Impacts of warm and cold situations in the Mediterranean basins on the West African monsoon: observed connection patterns (1979–2006) and climate simulations

3.1 Composite and correlation analyses through observed precipitation

3.2 Robustness of the relationship

The linear relationship is further analysed by computing several Singular Value decompositions (SVD) between temperature fields and OLR fields over the period 1979–2005. Interpolated OLR data are a practical estimate of convective activity over West Africa, since OLR values <240 W/m² depict the occurrence of high clouds closely linked to deep moist convective areas as those which are positioned within ITCZ. These SVDs have been performed both at monthly, 5-day and daily scales in isolation to be not confounded with longer-term variations and using different time filters. The main results are reported in Figs. 1 and 2 and Table 3.

Open image in new window

Open image in new window

Table 3

Percentages of covariance and covariance fractions explained by the leading July–September mode (SVD1) at a daily timescale and time correlations (Corr SVD1) with and without linear trend (detrend) and as a function of time filtering in different ranges for capturing the 3–5 days (AEWs), < 10-day (synoptic), 10–90 days (intraseasonal) variability over the period 1979–2006

T, OLR, days	With trend	Detrended	3–5 days	<10 days	10–90 days
% cov/var	11.5	17	10.7	8.5	9.9
% SVD1	51.9	29	46.2	45.6	43.9
Corr SVD1	+51	+42	+48	+41	+26

Figure 1 refers to a synchronous (no lag) analysis performed between monthly skin temperatures observed over an extended Mediterranean region (5°W–40°E; 20–45°N) and monthly OLR values referring to an extended WAM region (20°W–40°E; 10°S–20°N). The total covariance equals 32.7% of the total variance and the squared covariance fractions of the two leading modes match 33.4 and 22.5% respectively. The first heterogeneous mode and its related time coefficients are displayed in Fig. 1a, b. They clearly attest the statistical linkage between warm anomalies restricted to Eastern Mediterranean and low OLR anomalies expanding along 10–15°N and eastward to about 10°W. As noticed above, the Eastern Mediterranean concentrates the highest positive loadings, while the western basin and continental regions somewhat northward to the mean ITCZ position accumulates the strongest negative loadings (meaning enhanced deep convection) with the maximum located in the central and eastern Sahel. Such a result agrees well with those shown by Polo et al. (2008) at monthly-interannual scale, suggesting an active role of the eastern Mediterranean SST into the Sahelian variability. In fact, the coupled OLR and temperature time series in Fig. 1b are largely correlated (r = +0.70, significant at p = 0.01 taking into account autocorrelation in the series); they display both an in-phase interannual signal and the recent rainfall recovering since the 90 s. When these coefficients are averaged over the successive JAS season its correlation increases to +0.87, while the temperature and OLR time series are positively correlated with JAS rainfall averaged over central (10°W–10°E) Sahel: +0.50 and +0.67, significant at p = 0.05 and p = 0.01 respectively, taking into account autocorrelation in the series.

By construction the second skin temperature-OLR mode (Fig. 1c, d) is less energetic and registers lower serial correlation between time series (r = +0.61, significant at p = 0.01 taking into account autocorrelation in the series). However it is the direct complement of the first one: the strongest (positive) temperature loadings are here restricted to the West Mediterranean. A surface warming in this region is linked to lower OLR values (deeper convection) extending both in the Western parts of the Sahel (Senegal, Mali) and over equatorial eastern Atlantic (Gulf of Guinea). When time coefficients are averaged over the JAS seasons the correlation increase to +0.70 (p = 0.01), and the temperature and OLR time series are positively correlated with JAS rainfall over the western parts (18–10°W) of the Sahel: +0.31 and +0.43 respectively, this last value being significant at p = 0.05.

In fact the Mediterranean/WAM relationship is not purely synchronous as illustrated in Fig. 1c. This diagram displays the lead/lag correlation coefficients computed between spatial indexes averaging OLR values in the Sahel zone (10–20°N; 15°W–30°E) in JAS, at the heart of the rainy season and 3-month surface temperatures in the whole Mediterranean basin (32–40°N; 6–36°W) moving with a 5-day (pentad) time-step before and after JAS. Near all the values are significant at p = 0.05 but the strongest correlation is registered when the Mediterranean temperatures are leading OLR by 6 pentads (1 month).

