Evidence of two independent modes of convection at intraseasonal timescale in the West African summer monsoon



[1] Two main independent modes of convection at intraseasonal timescale in the West African summer monsoon are highlighted. One depicts a meridional dipole of convection over West Africa linked to a modulation of the latitude of the inter-tropical convergence zone (ITCZ); it is associated with a westward propagative signal of the convection over the Sahel, and to a modulation of cyclonic vorticity at these latitudes. The other one is characterized by a stationary uniform modulation of convection in the ITCZ without any significant modulation of its latitudinal location; it is associated with a modulation of the zonal wind component over the eastern equatorial Atlantic. The dominant periodicity of these two modes is around 15 days. It is necessary to take into account the relative parts of these two modes in order to accurately analyze the mechanisms involved in intraseasonal variability of the African monsoon.

1. Introduction

[2] Only few studies have addressed the question of the intraseasonal timescale variability of convection in the West African monsoon (WAM). Kiladis and Weickmann [1997] showed connections at the 6–30-day time scale between convection in the region 5–15°N/10–20°E and moisture advection over West Africa during northern summer. More recently, Sultan et al. [2003] highlighted a westward propagating signal of convection along the Sahelian latitudes, with a dominant period around 15 days, based on a regional rainfall index computed on the area 12.5–15°N/10°W–10°E and filtered between 10 and 60 days. On the other hand, Matthews [2002, 2004] focused on the 20–200 period range and identified by an Empirical Orthogonal Function (EOF) analysis a dominant mode over the whole African monsoon region which might arise as a remote response to the intraseasonal Madden-Julian Oscillation (MJO) over the warm pool region. Finally, Grodsky and Carton [2001] showed that intraseasonal modulation of convection may also occur during northern spring in the ITCZ over the tropical Atlantic with a dominant period in the 10–15-day band.

[3] A more complete investigation of intraseasonal variability of the summer WAM is presented here over the period 1979–2000, a dry one over West Africa compared to the long-term mean [Hastenrath, 1991]. It is based on three datasets. First, the daily Interpolated OLR dataset of 2.5° latitude/longitude resolution [Liebmann and Smith, 1996], from the Climate Diagnostics Center, which has been widely used for tropical studies where deep convection and rainfall can be estimated through low OLR values. Second, to validate this OLR-based study, daily interpolated in-situ rainfall dataset is available from IRD (Institut de Recherches pour le Développement) at the same resolution over the domain 3–20°N/18°W–25°E for the period 1979–1990, with void areas in the SW corner of the domain and in Nigeria and a part of Central Africa. Finally the NCEP-DOE AMIP-II Reanalysis (R-2) dataset, an updated version of the NCEP/NCAR reanalysis has been used to document the low-level circulation associated with the different intraseasonal modes of convection in the African monsoon. Several studies [Diedhiou et al., 1999; Sultan et al., 2003] have demonstrated the accuracy of the first version [Kalnay et al., 1996] of these reanalyzes in the description of the atmospheric circulation over West Africa, as well as high consistency with the IRD-rainfall and NOAA-OLR fields from synoptic to interannual timescales. The updated version of the reanalyzes shows significant improvements upon the first one in land surface parameters and land-ocean fluxes [Kanamitsu et al., 2002].

2. EOF Analysis of Intraseasonal Variability of Convection Over West Africa

[4] To document intraseasonal variability of convection over West and Central Africa during the summer monsoon (June to September), a Spatial EOF (SEOF; see Richman [1986]) analysis was performed on the 10–60-day bandpass filtered OLR values, using a non-recursive filter [Scavuzzo et al., 1998], over the domain shown on Figure 1 (10°S–30°N/30°W–30°E). We used the same filtering procedure as in Sultan et al. [2003] since it enabled to extract a dominant Sahelian signal around 15 days and a less marked one around 40 days. Moreover, we performed similar SEOF analyses on other period ranges (10–25, 10–90, 10–200-day) to validate the results on the 10–60-day range and to consider also possible MJO-like periodicities as done in Matthews [2002, 2004]. Resulting spatial patterns and composites derived from the principal component (PC) time series are fairly similar. Furthermore as we showed previously the dominance of the 10–25-day variability at least over the Sahel [Sultan et al., 2003], this study will provide a more complete analysis than the Matthews' ones, which considered lower frequency variability. Figure 1 (contours) shows 10–60-days filtered OLR composite maps of strong minus weak convective events related to each of the four first leading eigenvectors of the SEOF: for each respective PC time series, strong (resp. weak) events were selected from maximum (resp. minimum) greater (resp. lower) than at least one standard deviation apart of the mean. Then positive rainfall anomaly (high rainfall rate) coincides with negative OLR one (low OLR levels). The four modes explain respectively 12.6, 8.13, 5.75 and 5.27% of the filtered variance. To determine the number of PC to be retained, the scree test [Cattell, 1966] and the North et al. [1982] rule of thumb were used after taking into account the effect of the autocorrelation on the number of independent data. PC1/PC2 are resolved, but PC3/PC4 formed an “effectively degenerate multiplet”. However, as the four first PCs stay consistent when rotation with varimax criterion is applied (not shown), and as they are before the term scree, they are considered in the following.

