.
Sperber
,
2006
:
ENSO simulation in coupled ocean–atmosphere models: Are the current models better?
Climate Dyn.
,
27
,
1
15
.
doi:10.1007/s00382-006-0119-7
.
Biasutti
,
M.
, and
A.
Giannini
,
2006
:
Robust Sahel drying in response to late 20th century forcings.
Geophys. Res. Lett.
,
33
,
L11706
.
doi:10.1029/2006GL026067
.
Bracco
,
A.
,
F.
Kucharski
,
F.
Molteni
,
W.
Hazeleger
, and
C.
Severijns
,
2007
:
A recipe for simulating the interannual variability of the Asian summer monsoon and its relation with ENSO.
Climate Dyn.
,
28
,
441
460
.
doi:10.1007/s00382-006-0190-0
.
Bretherton
,
C. S.
,
C.
Smith
, and
J. M.
Wallace
,
1992
:
An intercomparison of methods for finding coupled patterns in climate data.
J. Climate
,
5
,
541
560
.
Capotondi
,
A.
,
A.
Wittenberg
, and
S.
Masina
,
2006
:
Spatial and temporal structure of tropical Pacific interannual variability in 20th century coupled simulations.
Ocean Modell.
,
15
,
274
298
.
Chang
,
C-P.
,
B.
Wang
, and
N-C. G.
Lau
,
Eds.,
.
2004
:
The global monsoon system: Research and forecast (IWM-III).
World Meteorological Organization, WMO/TD No. 1266, TMRP Rep. 70
.
Chen
,
T-C.
,
1985
:
Global water vapor flux and maintenance during FGGE.
Mon. Wea. Rev.
,
113
,
1801
1819
.
Chiang
,
J. C. H.
,
Y.
Kushnir
, and
A.
Giannini
,
2002
:
Deconstructing Atlantic intertropical convergence zone variability: Influence of the local cross-equatorial sea surface temperature gradient and remote forcing from the eastern equatorial Pacific.
J. Geophys. Res.
,
107
,
4004
.
doi:10.1029/2000JD000307
.
Cook
,
K. H.
, and
E. K.
Vizy
,
2006
:
Coupled model simulations of the West African monsoon system: Twentieth- and twenty-first-century simulations.
J. Climate
,
19
,
3681
3703
.
D’Abreton
,
P. C.
, and
P. D.
Tyson
,
1995
:
Divergent and non-divergent water vapour transport over southern Africa during wet and dry conditions.
Meteor. Atmos. Phys.
,
55
,
47
59
.
doi:10.1007/BF01029601
.
dell’Aquila
,
A.
,
V.
Lucarini
,
P. M.
Ruti
, and
S.
Calmanti
,
2005
:
Hayashi spectra of the northern hemisphere mid-latitude atmospheric variability in the NCEP/NCAR and ECMWF reanalyses.
Climate Dyn.
,
25
,
639
652
.
doi:10.1007/s00382-005-0048-x
.
Douville
,
H.
,
D.
Salas-Mélia
, and
S.
Tyteca
,
2006
:
On the tropical origin of uncertainties in the global land precipitation response to global warming.
Climate Dyn.
,
26
,
367
385
.
doi:10.1007/s00382-005-0088-2
.
Fedorov
,
A. V.
, and
S. G.
Philander
,
2001
:
A stability analysis of tropical ocean–atmosphere interactions: Bridging measurements and theory for El Niño.
J. Climate
,
14
,
3086
3101
.
Giannini
,
A.
,
R.
Saravanan
, and
P.
Chang
,
2005
:
Dynamics of the boreal summer African monsoon in the NSIPP1 atmospheric model.
Climate Dyn.
,
25
,
517
535
.
doi:10.1007/s00382-005-0056-x
.
Giannini
,
A.
,
A. W.
Robertson
, and
J-H.
Qian
,
2007
:
A role for tropical tropospheric temperature adjustment to El Niño–Southern Oscillation in the seasonality of monsoonal Indonesia precipitation predictability.
J. Geophys. Res.
,
112
,
D16110
.
doi:10.1029/2007JD008519
.
Gill
,
A. E.
,
1980
:
Some simple solutions for heat-induced tropical circulation.
Quart. J. Roy. Meteor. Soc.
,
106
,
447
462
.
Guilyardi
,
E.
,
2006
:
El Niño–mean state–seasonal cycle interactions in a multi-model ensemble.
Climate Dyn.
,
26
,
329
348
.
doi:10.1007/s00382-005-0084-6
.
Hannachi
,
A.
,
I. T.
Jolliffe
, and
D. B.
Stephenson
,
2007
:
Empirical orthogonal functions and related techniques in atmospheric science: A review.