3.3 Evolution of the leading skin temperature/OLR mode at the monthly scale

In order to assess better the robustness of the Mediterranean/WAM lagged relationship, Fig. 2 refers to SVD analyses performed at the monthly scale between Mediterranean skin temperatures and WAM OLR values 1 month later referring to the same domain that in Fig. 1. Here linear trends in the series have been removed and the display focuses on the May/June to October/November evolution of the lagged temperature/OLR relationship. The results confirm first the statistical linkage between the East Mediterranean and WAM domains shown above (Fig. 1a). Second they show that there is no negative OLR anomaly over the Sahel before June/July (Fig. 2b), confirming the previous dry regime as indicated from Table 1 (in May/June; Fig. 2a). There, warm anomalies are associated with positive values of OLR over Sudan-Sahel, meaning decreased convection. From Jul/Aug to Oct/Nov temperature loadings reinforce on the eastern Mediterranean according to the leading monthly coupled mode (Fig. 1a), while OLR anomalies are related to increased deep motions (Fig. 2c–f).

Once again, the East Mediterranean concentrates the highest positive temperature loadings in connection with WAM deep convection, which points out largest anomalies along 10–15°N and mainly located over inland regions. Also at daily scales, Table 3 summarizes the percentages explained by each leading JAS-mode, time correlation between time series before and after removing all linear trends at each grid point (columns 1–2). In fact the field coupling is slightly lower when the series are detrended with a leading mode, spatially similar to Fig. 2a but less energetic (29 vs. 52% of squared covariance fraction) and positively correlated (+0.42 vs. +0.51). Other columns underline possible connections with oscillatory modes at synoptic scale (<10 days) such as African Easterly Waves (AEW) in the 3–5 day range, when longer-term mean coefficients are removed.

To test the linearity of the connection the next section contrasts the signals observed in warm and cold situations over the Western and Eastern Mediterranean.

4.1 OLR signal

The mean July–September OLR field displayed in Fig. 3a exhibits high OLR values in the Mediterranean region and low OLR values near the ITCZ mean location since in dry subtropics (no cloud), OLR depends mainly on surface temperatures and is generally greater than 260 W/m²: any surface warming (cooling) in the lowest levels of the troposphere over the Mediterranean tends to enhance (reduce) OLR values. By contrast, in the vicinity of the ITCZ which concentrates the lowest OLR values (<240 W/m² in Fig. 3a) OLR is directly linked to the altitude of clouds (convective towers).

Open image in new window

The warm (cold) composites relative to the Western and Eastern Mediterranean are shown in Fig. 3b, c, respectively (Fig. 3d, e). They contrast the 8 warmest and the 8 coldest seasons versus all the remainders (caption in Fig. 3 indicates the selected years). Once again the signal is more significant in Fig. 3c and e, i.e. when the possible thermal forcing involves the East Mediterranean: warm (cold) anomalies in the Eastern basin are associated with lower (higher) OLR values. This denotes an enhancement (a reduction) of deep convection in the monsoon region when the East basin is abnormally warm (cold) and suggests hence the existence of associated signals in atmospheric dynamics.

4.2 Atmospheric dynamics

Vertical cross sections of wind and specific humidity along the 20°–0°W and 0°–30°E meridional planes and relative to the West and East Mediterranean are presented in Fig. 4. Panels a, b show the mean location and intensity of cell circulations along with the regions where specific humidity exceeds 10 g/kg. The northern and southern overturning cells are easily recognizable between the humid Mediterranean and monsoon regions (shading). The southern cell, where the monsoon takes place, includes southerly/northerly horizontal branches in low/high levels and ascending/subsiding branches located respectively by 8–10°N and 20°S. Westward to the 0° meridian the strong ascents are associated with strong northerlies at 200 hPa southward to the equator (Fig. 4a) which tends to fuel monsoon overturning in the southern cell. By contrast eastward to the 0° meridian the ascents fuel preferentially the northerlies in low levels from the Eastern basin and hence the northern cell (Fig. 4b).