Figure 1.

Contours: the strong minus weak convective intraseasonal events composite of 10–60-day filtered OLR over West and Central Africa for JJAS 1979–2000 based on the PC time series (see details in the text); OLR differences are expressed in W.m−2; (a) SEOF1; (b) SEOF2; (c) SEOF3; (d) SEOF4. Shaded: same as contours but for unfiltered rainfall on the period JJAS 1979–1990; rainfall differences are expressed in mm.d−1.

[5] The first mode can be interpreted as an enhancement, up to 25 W.m−2, of the mean convection over the African monsoon domain since its spatial pattern is superimposed to the climatological mean of the ITCZ OLR field (not shown; in summer the ITCZ is centered along 10°N, see for instance Figure 6 of Sultan et al. [2003]). This “Guinean” mode is similar to the dominant mode computed by Matthews [2004] while the filtering procedure is slightly different. The second and third modes, which have not been considered by Matthews, depict respectively a zonal and a meridional convection dipole, characterizing mainly a high convection modulation over the Sahel. These two modes are in spatial quadrature, and the PC time series are in phase quadrature, with a peak correlation of 0.4, significant at the level 0.001, for a lag of 4 days (not shown). So SEOF2 and SEOF3 tend to depict a zonally westward propagative signal of convection over the Sahel, which is concomitant with a variation of the ITCZ latitude depicted by the meridional dipole configuration of these two patterns. This “Sahelian” mode can be associated with the intraseasonal variability pattern highlighted by Sultan et al. [2003] and based on a regional Sahel rainfall index. The fourth mode depicts a positive pole located over the Atlantic along 10°N and a negative pole centered at 25°N–5°W. To validate these results based on SEOF, other approaches have been performed: Varimax rotated SEOF (RSEOF), Extended SEOF (ESEOF), Varimax rotated ESEOF (RESEOF), as well as Temporal EOF (TEOF), RTEOF, ETEOF and RETEOF. All these analyses confirm the patterns shown by the SEOF. In particular, the ESEOF analysis has confirmed the westward propagative character of the Sahelian mode and the stationary character of the Guinean mode; this will be shown further by another approach (see Figure 2). Finally, as the SEOF has been validated, it will be used in the following analysis of the reconstructed signal of the convection in the ITCZ, since the SEOF has the property to extract the highest variance at each step of the SEOF decomposition.

Figure 2.

Composite time sequences based on the OLR 10–60-day filtered index averaged on the area 7.5–12.5°N/10°W–10°E reconstructed by respectively SEOF1 and SEOF234. For each of the two reconstructed ITCZ indexes time series, we retained the dates (called t0) where this index is maximum (minimum) and its deviation from its mean is greater (lower) than its standard deviation to define a dry (wet) phase. The respective wet minus dry composite sequences are shown for the 10–60-day filtered OLR and 925 hPa wind fields. These sequences go (top to bottom) from t0 minus 6 days to t0 plus 6 days with a 3 days lag. On the left, the composite fields associated with the SEOF1-reconstructed ITCZ index; on the right same fields but associated with the SEOF234-reconstructed ITCZ index. Shaded areas represent the OLR anomalies expressed in W.m−2. Vector scale (m.s−1) is displayed below. The ITCZ box is displayed at t0.

[6] To validate the previous results based on OLR as a proxy for convection, we have computed composites of unfiltered rainfall fields based on the four OLR PC time series over the period June–September from 1979 to 1990, the common period between the NOAA-OLR and the IRD-rainfall datasets, by using the same procedure as for OLR. Figure 1 (shaded) shows the wet minus dry composite associated to the four main modes of intraseasonal convection over West and Central Africa. The resulting composite rainfall fields confirm the interpretation based on OLR. The first mode is associated with a modulation of precipitation including the whole ITCZ with high differences in precipitation amounts (the highest mean precipitation in the heart of the ITCZ is between 6 and 8 mm.day−1 outside of the mountains; see Figure 6 of Sultan et al. [2003]), but without any significant changes in the ITCZ latitudinal location. The second and third modes are associated with a meridional rainfall dipole whose poles are located north and south of the mean ITCZ latitude, depicting a modulation in latitude of the ITCZ location. The fourth mode shows the same polarities as for OLR with greater rainfall over the Fouta-Djalon massifs in the southwest area and in the middle of West Africa, and lower rainfall in the northwestern and the southeastern parts of the domain.