Int. J. Climatol.
,
27
,
1119
1152
.
doi:10.1002/joc.1499
.
Hoerling
,
M.
,
J.
Hurrell
,
J.
Eischeid
, and
A.
Phillips
,
2006
:
Detection and attribution of twentieth-century northern and southern African rainfall change.
J. Climate
,
19
,
3989
4008
.
Huang
,
B.
, and
J.
Shukla
,
2007
:
Mechanisms for the interannual variability in the tropical Indian Ocean. Part I: The role of remote forcing from the tropical Pacific.
J. Climate
,
20
,
2917
2936
.
Janicot
,
S.
,
V.
Moron
, and
B.
Fontaine
,
1996
:
Sahel droughts and ENSO dynamics.
Geophys. Res. Lett.
,
23
,
515
518
.
Janicot
,
S.
,
S.
Trzaska
, and
I.
Poccard
,
2001
:
Summer Sahel-ENSO teleconnection and decadal time scale SST variations.
Climate Dyn.
,
18
,
303
320
.
Joly
,
M.
,
A.
Voldoire
,
H.
Douville
,
P.
Terray
, and
J-F.
Royer
,
2007
:
African monsoon teleconnections with tropical SSTs: Validation and evolution in a set of IPCC4 simulations.
Climate Dyn.
,
29
,
1
20
.
doi:10.1007/s00382-006-0215-8
.
Joseph
,
R.
, and
S.
Nigam
,
2006
:
ENSO evolution and teleconnections in IPCC’s twentieth-century climate simulations: Realistic representation?
J. Climate
,
19
,
4360
4377
.
Jury
,
M. R.
,
D. B.
Enfield
, and
J-L.
Mélice
,
2002
:
Tropical monsoons around Africa: Stability of El Niño–Southern Oscillation associations and links with continental climate.
J. Geophys. Res.
,
107
,
3151
.
doi:10.1029/2000JC000507
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
471
.
Krishna Kumar
,
K.
,
M.
Hoerling
, and
B.
Rajagopalan
,
2005
:
Advancing dynamical prediction of Indian monsoon rainfall.
Geophys. Res. Lett.
,
32
,
L08704
.
doi:10.1029/2004GL021979
.
Lau
,
K. M.
,
S. S. P.
Shen
,
K-M.
Kim
, and
H.
Wang
,
2006
:
A multimodel study of the twentieth-century simulations of Sahel drought from the 1970s to 1990s.
J. Geophys. Res.
,
111
,
D07111
.
doi:10.1029/2005JD006281
.
Leloup
,
J.
,
M.
Lengaigne
, and
J-P.
Boulanger
,
2007
:
Twentieth century ENSO characteristics in the IPCC database.
Climate Dyn.
,
30
,
277
291
.
doi:10.1007/s00382-007-0284-3
.
Lin
,
J-L.
,
2007
:
The double-ITCZ problem in IPCC AR4 coupled GCMs: Ocean–atmosphere feedback analysis.
J. Climate
,
20
,
4497
4525
.
Lin
,
J-L.
, and
Coauthors
,
2006
:
Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals.
J. Climate
,
19
,
2665
2690
.
Lyon
,
B.
, and
S. J.
Mason
,
2007
:
The 1997–98 summer rainfall season in southern Africa. Part I: Observations.
J. Climate
,
20
,
5134
5148
.
Mitchell
,
T. D.
, and
P. D.
Jones
,
2005
:
An improved method of constructing a database of monthly climate observations and associated high-resolution grids.
Int. J. Climatol.
,
25
,
693
712
.
Neelin
,
J. D.
,
C.
Chou
, and
H.
Su
,
2003
:
Tropical drought regions in global warming and El Niño teleconnections.
Geophys. Res. Lett.
,
30
,
2275
.
doi:10.1029/2003GL018625
.
Nicholls
,
N.
,
2001
:
The insignificance of significance testing.
Bull. Amer. Meteor. Soc.
,
82
,
981
986
.
Nicholson
,
S. E.
, and
J. P.
Grist
,
2003
:
The seasonal evolution of the atmospheric circulation over West Africa and equatorial Africa.
J. Climate
,
16
,
1013
1030
.
Peña
,
M.
,
M.
Cai
, and
E.
Kalnay
,
2004
:
Life span of subseasonal coupled anomalies.
J. Climate
,
17
,
1597
1604
.
Poccard
,
I.
,
S.
Janicot
, and
P.
Camberlin
,
2000
:
Comparison of rainfall structures between NCEP/NCAR reanalyses and observed data over tropical Africa.