Open image in new window

Warm composites in Fig. 4c–f exhibit significant signals in both meridional circulation and specific humidity especially. Warm conditions in the Western basin are associated, westward to 0°, with lower-levels convergence and upper-levels ascending motions in the northern subtropical latitudes, and with significant moisture increase (shadings in Fig. 4c) in the low and mid troposphere above the equatorial Atlantic. However only 3% of the area can be considered as significant and there is no clear meridional-connection between the former and the latter. Neither a clear signal appears eastward to 0°. By contrast warm conditions in the Eastern basin reinforce the southern (monsoon) cell circulation between 0° and 30°E (Fig. 4f) with 5% of global field significance): low level advections from the South in temperature and moisture are significantly enhanced, while air ascents within the ITCZ and air descents above the Southern Tropics strengthen. This is accompanied by a significant moisture increase (shading in Fig. 4f) under 400 hPa over the continent (5–20°N). To a lesser extent, but of equal statistical significance, the northern Hadley cell is also reinforced showing a closed circulation.

The results reported in Sects. 3 and 4 are thus compatible with the existence of a significant statistical relationship between the East Mediterranean and the WAM-Sahel region. In order to check whether this is interpretable in terms of Mediterranean forcing, we will analyze multi-model outputs using SST sensitivity experiments performed on 4 AGCMs. This approach allows production of more reliable climate features than a single model method since climate simulations performed on any given model are sensitive to systematic errors, as previously shown by, among others, Bader and Latif (2003) and Giannini et al. (2003).

5.1 Description of the sensitivity experiments

In these experiments the SST anomaly patterns were calculated from observed ERSST dataset (Smith and Reynolds 2004) and over the period 1979–2005. The African monsoon multidisciplinary analyses (AMMA) community decided that the boundary conditions were based on the expansion coefficient of the leading Extended Maximum Covariance mode between WAM-precipitation (CMAP dataset) and Mediterranean-SST described in Polo et al. (2008). This mode shows strong positive links between the Sahelian rainfall and the spring to summer evolution of the SST anomalies in the eastern Mediterranean. In order to introduce some physical considerations in the SST patterns definitions, a SST composite was constructed by averaging years for which the phasing between SST and African precipitation was maximal. They have been computed chosen those years in which the SST and June–September precipitation expansion coefficients were higher than 1 standard deviation (1984, 1987, 1991, 1992, 1997) and those in which the expansion coefficients were less that −1 standard deviation (1979, 1994, 1999, 2001). In order to amplify the SST signals, it has chosen to compute the difference between the positive-composite and the negative-composite of SST anomalies for defining the positive phase of the thermal forcing; while the negative phase was computed by multiplying the positive phase by −1. The SST anomalous patterns were finally multiplied by two in order to amplify the atmospheric response (Fig. 5). SST conditions have then been prescribed with these anomaly patterns added to the climatology of observed global SST for the period 1979–2005.

Open image in new window

The ENEA, IPSL and UCM institutions performed 10 simulations with different initial conditions for each experiment (cold, warm, and control with no superimposed SST anomaly) while CNRM performed 20 simulations, allowing available a total of 150 simulations. However to give each model the same weight, we will consider 10 simulations for each model and each experiment (control, cold and warm), hence 10 × 4 × 3 = 120 simulations. Experiments have been run from 15th of April to 15th of October and each SST pattern have led to a set of 10 runs (20 runs for CNRM). Initial conditions have been taken from a long-term AMIP-type run with conditions representative of the end of the 20th century. A good way to do this was to take conditions around the 15th of April for 10 (or 20 for CNRM) different years in the set of possible initial conditions. The 10 (or 20) years for initial conditions have been taken from years 1979 to 1988 (or 1979–1998 for CNRM).

It is noteworthy that using such prescribed SST as a lower boundary condition implies that the sea acts as a reservoir of infinite heat capacity which is not the case in the real world where the Mediterranean has finite heat capacity. This tends to damp the surface heat fluxes and therefore to decrease the amplitude resulting from the model runs.