3. Reconstruction of an ITCZ Index by the Two Main Modes and Associated Patterns

[7] To evaluate the impact of each of the SEOF modes on the ITCZ convection, as well as to point out the convection and atmospheric circulation patterns associated to these modes, we have first computed 10°W–10°E averaged OLR 10–60-day filtered indexes at different latitude bands reconstructed by SEOF1, SEOF2+SEOF3+SEOF4 (SEOF234), and SEOF1+SEOF2+SEOF3+SEOF4 (SEOF1234), and compared them with the corresponding raw averaged OLR 10–60-day filtered indexes. Table 1 shows the corresponding correlation coefficients. The high correlation coefficients between OLR and SEOF1234 indexes contrast with the relatively low variance percentage related to the four SEOF (32%). This emphasizes the validity of the four first reconstructed modes on the ITCZ area. This is especially verified for SEOF1 and SEOF3, the part of SEOF2 being less significant at the longitudes 10°W–10°E. Moreover, high correlation coefficients with SEOF1 reinforce its central function within the ITCZ (with correlations up to 0.860 within 5–10°N), and its shared role with SEOF234 on Sahelian latitudes within 10–17.5°N. Northward of SEOF1 activity, SEOF234 is clearly dominant.

Table 1. Correlation Coefficients Computed Between 10°W–10°E Averaged OLR 10–60-Day Filtered Indexes at Different Latitude Bands and Their Reconstruction by SEOF1, SEOF234, SEOF1234, on the Period JJAS 1979–2000

[8] As the ITCZ is centered along 10°N during the summer monsoon, we focus now on the OLR 10–60-day filtered index averaged on the area 7.5–12.5°N/10°W–10°E and on the associated SEOF1 and SEOF234 reconstructed indexes. For each of the two reconstructed ITCZ indexes time series, we retained the dates (called t0) where this index is maximum (resp. minimum) and its deviation from its mean for each year is greater (resp. lower) than its standard deviation to define a dry (resp. wet) phase. As the ITCZ indexes are expressed through OLR values, a maximum in the corresponding time series means weaker convection. Figure 2 shows the wet minus dry composite sequence associated to these two reconstructed indexes, for the 10–60-day filtered OLR and 925 hPa wind fields. These sequences go from t0 minus 6 days to t0 plus 6 days with a 3 days lag. The sequence associated with the SEOF1-reconstructed index (left column of Figure 2) shows that the ITCZ convection increase is associated with a standing oscillation growing and decreasing over the southern coast of West Africa and Central Africa. An opposite polarity is visible west of 30°W. Between these two poles of convection, the positive zonal component of the 925 hPa wind increases when the convection grows in the African ITCZ, driving more moisture towards the convective area. The maximum of this westerly wind modulation is observed at t0 at 25°W–7.5°N. This time sequence enables to estimate the dominant periodicity of these fluctuations since we get a rather similar OLR anomaly field at t0−6 and t0+6. In fact, a more precise computation using the time sequence of the ITCZ reconstructed index points out a mean periodicity of 15 days (not shown). This pattern should be linked to the one depicted by Grodsky and Carton [2001] leading to a modulation of the wind direction of the monsoon flow whose mean direction is southwesterly. It might also be associated with the results of Matthews [2002, 2004], but as the dominant periodicity is different from the MJO mode, this point needs more investigation. On the other hand, the sequence associated with the SEOF234-reconstructed index (right of Figure 2) shows that the convection increase in the African ITCZ is associated with a propagating mode appearing first over Central Africa, moving to the north towards the Sahelian latitudes, and then propagating westward towards the eastern tropical Atlantic. This pattern is associated with a cyclonic circulation positioned ahead of the enhanced convective pole increasing the advection of moisture towards this pole. The maximum of the wind modulation is a southwesterly component observed at t0+3 along the western coast of West Africa. This pattern sequence is consistent with the results of Sultan et al. [2003], and we find as in Sultan et al. a dominant periodicity around 15 days.

4. Conclusion

[9] Intraseasonal variability of convection in the African ITCZ during the summer monsoon can be summed up by two independent modes. One is characterized by a stationary monopole of convection over West and Central Africa, and modulates the convective activity in the ITCZ but not its latitudinal location; it is associated with fluctuations in zonal advection of moisture over the eastern equatorial Atlantic toward Africa. Whether these fluctuations are an inherent feature of the WAM or a larger scale one independent dynamically to the WAM is under consideration. The other one is characterized by a westward propagative signal of the convection over the Sahel, and constitutes a meridional dipole of convection over West Africa consistent with an almost northward location of the ITCZ; it is associated with a modulation of cyclonic vorticity at the Sahelian latitudes controlling moisture advection towards the Sahel. The dominant periodicity of these two modes is around 15 days. Understanding mechanisms of rainfall variability during the summer WAM associated to this intraseasonal time scale implies to discriminate first the relative contribution of these two modes. Finally, interactions between these modes of intraseasonal variability and both the timing of the seasonal cycle and interannual variability is also a key question dealing with monsoon predictability, which need to be investigated.


[10] We are thankful to H. Laurent for providing the IRD rainfall data and to NOAA-CIRES Climate Diagnostics Center (Boulder CO) for providing the NCEP-DOE Reanalysis dataset, and the interpolated OLR dataset from (http://www.cdc.noaa.gov) and to the reviewers, Dr Grodsky and an anonymous one, for useful comments.