Climate Dyn.
,
16
,
897
915
.
Press
,
W. H.
,
S. A.
Teukolsky
,
W. T.
Vetterling
, and
B. P.
Flannery
,
1992
:
Numerical Recipes in FORTRAN: The Art of Scientific Computing.
2nd ed. Cambridge University Press, 963 pp
.
Randall
,
D.
, and
Coauthors
,
2007
:
Climate models and their evaluation.
Climate Change 2007: The Physical Science Basis, S. Solomon et al., Eds., Cambridge University Press, 589–662
.
Rayner
,
N. A.
,
D. E.
Parker
,
E. B.
Horton
,
C. K.
Folland
,
L. V.
Alexander
,
D. P.
Rowell
,
E. C.
Kent
, and
A.
Kaplan
,
2003
:
Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century.
J. Geophys. Res.
,
108
,
4407
.
doi:10.1029/2002JD002670
.
Rowell
,
D. P.
,
2001
:
Teleconnections between the tropical Pacific and the Sahel.
Quart. J. Roy. Meteor. Soc.
,
127
,
1683
1706
.
Rowell
,
D. P.
,
C. K.
Folland
,
K.
Maskell
,
J. A.
Owen
, and
M. N.
Ward
,
1992
:
Modelling the influence of global sea surface temperatures on the variability and predictability of seasonal Sahel rainfall.
Geophys. Res. Lett.
,
19
,
905
908
.
Sardeshmukh
,
P. D.
, and
B. J.
Hoskins
,
1985
:
Vorticity balances in the tropics during the 1982–83 El Niño–Southern Oscillation event.
Quart. J. Roy. Meteor. Soc.
,
111
,
261
278
.
Semazzi
,
F. H. M.
,
V.
Mehta
, and
Y. C.
Sud
,
1988
:
An investigation of the relationship between sub-Saharan rainfall and global sea surface temperatures.
Atmos.–Ocean
,
26
,
118
138
.
Sturaro
,
G.
,
2003
:
A closer look at the climatological discontinuities present in the NCEP/NCAR reanalysis temperature due to the introduction of satellite data.
Climate Dyn.
,
21
,
309
316
.
doi:10.1007/s00382-003-0334-4
.
Su
,
H.
,
J. D.
Neelin
, and
J. E.
Meyerson
,
2003
:
Sensitivity of tropical tropospheric temperature to sea surface temperature forcing.
J. Climate
,
16
,
1283
1301
.
Tarhule
,
A.
, and
P. J.
Lamb
,
2003
:
Climate research and seasonal forecasting for West Africans: Perceptions, dissemination, and use?
Bull. Amer. Meteor. Soc.
,
84
,
1741
1759
.
Terray
,
P.
,
P.
Delecluse
,
S.
Labattu
, and
L.
Terray
,
2003
:
Sea surface temperature associations with the late Indian summer monsoon.
Climate Dyn.
,
21
,
593
618
.
doi:10.1007/s00382-003-0354-0
.
Trenberth
,
K. E.
,
1997
:
The definition of El Niño.
Bull. Amer. Meteor. Soc.
,
78
,
2771
2777
.
Uppala
,
S. M.
, and
Coauthors
,
2005
:
The ERA-40 Re-Analysis.
Quart. J. Roy. Meteor. Soc.
,
131
,
2961
3012
.
van Oldenborgh
,
G. J.
,
S. Y.
Philip
, and
M.
Collins
,
2005
:
El Niño in a changing climate: A multi-model study.
Ocean Sci.
,
1
,
81
95
.
Wang
,
B.
,
1995
:
Interdecadal changes in El Niño onset in the last four decades.
J. Climate
,
8
,
267
285
.
Wang
,
B.
, and
S. I.
An
,
2002
:
A mechanism for decadal changes of ENSO behavior: Roles of background wind changes.
Climate Dyn.
,
18
,
475
486
.
doi:10.1007/s00382-001-0189-5
.
Wilks
,
D. S.
,
1997
:
Resampling hypothesis tests for autocorrelated fields.
J. Climate
,
10
,
65
82
.
Wu
,
W.
,
A. E.
Dessler
, and
G. R.
North
,
2006
:
Analysis of the correlations between atmospheric boundary-layer and free-tropospheric temperatures in the tropics.
Geophys. Res. Lett.
,
33
,
L20707
.
doi:10.1029/2006GL026708
.

Footnotes

Corresponding author address: Mathieu Joly, Météo-France, CNRM/GMGEC/UDC, 42 Ave. G. Coriolis, 31057 Toulouse CEDEX 1, France. Email: mathieu.joly@meteo.fr