5.2 Multi-model rainfall outputs

In order to validate models outputs, Fig. 6 shows the May to September mean monthly rainfall fields averaged over the period 1979–2001 in the CPC merged precipitation and in the control experiments of each model. Nearly all models reproduce well seasonal evolution of the rainbelt both in terms of meridional displacements and of location of rainfall maxima (amounts > 6 mm/day are shaded). However systematic biases exist in each model. For example, the CNRM, IPSL and ENEA simulations tend to underestimate the rainfall amounts while UCLA overestimates. Notice also that the seasonal excursion of the rainbelt in latitude is better reproduced in CNRM and UCLA outputs (see also the Hovmoller diagrams for the control runs in Losada et al. 2009; this issue).

Open image in new window

The 40 warm (W) and 40 cold (C) experiments limited to the Mediterranean are now compared to the 40 control simulations (CTL = climatology) regarding rainfall outputs from each model. The focus is made on the two main West African rainy seasons, i.e., the first Guinean rainy season in May–June (MJ) and the northern tropical rainy season in July–September (JAS).

Figure 7 displays the W-CTL and C-CTL field differences with superimposed shadings where they are significant. In MJ there is no coherent signal: i.e., the IPSL model produces significant rainfall differences (Fig. 7i, j, with 11 and 5% significant grid points, respectively), but with the same sign in both experiments. In JAS the responses are larger and increase in magnitude and significance. Even so, cold experiments generate fairly unclear results and are not conclusive at all, except with UCLA in Fig. 7o (6%). In contrast, warm experiments produce significant Sahelian rainfall excesses in 3 out 4 models (CNRM, IPSL, UCLA with respectively 10, 18 and 9% significant points), which outputs are in good agreement with our previous empirical results. The precipitation response from ENEA does not yield any significant anomaly, neither from C-CTL (Fig. 7g) nor from W-CTL (Fig. 7h). As shown in Fig. 6k–o, ENEA climatology is unable to simulate the latitudinal migration of the ITCZ-WAM system, and also fails in producing the maximum monsoonal rainfall, i.e. located in MJ instead of JAS (also in Losada et al. 2009, this issue). These features could largely explain the diffuse ENEA response; and, since this model does not reproduce the WAM seasonal cycle we will exclude its outputs in the following sections.

Open image in new window

In Fig. 8 statistical distributions of the MJ and JAS rainfall outputs are displayed in standardized values both model by model (Fig. 8a, b), and for the mean-ensemble including all the models (Fig. 8c, d). In order to reduce the systematic errors due to each model, the biases have been considered separately for each model and anomalies with respect to the corresponding model climatology have been computed. The histograms refer to a classification in three categories illustrating the relative occurrences of below normal, quasi normal and above normal situations in warm and cold experiments (see the legend).

Open image in new window

In MJ (Fig. 8a, c) there is no significant rainfall difference: no model causes a significant dispersion of the rainfall amounts produced by the cold and warm experiments (circles and asters in 8a, respectively). This is similar when using a model-ensemble approach: the histograms relative to the cold and warm simulations (white and blue bars in Fig. 8c, respectively) are alike. In JAS, at the opposite, the same experiments create contrasted responses, i.e., less (more) rainfall in cold (warm) situations. However, there is a more consistent signal in warm outputs than in cold ones: (1) the latter fail 32.5% in distinguishing the precipitation sign, against only 12.5% in the former (Fig. 8b); (2) and, considering 4-models together, the cold (warm) experiments model-ensemble generate a ratio of 9/18 (27/2) between excess/deficits (Fig. 8d). Warm situations in the Mediterranean seem therefore to impact more on West African precipitation in contrast with empirical findings, which showed a symmetrical signal (Fig. 3c, e). For all above, the rest of the work will focus on both the rainy season (JAS) and the forced-response in warm conditions.

5.3 Large-scale wind circulation

Multi-model composites using the anomaly method are now presented for contrasting respectively the 30 cold and 30 warm experiments to the 30 control runs in terms of circulation along the vertical meridional plane (CNRM, UCM, IPSL; excluding ENEA). Figure 9 presents vertical cross sections of wind components (u,v, omega) as simulated by each model averaged along 20°W–30°E where arrows refer to v and omega and contours to the zonal component: left panels display mean control simulations for each model while diagrams at right refer to the significant warm minus control differences tested with a Student t test at p = 0.05. Notice first that several basic elements of the mean monsoon circulation are well simulated by the GCMs (Fig. 9a, c, e), i.e., the monsoon cell southward to 20°N fuelled by air ascents over the Sahel domain and SW winds, the northern overturning circulation, the northern subtropical Westerly Jet in high troposphere and also subsidence over the northern subtropics. However there are some discrepancies with the easterly circulation. For example the African Easterly Jet (AEJ) in mid troposphere is strongly underestimated by UCM and CNRM while IPSL underestimates tropical easterly jet (TEJ) maxima in high levels.

Open image in new window

Regarding to the W-CTL model responses, the more significant signals are observed in CNRM and UCM simulations (Fig. 9b, d); i.e., note the unclear circulation in IPSL run (Fig. 9f). As shown in observations (Sect. 4.2), it can be seen how CNRM experiment produces a significant reinforcement of both the northern and southern Hadley-cells; while UCM only yields the overturning anomaly of the southern meridional circulation. These features reflect the enhancement of the ITCZ activity, according to one of the Rowell’s (2003) feedback mechanism which involves reinforced monsoonal circulation (low-level westerlies and rising motion) responding to the enhanced moisture transport from the Mediterranean. An additional northward displacement of this deeper-ITCZ is also apparent: from 10 to 15°N in CTL to 15–20°N in W-CTL (Fig. 9). In addition to strengthened southwesterly flow associated with increased latent heat (diabatic heating) in the Sahel region, forced by the above anomalous moisture convergence (Rowell 2003); Peyrillé et al. (2007), by using an idealized 2-D model, have suggested that the poleward extent of continental convection is partially explained by the weakening of the climatological northeasterly flow from the Mediterranean (see below, Sect. 5.6). Here, in order to gain insight into the proposed mechanism from 3-D AGCM simulations, Fig. 10 presents jointly the upper (200 hPa, second row) and lower (850 hPa, fourth row) streamfunction anomalies. In 850 hPa response are apparent the alterations in the local-ITCZ (Fig. 10d, h, l): an anomalous cyclone in the WAM domain at CNRM and UCM forcing deep-upward motions; but an anticyclonic anomaly along Central African Republic and Guinean coastline in IPSL. It is also noticeable the low-level negative anomalies surrounding the eastern Mediterranean basin, associated with negative pressure anomalies there and in accordance with the observational work by Raicich et al. (2003). This feature, the low-level convergence anomaly, would go along with Peyrillé et al. (2007) regarding the decrease of NE winds in response to warming of Mediterranean SST (see Sect. 5.6); and partially in contrast to Rowell (2003) and Jung et al. (2006) who found no change in the circulation leading to enhanced monsoon precipitation.

Open image in new window

Additionally, this anomalous low-level convergent inflow is associated with positive streamfunction anomalies at upper-levels, revealing a baroclinic structure in the atmosphere (Fig. 10b, f, j): along the whole subtropical belt in CNRM, confined into the eastern Mediterranean in UCM, and isolated over the Arabian Peninsula in IPSL. Similar thermally-driven response to changes in the Mediterranean SST has been reported by Li (2006), who found a circumglobal wintertime-teleconnection initiated in the Mediterranean basin. Experiments analyzed here seem to corroborate this direct global-response to Mediterranean summer forcing (García-Serrano et al. manuscript in preparation); while previous evidences support this finding (Jung et al. 2006).

5.4 Moisture fluxes

Figure 11 a,b focuses better on the zonal and meridional mean moisture fluxes in low and mid troposheric layers using all the CNRM runs, i.e., 20 control and 20 warm simulations. Only the CNRM is chosen for taking advantage of the good vertical resolution of this model (45 vertical levels) particularly in low layers (Salas-Mélia et al. 2005). This approach is not extrapolable to other models but it provides valuable insights.

Open image in new window

The moisture transport is mainly organized with a SW maximum at 925 hPa by 10°N (positive values in panels a, b) and an easterly maximum at 700 hPa around 10°N, hence just under and southward to the main AEJ axis, as shown by the negative values in Fig. 11a. Warm-CTL composites in Fig. 11c, d present larger significant signals. In particular the zonal moisture transport by the AEJ is shifted northward (negative/positive dipolar patterns at 700 hPa in Fig. 11c) while in lower levels, negative/positive differences northward/southward to 20°N in Fig. 11c, d reveal a significant increase of northeasterly and southwesterly monsoonal flux. This finding implies an enhanced deep convection in the WAM area. Additional analyzes (not reported here) have shown that the increase in northeasterly moisture flux is mainly due to an augmentation of the specific humidity in low-level over the warmer eastern Mediterranean. This strengthens the flux and together with the reinforcement of the southwest monsoon augments moisture convergence in low levels, according to Rowell (2003). This feedback mechanism would also help to the anomalous northward displacement of the ITCZ, providing above-average precipitation over the Sahel.

5.5 Response in low levels

To assess better the capabilities of the 3 models to reproduce correctly the tropical-subtropical circulation over the Mediterranean-African window, Fig. 12a, c, e displays the mean low level wind fields as produced in the control simulations. The skill of simulations is rather good since several key-elements are reproduced and well located, i.e., Azores and Ste Helena anticyclonic gyres, monsoon and Harmattan systems, ITCZ…

Open image in new window

The Warm-CTL significant differences at p = 0.10 are presented in Fig. 12b, d, f with shadings when the v component is positive. The 3 models give anomalous surface westerlies over the Sahelian area, although the location of the signals is different from one model to another. Thus, CNRM concentrates the strongest signals over the Atlantic (under 12°N) and the eastern Sahel, UCM over a large part of the West African continent north to the equator and subtropical Atlantic (12–24°N), while IPSL produces significant westerlies in the Sahelian domain but with a similar amplitude south of the equator. Despite these differences, common features emerge from the multi-model approach (Fig. 12g, h). This multi-model ensemble retains large anomalous westerlies coming inland from the Atlantic (Rowell 2003) and reaching the Sahel region, but does not take into account the southerly component from Gulf of Guinea. Additionally, as also shown in each model (Fig. 12b, d, f), the weakening of the climatological northeasterly flow from the eastern Mediterranean is apparent. These anomalous surface winds have a clear southerly component east of 10°E over Egypt. This feature is consistent with the anomalous low-level convergence in the southeastern Mediterranean basin, as discussed above (Fig. 10d, h, l), and is in agreement with Peyrillé et al. (2007) regarding the penetration decrease of the mean circulation depending on the thermal contrast between the African continent and the Mediterranean Sea (or SST-forced in this case).

5.6 Response in mid and high levels

Figures 13 and 14 present the mean and composite wind fields at 600 and 200 hPa, respectively, with shadings where the zonal component is positive (blows eastward). In the top-middle panels are ensemble responses for each model, while in bottom panels are presented the multi-model approach (Figs. 13, 14g, h). At 600–200 hPa the control experiment (Figs. 13, 14a, c, e) reproduces correctly several specific features of the upper and mid level circulation, i.e., the AEJ and TEJ main axes over West Africa, the subtropical westerly jet pattern over the Mediterranean, and at 600 hPa northerlies from the Mediterranean. Warm experiments (Figs. 13, 14b, d, f) cause (1) a significant reduction of subtropical westerlies in high levels south of the equator; (2) a significant increase in TEJ, more clearly seen in each model behaviour than multi-model ensemble, confined to 0 and 15°E; and (3) a relative decrease in AEJ, with clear distinct location in each model which causes a diffuse signal in the multi-model ensemble. All of these features are known to be favourable to the monsoon circulation (Newell and Kidson 1984; Fontaine et al., 1995).

Open image in new window

Open image in new window